OpenCloudOS-Kernel/fs/nfs/client.c

1643 lines
41 KiB
C
Raw Normal View History

/* client.c: NFS client sharing and management code
*
* Copyright (C) 2006 Red Hat, Inc. All Rights Reserved.
* Written by David Howells (dhowells@redhat.com)
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License
* as published by the Free Software Foundation; either version
* 2 of the License, or (at your option) any later version.
*/
#include <linux/module.h>
#include <linux/init.h>
#include <linux/sched.h>
#include <linux/time.h>
#include <linux/kernel.h>
#include <linux/mm.h>
#include <linux/string.h>
#include <linux/stat.h>
#include <linux/errno.h>
#include <linux/unistd.h>
#include <linux/sunrpc/clnt.h>
#include <linux/sunrpc/stats.h>
#include <linux/sunrpc/metrics.h>
#include <linux/sunrpc/xprtsock.h>
#include <linux/sunrpc/xprtrdma.h>
#include <linux/nfs_fs.h>
#include <linux/nfs_mount.h>
#include <linux/nfs4_mount.h>
#include <linux/lockd/bind.h>
#include <linux/seq_file.h>
#include <linux/mount.h>
#include <linux/nfs_idmap.h>
#include <linux/vfs.h>
#include <linux/inet.h>
#include <linux/in6.h>
#include <net/ipv6.h>
#include <linux/nfs_xdr.h>
#include <asm/system.h>
#include "nfs4_fs.h"
#include "callback.h"
#include "delegation.h"
#include "iostat.h"
#include "internal.h"
#define NFSDBG_FACILITY NFSDBG_CLIENT
static DEFINE_SPINLOCK(nfs_client_lock);
static LIST_HEAD(nfs_client_list);
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
static LIST_HEAD(nfs_volume_list);
static DECLARE_WAIT_QUEUE_HEAD(nfs_client_active_wq);
/*
* RPC cruft for NFS
*/
static struct rpc_version *nfs_version[5] = {
[2] = &nfs_version2,
#ifdef CONFIG_NFS_V3
[3] = &nfs_version3,
#endif
#ifdef CONFIG_NFS_V4
[4] = &nfs_version4,
#endif
};
struct rpc_program nfs_program = {
.name = "nfs",
.number = NFS_PROGRAM,
.nrvers = ARRAY_SIZE(nfs_version),
.version = nfs_version,
.stats = &nfs_rpcstat,
.pipe_dir_name = "/nfs",
};
struct rpc_stat nfs_rpcstat = {
.program = &nfs_program
};
#ifdef CONFIG_NFS_V3_ACL
static struct rpc_stat nfsacl_rpcstat = { &nfsacl_program };
static struct rpc_version * nfsacl_version[] = {
[3] = &nfsacl_version3,
};
struct rpc_program nfsacl_program = {
.name = "nfsacl",
.number = NFS_ACL_PROGRAM,
.nrvers = ARRAY_SIZE(nfsacl_version),
.version = nfsacl_version,
.stats = &nfsacl_rpcstat,
};
#endif /* CONFIG_NFS_V3_ACL */
struct nfs_client_initdata {
const char *hostname;
const struct sockaddr *addr;
size_t addrlen;
const struct nfs_rpc_ops *rpc_ops;
int proto;
};
/*
* Allocate a shared client record
*
* Since these are allocated/deallocated very rarely, we don't
* bother putting them in a slab cache...
*/
static struct nfs_client *nfs_alloc_client(const struct nfs_client_initdata *cl_init)
{
struct nfs_client *clp;
struct rpc_cred *cred;
if ((clp = kzalloc(sizeof(*clp), GFP_KERNEL)) == NULL)
goto error_0;
clp->rpc_ops = cl_init->rpc_ops;
if (cl_init->rpc_ops->version == 4) {
if (nfs_callback_up() < 0)
goto error_2;
__set_bit(NFS_CS_CALLBACK, &clp->cl_res_state);
}
atomic_set(&clp->cl_count, 1);
clp->cl_cons_state = NFS_CS_INITING;
memcpy(&clp->cl_addr, cl_init->addr, cl_init->addrlen);
clp->cl_addrlen = cl_init->addrlen;
if (cl_init->hostname) {
clp->cl_hostname = kstrdup(cl_init->hostname, GFP_KERNEL);
if (!clp->cl_hostname)
goto error_3;
}
INIT_LIST_HEAD(&clp->cl_superblocks);
clp->cl_rpcclient = ERR_PTR(-EINVAL);
clp->cl_proto = cl_init->proto;
#ifdef CONFIG_NFS_V4
INIT_LIST_HEAD(&clp->cl_delegations);
spin_lock_init(&clp->cl_lock);
2006-11-22 22:55:48 +08:00
INIT_DELAYED_WORK(&clp->cl_renewd, nfs4_renew_state);
rpc_init_wait_queue(&clp->cl_rpcwaitq, "NFS client");
clp->cl_boot_time = CURRENT_TIME;
clp->cl_state = 1 << NFS4CLNT_LEASE_EXPIRED;
#endif
cred = rpc_lookup_machine_cred();
if (!IS_ERR(cred))
clp->cl_machine_cred = cred;
return clp;
error_3:
if (__test_and_clear_bit(NFS_CS_CALLBACK, &clp->cl_res_state))
nfs_callback_down();
error_2:
kfree(clp);
error_0:
return NULL;
}
static void nfs4_shutdown_client(struct nfs_client *clp)
{
#ifdef CONFIG_NFS_V4
if (__test_and_clear_bit(NFS_CS_RENEWD, &clp->cl_res_state))
nfs4_kill_renewd(clp);
BUG_ON(!RB_EMPTY_ROOT(&clp->cl_state_owners));
if (__test_and_clear_bit(NFS_CS_IDMAP, &clp->cl_res_state))
nfs_idmap_delete(clp);
rpc_destroy_wait_queue(&clp->cl_rpcwaitq);
#endif
}
/*
* Destroy a shared client record
*/
static void nfs_free_client(struct nfs_client *clp)
{
dprintk("--> nfs_free_client(%u)\n", clp->rpc_ops->version);
nfs4_shutdown_client(clp);
/* -EIO all pending I/O */
if (!IS_ERR(clp->cl_rpcclient))
rpc_shutdown_client(clp->cl_rpcclient);
if (__test_and_clear_bit(NFS_CS_CALLBACK, &clp->cl_res_state))
nfs_callback_down();
if (clp->cl_machine_cred != NULL)
put_rpccred(clp->cl_machine_cred);
kfree(clp->cl_hostname);
kfree(clp);
dprintk("<-- nfs_free_client()\n");
}
/*
* Release a reference to a shared client record
*/
void nfs_put_client(struct nfs_client *clp)
{
if (!clp)
return;
dprintk("--> nfs_put_client({%d})\n", atomic_read(&clp->cl_count));
if (atomic_dec_and_lock(&clp->cl_count, &nfs_client_lock)) {
list_del(&clp->cl_share_link);
spin_unlock(&nfs_client_lock);
BUG_ON(!list_empty(&clp->cl_superblocks));
nfs_free_client(clp);
}
}
#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
static const struct in6_addr *nfs_map_ipv4_addr(const struct sockaddr *sa, struct in6_addr *addr_mapped)
{
switch (sa->sa_family) {
default:
return NULL;
case AF_INET6:
return &((const struct sockaddr_in6 *)sa)->sin6_addr;
break;
case AF_INET:
ipv6_addr_set_v4mapped(((const struct sockaddr_in *)sa)->sin_addr.s_addr,
addr_mapped);
return addr_mapped;
}
}
static int nfs_sockaddr_match_ipaddr(const struct sockaddr *sa1,
const struct sockaddr *sa2)
{
const struct in6_addr *addr1;
const struct in6_addr *addr2;
struct in6_addr addr1_mapped;
struct in6_addr addr2_mapped;
addr1 = nfs_map_ipv4_addr(sa1, &addr1_mapped);
if (likely(addr1 != NULL)) {
addr2 = nfs_map_ipv4_addr(sa2, &addr2_mapped);
if (likely(addr2 != NULL))
return ipv6_addr_equal(addr1, addr2);
}
return 0;
}
/*
* Test if two ip6 socket addresses refer to the same socket by
* comparing relevant fields. The padding bytes specifically, are not
* compared. sin6_flowinfo is not compared because it only affects QoS
* and sin6_scope_id is only compared if the address is "link local"
* because "link local" addresses need only be unique to a specific
* link. Conversely, ordinary unicast addresses might have different
* sin6_scope_id.
*
* The caller should ensure both socket addresses are AF_INET6.
*/
static int nfs_sockaddr_cmp_ip6(const struct sockaddr *sa1,
const struct sockaddr *sa2)
{
const struct sockaddr_in6 *saddr1 = (const struct sockaddr_in6 *)sa1;
const struct sockaddr_in6 *saddr2 = (const struct sockaddr_in6 *)sa2;
if (!ipv6_addr_equal(&saddr1->sin6_addr,
&saddr1->sin6_addr))
return 0;
if (ipv6_addr_scope(&saddr1->sin6_addr) == IPV6_ADDR_SCOPE_LINKLOCAL &&
saddr1->sin6_scope_id != saddr2->sin6_scope_id)
return 0;
return saddr1->sin6_port == saddr2->sin6_port;
}
#else
static int nfs_sockaddr_match_ipaddr4(const struct sockaddr_in *sa1,
const struct sockaddr_in *sa2)
{
return sa1->sin_addr.s_addr == sa2->sin_addr.s_addr;
}
static int nfs_sockaddr_match_ipaddr(const struct sockaddr *sa1,
const struct sockaddr *sa2)
{
if (unlikely(sa1->sa_family != AF_INET || sa2->sa_family != AF_INET))
return 0;
return nfs_sockaddr_match_ipaddr4((const struct sockaddr_in *)sa1,
(const struct sockaddr_in *)sa2);
}
static int nfs_sockaddr_cmp_ip6(const struct sockaddr * sa1,
const struct sockaddr * sa2)
{
return 0;
}
#endif
/*
* Test if two ip4 socket addresses refer to the same socket, by
* comparing relevant fields. The padding bytes specifically, are
* not compared.
*
* The caller should ensure both socket addresses are AF_INET.
*/
static int nfs_sockaddr_cmp_ip4(const struct sockaddr *sa1,
const struct sockaddr *sa2)
{
const struct sockaddr_in *saddr1 = (const struct sockaddr_in *)sa1;
const struct sockaddr_in *saddr2 = (const struct sockaddr_in *)sa2;
if (saddr1->sin_addr.s_addr != saddr2->sin_addr.s_addr)
return 0;
return saddr1->sin_port == saddr2->sin_port;
}
/*
* Test if two socket addresses represent the same actual socket,
* by comparing (only) relevant fields.
*/
static int nfs_sockaddr_cmp(const struct sockaddr *sa1,
const struct sockaddr *sa2)
{
if (sa1->sa_family != sa2->sa_family)
return 0;
switch (sa1->sa_family) {
case AF_INET:
return nfs_sockaddr_cmp_ip4(sa1, sa2);
case AF_INET6:
return nfs_sockaddr_cmp_ip6(sa1, sa2);
}
return 0;
}
/*
* Find a client by IP address and protocol version
* - returns NULL if no such client
*/
struct nfs_client *nfs_find_client(const struct sockaddr *addr, u32 nfsversion)
{
struct nfs_client *clp;
spin_lock(&nfs_client_lock);
list_for_each_entry(clp, &nfs_client_list, cl_share_link) {
struct sockaddr *clap = (struct sockaddr *)&clp->cl_addr;
/* Don't match clients that failed to initialise properly */
if (clp->cl_cons_state != NFS_CS_READY)
continue;
/* Different NFS versions cannot share the same nfs_client */
if (clp->rpc_ops->version != nfsversion)
continue;
/* Match only the IP address, not the port number */
if (!nfs_sockaddr_match_ipaddr(addr, clap))
continue;
atomic_inc(&clp->cl_count);
spin_unlock(&nfs_client_lock);
return clp;
}
spin_unlock(&nfs_client_lock);
return NULL;
}
/*
* Find a client by IP address and protocol version
* - returns NULL if no such client
*/
struct nfs_client *nfs_find_client_next(struct nfs_client *clp)
{
struct sockaddr *sap = (struct sockaddr *)&clp->cl_addr;
u32 nfsvers = clp->rpc_ops->version;
spin_lock(&nfs_client_lock);
list_for_each_entry_continue(clp, &nfs_client_list, cl_share_link) {
struct sockaddr *clap = (struct sockaddr *)&clp->cl_addr;
/* Don't match clients that failed to initialise properly */
if (clp->cl_cons_state != NFS_CS_READY)
continue;
/* Different NFS versions cannot share the same nfs_client */
if (clp->rpc_ops->version != nfsvers)
continue;
/* Match only the IP address, not the port number */
if (!nfs_sockaddr_match_ipaddr(sap, clap))
continue;
atomic_inc(&clp->cl_count);
spin_unlock(&nfs_client_lock);
return clp;
}
spin_unlock(&nfs_client_lock);
return NULL;
}
/*
* Find an nfs_client on the list that matches the initialisation data
* that is supplied.
*/
static struct nfs_client *nfs_match_client(const struct nfs_client_initdata *data)
{
struct nfs_client *clp;
const struct sockaddr *sap = data->addr;
list_for_each_entry(clp, &nfs_client_list, cl_share_link) {
const struct sockaddr *clap = (struct sockaddr *)&clp->cl_addr;
/* Don't match clients that failed to initialise properly */
if (clp->cl_cons_state < 0)
continue;
/* Different NFS versions cannot share the same nfs_client */
if (clp->rpc_ops != data->rpc_ops)
continue;
if (clp->cl_proto != data->proto)
continue;
/* Match the full socket address */
if (!nfs_sockaddr_cmp(sap, clap))
continue;
atomic_inc(&clp->cl_count);
return clp;
}
return NULL;
}
/*
* Look up a client by IP address and protocol version
* - creates a new record if one doesn't yet exist
*/
static struct nfs_client *nfs_get_client(const struct nfs_client_initdata *cl_init)
{
struct nfs_client *clp, *new = NULL;
int error;
dprintk("--> nfs_get_client(%s,v%u)\n",
cl_init->hostname ?: "", cl_init->rpc_ops->version);
/* see if the client already exists */
do {
spin_lock(&nfs_client_lock);
clp = nfs_match_client(cl_init);
if (clp)
goto found_client;
if (new)
goto install_client;
spin_unlock(&nfs_client_lock);
new = nfs_alloc_client(cl_init);
} while (new);
return ERR_PTR(-ENOMEM);
/* install a new client and return with it unready */
install_client:
clp = new;
list_add(&clp->cl_share_link, &nfs_client_list);
spin_unlock(&nfs_client_lock);
dprintk("--> nfs_get_client() = %p [new]\n", clp);
return clp;
/* found an existing client
* - make sure it's ready before returning
*/
found_client:
spin_unlock(&nfs_client_lock);
if (new)
nfs_free_client(new);
error = wait_event_killable(nfs_client_active_wq,
clp->cl_cons_state != NFS_CS_INITING);
if (error < 0) {
nfs_put_client(clp);
return ERR_PTR(-ERESTARTSYS);
}
if (clp->cl_cons_state < NFS_CS_READY) {
error = clp->cl_cons_state;
nfs_put_client(clp);
return ERR_PTR(error);
}
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
BUG_ON(clp->cl_cons_state != NFS_CS_READY);
dprintk("--> nfs_get_client() = %p [share]\n", clp);
return clp;
}
/*
* Mark a server as ready or failed
*/
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
static void nfs_mark_client_ready(struct nfs_client *clp, int state)
{
clp->cl_cons_state = state;
wake_up_all(&nfs_client_active_wq);
}
/*
* Initialise the timeout values for a connection
*/
static void nfs_init_timeout_values(struct rpc_timeout *to, int proto,
unsigned int timeo, unsigned int retrans)
{
to->to_initval = timeo * HZ / 10;
to->to_retries = retrans;
switch (proto) {
case XPRT_TRANSPORT_TCP:
case XPRT_TRANSPORT_RDMA:
if (to->to_retries == 0)
to->to_retries = NFS_DEF_TCP_RETRANS;
if (to->to_initval == 0)
to->to_initval = NFS_DEF_TCP_TIMEO * HZ / 10;
if (to->to_initval > NFS_MAX_TCP_TIMEOUT)
to->to_initval = NFS_MAX_TCP_TIMEOUT;
to->to_increment = to->to_initval;
to->to_maxval = to->to_initval + (to->to_increment * to->to_retries);
if (to->to_maxval > NFS_MAX_TCP_TIMEOUT)
to->to_maxval = NFS_MAX_TCP_TIMEOUT;
if (to->to_maxval < to->to_initval)
to->to_maxval = to->to_initval;
to->to_exponential = 0;
break;
case XPRT_TRANSPORT_UDP:
if (to->to_retries == 0)
to->to_retries = NFS_DEF_UDP_RETRANS;
if (!to->to_initval)
to->to_initval = NFS_DEF_UDP_TIMEO * HZ / 10;
if (to->to_initval > NFS_MAX_UDP_TIMEOUT)
to->to_initval = NFS_MAX_UDP_TIMEOUT;
to->to_maxval = NFS_MAX_UDP_TIMEOUT;
to->to_exponential = 1;
break;
default:
BUG();
}
}
/*
* Create an RPC client handle
*/
static int nfs_create_rpc_client(struct nfs_client *clp,
const struct rpc_timeout *timeparms,
rpc_authflavor_t flavor,
int discrtry, int noresvport)
{
struct rpc_clnt *clnt = NULL;
struct rpc_create_args args = {
.protocol = clp->cl_proto,
.address = (struct sockaddr *)&clp->cl_addr,
.addrsize = clp->cl_addrlen,
.timeout = timeparms,
.servername = clp->cl_hostname,
.program = &nfs_program,
.version = clp->rpc_ops->version,
.authflavor = flavor,
};
if (discrtry)
args.flags |= RPC_CLNT_CREATE_DISCRTRY;
if (noresvport)
args.flags |= RPC_CLNT_CREATE_NONPRIVPORT;
if (!IS_ERR(clp->cl_rpcclient))
return 0;
clnt = rpc_create(&args);
if (IS_ERR(clnt)) {
dprintk("%s: cannot create RPC client. Error = %ld\n",
__func__, PTR_ERR(clnt));
return PTR_ERR(clnt);
}
clp->cl_rpcclient = clnt;
return 0;
}
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
/*
* Version 2 or 3 client destruction
*/
static void nfs_destroy_server(struct nfs_server *server)
{
if (!(server->flags & NFS_MOUNT_NONLM))
nlmclnt_done(server->nlm_host);
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
}
/*
* Version 2 or 3 lockd setup
*/
static int nfs_start_lockd(struct nfs_server *server)
{
struct nlm_host *host;
struct nfs_client *clp = server->nfs_client;
struct nlmclnt_initdata nlm_init = {
.hostname = clp->cl_hostname,
.address = (struct sockaddr *)&clp->cl_addr,
.addrlen = clp->cl_addrlen,
.protocol = server->flags & NFS_MOUNT_TCP ?
IPPROTO_TCP : IPPROTO_UDP,
.nfs_version = clp->rpc_ops->version,
.noresvport = server->flags & NFS_MOUNT_NORESVPORT ?
1 : 0,
};
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
if (nlm_init.nfs_version > 3)
return 0;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
if (server->flags & NFS_MOUNT_NONLM)
return 0;
host = nlmclnt_init(&nlm_init);
if (IS_ERR(host))
return PTR_ERR(host);
server->nlm_host = host;
server->destroy = nfs_destroy_server;
return 0;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
}
/*
* Initialise an NFSv3 ACL client connection
*/
#ifdef CONFIG_NFS_V3_ACL
static void nfs_init_server_aclclient(struct nfs_server *server)
{
if (server->nfs_client->rpc_ops->version != 3)
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
goto out_noacl;
if (server->flags & NFS_MOUNT_NOACL)
goto out_noacl;
server->client_acl = rpc_bind_new_program(server->client, &nfsacl_program, 3);
if (IS_ERR(server->client_acl))
goto out_noacl;
/* No errors! Assume that Sun nfsacls are supported */
server->caps |= NFS_CAP_ACLS;
return;
out_noacl:
server->caps &= ~NFS_CAP_ACLS;
}
#else
static inline void nfs_init_server_aclclient(struct nfs_server *server)
{
server->flags &= ~NFS_MOUNT_NOACL;
server->caps &= ~NFS_CAP_ACLS;
}
#endif
/*
* Create a general RPC client
*/
static int nfs_init_server_rpcclient(struct nfs_server *server,
const struct rpc_timeout *timeo,
rpc_authflavor_t pseudoflavour)
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
{
struct nfs_client *clp = server->nfs_client;
server->client = rpc_clone_client(clp->cl_rpcclient);
if (IS_ERR(server->client)) {
dprintk("%s: couldn't create rpc_client!\n", __func__);
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
return PTR_ERR(server->client);
}
memcpy(&server->client->cl_timeout_default,
timeo,
sizeof(server->client->cl_timeout_default));
server->client->cl_timeout = &server->client->cl_timeout_default;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
if (pseudoflavour != clp->cl_rpcclient->cl_auth->au_flavor) {
struct rpc_auth *auth;
auth = rpcauth_create(pseudoflavour, server->client);
if (IS_ERR(auth)) {
dprintk("%s: couldn't create credcache!\n", __func__);
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
return PTR_ERR(auth);
}
}
server->client->cl_softrtry = 0;
if (server->flags & NFS_MOUNT_SOFT)
server->client->cl_softrtry = 1;
return 0;
}
/*
* Initialise an NFS2 or NFS3 client
*/
static int nfs_init_client(struct nfs_client *clp,
const struct rpc_timeout *timeparms,
const struct nfs_parsed_mount_data *data)
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
{
int error;
if (clp->cl_cons_state == NFS_CS_READY) {
/* the client is already initialised */
dprintk("<-- nfs_init_client() = 0 [already %p]\n", clp);
return 0;
}
/*
* Create a client RPC handle for doing FSSTAT with UNIX auth only
* - RFC 2623, sec 2.3.2
*/
NFS: add "[no]resvport" mount option The standard default security setting for NFS is AUTH_SYS. An NFS client connects to NFS servers via a privileged source port and a fixed standard destination port (2049). The client sends raw uid and gid numbers to identify users making NFS requests, and the server assumes an appropriate authority on the client has vetted these values because the source port is privileged. On Linux, by default in-kernel RPC services use a privileged port in the range between 650 and 1023 to avoid using source ports of well- known IP services. Using such a small range limits the number of NFS mount points and the number of unique NFS servers to which a client can connect concurrently. An NFS client can use unprivileged source ports to expand the range of source port numbers, allowing more concurrent server connections and more NFS mount points. Servers must explicitly allow NFS connections from unprivileged ports for this to work. In the past, bumping the value of the sunrpc.max_resvport sysctl on the client would permit the NFS client to use unprivileged ports. Bumping this setting also changes the maximum port number used by other in-kernel RPC services, some of which still required a port number less than 1023. This is exacerbated by the way source port numbers are chosen by the Linux RPC client, which starts at the top of the range and works downwards. It means that bumping the maximum means all RPC services requesting a source port will likely get an unprivileged port instead of a privileged one. Changing this setting effects all NFS mount points on a client. A sysadmin could not selectively choose which mount points would use non-privileged ports and which could not. Lastly, this mechanism of expanding the limit on the number of NFS mount points was entirely undocumented. To address the need for the NFS client to use a large range of source ports without interfering with the activity of other in-kernel RPC services, we introduce a new NFS mount option. This option explicitly tells only the NFS client to use a non-privileged source port when communicating with the NFS server for one specific mount point. This new mount option is called "resvport," like the similar NFS mount option on FreeBSD and Mac OS X. A sister patch for nfs-utils will be submitted that documents this new option in nfs(5). The default setting for this new mount option requires the NFS client to use a privileged port, as before. Explicitly specifying the "noresvport" mount option allows the NFS client to use an unprivileged source port for this mount point when connecting to the NFS server port. This mount option is supported only for text-based NFS mounts. [ Sidebar: it is widely known that security mechanisms based on the use of privileged source ports are ineffective. However, the NFS client can combine the use of unprivileged ports with the use of secure authentication mechanisms, such as Kerberos. This allows a large number of connections and mount points while ensuring a useful level of security. Eventually we may change the default setting for this option depending on the security flavor used for the mount. For example, if the mount is using only AUTH_SYS, then the default setting will be "resvport;" if the mount is using a strong security flavor such as krb5, the default setting will be "noresvport." ] Signed-off-by: Chuck Lever <chuck.lever@oracle.com> [Trond.Myklebust@netapp.com: Fixed a bug whereby nfs4_init_client() was being called with incorrect arguments.] Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2008-12-24 04:21:37 +08:00
error = nfs_create_rpc_client(clp, timeparms, RPC_AUTH_UNIX,
0, data->flags & NFS_MOUNT_NORESVPORT);
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
if (error < 0)
goto error;
nfs_mark_client_ready(clp, NFS_CS_READY);
return 0;
error:
nfs_mark_client_ready(clp, error);
dprintk("<-- nfs_init_client() = xerror %d\n", error);
return error;
}
/*
* Create a version 2 or 3 client
*/
static int nfs_init_server(struct nfs_server *server,
const struct nfs_parsed_mount_data *data)
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
{
struct nfs_client_initdata cl_init = {
.hostname = data->nfs_server.hostname,
.addr = (const struct sockaddr *)&data->nfs_server.address,
.addrlen = data->nfs_server.addrlen,
.rpc_ops = &nfs_v2_clientops,
.proto = data->nfs_server.protocol,
};
struct rpc_timeout timeparms;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
struct nfs_client *clp;
int error;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
dprintk("--> nfs_init_server()\n");
#ifdef CONFIG_NFS_V3
if (data->flags & NFS_MOUNT_VER3)
cl_init.rpc_ops = &nfs_v3_clientops;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
#endif
/* Allocate or find a client reference we can use */
clp = nfs_get_client(&cl_init);
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
if (IS_ERR(clp)) {
dprintk("<-- nfs_init_server() = error %ld\n", PTR_ERR(clp));
return PTR_ERR(clp);
}
nfs_init_timeout_values(&timeparms, data->nfs_server.protocol,
data->timeo, data->retrans);
error = nfs_init_client(clp, &timeparms, data);
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
if (error < 0)
goto error;
server->nfs_client = clp;
/* Initialise the client representation from the mount data */
server->flags = data->flags;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
if (data->rsize)
server->rsize = nfs_block_size(data->rsize, NULL);
if (data->wsize)
server->wsize = nfs_block_size(data->wsize, NULL);
server->acregmin = data->acregmin * HZ;
server->acregmax = data->acregmax * HZ;
server->acdirmin = data->acdirmin * HZ;
server->acdirmax = data->acdirmax * HZ;
/* Start lockd here, before we might error out */
error = nfs_start_lockd(server);
if (error < 0)
goto error;
server->port = data->nfs_server.port;
error = nfs_init_server_rpcclient(server, &timeparms, data->auth_flavors[0]);
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
if (error < 0)
goto error;
/* Preserve the values of mount_server-related mount options */
if (data->mount_server.addrlen) {
memcpy(&server->mountd_address, &data->mount_server.address,
data->mount_server.addrlen);
server->mountd_addrlen = data->mount_server.addrlen;
}
server->mountd_version = data->mount_server.version;
server->mountd_port = data->mount_server.port;
server->mountd_protocol = data->mount_server.protocol;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
server->namelen = data->namlen;
/* Create a client RPC handle for the NFSv3 ACL management interface */
nfs_init_server_aclclient(server);
dprintk("<-- nfs_init_server() = 0 [new %p]\n", clp);
return 0;
error:
server->nfs_client = NULL;
nfs_put_client(clp);
dprintk("<-- nfs_init_server() = xerror %d\n", error);
return error;
}
/*
* Load up the server record from information gained in an fsinfo record
*/
static void nfs_server_set_fsinfo(struct nfs_server *server, struct nfs_fsinfo *fsinfo)
{
unsigned long max_rpc_payload;
/* Work out a lot of parameters */
if (server->rsize == 0)
server->rsize = nfs_block_size(fsinfo->rtpref, NULL);
if (server->wsize == 0)
server->wsize = nfs_block_size(fsinfo->wtpref, NULL);
if (fsinfo->rtmax >= 512 && server->rsize > fsinfo->rtmax)
server->rsize = nfs_block_size(fsinfo->rtmax, NULL);
if (fsinfo->wtmax >= 512 && server->wsize > fsinfo->wtmax)
server->wsize = nfs_block_size(fsinfo->wtmax, NULL);
max_rpc_payload = nfs_block_size(rpc_max_payload(server->client), NULL);
if (server->rsize > max_rpc_payload)
server->rsize = max_rpc_payload;
if (server->rsize > NFS_MAX_FILE_IO_SIZE)
server->rsize = NFS_MAX_FILE_IO_SIZE;
server->rpages = (server->rsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
server->backing_dev_info.ra_pages = server->rpages * NFS_MAX_READAHEAD;
if (server->wsize > max_rpc_payload)
server->wsize = max_rpc_payload;
if (server->wsize > NFS_MAX_FILE_IO_SIZE)
server->wsize = NFS_MAX_FILE_IO_SIZE;
server->wpages = (server->wsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
server->wtmult = nfs_block_bits(fsinfo->wtmult, NULL);
server->dtsize = nfs_block_size(fsinfo->dtpref, NULL);
if (server->dtsize > PAGE_CACHE_SIZE)
server->dtsize = PAGE_CACHE_SIZE;
if (server->dtsize > server->rsize)
server->dtsize = server->rsize;
if (server->flags & NFS_MOUNT_NOAC) {
server->acregmin = server->acregmax = 0;
server->acdirmin = server->acdirmax = 0;
}
server->maxfilesize = fsinfo->maxfilesize;
/* We're airborne Set socket buffersize */
rpc_setbufsize(server->client, server->wsize + 100, server->rsize + 100);
}
/*
* Probe filesystem information, including the FSID on v2/v3
*/
static int nfs_probe_fsinfo(struct nfs_server *server, struct nfs_fh *mntfh, struct nfs_fattr *fattr)
{
struct nfs_fsinfo fsinfo;
struct nfs_client *clp = server->nfs_client;
int error;
dprintk("--> nfs_probe_fsinfo()\n");
if (clp->rpc_ops->set_capabilities != NULL) {
error = clp->rpc_ops->set_capabilities(server, mntfh);
if (error < 0)
goto out_error;
}
fsinfo.fattr = fattr;
nfs_fattr_init(fattr);
error = clp->rpc_ops->fsinfo(server, mntfh, &fsinfo);
if (error < 0)
goto out_error;
nfs_server_set_fsinfo(server, &fsinfo);
error = bdi_init(&server->backing_dev_info);
if (error)
goto out_error;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
/* Get some general file system info */
if (server->namelen == 0) {
struct nfs_pathconf pathinfo;
pathinfo.fattr = fattr;
nfs_fattr_init(fattr);
if (clp->rpc_ops->pathconf(server, mntfh, &pathinfo) >= 0)
server->namelen = pathinfo.max_namelen;
}
dprintk("<-- nfs_probe_fsinfo() = 0\n");
return 0;
out_error:
dprintk("nfs_probe_fsinfo: error = %d\n", -error);
return error;
}
/*
* Copy useful information when duplicating a server record
*/
static void nfs_server_copy_userdata(struct nfs_server *target, struct nfs_server *source)
{
target->flags = source->flags;
target->acregmin = source->acregmin;
target->acregmax = source->acregmax;
target->acdirmin = source->acdirmin;
target->acdirmax = source->acdirmax;
target->caps = source->caps;
}
/*
* Allocate and initialise a server record
*/
static struct nfs_server *nfs_alloc_server(void)
{
struct nfs_server *server;
server = kzalloc(sizeof(struct nfs_server), GFP_KERNEL);
if (!server)
return NULL;
server->client = server->client_acl = ERR_PTR(-EINVAL);
/* Zero out the NFS state stuff */
INIT_LIST_HEAD(&server->client_link);
INIT_LIST_HEAD(&server->master_link);
atomic_set(&server->active, 0);
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
server->io_stats = nfs_alloc_iostats();
if (!server->io_stats) {
kfree(server);
return NULL;
}
return server;
}
/*
* Free up a server record
*/
void nfs_free_server(struct nfs_server *server)
{
dprintk("--> nfs_free_server()\n");
spin_lock(&nfs_client_lock);
list_del(&server->client_link);
list_del(&server->master_link);
spin_unlock(&nfs_client_lock);
if (server->destroy != NULL)
server->destroy(server);
if (!IS_ERR(server->client_acl))
rpc_shutdown_client(server->client_acl);
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
if (!IS_ERR(server->client))
rpc_shutdown_client(server->client);
nfs_put_client(server->nfs_client);
nfs_free_iostats(server->io_stats);
bdi_destroy(&server->backing_dev_info);
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
kfree(server);
nfs_release_automount_timer();
dprintk("<-- nfs_free_server()\n");
}
/*
* Create a version 2 or 3 volume record
* - keyed on server and FSID
*/
struct nfs_server *nfs_create_server(const struct nfs_parsed_mount_data *data,
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
struct nfs_fh *mntfh)
{
struct nfs_server *server;
struct nfs_fattr fattr;
int error;
server = nfs_alloc_server();
if (!server)
return ERR_PTR(-ENOMEM);
/* Get a client representation */
error = nfs_init_server(server, data);
if (error < 0)
goto error;
BUG_ON(!server->nfs_client);
BUG_ON(!server->nfs_client->rpc_ops);
BUG_ON(!server->nfs_client->rpc_ops->file_inode_ops);
/* Probe the root fh to retrieve its FSID */
error = nfs_probe_fsinfo(server, mntfh, &fattr);
if (error < 0)
goto error;
if (server->nfs_client->rpc_ops->version == 3) {
if (server->namelen == 0 || server->namelen > NFS3_MAXNAMLEN)
server->namelen = NFS3_MAXNAMLEN;
if (!(data->flags & NFS_MOUNT_NORDIRPLUS))
server->caps |= NFS_CAP_READDIRPLUS;
} else {
if (server->namelen == 0 || server->namelen > NFS2_MAXNAMLEN)
server->namelen = NFS2_MAXNAMLEN;
}
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
if (!(fattr.valid & NFS_ATTR_FATTR)) {
error = server->nfs_client->rpc_ops->getattr(server, mntfh, &fattr);
if (error < 0) {
dprintk("nfs_create_server: getattr error = %d\n", -error);
goto error;
}
}
memcpy(&server->fsid, &fattr.fsid, sizeof(server->fsid));
dprintk("Server FSID: %llx:%llx\n",
(unsigned long long) server->fsid.major,
(unsigned long long) server->fsid.minor);
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
BUG_ON(!server->nfs_client);
BUG_ON(!server->nfs_client->rpc_ops);
BUG_ON(!server->nfs_client->rpc_ops->file_inode_ops);
spin_lock(&nfs_client_lock);
list_add_tail(&server->client_link, &server->nfs_client->cl_superblocks);
list_add_tail(&server->master_link, &nfs_volume_list);
spin_unlock(&nfs_client_lock);
server->mount_time = jiffies;
return server;
error:
nfs_free_server(server);
return ERR_PTR(error);
}
#ifdef CONFIG_NFS_V4
/*
* Initialise an NFS4 client record
*/
static int nfs4_init_client(struct nfs_client *clp,
const struct rpc_timeout *timeparms,
const char *ip_addr,
NFS: add "[no]resvport" mount option The standard default security setting for NFS is AUTH_SYS. An NFS client connects to NFS servers via a privileged source port and a fixed standard destination port (2049). The client sends raw uid and gid numbers to identify users making NFS requests, and the server assumes an appropriate authority on the client has vetted these values because the source port is privileged. On Linux, by default in-kernel RPC services use a privileged port in the range between 650 and 1023 to avoid using source ports of well- known IP services. Using such a small range limits the number of NFS mount points and the number of unique NFS servers to which a client can connect concurrently. An NFS client can use unprivileged source ports to expand the range of source port numbers, allowing more concurrent server connections and more NFS mount points. Servers must explicitly allow NFS connections from unprivileged ports for this to work. In the past, bumping the value of the sunrpc.max_resvport sysctl on the client would permit the NFS client to use unprivileged ports. Bumping this setting also changes the maximum port number used by other in-kernel RPC services, some of which still required a port number less than 1023. This is exacerbated by the way source port numbers are chosen by the Linux RPC client, which starts at the top of the range and works downwards. It means that bumping the maximum means all RPC services requesting a source port will likely get an unprivileged port instead of a privileged one. Changing this setting effects all NFS mount points on a client. A sysadmin could not selectively choose which mount points would use non-privileged ports and which could not. Lastly, this mechanism of expanding the limit on the number of NFS mount points was entirely undocumented. To address the need for the NFS client to use a large range of source ports without interfering with the activity of other in-kernel RPC services, we introduce a new NFS mount option. This option explicitly tells only the NFS client to use a non-privileged source port when communicating with the NFS server for one specific mount point. This new mount option is called "resvport," like the similar NFS mount option on FreeBSD and Mac OS X. A sister patch for nfs-utils will be submitted that documents this new option in nfs(5). The default setting for this new mount option requires the NFS client to use a privileged port, as before. Explicitly specifying the "noresvport" mount option allows the NFS client to use an unprivileged source port for this mount point when connecting to the NFS server port. This mount option is supported only for text-based NFS mounts. [ Sidebar: it is widely known that security mechanisms based on the use of privileged source ports are ineffective. However, the NFS client can combine the use of unprivileged ports with the use of secure authentication mechanisms, such as Kerberos. This allows a large number of connections and mount points while ensuring a useful level of security. Eventually we may change the default setting for this option depending on the security flavor used for the mount. For example, if the mount is using only AUTH_SYS, then the default setting will be "resvport;" if the mount is using a strong security flavor such as krb5, the default setting will be "noresvport." ] Signed-off-by: Chuck Lever <chuck.lever@oracle.com> [Trond.Myklebust@netapp.com: Fixed a bug whereby nfs4_init_client() was being called with incorrect arguments.] Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2008-12-24 04:21:37 +08:00
rpc_authflavor_t authflavour,
int flags)
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
{
int error;
if (clp->cl_cons_state == NFS_CS_READY) {
/* the client is initialised already */
dprintk("<-- nfs4_init_client() = 0 [already %p]\n", clp);
return 0;
}
/* Check NFS protocol revision and initialize RPC op vector */
clp->rpc_ops = &nfs_v4_clientops;
error = nfs_create_rpc_client(clp, timeparms, authflavour,
NFS: add "[no]resvport" mount option The standard default security setting for NFS is AUTH_SYS. An NFS client connects to NFS servers via a privileged source port and a fixed standard destination port (2049). The client sends raw uid and gid numbers to identify users making NFS requests, and the server assumes an appropriate authority on the client has vetted these values because the source port is privileged. On Linux, by default in-kernel RPC services use a privileged port in the range between 650 and 1023 to avoid using source ports of well- known IP services. Using such a small range limits the number of NFS mount points and the number of unique NFS servers to which a client can connect concurrently. An NFS client can use unprivileged source ports to expand the range of source port numbers, allowing more concurrent server connections and more NFS mount points. Servers must explicitly allow NFS connections from unprivileged ports for this to work. In the past, bumping the value of the sunrpc.max_resvport sysctl on the client would permit the NFS client to use unprivileged ports. Bumping this setting also changes the maximum port number used by other in-kernel RPC services, some of which still required a port number less than 1023. This is exacerbated by the way source port numbers are chosen by the Linux RPC client, which starts at the top of the range and works downwards. It means that bumping the maximum means all RPC services requesting a source port will likely get an unprivileged port instead of a privileged one. Changing this setting effects all NFS mount points on a client. A sysadmin could not selectively choose which mount points would use non-privileged ports and which could not. Lastly, this mechanism of expanding the limit on the number of NFS mount points was entirely undocumented. To address the need for the NFS client to use a large range of source ports without interfering with the activity of other in-kernel RPC services, we introduce a new NFS mount option. This option explicitly tells only the NFS client to use a non-privileged source port when communicating with the NFS server for one specific mount point. This new mount option is called "resvport," like the similar NFS mount option on FreeBSD and Mac OS X. A sister patch for nfs-utils will be submitted that documents this new option in nfs(5). The default setting for this new mount option requires the NFS client to use a privileged port, as before. Explicitly specifying the "noresvport" mount option allows the NFS client to use an unprivileged source port for this mount point when connecting to the NFS server port. This mount option is supported only for text-based NFS mounts. [ Sidebar: it is widely known that security mechanisms based on the use of privileged source ports are ineffective. However, the NFS client can combine the use of unprivileged ports with the use of secure authentication mechanisms, such as Kerberos. This allows a large number of connections and mount points while ensuring a useful level of security. Eventually we may change the default setting for this option depending on the security flavor used for the mount. For example, if the mount is using only AUTH_SYS, then the default setting will be "resvport;" if the mount is using a strong security flavor such as krb5, the default setting will be "noresvport." ] Signed-off-by: Chuck Lever <chuck.lever@oracle.com> [Trond.Myklebust@netapp.com: Fixed a bug whereby nfs4_init_client() was being called with incorrect arguments.] Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2008-12-24 04:21:37 +08:00
1, flags & NFS_MOUNT_NORESVPORT);
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
if (error < 0)
goto error;
memcpy(clp->cl_ipaddr, ip_addr, sizeof(clp->cl_ipaddr));
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
error = nfs_idmap_new(clp);
if (error < 0) {
dprintk("%s: failed to create idmapper. Error = %d\n",
__func__, error);
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
goto error;
}
__set_bit(NFS_CS_IDMAP, &clp->cl_res_state);
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
nfs_mark_client_ready(clp, NFS_CS_READY);
return 0;
error:
nfs_mark_client_ready(clp, error);
dprintk("<-- nfs4_init_client() = xerror %d\n", error);
return error;
}
/*
* Set up an NFS4 client
*/
static int nfs4_set_client(struct nfs_server *server,
const char *hostname,
const struct sockaddr *addr,
const size_t addrlen,
const char *ip_addr,
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
rpc_authflavor_t authflavour,
int proto, const struct rpc_timeout *timeparms)
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
{
struct nfs_client_initdata cl_init = {
.hostname = hostname,
.addr = addr,
.addrlen = addrlen,
.rpc_ops = &nfs_v4_clientops,
.proto = proto,
};
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
struct nfs_client *clp;
int error;
dprintk("--> nfs4_set_client()\n");
/* Allocate or find a client reference we can use */
clp = nfs_get_client(&cl_init);
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
if (IS_ERR(clp)) {
error = PTR_ERR(clp);
goto error;
}
NFS: add "[no]resvport" mount option The standard default security setting for NFS is AUTH_SYS. An NFS client connects to NFS servers via a privileged source port and a fixed standard destination port (2049). The client sends raw uid and gid numbers to identify users making NFS requests, and the server assumes an appropriate authority on the client has vetted these values because the source port is privileged. On Linux, by default in-kernel RPC services use a privileged port in the range between 650 and 1023 to avoid using source ports of well- known IP services. Using such a small range limits the number of NFS mount points and the number of unique NFS servers to which a client can connect concurrently. An NFS client can use unprivileged source ports to expand the range of source port numbers, allowing more concurrent server connections and more NFS mount points. Servers must explicitly allow NFS connections from unprivileged ports for this to work. In the past, bumping the value of the sunrpc.max_resvport sysctl on the client would permit the NFS client to use unprivileged ports. Bumping this setting also changes the maximum port number used by other in-kernel RPC services, some of which still required a port number less than 1023. This is exacerbated by the way source port numbers are chosen by the Linux RPC client, which starts at the top of the range and works downwards. It means that bumping the maximum means all RPC services requesting a source port will likely get an unprivileged port instead of a privileged one. Changing this setting effects all NFS mount points on a client. A sysadmin could not selectively choose which mount points would use non-privileged ports and which could not. Lastly, this mechanism of expanding the limit on the number of NFS mount points was entirely undocumented. To address the need for the NFS client to use a large range of source ports without interfering with the activity of other in-kernel RPC services, we introduce a new NFS mount option. This option explicitly tells only the NFS client to use a non-privileged source port when communicating with the NFS server for one specific mount point. This new mount option is called "resvport," like the similar NFS mount option on FreeBSD and Mac OS X. A sister patch for nfs-utils will be submitted that documents this new option in nfs(5). The default setting for this new mount option requires the NFS client to use a privileged port, as before. Explicitly specifying the "noresvport" mount option allows the NFS client to use an unprivileged source port for this mount point when connecting to the NFS server port. This mount option is supported only for text-based NFS mounts. [ Sidebar: it is widely known that security mechanisms based on the use of privileged source ports are ineffective. However, the NFS client can combine the use of unprivileged ports with the use of secure authentication mechanisms, such as Kerberos. This allows a large number of connections and mount points while ensuring a useful level of security. Eventually we may change the default setting for this option depending on the security flavor used for the mount. For example, if the mount is using only AUTH_SYS, then the default setting will be "resvport;" if the mount is using a strong security flavor such as krb5, the default setting will be "noresvport." ] Signed-off-by: Chuck Lever <chuck.lever@oracle.com> [Trond.Myklebust@netapp.com: Fixed a bug whereby nfs4_init_client() was being called with incorrect arguments.] Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2008-12-24 04:21:37 +08:00
error = nfs4_init_client(clp, timeparms, ip_addr, authflavour,
server->flags);
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
if (error < 0)
goto error_put;
server->nfs_client = clp;
dprintk("<-- nfs4_set_client() = 0 [new %p]\n", clp);
return 0;
error_put:
nfs_put_client(clp);
error:
dprintk("<-- nfs4_set_client() = xerror %d\n", error);
return error;
}
/*
* Create a version 4 volume record
*/
static int nfs4_init_server(struct nfs_server *server,
const struct nfs_parsed_mount_data *data)
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
{
struct rpc_timeout timeparms;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
int error;
dprintk("--> nfs4_init_server()\n");
nfs_init_timeout_values(&timeparms, data->nfs_server.protocol,
data->timeo, data->retrans);
/* Initialise the client representation from the mount data */
server->flags = data->flags;
server->caps |= NFS_CAP_ATOMIC_OPEN;
/* Get a client record */
error = nfs4_set_client(server,
data->nfs_server.hostname,
(const struct sockaddr *)&data->nfs_server.address,
data->nfs_server.addrlen,
data->client_address,
data->auth_flavors[0],
data->nfs_server.protocol,
&timeparms);
if (error < 0)
goto error;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
if (data->rsize)
server->rsize = nfs_block_size(data->rsize, NULL);
if (data->wsize)
server->wsize = nfs_block_size(data->wsize, NULL);
server->acregmin = data->acregmin * HZ;
server->acregmax = data->acregmax * HZ;
server->acdirmin = data->acdirmin * HZ;
server->acdirmax = data->acdirmax * HZ;
server->port = data->nfs_server.port;
error = nfs_init_server_rpcclient(server, &timeparms, data->auth_flavors[0]);
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
error:
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
/* Done */
dprintk("<-- nfs4_init_server() = %d\n", error);
return error;
}
/*
* Create a version 4 volume record
* - keyed on server and FSID
*/
struct nfs_server *nfs4_create_server(const struct nfs_parsed_mount_data *data,
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
struct nfs_fh *mntfh)
{
struct nfs_fattr fattr;
struct nfs_server *server;
int error;
dprintk("--> nfs4_create_server()\n");
server = nfs_alloc_server();
if (!server)
return ERR_PTR(-ENOMEM);
/* set up the general RPC client */
error = nfs4_init_server(server, data);
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
if (error < 0)
goto error;
BUG_ON(!server->nfs_client);
BUG_ON(!server->nfs_client->rpc_ops);
BUG_ON(!server->nfs_client->rpc_ops->file_inode_ops);
/* Probe the root fh to retrieve its FSID */
error = nfs4_path_walk(server, mntfh, data->nfs_server.export_path);
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
if (error < 0)
goto error;
dprintk("Server FSID: %llx:%llx\n",
(unsigned long long) server->fsid.major,
(unsigned long long) server->fsid.minor);
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
dprintk("Mount FH: %d\n", mntfh->size);
error = nfs_probe_fsinfo(server, mntfh, &fattr);
if (error < 0)
goto error;
if (server->namelen == 0 || server->namelen > NFS4_MAXNAMLEN)
server->namelen = NFS4_MAXNAMLEN;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
BUG_ON(!server->nfs_client);
BUG_ON(!server->nfs_client->rpc_ops);
BUG_ON(!server->nfs_client->rpc_ops->file_inode_ops);
spin_lock(&nfs_client_lock);
list_add_tail(&server->client_link, &server->nfs_client->cl_superblocks);
list_add_tail(&server->master_link, &nfs_volume_list);
spin_unlock(&nfs_client_lock);
server->mount_time = jiffies;
dprintk("<-- nfs4_create_server() = %p\n", server);
return server;
error:
nfs_free_server(server);
dprintk("<-- nfs4_create_server() = error %d\n", error);
return ERR_PTR(error);
}
/*
* Create an NFS4 referral server record
*/
struct nfs_server *nfs4_create_referral_server(struct nfs_clone_mount *data,
struct nfs_fh *mntfh)
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
{
struct nfs_client *parent_client;
struct nfs_server *server, *parent_server;
struct nfs_fattr fattr;
int error;
dprintk("--> nfs4_create_referral_server()\n");
server = nfs_alloc_server();
if (!server)
return ERR_PTR(-ENOMEM);
parent_server = NFS_SB(data->sb);
parent_client = parent_server->nfs_client;
/* Initialise the client representation from the parent server */
nfs_server_copy_userdata(server, parent_server);
server->caps |= NFS_CAP_ATOMIC_OPEN;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
/* Get a client representation.
* Note: NFSv4 always uses TCP, */
error = nfs4_set_client(server, data->hostname,
data->addr,
data->addrlen,
parent_client->cl_ipaddr,
data->authflavor,
parent_server->client->cl_xprt->prot,
parent_server->client->cl_timeout);
if (error < 0)
goto error;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
error = nfs_init_server_rpcclient(server, parent_server->client->cl_timeout, data->authflavor);
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
if (error < 0)
goto error;
BUG_ON(!server->nfs_client);
BUG_ON(!server->nfs_client->rpc_ops);
BUG_ON(!server->nfs_client->rpc_ops->file_inode_ops);
/* Probe the root fh to retrieve its FSID and filehandle */
error = nfs4_path_walk(server, mntfh, data->mnt_path);
if (error < 0)
goto error;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
/* probe the filesystem info for this server filesystem */
error = nfs_probe_fsinfo(server, mntfh, &fattr);
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
if (error < 0)
goto error;
if (server->namelen == 0 || server->namelen > NFS4_MAXNAMLEN)
server->namelen = NFS4_MAXNAMLEN;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
dprintk("Referral FSID: %llx:%llx\n",
(unsigned long long) server->fsid.major,
(unsigned long long) server->fsid.minor);
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
spin_lock(&nfs_client_lock);
list_add_tail(&server->client_link, &server->nfs_client->cl_superblocks);
list_add_tail(&server->master_link, &nfs_volume_list);
spin_unlock(&nfs_client_lock);
server->mount_time = jiffies;
dprintk("<-- nfs_create_referral_server() = %p\n", server);
return server;
error:
nfs_free_server(server);
dprintk("<-- nfs4_create_referral_server() = error %d\n", error);
return ERR_PTR(error);
}
#endif /* CONFIG_NFS_V4 */
/*
* Clone an NFS2, NFS3 or NFS4 server record
*/
struct nfs_server *nfs_clone_server(struct nfs_server *source,
struct nfs_fh *fh,
struct nfs_fattr *fattr)
{
struct nfs_server *server;
struct nfs_fattr fattr_fsinfo;
int error;
dprintk("--> nfs_clone_server(,%llx:%llx,)\n",
(unsigned long long) fattr->fsid.major,
(unsigned long long) fattr->fsid.minor);
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
server = nfs_alloc_server();
if (!server)
return ERR_PTR(-ENOMEM);
/* Copy data from the source */
server->nfs_client = source->nfs_client;
atomic_inc(&server->nfs_client->cl_count);
nfs_server_copy_userdata(server, source);
server->fsid = fattr->fsid;
error = nfs_init_server_rpcclient(server,
source->client->cl_timeout,
source->client->cl_auth->au_flavor);
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
if (error < 0)
goto out_free_server;
if (!IS_ERR(source->client_acl))
nfs_init_server_aclclient(server);
/* probe the filesystem info for this server filesystem */
error = nfs_probe_fsinfo(server, fh, &fattr_fsinfo);
if (error < 0)
goto out_free_server;
if (server->namelen == 0 || server->namelen > NFS4_MAXNAMLEN)
server->namelen = NFS4_MAXNAMLEN;
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
dprintk("Cloned FSID: %llx:%llx\n",
(unsigned long long) server->fsid.major,
(unsigned long long) server->fsid.minor);
NFS: Share NFS superblocks per-protocol per-server per-FSID The attached patch makes NFS share superblocks between mounts from the same server and FSID over the same protocol. It does this by creating each superblock with a false root and returning the real root dentry in the vfsmount presented by get_sb(). The root dentry set starts off as an anonymous dentry if we don't already have the dentry for its inode, otherwise it simply returns the dentry we already have. We may thus end up with several trees of dentries in the superblock, and if at some later point one of anonymous tree roots is discovered by normal filesystem activity to be located in another tree within the superblock, the anonymous root is named and materialises attached to the second tree at the appropriate point. Why do it this way? Why not pass an extra argument to the mount() syscall to indicate the subpath and then pathwalk from the server root to the desired directory? You can't guarantee this will work for two reasons: (1) The root and intervening nodes may not be accessible to the client. With NFS2 and NFS3, for instance, mountd is called on the server to get the filehandle for the tip of a path. mountd won't give us handles for anything we don't have permission to access, and so we can't set up NFS inodes for such nodes, and so can't easily set up dentries (we'd have to have ghost inodes or something). With this patch we don't actually create dentries until we get handles from the server that we can use to set up their inodes, and we don't actually bind them into the tree until we know for sure where they go. (2) Inaccessible symbolic links. If we're asked to mount two exports from the server, eg: mount warthog:/warthog/aaa/xxx /mmm mount warthog:/warthog/bbb/yyy /nnn We may not be able to access anything nearer the root than xxx and yyy, but we may find out later that /mmm/www/yyy, say, is actually the same directory as the one mounted on /nnn. What we might then find out, for example, is that /warthog/bbb was actually a symbolic link to /warthog/aaa/xxx/www, but we can't actually determine that by talking to the server until /warthog is made available by NFS. This would lead to having constructed an errneous dentry tree which we can't easily fix. We can end up with a dentry marked as a directory when it should actually be a symlink, or we could end up with an apparently hardlinked directory. With this patch we need not make assumptions about the type of a dentry for which we can't retrieve information, nor need we assume we know its place in the grand scheme of things until we actually see that place. This patch reduces the possibility of aliasing in the inode and page caches for inodes that may be accessed by more than one NFS export. It also reduces the number of superblocks required for NFS where there are many NFS exports being used from a server (home directory server + autofs for example). This in turn makes it simpler to do local caching of network filesystems, as it can then be guaranteed that there won't be links from multiple inodes in separate superblocks to the same cache file. Obviously, cache aliasing between different levels of NFS protocol could still be a problem, but at least that gives us another key to use when indexing the cache. This patch makes the following changes: (1) The server record construction/destruction has been abstracted out into its own set of functions to make things easier to get right. These have been moved into fs/nfs/client.c. All the code in fs/nfs/client.c has to do with the management of connections to servers, and doesn't touch superblocks in any way; the remaining code in fs/nfs/super.c has to do with VFS superblock management. (2) The sequence of events undertaken by NFS mount is now reordered: (a) A volume representation (struct nfs_server) is allocated. (b) A server representation (struct nfs_client) is acquired. This may be allocated or shared, and is keyed on server address, port and NFS version. (c) If allocated, the client representation is initialised. The state member variable of nfs_client is used to prevent a race during initialisation from two mounts. (d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we are given the root FH in advance. (e) The volume FSID is probed for on the root FH. (f) The volume representation is initialised from the FSINFO record retrieved on the root FH. (g) sget() is called to acquire a superblock. This may be allocated or shared, keyed on client pointer and FSID. (h) If allocated, the superblock is initialised. (i) If the superblock is shared, then the new nfs_server record is discarded. (j) The root dentry for this mount is looked up from the root FH. (k) The root dentry for this mount is assigned to the vfsmount. (3) nfs_readdir_lookup() creates dentries for each of the entries readdir() returns; this function now attaches disconnected trees from alternate roots that happen to be discovered attached to a directory being read (in the same way nfs_lookup() is made to do for lookup ops). The new d_materialise_unique() function is now used to do this, thus permitting the whole thing to be done under one set of locks, and thus avoiding any race between mount and lookup operations on the same directory. (4) The client management code uses a new debug facility: NFSDBG_CLIENT which is set by echoing 1024 to /proc/net/sunrpc/nfs_debug. (5) Clone mounts are now called xdev mounts. (6) Use the dentry passed to the statfs() op as the handle for retrieving fs statistics rather than the root dentry of the superblock (which is now a dummy). Signed-Off-By: David Howells <dhowells@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2006-08-23 08:06:13 +08:00
error = nfs_start_lockd(server);
if (error < 0)
goto out_free_server;
spin_lock(&nfs_client_lock);
list_add_tail(&server->client_link, &server->nfs_client->cl_superblocks);
list_add_tail(&server->master_link, &nfs_volume_list);
spin_unlock(&nfs_client_lock);
server->mount_time = jiffies;
dprintk("<-- nfs_clone_server() = %p\n", server);
return server;
out_free_server:
nfs_free_server(server);
dprintk("<-- nfs_clone_server() = error %d\n", error);
return ERR_PTR(error);
}
#ifdef CONFIG_PROC_FS
static struct proc_dir_entry *proc_fs_nfs;
static int nfs_server_list_open(struct inode *inode, struct file *file);
static void *nfs_server_list_start(struct seq_file *p, loff_t *pos);
static void *nfs_server_list_next(struct seq_file *p, void *v, loff_t *pos);
static void nfs_server_list_stop(struct seq_file *p, void *v);
static int nfs_server_list_show(struct seq_file *m, void *v);
static struct seq_operations nfs_server_list_ops = {
.start = nfs_server_list_start,
.next = nfs_server_list_next,
.stop = nfs_server_list_stop,
.show = nfs_server_list_show,
};
static const struct file_operations nfs_server_list_fops = {
.open = nfs_server_list_open,
.read = seq_read,
.llseek = seq_lseek,
.release = seq_release,
.owner = THIS_MODULE,
};
static int nfs_volume_list_open(struct inode *inode, struct file *file);
static void *nfs_volume_list_start(struct seq_file *p, loff_t *pos);
static void *nfs_volume_list_next(struct seq_file *p, void *v, loff_t *pos);
static void nfs_volume_list_stop(struct seq_file *p, void *v);
static int nfs_volume_list_show(struct seq_file *m, void *v);
static struct seq_operations nfs_volume_list_ops = {
.start = nfs_volume_list_start,
.next = nfs_volume_list_next,
.stop = nfs_volume_list_stop,
.show = nfs_volume_list_show,
};
static const struct file_operations nfs_volume_list_fops = {
.open = nfs_volume_list_open,
.read = seq_read,
.llseek = seq_lseek,
.release = seq_release,
.owner = THIS_MODULE,
};
/*
* open "/proc/fs/nfsfs/servers" which provides a summary of servers with which
* we're dealing
*/
static int nfs_server_list_open(struct inode *inode, struct file *file)
{
struct seq_file *m;
int ret;
ret = seq_open(file, &nfs_server_list_ops);
if (ret < 0)
return ret;
m = file->private_data;
m->private = PDE(inode)->data;
return 0;
}
/*
* set up the iterator to start reading from the server list and return the first item
*/
static void *nfs_server_list_start(struct seq_file *m, loff_t *_pos)
{
/* lock the list against modification */
spin_lock(&nfs_client_lock);
return seq_list_start_head(&nfs_client_list, *_pos);
}
/*
* move to next server
*/
static void *nfs_server_list_next(struct seq_file *p, void *v, loff_t *pos)
{
return seq_list_next(v, &nfs_client_list, pos);
}
/*
* clean up after reading from the transports list
*/
static void nfs_server_list_stop(struct seq_file *p, void *v)
{
spin_unlock(&nfs_client_lock);
}
/*
* display a header line followed by a load of call lines
*/
static int nfs_server_list_show(struct seq_file *m, void *v)
{
struct nfs_client *clp;
/* display header on line 1 */
if (v == &nfs_client_list) {
seq_puts(m, "NV SERVER PORT USE HOSTNAME\n");
return 0;
}
/* display one transport per line on subsequent lines */
clp = list_entry(v, struct nfs_client, cl_share_link);
seq_printf(m, "v%u %s %s %3d %s\n",
clp->rpc_ops->version,
rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_HEX_ADDR),
rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_HEX_PORT),
atomic_read(&clp->cl_count),
clp->cl_hostname);
return 0;
}
/*
* open "/proc/fs/nfsfs/volumes" which provides a summary of extant volumes
*/
static int nfs_volume_list_open(struct inode *inode, struct file *file)
{
struct seq_file *m;
int ret;
ret = seq_open(file, &nfs_volume_list_ops);
if (ret < 0)
return ret;
m = file->private_data;
m->private = PDE(inode)->data;
return 0;
}
/*
* set up the iterator to start reading from the volume list and return the first item
*/
static void *nfs_volume_list_start(struct seq_file *m, loff_t *_pos)
{
/* lock the list against modification */
spin_lock(&nfs_client_lock);
return seq_list_start_head(&nfs_volume_list, *_pos);
}
/*
* move to next volume
*/
static void *nfs_volume_list_next(struct seq_file *p, void *v, loff_t *pos)
{
return seq_list_next(v, &nfs_volume_list, pos);
}
/*
* clean up after reading from the transports list
*/
static void nfs_volume_list_stop(struct seq_file *p, void *v)
{
spin_unlock(&nfs_client_lock);
}
/*
* display a header line followed by a load of call lines
*/
static int nfs_volume_list_show(struct seq_file *m, void *v)
{
struct nfs_server *server;
struct nfs_client *clp;
char dev[8], fsid[17];
/* display header on line 1 */
if (v == &nfs_volume_list) {
seq_puts(m, "NV SERVER PORT DEV FSID\n");
return 0;
}
/* display one transport per line on subsequent lines */
server = list_entry(v, struct nfs_server, master_link);
clp = server->nfs_client;
snprintf(dev, 8, "%u:%u",
MAJOR(server->s_dev), MINOR(server->s_dev));
snprintf(fsid, 17, "%llx:%llx",
(unsigned long long) server->fsid.major,
(unsigned long long) server->fsid.minor);
seq_printf(m, "v%u %s %s %-7s %-17s\n",
clp->rpc_ops->version,
rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_HEX_ADDR),
rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_HEX_PORT),
dev,
fsid);
return 0;
}
/*
* initialise the /proc/fs/nfsfs/ directory
*/
int __init nfs_fs_proc_init(void)
{
struct proc_dir_entry *p;
proc_fs_nfs = proc_mkdir("fs/nfsfs", NULL);
if (!proc_fs_nfs)
goto error_0;
proc_fs_nfs->owner = THIS_MODULE;
/* a file of servers with which we're dealing */
p = proc_create("servers", S_IFREG|S_IRUGO,
proc_fs_nfs, &nfs_server_list_fops);
if (!p)
goto error_1;
/* a file of volumes that we have mounted */
p = proc_create("volumes", S_IFREG|S_IRUGO,
proc_fs_nfs, &nfs_volume_list_fops);
if (!p)
goto error_2;
return 0;
error_2:
remove_proc_entry("servers", proc_fs_nfs);
error_1:
remove_proc_entry("fs/nfsfs", NULL);
error_0:
return -ENOMEM;
}
/*
* clean up the /proc/fs/nfsfs/ directory
*/
void nfs_fs_proc_exit(void)
{
remove_proc_entry("volumes", proc_fs_nfs);
remove_proc_entry("servers", proc_fs_nfs);
remove_proc_entry("fs/nfsfs", NULL);
}
#endif /* CONFIG_PROC_FS */