2005-04-17 06:20:36 +08:00
|
|
|
/* dir.c: AFS filesystem directory handling
|
|
|
|
*
|
|
|
|
* Copyright (C) 2002 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/kernel.h>
|
|
|
|
#include <linux/module.h>
|
|
|
|
#include <linux/init.h>
|
|
|
|
#include <linux/fs.h>
|
2011-01-07 14:49:57 +08:00
|
|
|
#include <linux/namei.h>
|
2005-04-17 06:20:36 +08:00
|
|
|
#include <linux/pagemap.h>
|
2007-04-27 06:57:07 +08:00
|
|
|
#include <linux/ctype.h>
|
Detach sched.h from mm.h
First thing mm.h does is including sched.h solely for can_do_mlock() inline
function which has "current" dereference inside. By dealing with can_do_mlock()
mm.h can be detached from sched.h which is good. See below, why.
This patch
a) removes unconditional inclusion of sched.h from mm.h
b) makes can_do_mlock() normal function in mm/mlock.c
c) exports can_do_mlock() to not break compilation
d) adds sched.h inclusions back to files that were getting it indirectly.
e) adds less bloated headers to some files (asm/signal.h, jiffies.h) that were
getting them indirectly
Net result is:
a) mm.h users would get less code to open, read, preprocess, parse, ... if
they don't need sched.h
b) sched.h stops being dependency for significant number of files:
on x86_64 allmodconfig touching sched.h results in recompile of 4083 files,
after patch it's only 3744 (-8.3%).
Cross-compile tested on
all arm defconfigs, all mips defconfigs, all powerpc defconfigs,
alpha alpha-up
arm
i386 i386-up i386-defconfig i386-allnoconfig
ia64 ia64-up
m68k
mips
parisc parisc-up
powerpc powerpc-up
s390 s390-up
sparc sparc-up
sparc64 sparc64-up
um-x86_64
x86_64 x86_64-up x86_64-defconfig x86_64-allnoconfig
as well as my two usual configs.
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-21 05:22:52 +08:00
|
|
|
#include <linux/sched.h>
|
2018-02-06 14:26:30 +08:00
|
|
|
#include <linux/dns_resolver.h>
|
2005-04-17 06:20:36 +08:00
|
|
|
#include "internal.h"
|
|
|
|
|
2007-04-27 06:59:35 +08:00
|
|
|
static struct dentry *afs_lookup(struct inode *dir, struct dentry *dentry,
|
2012-06-11 05:13:09 +08:00
|
|
|
unsigned int flags);
|
2018-02-06 14:26:30 +08:00
|
|
|
static struct dentry *afs_dynroot_lookup(struct inode *dir, struct dentry *dentry,
|
|
|
|
unsigned int flags);
|
2005-04-17 06:20:36 +08:00
|
|
|
static int afs_dir_open(struct inode *inode, struct file *file);
|
2013-05-23 04:31:14 +08:00
|
|
|
static int afs_readdir(struct file *file, struct dir_context *ctx);
|
2012-06-11 04:03:43 +08:00
|
|
|
static int afs_d_revalidate(struct dentry *dentry, unsigned int flags);
|
2011-01-07 14:49:23 +08:00
|
|
|
static int afs_d_delete(const struct dentry *dentry);
|
2007-04-27 06:59:35 +08:00
|
|
|
static void afs_d_release(struct dentry *dentry);
|
2018-04-10 04:12:31 +08:00
|
|
|
static int afs_lookup_one_filldir(struct dir_context *ctx, const char *name, int nlen,
|
2006-10-03 16:13:46 +08:00
|
|
|
loff_t fpos, u64 ino, unsigned dtype);
|
2018-04-10 04:12:31 +08:00
|
|
|
static int afs_lookup_filldir(struct dir_context *ctx, const char *name, int nlen,
|
|
|
|
loff_t fpos, u64 ino, unsigned dtype);
|
2011-07-26 13:42:34 +08:00
|
|
|
static int afs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
|
2012-06-11 06:05:36 +08:00
|
|
|
bool excl);
|
2011-07-26 13:41:39 +08:00
|
|
|
static int afs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode);
|
2007-04-27 06:59:35 +08:00
|
|
|
static int afs_rmdir(struct inode *dir, struct dentry *dentry);
|
|
|
|
static int afs_unlink(struct inode *dir, struct dentry *dentry);
|
|
|
|
static int afs_link(struct dentry *from, struct inode *dir,
|
|
|
|
struct dentry *dentry);
|
|
|
|
static int afs_symlink(struct inode *dir, struct dentry *dentry,
|
|
|
|
const char *content);
|
|
|
|
static int afs_rename(struct inode *old_dir, struct dentry *old_dentry,
|
fs: make remaining filesystems use .rename2
This is trivial to do:
- add flags argument to foo_rename()
- check if flags is zero
- assign foo_rename() to .rename2 instead of .rename
This doesn't mean it's impossible to support RENAME_NOREPLACE for these
filesystems, but it is not trivial, like for local filesystems.
RENAME_NOREPLACE must guarantee atomicity (i.e. it shouldn't be possible
for a file to be created on one host while it is overwritten by rename on
another host).
Filesystems converted:
9p, afs, ceph, coda, ecryptfs, kernfs, lustre, ncpfs, nfs, ocfs2, orangefs.
After this, we can get rid of the duplicate interfaces for rename.
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Acked-by: David Howells <dhowells@redhat.com> [AFS]
Acked-by: Mike Marshall <hubcap@omnibond.com>
Cc: Eric Van Hensbergen <ericvh@gmail.com>
Cc: Ilya Dryomov <idryomov@gmail.com>
Cc: Jan Harkes <jaharkes@cs.cmu.edu>
Cc: Tyler Hicks <tyhicks@canonical.com>
Cc: Oleg Drokin <oleg.drokin@intel.com>
Cc: Trond Myklebust <trond.myklebust@primarydata.com>
Cc: Mark Fasheh <mfasheh@suse.com>
2016-09-27 17:03:58 +08:00
|
|
|
struct inode *new_dir, struct dentry *new_dentry,
|
|
|
|
unsigned int flags);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2006-03-28 17:56:42 +08:00
|
|
|
const struct file_operations afs_dir_file_operations = {
|
2005-04-17 06:20:36 +08:00
|
|
|
.open = afs_dir_open,
|
2007-04-27 06:57:07 +08:00
|
|
|
.release = afs_release,
|
2016-05-11 02:27:44 +08:00
|
|
|
.iterate_shared = afs_readdir,
|
2007-07-16 14:40:12 +08:00
|
|
|
.lock = afs_lock,
|
2008-09-04 03:53:01 +08:00
|
|
|
.llseek = generic_file_llseek,
|
2005-04-17 06:20:36 +08:00
|
|
|
};
|
|
|
|
|
2007-02-12 16:55:38 +08:00
|
|
|
const struct inode_operations afs_dir_inode_operations = {
|
2007-04-27 06:59:35 +08:00
|
|
|
.create = afs_create,
|
|
|
|
.lookup = afs_lookup,
|
|
|
|
.link = afs_link,
|
|
|
|
.unlink = afs_unlink,
|
|
|
|
.symlink = afs_symlink,
|
|
|
|
.mkdir = afs_mkdir,
|
|
|
|
.rmdir = afs_rmdir,
|
2016-09-27 17:03:58 +08:00
|
|
|
.rename = afs_rename,
|
2007-04-27 06:57:07 +08:00
|
|
|
.permission = afs_permission,
|
2007-05-09 17:33:45 +08:00
|
|
|
.getattr = afs_getattr,
|
AFS: implement basic file write support
Implement support for writing to regular AFS files, including:
(1) write
(2) truncate
(3) fsync, fdatasync
(4) chmod, chown, chgrp, utime.
AFS writeback attempts to batch writes into as chunks as large as it can manage
up to the point that it writes back 65535 pages in one chunk or it meets a
locked page.
Furthermore, if a page has been written to using a particular key, then should
another write to that page use some other key, the first write will be flushed
before the second is allowed to take place. If the first write fails due to a
security error, then the page will be scrapped and reread before the second
write takes place.
If a page is dirty and the callback on it is broken by the server, then the
dirty data is not discarded (same behaviour as NFS).
Shared-writable mappings are not supported by this patch.
[akpm@linux-foundation.org: fix a bunch of warnings]
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-09 17:33:46 +08:00
|
|
|
.setattr = afs_setattr,
|
2017-07-06 22:50:27 +08:00
|
|
|
.listxattr = afs_listxattr,
|
2005-04-17 06:20:36 +08:00
|
|
|
};
|
|
|
|
|
2018-02-06 14:26:30 +08:00
|
|
|
const struct file_operations afs_dynroot_file_operations = {
|
|
|
|
.open = dcache_dir_open,
|
|
|
|
.release = dcache_dir_close,
|
|
|
|
.iterate_shared = dcache_readdir,
|
|
|
|
.llseek = dcache_dir_lseek,
|
|
|
|
};
|
|
|
|
|
|
|
|
const struct inode_operations afs_dynroot_inode_operations = {
|
|
|
|
.lookup = afs_dynroot_lookup,
|
|
|
|
};
|
|
|
|
|
2011-01-13 09:04:20 +08:00
|
|
|
const struct dentry_operations afs_fs_dentry_operations = {
|
2005-04-17 06:20:36 +08:00
|
|
|
.d_revalidate = afs_d_revalidate,
|
|
|
|
.d_delete = afs_d_delete,
|
2007-04-27 06:59:35 +08:00
|
|
|
.d_release = afs_d_release,
|
2011-01-15 03:04:05 +08:00
|
|
|
.d_automount = afs_d_automount,
|
2005-04-17 06:20:36 +08:00
|
|
|
};
|
|
|
|
|
|
|
|
#define AFS_DIR_HASHTBL_SIZE 128
|
|
|
|
#define AFS_DIR_DIRENT_SIZE 32
|
|
|
|
#define AFS_DIRENT_PER_BLOCK 64
|
|
|
|
|
|
|
|
union afs_dirent {
|
|
|
|
struct {
|
|
|
|
uint8_t valid;
|
|
|
|
uint8_t unused[1];
|
|
|
|
__be16 hash_next;
|
|
|
|
__be32 vnode;
|
|
|
|
__be32 unique;
|
|
|
|
uint8_t name[16];
|
|
|
|
uint8_t overflow[4]; /* if any char of the name (inc
|
|
|
|
* NUL) reaches here, consume
|
|
|
|
* the next dirent too */
|
|
|
|
} u;
|
|
|
|
uint8_t extended_name[32];
|
|
|
|
};
|
|
|
|
|
|
|
|
/* AFS directory page header (one at the beginning of every 2048-byte chunk) */
|
|
|
|
struct afs_dir_pagehdr {
|
|
|
|
__be16 npages;
|
|
|
|
__be16 magic;
|
|
|
|
#define AFS_DIR_MAGIC htons(1234)
|
|
|
|
uint8_t nentries;
|
|
|
|
uint8_t bitmap[8];
|
|
|
|
uint8_t pad[19];
|
|
|
|
};
|
|
|
|
|
|
|
|
/* directory block layout */
|
|
|
|
union afs_dir_block {
|
|
|
|
|
|
|
|
struct afs_dir_pagehdr pagehdr;
|
|
|
|
|
|
|
|
struct {
|
|
|
|
struct afs_dir_pagehdr pagehdr;
|
|
|
|
uint8_t alloc_ctrs[128];
|
|
|
|
/* dir hash table */
|
|
|
|
uint16_t hashtable[AFS_DIR_HASHTBL_SIZE];
|
|
|
|
} hdr;
|
|
|
|
|
|
|
|
union afs_dirent dirents[AFS_DIRENT_PER_BLOCK];
|
|
|
|
};
|
|
|
|
|
|
|
|
/* layout on a linux VM page */
|
|
|
|
struct afs_dir_page {
|
|
|
|
union afs_dir_block blocks[PAGE_SIZE / sizeof(union afs_dir_block)];
|
|
|
|
};
|
|
|
|
|
2018-04-10 04:12:31 +08:00
|
|
|
struct afs_lookup_one_cookie {
|
|
|
|
struct dir_context ctx;
|
|
|
|
struct qstr name;
|
|
|
|
bool found;
|
|
|
|
struct afs_fid fid;
|
|
|
|
};
|
|
|
|
|
2007-04-27 06:59:35 +08:00
|
|
|
struct afs_lookup_cookie {
|
2018-04-10 04:12:31 +08:00
|
|
|
struct dir_context ctx;
|
|
|
|
struct qstr name;
|
|
|
|
bool found;
|
|
|
|
bool one_only;
|
|
|
|
unsigned short nr_fids;
|
|
|
|
struct afs_file_status *statuses;
|
|
|
|
struct afs_callback *callbacks;
|
|
|
|
struct afs_fid fids[50];
|
2005-04-17 06:20:36 +08:00
|
|
|
};
|
|
|
|
|
|
|
|
/*
|
|
|
|
* check that a directory page is valid
|
|
|
|
*/
|
2017-11-02 23:27:52 +08:00
|
|
|
bool afs_dir_check_page(struct inode *dir, struct page *page)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
|
|
|
struct afs_dir_page *dbuf;
|
2017-11-02 23:27:52 +08:00
|
|
|
struct afs_vnode *vnode = AFS_FS_I(dir);
|
|
|
|
loff_t latter, i_size, off;
|
2005-04-17 06:20:36 +08:00
|
|
|
int tmp, qty;
|
|
|
|
|
|
|
|
#if 0
|
|
|
|
/* check the page count */
|
|
|
|
qty = desc.size / sizeof(dbuf->blocks[0]);
|
|
|
|
if (qty == 0)
|
|
|
|
goto error;
|
|
|
|
|
2007-04-27 06:55:03 +08:00
|
|
|
if (page->index == 0 && qty != ntohs(dbuf->blocks[0].pagehdr.npages)) {
|
2005-04-17 06:20:36 +08:00
|
|
|
printk("kAFS: %s(%lu): wrong number of dir blocks %d!=%hu\n",
|
2008-04-30 15:55:09 +08:00
|
|
|
__func__, dir->i_ino, qty,
|
2007-04-27 06:55:03 +08:00
|
|
|
ntohs(dbuf->blocks[0].pagehdr.npages));
|
2005-04-17 06:20:36 +08:00
|
|
|
goto error;
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
2017-11-02 23:27:52 +08:00
|
|
|
/* Determine how many magic numbers there should be in this page, but
|
|
|
|
* we must take care because the directory may change size under us.
|
|
|
|
*/
|
|
|
|
off = page_offset(page);
|
|
|
|
i_size = i_size_read(dir);
|
|
|
|
if (i_size <= off)
|
|
|
|
goto checked;
|
|
|
|
|
|
|
|
latter = i_size - off;
|
2005-04-17 06:20:36 +08:00
|
|
|
if (latter >= PAGE_SIZE)
|
|
|
|
qty = PAGE_SIZE;
|
|
|
|
else
|
|
|
|
qty = latter;
|
|
|
|
qty /= sizeof(union afs_dir_block);
|
|
|
|
|
|
|
|
/* check them */
|
|
|
|
dbuf = page_address(page);
|
|
|
|
for (tmp = 0; tmp < qty; tmp++) {
|
|
|
|
if (dbuf->blocks[tmp].pagehdr.magic != AFS_DIR_MAGIC) {
|
2017-11-02 23:27:52 +08:00
|
|
|
printk("kAFS: %s(%lx): bad magic %d/%d is %04hx\n",
|
2008-04-30 15:55:09 +08:00
|
|
|
__func__, dir->i_ino, tmp, qty,
|
2005-04-17 06:20:36 +08:00
|
|
|
ntohs(dbuf->blocks[tmp].pagehdr.magic));
|
2017-11-02 23:27:52 +08:00
|
|
|
trace_afs_dir_check_failed(vnode, off, i_size);
|
2005-04-17 06:20:36 +08:00
|
|
|
goto error;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2017-11-02 23:27:52 +08:00
|
|
|
checked:
|
2005-04-17 06:20:36 +08:00
|
|
|
SetPageChecked(page);
|
2016-04-23 03:06:44 +08:00
|
|
|
return true;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2007-04-27 06:49:28 +08:00
|
|
|
error:
|
2005-04-17 06:20:36 +08:00
|
|
|
SetPageError(page);
|
2016-04-23 03:06:44 +08:00
|
|
|
return false;
|
2007-04-27 06:49:28 +08:00
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* discard a page cached in the pagecache
|
|
|
|
*/
|
|
|
|
static inline void afs_dir_put_page(struct page *page)
|
|
|
|
{
|
|
|
|
kunmap(page);
|
2017-11-02 23:27:52 +08:00
|
|
|
unlock_page(page);
|
mm, fs: get rid of PAGE_CACHE_* and page_cache_{get,release} macros
PAGE_CACHE_{SIZE,SHIFT,MASK,ALIGN} macros were introduced *long* time
ago with promise that one day it will be possible to implement page
cache with bigger chunks than PAGE_SIZE.
This promise never materialized. And unlikely will.
We have many places where PAGE_CACHE_SIZE assumed to be equal to
PAGE_SIZE. And it's constant source of confusion on whether
PAGE_CACHE_* or PAGE_* constant should be used in a particular case,
especially on the border between fs and mm.
Global switching to PAGE_CACHE_SIZE != PAGE_SIZE would cause to much
breakage to be doable.
Let's stop pretending that pages in page cache are special. They are
not.
The changes are pretty straight-forward:
- <foo> << (PAGE_CACHE_SHIFT - PAGE_SHIFT) -> <foo>;
- <foo> >> (PAGE_CACHE_SHIFT - PAGE_SHIFT) -> <foo>;
- PAGE_CACHE_{SIZE,SHIFT,MASK,ALIGN} -> PAGE_{SIZE,SHIFT,MASK,ALIGN};
- page_cache_get() -> get_page();
- page_cache_release() -> put_page();
This patch contains automated changes generated with coccinelle using
script below. For some reason, coccinelle doesn't patch header files.
I've called spatch for them manually.
The only adjustment after coccinelle is revert of changes to
PAGE_CAHCE_ALIGN definition: we are going to drop it later.
There are few places in the code where coccinelle didn't reach. I'll
fix them manually in a separate patch. Comments and documentation also
will be addressed with the separate patch.
virtual patch
@@
expression E;
@@
- E << (PAGE_CACHE_SHIFT - PAGE_SHIFT)
+ E
@@
expression E;
@@
- E >> (PAGE_CACHE_SHIFT - PAGE_SHIFT)
+ E
@@
@@
- PAGE_CACHE_SHIFT
+ PAGE_SHIFT
@@
@@
- PAGE_CACHE_SIZE
+ PAGE_SIZE
@@
@@
- PAGE_CACHE_MASK
+ PAGE_MASK
@@
expression E;
@@
- PAGE_CACHE_ALIGN(E)
+ PAGE_ALIGN(E)
@@
expression E;
@@
- page_cache_get(E)
+ get_page(E)
@@
expression E;
@@
- page_cache_release(E)
+ put_page(E)
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-04-01 20:29:47 +08:00
|
|
|
put_page(page);
|
2007-04-27 06:49:28 +08:00
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* get a page into the pagecache
|
|
|
|
*/
|
2007-04-27 06:57:07 +08:00
|
|
|
static struct page *afs_dir_get_page(struct inode *dir, unsigned long index,
|
|
|
|
struct key *key)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
|
|
|
struct page *page;
|
|
|
|
_enter("{%lu},%lu", dir->i_ino, index);
|
|
|
|
|
2010-05-21 22:27:09 +08:00
|
|
|
page = read_cache_page(dir->i_mapping, index, afs_page_filler, key);
|
2005-04-17 06:20:36 +08:00
|
|
|
if (!IS_ERR(page)) {
|
2017-11-02 23:27:52 +08:00
|
|
|
lock_page(page);
|
2005-04-17 06:20:36 +08:00
|
|
|
kmap(page);
|
2016-04-23 03:06:44 +08:00
|
|
|
if (unlikely(!PageChecked(page))) {
|
2017-11-02 23:27:52 +08:00
|
|
|
if (PageError(page))
|
2016-04-23 03:06:44 +08:00
|
|
|
goto fail;
|
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
return page;
|
|
|
|
|
2007-04-27 06:49:28 +08:00
|
|
|
fail:
|
2005-04-17 06:20:36 +08:00
|
|
|
afs_dir_put_page(page);
|
2007-04-27 06:55:03 +08:00
|
|
|
_leave(" = -EIO");
|
2005-04-17 06:20:36 +08:00
|
|
|
return ERR_PTR(-EIO);
|
2007-04-27 06:49:28 +08:00
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* open an AFS directory file
|
|
|
|
*/
|
|
|
|
static int afs_dir_open(struct inode *inode, struct file *file)
|
|
|
|
{
|
|
|
|
_enter("{%lu}", inode->i_ino);
|
|
|
|
|
2006-10-11 16:22:05 +08:00
|
|
|
BUILD_BUG_ON(sizeof(union afs_dir_block) != 2048);
|
|
|
|
BUILD_BUG_ON(sizeof(union afs_dirent) != 32);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2007-04-27 06:55:03 +08:00
|
|
|
if (test_bit(AFS_VNODE_DELETED, &AFS_FS_I(inode)->flags))
|
2005-04-17 06:20:36 +08:00
|
|
|
return -ENOENT;
|
|
|
|
|
2007-04-27 06:57:07 +08:00
|
|
|
return afs_open(inode, file);
|
2007-04-27 06:49:28 +08:00
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* deal with one block in an AFS directory
|
|
|
|
*/
|
2013-05-23 04:31:14 +08:00
|
|
|
static int afs_dir_iterate_block(struct dir_context *ctx,
|
2005-04-17 06:20:36 +08:00
|
|
|
union afs_dir_block *block,
|
2013-05-23 04:31:14 +08:00
|
|
|
unsigned blkoff)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
|
|
|
union afs_dirent *dire;
|
|
|
|
unsigned offset, next, curr;
|
|
|
|
size_t nlen;
|
2013-05-23 04:31:14 +08:00
|
|
|
int tmp;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2013-05-23 04:31:14 +08:00
|
|
|
_enter("%u,%x,%p,,",(unsigned)ctx->pos,blkoff,block);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2013-05-23 04:31:14 +08:00
|
|
|
curr = (ctx->pos - blkoff) / sizeof(union afs_dirent);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
/* walk through the block, an entry at a time */
|
|
|
|
for (offset = AFS_DIRENT_PER_BLOCK - block->pagehdr.nentries;
|
|
|
|
offset < AFS_DIRENT_PER_BLOCK;
|
|
|
|
offset = next
|
|
|
|
) {
|
|
|
|
next = offset + 1;
|
|
|
|
|
|
|
|
/* skip entries marked unused in the bitmap */
|
|
|
|
if (!(block->pagehdr.bitmap[offset / 8] &
|
|
|
|
(1 << (offset % 8)))) {
|
2017-02-28 06:30:02 +08:00
|
|
|
_debug("ENT[%zu.%u]: unused",
|
2005-04-17 06:20:36 +08:00
|
|
|
blkoff / sizeof(union afs_dir_block), offset);
|
|
|
|
if (offset >= curr)
|
2013-05-23 04:31:14 +08:00
|
|
|
ctx->pos = blkoff +
|
2005-04-17 06:20:36 +08:00
|
|
|
next * sizeof(union afs_dirent);
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* got a valid entry */
|
|
|
|
dire = &block->dirents[offset];
|
|
|
|
nlen = strnlen(dire->u.name,
|
|
|
|
sizeof(*block) -
|
|
|
|
offset * sizeof(union afs_dirent));
|
|
|
|
|
2017-02-28 06:30:02 +08:00
|
|
|
_debug("ENT[%zu.%u]: %s %zu \"%s\"",
|
2005-04-17 06:20:36 +08:00
|
|
|
blkoff / sizeof(union afs_dir_block), offset,
|
|
|
|
(offset < curr ? "skip" : "fill"),
|
|
|
|
nlen, dire->u.name);
|
|
|
|
|
|
|
|
/* work out where the next possible entry is */
|
|
|
|
for (tmp = nlen; tmp > 15; tmp -= sizeof(union afs_dirent)) {
|
|
|
|
if (next >= AFS_DIRENT_PER_BLOCK) {
|
2017-02-28 06:30:02 +08:00
|
|
|
_debug("ENT[%zu.%u]:"
|
2005-04-17 06:20:36 +08:00
|
|
|
" %u travelled beyond end dir block"
|
2017-02-28 06:30:02 +08:00
|
|
|
" (len %u/%zu)",
|
2005-04-17 06:20:36 +08:00
|
|
|
blkoff / sizeof(union afs_dir_block),
|
|
|
|
offset, next, tmp, nlen);
|
|
|
|
return -EIO;
|
|
|
|
}
|
|
|
|
if (!(block->pagehdr.bitmap[next / 8] &
|
|
|
|
(1 << (next % 8)))) {
|
2017-02-28 06:30:02 +08:00
|
|
|
_debug("ENT[%zu.%u]:"
|
|
|
|
" %u unmarked extension (len %u/%zu)",
|
2005-04-17 06:20:36 +08:00
|
|
|
blkoff / sizeof(union afs_dir_block),
|
|
|
|
offset, next, tmp, nlen);
|
|
|
|
return -EIO;
|
|
|
|
}
|
|
|
|
|
2017-02-28 06:30:02 +08:00
|
|
|
_debug("ENT[%zu.%u]: ext %u/%zu",
|
2005-04-17 06:20:36 +08:00
|
|
|
blkoff / sizeof(union afs_dir_block),
|
|
|
|
next, tmp, nlen);
|
|
|
|
next++;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* skip if starts before the current position */
|
|
|
|
if (offset < curr)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
/* found the next entry */
|
2013-05-23 04:31:14 +08:00
|
|
|
if (!dir_emit(ctx, dire->u.name, nlen,
|
2005-04-17 06:20:36 +08:00
|
|
|
ntohl(dire->u.vnode),
|
2018-04-10 04:12:31 +08:00
|
|
|
(ctx->actor == afs_lookup_filldir ||
|
|
|
|
ctx->actor == afs_lookup_one_filldir)?
|
2013-05-23 04:31:14 +08:00
|
|
|
ntohl(dire->u.unique) : DT_UNKNOWN)) {
|
2005-04-17 06:20:36 +08:00
|
|
|
_leave(" = 0 [full]");
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2013-05-23 04:31:14 +08:00
|
|
|
ctx->pos = blkoff + next * sizeof(union afs_dirent);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
_leave(" = 1 [more]");
|
|
|
|
return 1;
|
2007-04-27 06:49:28 +08:00
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
/*
|
2007-04-27 06:55:03 +08:00
|
|
|
* iterate through the data blob that lists the contents of an AFS directory
|
2005-04-17 06:20:36 +08:00
|
|
|
*/
|
2013-05-23 04:31:14 +08:00
|
|
|
static int afs_dir_iterate(struct inode *dir, struct dir_context *ctx,
|
|
|
|
struct key *key)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2007-04-27 06:55:03 +08:00
|
|
|
union afs_dir_block *dblock;
|
2005-04-17 06:20:36 +08:00
|
|
|
struct afs_dir_page *dbuf;
|
|
|
|
struct page *page;
|
|
|
|
unsigned blkoff, limit;
|
|
|
|
int ret;
|
|
|
|
|
2013-05-23 04:31:14 +08:00
|
|
|
_enter("{%lu},%u,,", dir->i_ino, (unsigned)ctx->pos);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2007-04-27 06:55:03 +08:00
|
|
|
if (test_bit(AFS_VNODE_DELETED, &AFS_FS_I(dir)->flags)) {
|
2005-04-17 06:20:36 +08:00
|
|
|
_leave(" = -ESTALE");
|
|
|
|
return -ESTALE;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* round the file position up to the next entry boundary */
|
2013-05-23 04:31:14 +08:00
|
|
|
ctx->pos += sizeof(union afs_dirent) - 1;
|
|
|
|
ctx->pos &= ~(sizeof(union afs_dirent) - 1);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
/* walk through the blocks in sequence */
|
|
|
|
ret = 0;
|
2013-05-23 04:31:14 +08:00
|
|
|
while (ctx->pos < dir->i_size) {
|
|
|
|
blkoff = ctx->pos & ~(sizeof(union afs_dir_block) - 1);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
/* fetch the appropriate page from the directory */
|
2007-04-27 06:57:07 +08:00
|
|
|
page = afs_dir_get_page(dir, blkoff / PAGE_SIZE, key);
|
2005-04-17 06:20:36 +08:00
|
|
|
if (IS_ERR(page)) {
|
|
|
|
ret = PTR_ERR(page);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
limit = blkoff & ~(PAGE_SIZE - 1);
|
|
|
|
|
|
|
|
dbuf = page_address(page);
|
|
|
|
|
|
|
|
/* deal with the individual blocks stashed on this page */
|
|
|
|
do {
|
|
|
|
dblock = &dbuf->blocks[(blkoff % PAGE_SIZE) /
|
|
|
|
sizeof(union afs_dir_block)];
|
2013-05-23 04:31:14 +08:00
|
|
|
ret = afs_dir_iterate_block(ctx, dblock, blkoff);
|
2005-04-17 06:20:36 +08:00
|
|
|
if (ret != 1) {
|
|
|
|
afs_dir_put_page(page);
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
blkoff += sizeof(union afs_dir_block);
|
|
|
|
|
2013-05-23 04:31:14 +08:00
|
|
|
} while (ctx->pos < dir->i_size && blkoff < limit);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
afs_dir_put_page(page);
|
|
|
|
ret = 0;
|
|
|
|
}
|
|
|
|
|
2007-04-27 06:49:28 +08:00
|
|
|
out:
|
2005-04-17 06:20:36 +08:00
|
|
|
_leave(" = %d", ret);
|
|
|
|
return ret;
|
2007-04-27 06:49:28 +08:00
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* read an AFS directory
|
|
|
|
*/
|
2013-05-23 04:31:14 +08:00
|
|
|
static int afs_readdir(struct file *file, struct dir_context *ctx)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2017-11-02 23:27:52 +08:00
|
|
|
return afs_dir_iterate(file_inode(file), ctx, afs_file_key(file));
|
2007-04-27 06:49:28 +08:00
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
/*
|
2018-04-10 04:12:31 +08:00
|
|
|
* Search the directory for a single name
|
2005-04-17 06:20:36 +08:00
|
|
|
* - if afs_dir_iterate_block() spots this function, it'll pass the FID
|
|
|
|
* uniquifier through dtype
|
|
|
|
*/
|
2018-04-10 04:12:31 +08:00
|
|
|
static int afs_lookup_one_filldir(struct dir_context *ctx, const char *name,
|
|
|
|
int nlen, loff_t fpos, u64 ino, unsigned dtype)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2018-04-10 04:12:31 +08:00
|
|
|
struct afs_lookup_one_cookie *cookie =
|
|
|
|
container_of(ctx, struct afs_lookup_one_cookie, ctx);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2013-05-23 04:31:14 +08:00
|
|
|
_enter("{%s,%u},%s,%u,,%llu,%u",
|
|
|
|
cookie->name.name, cookie->name.len, name, nlen,
|
2007-04-27 07:06:22 +08:00
|
|
|
(unsigned long long) ino, dtype);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2007-04-27 06:55:03 +08:00
|
|
|
/* insanity checks first */
|
|
|
|
BUILD_BUG_ON(sizeof(union afs_dir_block) != 2048);
|
|
|
|
BUILD_BUG_ON(sizeof(union afs_dirent) != 32);
|
|
|
|
|
2013-05-23 04:31:14 +08:00
|
|
|
if (cookie->name.len != nlen ||
|
|
|
|
memcmp(cookie->name.name, name, nlen) != 0) {
|
2005-04-17 06:20:36 +08:00
|
|
|
_leave(" = 0 [no]");
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
cookie->fid.vnode = ino;
|
|
|
|
cookie->fid.unique = dtype;
|
|
|
|
cookie->found = 1;
|
|
|
|
|
|
|
|
_leave(" = -1 [found]");
|
|
|
|
return -1;
|
2007-04-27 06:49:28 +08:00
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
/*
|
2018-04-10 04:12:31 +08:00
|
|
|
* Do a lookup of a single name in a directory
|
2007-04-27 06:59:35 +08:00
|
|
|
* - just returns the FID the dentry name maps to if found
|
2005-04-17 06:20:36 +08:00
|
|
|
*/
|
2018-04-10 04:12:31 +08:00
|
|
|
static int afs_do_lookup_one(struct inode *dir, struct dentry *dentry,
|
|
|
|
struct afs_fid *fid, struct key *key)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2013-05-23 04:31:14 +08:00
|
|
|
struct afs_super_info *as = dir->i_sb->s_fs_info;
|
2018-04-10 04:12:31 +08:00
|
|
|
struct afs_lookup_one_cookie cookie = {
|
|
|
|
.ctx.actor = afs_lookup_one_filldir,
|
2013-05-23 04:31:14 +08:00
|
|
|
.name = dentry->d_name,
|
|
|
|
.fid.vid = as->volume->vid
|
|
|
|
};
|
2005-04-17 06:20:36 +08:00
|
|
|
int ret;
|
|
|
|
|
2014-10-22 08:11:25 +08:00
|
|
|
_enter("{%lu},%p{%pd},", dir->i_ino, dentry, dentry);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
/* search the directory */
|
2013-05-23 04:31:14 +08:00
|
|
|
ret = afs_dir_iterate(dir, &cookie.ctx, key);
|
2005-04-17 06:20:36 +08:00
|
|
|
if (ret < 0) {
|
2007-04-27 06:55:03 +08:00
|
|
|
_leave(" = %d [iter]", ret);
|
|
|
|
return ret;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
ret = -ENOENT;
|
|
|
|
if (!cookie.found) {
|
2007-04-27 06:55:03 +08:00
|
|
|
_leave(" = -ENOENT [not found]");
|
|
|
|
return -ENOENT;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
2007-04-27 06:55:03 +08:00
|
|
|
*fid = cookie.fid;
|
|
|
|
_leave(" = 0 { vn=%u u=%u }", fid->vnode, fid->unique);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2018-04-10 04:12:31 +08:00
|
|
|
/*
|
|
|
|
* search the directory for a name
|
|
|
|
* - if afs_dir_iterate_block() spots this function, it'll pass the FID
|
|
|
|
* uniquifier through dtype
|
|
|
|
*/
|
|
|
|
static int afs_lookup_filldir(struct dir_context *ctx, const char *name,
|
|
|
|
int nlen, loff_t fpos, u64 ino, unsigned dtype)
|
|
|
|
{
|
|
|
|
struct afs_lookup_cookie *cookie =
|
|
|
|
container_of(ctx, struct afs_lookup_cookie, ctx);
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
_enter("{%s,%u},%s,%u,,%llu,%u",
|
|
|
|
cookie->name.name, cookie->name.len, name, nlen,
|
|
|
|
(unsigned long long) ino, dtype);
|
|
|
|
|
|
|
|
/* insanity checks first */
|
|
|
|
BUILD_BUG_ON(sizeof(union afs_dir_block) != 2048);
|
|
|
|
BUILD_BUG_ON(sizeof(union afs_dirent) != 32);
|
|
|
|
|
|
|
|
if (cookie->found) {
|
|
|
|
if (cookie->nr_fids < 50) {
|
|
|
|
cookie->fids[cookie->nr_fids].vnode = ino;
|
|
|
|
cookie->fids[cookie->nr_fids].unique = dtype;
|
|
|
|
cookie->nr_fids++;
|
|
|
|
}
|
|
|
|
} else if (cookie->name.len == nlen &&
|
|
|
|
memcmp(cookie->name.name, name, nlen) == 0) {
|
|
|
|
cookie->fids[0].vnode = ino;
|
|
|
|
cookie->fids[0].unique = dtype;
|
|
|
|
cookie->found = 1;
|
|
|
|
if (cookie->one_only)
|
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
|
|
|
|
ret = cookie->nr_fids >= 50 ? -1 : 0;
|
|
|
|
_leave(" = %d", ret);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Do a lookup in a directory. We make use of bulk lookup to query a slew of
|
|
|
|
* files in one go and create inodes for them. The inode of the file we were
|
|
|
|
* asked for is returned.
|
|
|
|
*/
|
|
|
|
static struct inode *afs_do_lookup(struct inode *dir, struct dentry *dentry,
|
|
|
|
struct key *key)
|
|
|
|
{
|
|
|
|
struct afs_lookup_cookie *cookie;
|
|
|
|
struct afs_cb_interest *cbi = NULL;
|
|
|
|
struct afs_super_info *as = dir->i_sb->s_fs_info;
|
|
|
|
struct afs_iget_data data;
|
|
|
|
struct afs_fs_cursor fc;
|
|
|
|
struct afs_vnode *dvnode = AFS_FS_I(dir);
|
|
|
|
struct inode *inode = NULL;
|
|
|
|
int ret, i;
|
|
|
|
|
|
|
|
_enter("{%lu},%p{%pd},", dir->i_ino, dentry, dentry);
|
|
|
|
|
|
|
|
cookie = kzalloc(sizeof(struct afs_lookup_cookie), GFP_KERNEL);
|
|
|
|
if (!cookie)
|
|
|
|
return ERR_PTR(-ENOMEM);
|
|
|
|
|
|
|
|
cookie->ctx.actor = afs_lookup_filldir;
|
|
|
|
cookie->name = dentry->d_name;
|
|
|
|
cookie->nr_fids = 1; /* slot 0 is saved for the fid we actually want */
|
|
|
|
|
|
|
|
read_seqlock_excl(&dvnode->cb_lock);
|
|
|
|
if (dvnode->cb_interest &&
|
|
|
|
dvnode->cb_interest->server &&
|
|
|
|
test_bit(AFS_SERVER_FL_NO_IBULK, &dvnode->cb_interest->server->flags))
|
|
|
|
cookie->one_only = true;
|
|
|
|
read_sequnlock_excl(&dvnode->cb_lock);
|
|
|
|
|
|
|
|
for (i = 0; i < 50; i++)
|
|
|
|
cookie->fids[i].vid = as->volume->vid;
|
|
|
|
|
|
|
|
/* search the directory */
|
|
|
|
ret = afs_dir_iterate(dir, &cookie->ctx, key);
|
|
|
|
if (ret < 0) {
|
|
|
|
inode = ERR_PTR(ret);
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
inode = ERR_PTR(-ENOENT);
|
|
|
|
if (!cookie->found)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
/* Check to see if we already have an inode for the primary fid. */
|
|
|
|
data.volume = dvnode->volume;
|
|
|
|
data.fid = cookie->fids[0];
|
|
|
|
inode = ilookup5(dir->i_sb, cookie->fids[0].vnode, afs_iget5_test, &data);
|
|
|
|
if (inode)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
/* Need space for examining all the selected files */
|
|
|
|
inode = ERR_PTR(-ENOMEM);
|
|
|
|
cookie->statuses = kcalloc(cookie->nr_fids, sizeof(struct afs_file_status),
|
|
|
|
GFP_KERNEL);
|
|
|
|
if (!cookie->statuses)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
cookie->callbacks = kcalloc(cookie->nr_fids, sizeof(struct afs_callback),
|
|
|
|
GFP_KERNEL);
|
|
|
|
if (!cookie->callbacks)
|
|
|
|
goto out_s;
|
|
|
|
|
|
|
|
/* Try FS.InlineBulkStatus first. Abort codes for the individual
|
|
|
|
* lookups contained therein are stored in the reply without aborting
|
|
|
|
* the whole operation.
|
|
|
|
*/
|
|
|
|
if (cookie->one_only)
|
|
|
|
goto no_inline_bulk_status;
|
|
|
|
|
|
|
|
inode = ERR_PTR(-ERESTARTSYS);
|
|
|
|
if (afs_begin_vnode_operation(&fc, dvnode, key)) {
|
|
|
|
while (afs_select_fileserver(&fc)) {
|
|
|
|
if (test_bit(AFS_SERVER_FL_NO_IBULK,
|
|
|
|
&fc.cbi->server->flags)) {
|
|
|
|
fc.ac.abort_code = RX_INVALID_OPERATION;
|
|
|
|
fc.ac.error = -ECONNABORTED;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
afs_fs_inline_bulk_status(&fc,
|
|
|
|
afs_v2net(dvnode),
|
|
|
|
cookie->fids,
|
|
|
|
cookie->statuses,
|
|
|
|
cookie->callbacks,
|
|
|
|
cookie->nr_fids, NULL);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (fc.ac.error == 0)
|
|
|
|
cbi = afs_get_cb_interest(fc.cbi);
|
|
|
|
if (fc.ac.abort_code == RX_INVALID_OPERATION)
|
|
|
|
set_bit(AFS_SERVER_FL_NO_IBULK, &fc.cbi->server->flags);
|
|
|
|
inode = ERR_PTR(afs_end_vnode_operation(&fc));
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!IS_ERR(inode))
|
|
|
|
goto success;
|
|
|
|
if (fc.ac.abort_code != RX_INVALID_OPERATION)
|
|
|
|
goto out_c;
|
|
|
|
|
|
|
|
no_inline_bulk_status:
|
|
|
|
/* We could try FS.BulkStatus next, but this aborts the entire op if
|
|
|
|
* any of the lookups fails - so, for the moment, revert to
|
|
|
|
* FS.FetchStatus for just the primary fid.
|
|
|
|
*/
|
|
|
|
cookie->nr_fids = 1;
|
|
|
|
inode = ERR_PTR(-ERESTARTSYS);
|
|
|
|
if (afs_begin_vnode_operation(&fc, dvnode, key)) {
|
|
|
|
while (afs_select_fileserver(&fc)) {
|
|
|
|
afs_fs_fetch_status(&fc,
|
|
|
|
afs_v2net(dvnode),
|
|
|
|
cookie->fids,
|
|
|
|
cookie->statuses,
|
|
|
|
cookie->callbacks,
|
|
|
|
NULL);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (fc.ac.error == 0)
|
|
|
|
cbi = afs_get_cb_interest(fc.cbi);
|
|
|
|
inode = ERR_PTR(afs_end_vnode_operation(&fc));
|
|
|
|
}
|
|
|
|
|
|
|
|
if (IS_ERR(inode))
|
|
|
|
goto out_c;
|
|
|
|
|
|
|
|
for (i = 0; i < cookie->nr_fids; i++)
|
|
|
|
cookie->statuses[i].abort_code = 0;
|
|
|
|
|
|
|
|
success:
|
|
|
|
/* Turn all the files into inodes and save the first one - which is the
|
|
|
|
* one we actually want.
|
|
|
|
*/
|
|
|
|
if (cookie->statuses[0].abort_code != 0)
|
|
|
|
inode = ERR_PTR(afs_abort_to_error(cookie->statuses[0].abort_code));
|
|
|
|
|
|
|
|
for (i = 0; i < cookie->nr_fids; i++) {
|
|
|
|
struct inode *ti;
|
|
|
|
|
|
|
|
if (cookie->statuses[i].abort_code != 0)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
ti = afs_iget(dir->i_sb, key, &cookie->fids[i],
|
|
|
|
&cookie->statuses[i],
|
|
|
|
&cookie->callbacks[i],
|
|
|
|
cbi);
|
|
|
|
if (i == 0) {
|
|
|
|
inode = ti;
|
|
|
|
} else {
|
|
|
|
if (!IS_ERR(ti))
|
|
|
|
iput(ti);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
out_c:
|
|
|
|
afs_put_cb_interest(afs_v2net(dvnode), cbi);
|
|
|
|
kfree(cookie->callbacks);
|
|
|
|
out_s:
|
|
|
|
kfree(cookie->statuses);
|
|
|
|
out:
|
|
|
|
kfree(cookie);
|
|
|
|
return inode;
|
|
|
|
}
|
|
|
|
|
2018-02-06 14:26:30 +08:00
|
|
|
/*
|
|
|
|
* Probe to see if a cell may exist. This prevents positive dentries from
|
|
|
|
* being created unnecessarily.
|
|
|
|
*/
|
|
|
|
static int afs_probe_cell_name(struct dentry *dentry)
|
|
|
|
{
|
|
|
|
struct afs_cell *cell;
|
|
|
|
const char *name = dentry->d_name.name;
|
|
|
|
size_t len = dentry->d_name.len;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
/* Names prefixed with a dot are R/W mounts. */
|
|
|
|
if (name[0] == '.') {
|
|
|
|
if (len == 1)
|
|
|
|
return -EINVAL;
|
|
|
|
name++;
|
|
|
|
len--;
|
|
|
|
}
|
|
|
|
|
|
|
|
cell = afs_lookup_cell_rcu(afs_d2net(dentry), name, len);
|
|
|
|
if (!IS_ERR(cell)) {
|
|
|
|
afs_put_cell(afs_d2net(dentry), cell);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
ret = dns_query("afsdb", name, len, "ipv4", NULL, NULL);
|
|
|
|
if (ret == -ENODATA)
|
|
|
|
ret = -EDESTADDRREQ;
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2010-08-11 16:38:04 +08:00
|
|
|
/*
|
|
|
|
* Try to auto mount the mountpoint with pseudo directory, if the autocell
|
|
|
|
* operation is setted.
|
|
|
|
*/
|
2018-04-10 04:12:31 +08:00
|
|
|
static struct inode *afs_try_auto_mntpt(struct dentry *dentry, struct inode *dir)
|
2010-08-11 16:38:04 +08:00
|
|
|
{
|
|
|
|
struct afs_vnode *vnode = AFS_FS_I(dir);
|
|
|
|
struct inode *inode;
|
2018-02-06 14:26:30 +08:00
|
|
|
int ret = -ENOENT;
|
2010-08-11 16:38:04 +08:00
|
|
|
|
2018-02-06 14:26:30 +08:00
|
|
|
_enter("%p{%pd}, {%x:%u}",
|
|
|
|
dentry, dentry, vnode->fid.vid, vnode->fid.vnode);
|
|
|
|
|
|
|
|
if (!test_bit(AFS_VNODE_AUTOCELL, &vnode->flags))
|
|
|
|
goto out;
|
2010-08-11 16:38:04 +08:00
|
|
|
|
2018-02-06 14:26:30 +08:00
|
|
|
ret = afs_probe_cell_name(dentry);
|
|
|
|
if (ret < 0)
|
2010-08-11 16:38:04 +08:00
|
|
|
goto out;
|
|
|
|
|
2018-02-06 14:26:30 +08:00
|
|
|
inode = afs_iget_pseudo_dir(dir->i_sb, false);
|
2010-08-11 16:38:04 +08:00
|
|
|
if (IS_ERR(inode)) {
|
|
|
|
ret = PTR_ERR(inode);
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
_leave("= %p", inode);
|
|
|
|
return inode;
|
|
|
|
|
|
|
|
out:
|
|
|
|
_leave("= %d", ret);
|
|
|
|
return ERR_PTR(ret);
|
|
|
|
}
|
|
|
|
|
2018-04-10 04:12:31 +08:00
|
|
|
/*
|
|
|
|
* Look up an entry in a directory with @sys substitution.
|
|
|
|
*/
|
|
|
|
static struct dentry *afs_lookup_atsys(struct inode *dir, struct dentry *dentry,
|
|
|
|
struct key *key)
|
|
|
|
{
|
|
|
|
struct afs_sysnames *subs;
|
|
|
|
struct afs_net *net = afs_i2net(dir);
|
|
|
|
struct dentry *ret;
|
|
|
|
char *buf, *p, *name;
|
|
|
|
int len, i;
|
|
|
|
|
|
|
|
_enter("");
|
|
|
|
|
|
|
|
ret = ERR_PTR(-ENOMEM);
|
|
|
|
p = buf = kmalloc(AFSNAMEMAX, GFP_KERNEL);
|
|
|
|
if (!buf)
|
|
|
|
goto out_p;
|
|
|
|
if (dentry->d_name.len > 4) {
|
|
|
|
memcpy(p, dentry->d_name.name, dentry->d_name.len - 4);
|
|
|
|
p += dentry->d_name.len - 4;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* There is an ordered list of substitutes that we have to try. */
|
|
|
|
read_lock(&net->sysnames_lock);
|
|
|
|
subs = net->sysnames;
|
|
|
|
refcount_inc(&subs->usage);
|
|
|
|
read_unlock(&net->sysnames_lock);
|
|
|
|
|
|
|
|
for (i = 0; i < subs->nr; i++) {
|
|
|
|
name = subs->subs[i];
|
|
|
|
len = dentry->d_name.len - 4 + strlen(name);
|
|
|
|
if (len >= AFSNAMEMAX) {
|
|
|
|
ret = ERR_PTR(-ENAMETOOLONG);
|
|
|
|
goto out_s;
|
|
|
|
}
|
|
|
|
|
|
|
|
strcpy(p, name);
|
|
|
|
ret = lookup_one_len(buf, dentry->d_parent, len);
|
|
|
|
if (IS_ERR(ret) || d_is_positive(ret))
|
|
|
|
goto out_s;
|
|
|
|
dput(ret);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* We don't want to d_add() the @sys dentry here as we don't want to
|
|
|
|
* the cached dentry to hide changes to the sysnames list.
|
|
|
|
*/
|
|
|
|
ret = NULL;
|
|
|
|
out_s:
|
|
|
|
afs_put_sysnames(subs);
|
|
|
|
kfree(buf);
|
|
|
|
out_p:
|
|
|
|
key_put(key);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2007-04-27 06:55:03 +08:00
|
|
|
/*
|
|
|
|
* look up an entry in a directory
|
|
|
|
*/
|
2007-04-27 06:59:35 +08:00
|
|
|
static struct dentry *afs_lookup(struct inode *dir, struct dentry *dentry,
|
2012-06-11 05:13:09 +08:00
|
|
|
unsigned int flags)
|
2007-04-27 06:55:03 +08:00
|
|
|
{
|
2018-04-10 04:12:31 +08:00
|
|
|
struct afs_vnode *dvnode = AFS_FS_I(dir);
|
2007-04-27 06:55:03 +08:00
|
|
|
struct inode *inode;
|
2007-04-27 06:57:07 +08:00
|
|
|
struct key *key;
|
2007-04-27 06:55:03 +08:00
|
|
|
int ret;
|
|
|
|
|
2014-10-22 08:11:25 +08:00
|
|
|
_enter("{%x:%u},%p{%pd},",
|
2018-04-10 04:12:31 +08:00
|
|
|
dvnode->fid.vid, dvnode->fid.vnode, dentry, dentry);
|
2007-04-27 06:59:35 +08:00
|
|
|
|
2015-03-18 06:25:59 +08:00
|
|
|
ASSERTCMP(d_inode(dentry), ==, NULL);
|
2007-04-27 06:55:03 +08:00
|
|
|
|
2007-05-11 13:22:20 +08:00
|
|
|
if (dentry->d_name.len >= AFSNAMEMAX) {
|
2007-04-27 06:55:03 +08:00
|
|
|
_leave(" = -ENAMETOOLONG");
|
|
|
|
return ERR_PTR(-ENAMETOOLONG);
|
|
|
|
}
|
|
|
|
|
2018-04-10 04:12:31 +08:00
|
|
|
if (test_bit(AFS_VNODE_DELETED, &dvnode->flags)) {
|
2007-04-27 06:55:03 +08:00
|
|
|
_leave(" = -ESTALE");
|
|
|
|
return ERR_PTR(-ESTALE);
|
|
|
|
}
|
|
|
|
|
2018-04-10 04:12:31 +08:00
|
|
|
key = afs_request_key(dvnode->volume->cell);
|
2007-04-27 06:57:07 +08:00
|
|
|
if (IS_ERR(key)) {
|
|
|
|
_leave(" = %ld [key]", PTR_ERR(key));
|
2008-02-07 16:15:26 +08:00
|
|
|
return ERR_CAST(key);
|
2007-04-27 06:57:07 +08:00
|
|
|
}
|
|
|
|
|
2018-04-10 04:12:31 +08:00
|
|
|
ret = afs_validate(dvnode, key);
|
2007-04-27 06:59:35 +08:00
|
|
|
if (ret < 0) {
|
|
|
|
key_put(key);
|
|
|
|
_leave(" = %d [val]", ret);
|
|
|
|
return ERR_PTR(ret);
|
|
|
|
}
|
|
|
|
|
2018-04-10 04:12:31 +08:00
|
|
|
if (dentry->d_name.len >= 4 &&
|
|
|
|
dentry->d_name.name[dentry->d_name.len - 4] == '@' &&
|
|
|
|
dentry->d_name.name[dentry->d_name.len - 3] == 's' &&
|
|
|
|
dentry->d_name.name[dentry->d_name.len - 2] == 'y' &&
|
|
|
|
dentry->d_name.name[dentry->d_name.len - 1] == 's')
|
|
|
|
return afs_lookup_atsys(dir, dentry, key);
|
|
|
|
|
2018-04-10 04:12:31 +08:00
|
|
|
inode = afs_do_lookup(dir, dentry, key);
|
|
|
|
if (IS_ERR(inode)) {
|
|
|
|
ret = PTR_ERR(inode);
|
2018-02-06 14:26:30 +08:00
|
|
|
if (ret == -ENOENT) {
|
2018-04-10 04:12:31 +08:00
|
|
|
inode = afs_try_auto_mntpt(dentry, dir);
|
2018-02-06 14:26:30 +08:00
|
|
|
if (!IS_ERR(inode)) {
|
|
|
|
key_put(key);
|
|
|
|
goto success;
|
|
|
|
}
|
|
|
|
|
|
|
|
ret = PTR_ERR(inode);
|
2010-08-11 16:38:04 +08:00
|
|
|
}
|
|
|
|
|
2007-04-27 06:57:07 +08:00
|
|
|
key_put(key);
|
2007-04-27 06:59:35 +08:00
|
|
|
if (ret == -ENOENT) {
|
|
|
|
d_add(dentry, NULL);
|
|
|
|
_leave(" = NULL [negative]");
|
|
|
|
return NULL;
|
|
|
|
}
|
2007-04-27 06:55:03 +08:00
|
|
|
_leave(" = %d [do]", ret);
|
2005-04-17 06:20:36 +08:00
|
|
|
return ERR_PTR(ret);
|
|
|
|
}
|
2018-04-10 04:12:31 +08:00
|
|
|
dentry->d_fsdata = (void *)(unsigned long)dvnode->status.data_version;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2007-04-27 06:55:03 +08:00
|
|
|
/* instantiate the dentry */
|
2007-04-27 06:57:07 +08:00
|
|
|
key_put(key);
|
2007-04-27 06:55:03 +08:00
|
|
|
if (IS_ERR(inode)) {
|
|
|
|
_leave(" = %ld", PTR_ERR(inode));
|
2008-02-07 16:15:26 +08:00
|
|
|
return ERR_CAST(inode);
|
2007-04-27 06:55:03 +08:00
|
|
|
}
|
|
|
|
|
2010-08-11 16:38:04 +08:00
|
|
|
success:
|
2005-04-17 06:20:36 +08:00
|
|
|
d_add(dentry, inode);
|
2018-04-10 04:12:31 +08:00
|
|
|
_leave(" = 0 { ino=%lu v=%u }",
|
2015-03-18 06:25:59 +08:00
|
|
|
d_inode(dentry)->i_ino,
|
|
|
|
d_inode(dentry)->i_generation);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
return NULL;
|
2007-04-27 06:49:28 +08:00
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2018-04-06 21:17:23 +08:00
|
|
|
/*
|
|
|
|
* Look up @cell in a dynroot directory. This is a substitution for the
|
|
|
|
* local cell name for the net namespace.
|
|
|
|
*/
|
|
|
|
static struct dentry *afs_lookup_atcell(struct dentry *dentry)
|
|
|
|
{
|
|
|
|
struct afs_cell *cell;
|
|
|
|
struct afs_net *net = afs_d2net(dentry);
|
|
|
|
struct dentry *ret;
|
|
|
|
unsigned int seq = 0;
|
|
|
|
char *name;
|
|
|
|
int len;
|
|
|
|
|
|
|
|
if (!net->ws_cell)
|
|
|
|
return ERR_PTR(-ENOENT);
|
|
|
|
|
|
|
|
ret = ERR_PTR(-ENOMEM);
|
|
|
|
name = kmalloc(AFS_MAXCELLNAME + 1, GFP_KERNEL);
|
|
|
|
if (!name)
|
|
|
|
goto out_p;
|
|
|
|
|
|
|
|
rcu_read_lock();
|
|
|
|
do {
|
|
|
|
read_seqbegin_or_lock(&net->cells_lock, &seq);
|
|
|
|
cell = rcu_dereference_raw(net->ws_cell);
|
|
|
|
if (cell) {
|
|
|
|
len = cell->name_len;
|
|
|
|
memcpy(name, cell->name, len + 1);
|
|
|
|
}
|
|
|
|
} while (need_seqretry(&net->cells_lock, seq));
|
|
|
|
done_seqretry(&net->cells_lock, seq);
|
|
|
|
rcu_read_unlock();
|
|
|
|
|
|
|
|
ret = ERR_PTR(-ENOENT);
|
|
|
|
if (!cell)
|
|
|
|
goto out_n;
|
|
|
|
|
|
|
|
ret = lookup_one_len(name, dentry->d_parent, len);
|
|
|
|
|
|
|
|
/* We don't want to d_add() the @cell dentry here as we don't want to
|
|
|
|
* the cached dentry to hide changes to the local cell name.
|
|
|
|
*/
|
|
|
|
|
|
|
|
out_n:
|
|
|
|
kfree(name);
|
|
|
|
out_p:
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2018-02-06 14:26:30 +08:00
|
|
|
/*
|
|
|
|
* Look up an entry in a dynroot directory.
|
|
|
|
*/
|
|
|
|
static struct dentry *afs_dynroot_lookup(struct inode *dir, struct dentry *dentry,
|
|
|
|
unsigned int flags)
|
|
|
|
{
|
|
|
|
struct afs_vnode *vnode;
|
|
|
|
struct inode *inode;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
vnode = AFS_FS_I(dir);
|
|
|
|
|
|
|
|
_enter("%pd", dentry);
|
|
|
|
|
|
|
|
ASSERTCMP(d_inode(dentry), ==, NULL);
|
|
|
|
|
|
|
|
if (dentry->d_name.len >= AFSNAMEMAX) {
|
|
|
|
_leave(" = -ENAMETOOLONG");
|
|
|
|
return ERR_PTR(-ENAMETOOLONG);
|
|
|
|
}
|
|
|
|
|
2018-04-06 21:17:23 +08:00
|
|
|
if (dentry->d_name.len == 5 &&
|
|
|
|
memcmp(dentry->d_name.name, "@cell", 5) == 0)
|
|
|
|
return afs_lookup_atcell(dentry);
|
|
|
|
|
2018-04-10 04:12:31 +08:00
|
|
|
inode = afs_try_auto_mntpt(dentry, dir);
|
2018-02-06 14:26:30 +08:00
|
|
|
if (IS_ERR(inode)) {
|
|
|
|
ret = PTR_ERR(inode);
|
|
|
|
if (ret == -ENOENT) {
|
|
|
|
d_add(dentry, NULL);
|
|
|
|
_leave(" = NULL [negative]");
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
_leave(" = %d [do]", ret);
|
|
|
|
return ERR_PTR(ret);
|
|
|
|
}
|
|
|
|
|
|
|
|
d_add(dentry, inode);
|
|
|
|
_leave(" = 0 { ino=%lu v=%u }",
|
|
|
|
d_inode(dentry)->i_ino, d_inode(dentry)->i_generation);
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
/*
|
|
|
|
* check that a dentry lookup hit has found a valid entry
|
|
|
|
* - NOTE! the hit can be a negative hit too, so we can't assume we have an
|
|
|
|
* inode
|
|
|
|
*/
|
2012-06-11 04:03:43 +08:00
|
|
|
static int afs_d_revalidate(struct dentry *dentry, unsigned int flags)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2018-02-06 14:26:30 +08:00
|
|
|
struct afs_super_info *as = dentry->d_sb->s_fs_info;
|
2007-04-27 06:59:35 +08:00
|
|
|
struct afs_vnode *vnode, *dir;
|
2009-07-09 17:44:30 +08:00
|
|
|
struct afs_fid uninitialized_var(fid);
|
2005-04-17 06:20:36 +08:00
|
|
|
struct dentry *parent;
|
2017-11-02 23:27:49 +08:00
|
|
|
struct inode *inode;
|
2007-04-27 06:57:07 +08:00
|
|
|
struct key *key;
|
2007-04-27 06:59:35 +08:00
|
|
|
void *dir_version;
|
2005-04-17 06:20:36 +08:00
|
|
|
int ret;
|
|
|
|
|
2012-06-11 04:03:43 +08:00
|
|
|
if (flags & LOOKUP_RCU)
|
2011-01-07 14:49:57 +08:00
|
|
|
return -ECHILD;
|
|
|
|
|
2018-02-06 14:26:30 +08:00
|
|
|
if (as->dyn_root)
|
|
|
|
return 1;
|
|
|
|
|
2017-11-02 23:27:49 +08:00
|
|
|
if (d_really_is_positive(dentry)) {
|
|
|
|
vnode = AFS_FS_I(d_inode(dentry));
|
2014-10-22 08:11:25 +08:00
|
|
|
_enter("{v={%x:%u} n=%pd fl=%lx},",
|
|
|
|
vnode->fid.vid, vnode->fid.vnode, dentry,
|
2007-04-27 06:59:35 +08:00
|
|
|
vnode->flags);
|
2017-11-02 23:27:49 +08:00
|
|
|
} else {
|
2014-10-22 08:11:25 +08:00
|
|
|
_enter("{neg n=%pd}", dentry);
|
2017-11-02 23:27:49 +08:00
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2007-04-27 06:59:35 +08:00
|
|
|
key = afs_request_key(AFS_FS_S(dentry->d_sb)->volume->cell);
|
2007-04-27 06:57:07 +08:00
|
|
|
if (IS_ERR(key))
|
|
|
|
key = NULL;
|
|
|
|
|
2017-11-02 23:27:49 +08:00
|
|
|
if (d_really_is_positive(dentry)) {
|
|
|
|
inode = d_inode(dentry);
|
|
|
|
if (inode) {
|
|
|
|
vnode = AFS_FS_I(inode);
|
|
|
|
afs_validate(vnode, key);
|
|
|
|
if (test_bit(AFS_VNODE_DELETED, &vnode->flags))
|
|
|
|
goto out_bad;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
/* lock down the parent dentry so we can peer at it */
|
2007-04-27 06:55:03 +08:00
|
|
|
parent = dget_parent(dentry);
|
2015-03-18 06:25:59 +08:00
|
|
|
dir = AFS_FS_I(d_inode(parent));
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2007-04-27 06:59:35 +08:00
|
|
|
/* validate the parent directory */
|
2017-11-02 23:27:49 +08:00
|
|
|
afs_validate(dir, key);
|
2007-04-27 06:59:35 +08:00
|
|
|
|
|
|
|
if (test_bit(AFS_VNODE_DELETED, &dir->flags)) {
|
2014-10-22 08:11:25 +08:00
|
|
|
_debug("%pd: parent dir deleted", dentry);
|
2017-11-02 23:27:49 +08:00
|
|
|
goto out_bad_parent;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
2007-04-27 06:59:35 +08:00
|
|
|
dir_version = (void *) (unsigned long) dir->status.data_version;
|
|
|
|
if (dentry->d_fsdata == dir_version)
|
|
|
|
goto out_valid; /* the dir contents are unchanged */
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2007-04-27 06:59:35 +08:00
|
|
|
_debug("dir modified");
|
|
|
|
|
|
|
|
/* search the directory for this vnode */
|
2018-04-10 04:12:31 +08:00
|
|
|
ret = afs_do_lookup_one(&dir->vfs_inode, dentry, &fid, key);
|
2007-04-27 06:59:35 +08:00
|
|
|
switch (ret) {
|
|
|
|
case 0:
|
|
|
|
/* the filename maps to something */
|
2015-03-18 06:25:59 +08:00
|
|
|
if (d_really_is_negative(dentry))
|
2017-11-02 23:27:49 +08:00
|
|
|
goto out_bad_parent;
|
|
|
|
inode = d_inode(dentry);
|
|
|
|
if (is_bad_inode(inode)) {
|
2014-10-22 08:11:25 +08:00
|
|
|
printk("kAFS: afs_d_revalidate: %pd2 has bad inode\n",
|
|
|
|
dentry);
|
2017-11-02 23:27:49 +08:00
|
|
|
goto out_bad_parent;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
2017-11-02 23:27:49 +08:00
|
|
|
vnode = AFS_FS_I(inode);
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
/* if the vnode ID has changed, then the dirent points to a
|
|
|
|
* different file */
|
2007-04-27 06:55:03 +08:00
|
|
|
if (fid.vnode != vnode->fid.vnode) {
|
2014-10-22 08:11:25 +08:00
|
|
|
_debug("%pd: dirent changed [%u != %u]",
|
|
|
|
dentry, fid.vnode,
|
2007-04-27 06:55:03 +08:00
|
|
|
vnode->fid.vnode);
|
2005-04-17 06:20:36 +08:00
|
|
|
goto not_found;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* if the vnode ID uniqifier has changed, then the file has
|
2007-04-27 06:59:35 +08:00
|
|
|
* been deleted and replaced, and the original vnode ID has
|
|
|
|
* been reused */
|
2007-04-27 06:55:03 +08:00
|
|
|
if (fid.unique != vnode->fid.unique) {
|
2014-10-22 08:11:25 +08:00
|
|
|
_debug("%pd: file deleted (uq %u -> %u I:%u)",
|
|
|
|
dentry, fid.unique,
|
2008-01-29 12:58:27 +08:00
|
|
|
vnode->fid.unique,
|
2017-11-02 23:27:49 +08:00
|
|
|
vnode->vfs_inode.i_generation);
|
|
|
|
write_seqlock(&vnode->cb_lock);
|
2007-04-27 06:55:03 +08:00
|
|
|
set_bit(AFS_VNODE_DELETED, &vnode->flags);
|
2017-11-02 23:27:49 +08:00
|
|
|
write_sequnlock(&vnode->cb_lock);
|
2007-04-27 06:59:35 +08:00
|
|
|
goto not_found;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
2007-04-27 06:59:35 +08:00
|
|
|
goto out_valid;
|
2007-04-27 06:55:03 +08:00
|
|
|
|
2007-04-27 06:59:35 +08:00
|
|
|
case -ENOENT:
|
|
|
|
/* the filename is unknown */
|
2014-10-22 08:11:25 +08:00
|
|
|
_debug("%pd: dirent not found", dentry);
|
2015-03-18 06:25:59 +08:00
|
|
|
if (d_really_is_positive(dentry))
|
2007-04-27 06:59:35 +08:00
|
|
|
goto not_found;
|
|
|
|
goto out_valid;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2007-04-27 06:59:35 +08:00
|
|
|
default:
|
2014-10-22 08:11:25 +08:00
|
|
|
_debug("failed to iterate dir %pd: %d",
|
|
|
|
parent, ret);
|
2017-11-02 23:27:49 +08:00
|
|
|
goto out_bad_parent;
|
2007-04-27 06:55:03 +08:00
|
|
|
}
|
|
|
|
|
2007-04-27 06:49:28 +08:00
|
|
|
out_valid:
|
2007-04-27 06:59:35 +08:00
|
|
|
dentry->d_fsdata = dir_version;
|
2005-04-17 06:20:36 +08:00
|
|
|
dput(parent);
|
2007-04-27 06:57:07 +08:00
|
|
|
key_put(key);
|
2005-04-17 06:20:36 +08:00
|
|
|
_leave(" = 1 [valid]");
|
|
|
|
return 1;
|
|
|
|
|
|
|
|
/* the dirent, if it exists, now points to a different vnode */
|
2007-04-27 06:49:28 +08:00
|
|
|
not_found:
|
2005-04-17 06:20:36 +08:00
|
|
|
spin_lock(&dentry->d_lock);
|
|
|
|
dentry->d_flags |= DCACHE_NFSFS_RENAMED;
|
|
|
|
spin_unlock(&dentry->d_lock);
|
|
|
|
|
2017-11-02 23:27:49 +08:00
|
|
|
out_bad_parent:
|
2014-10-22 08:11:25 +08:00
|
|
|
_debug("dropping dentry %pd2", dentry);
|
2005-04-17 06:20:36 +08:00
|
|
|
dput(parent);
|
2017-11-02 23:27:49 +08:00
|
|
|
out_bad:
|
2007-04-27 06:57:07 +08:00
|
|
|
key_put(key);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
_leave(" = 0 [bad]");
|
|
|
|
return 0;
|
2007-04-27 06:49:28 +08:00
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* allow the VFS to enquire as to whether a dentry should be unhashed (mustn't
|
|
|
|
* sleep)
|
|
|
|
* - called from dput() when d_count is going to 0.
|
|
|
|
* - return 1 to request dentry be unhashed, 0 otherwise
|
|
|
|
*/
|
2011-01-07 14:49:23 +08:00
|
|
|
static int afs_d_delete(const struct dentry *dentry)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2014-10-22 08:11:25 +08:00
|
|
|
_enter("%pd", dentry);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
|
|
|
|
goto zap;
|
|
|
|
|
2015-03-18 06:25:59 +08:00
|
|
|
if (d_really_is_positive(dentry) &&
|
|
|
|
(test_bit(AFS_VNODE_DELETED, &AFS_FS_I(d_inode(dentry))->flags) ||
|
|
|
|
test_bit(AFS_VNODE_PSEUDODIR, &AFS_FS_I(d_inode(dentry))->flags)))
|
2010-08-11 16:38:04 +08:00
|
|
|
goto zap;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
_leave(" = 0 [keep]");
|
|
|
|
return 0;
|
|
|
|
|
2007-04-27 06:49:28 +08:00
|
|
|
zap:
|
2005-04-17 06:20:36 +08:00
|
|
|
_leave(" = 1 [zap]");
|
|
|
|
return 1;
|
2007-04-27 06:49:28 +08:00
|
|
|
}
|
2007-04-27 06:59:35 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* handle dentry release
|
|
|
|
*/
|
|
|
|
static void afs_d_release(struct dentry *dentry)
|
|
|
|
{
|
2014-10-22 08:11:25 +08:00
|
|
|
_enter("%pd", dentry);
|
2007-04-27 06:59:35 +08:00
|
|
|
}
|
|
|
|
|
afs: Overhaul volume and server record caching and fileserver rotation
The current code assumes that volumes and servers are per-cell and are
never shared, but this is not enforced, and, indeed, public cells do exist
that are aliases of each other. Further, an organisation can, say, set up
a public cell and a private cell with overlapping, but not identical, sets
of servers. The difference is purely in the database attached to the VL
servers.
The current code will malfunction if it sees a server in two cells as it
assumes global address -> server record mappings and that each server is in
just one cell.
Further, each server may have multiple addresses - and may have addresses
of different families (IPv4 and IPv6, say).
To this end, the following structural changes are made:
(1) Server record management is overhauled:
(a) Server records are made independent of cell. The namespace keeps
track of them, volume records have lists of them and each vnode
has a server on which its callback interest currently resides.
(b) The cell record no longer keeps a list of servers known to be in
that cell.
(c) The server records are now kept in a flat list because there's no
single address to sort on.
(d) Server records are now keyed by their UUID within the namespace.
(e) The addresses for a server are obtained with the VL.GetAddrsU
rather than with VL.GetEntryByName, using the server's UUID as a
parameter.
(f) Cached server records are garbage collected after a period of
non-use and are counted out of existence before purging is allowed
to complete. This protects the work functions against rmmod.
(g) The servers list is now in /proc/fs/afs/servers.
(2) Volume record management is overhauled:
(a) An RCU-replaceable server list is introduced. This tracks both
servers and their coresponding callback interests.
(b) The superblock is now keyed on cell record and numeric volume ID.
(c) The volume record is now tied to the superblock which mounts it,
and is activated when mounted and deactivated when unmounted.
This makes it easier to handle the cache cookie without causing a
double-use in fscache.
(d) The volume record is loaded from the VLDB using VL.GetEntryByNameU
to get the server UUID list.
(e) The volume name is updated if it is seen to have changed when the
volume is updated (the update is keyed on the volume ID).
(3) The vlocation record is got rid of and VLDB records are no longer
cached. Sufficient information is stored in the volume record, though
an update to a volume record is now no longer shared between related
volumes (volumes come in bundles of three: R/W, R/O and backup).
and the following procedural changes are made:
(1) The fileserver cursor introduced previously is now fleshed out and
used to iterate over fileservers and their addresses.
(2) Volume status is checked during iteration, and the server list is
replaced if a change is detected.
(3) Server status is checked during iteration, and the address list is
replaced if a change is detected.
(4) The abort code is saved into the address list cursor and -ECONNABORTED
returned in afs_make_call() if a remote abort happened rather than
translating the abort into an error message. This allows actions to
be taken depending on the abort code more easily.
(a) If a VMOVED abort is seen then this is handled by rechecking the
volume and restarting the iteration.
(b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is
handled by sleeping for a short period and retrying and/or trying
other servers that might serve that volume. A message is also
displayed once until the condition has cleared.
(c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the
moment.
(d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to
see if it has been deleted; if not, the fileserver is probably
indicating that the volume couldn't be attached and needs
salvaging.
(e) If statfs() sees one of these aborts, it does not sleep, but
rather returns an error, so as not to block the umount program.
(5) The fileserver iteration functions in vnode.c are now merged into
their callers and more heavily macroised around the cursor. vnode.c
is removed.
(6) Operations on a particular vnode are serialised on that vnode because
the server will lock that vnode whilst it operates on it, so a second
op sent will just have to wait.
(7) Fileservers are probed with FS.GetCapabilities before being used.
This is where service upgrade will be done.
(8) A callback interest on a fileserver is set up before an FS operation
is performed and passed through to afs_make_call() so that it can be
set on the vnode if the operation returns a callback. The callback
interest is passed through to afs_iget() also so that it can be set
there too.
In general, record updating is done on an as-needed basis when we try to
access servers, volumes or vnodes rather than offloading it to work items
and special threads.
Notes:
(1) Pre AFS-3.4 servers are no longer supported, though this can be added
back if necessary (AFS-3.4 was released in 1998).
(2) VBUSY is retried forever for the moment at intervals of 1s.
(3) /proc/fs/afs/<cell>/servers no longer exists.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 23:27:50 +08:00
|
|
|
/*
|
|
|
|
* Create a new inode for create/mkdir/symlink
|
|
|
|
*/
|
|
|
|
static void afs_vnode_new_inode(struct afs_fs_cursor *fc,
|
|
|
|
struct dentry *new_dentry,
|
|
|
|
struct afs_fid *newfid,
|
|
|
|
struct afs_file_status *newstatus,
|
|
|
|
struct afs_callback *newcb)
|
|
|
|
{
|
|
|
|
struct inode *inode;
|
|
|
|
|
|
|
|
if (fc->ac.error < 0)
|
|
|
|
return;
|
|
|
|
|
2017-11-21 07:04:08 +08:00
|
|
|
d_drop(new_dentry);
|
|
|
|
|
afs: Overhaul volume and server record caching and fileserver rotation
The current code assumes that volumes and servers are per-cell and are
never shared, but this is not enforced, and, indeed, public cells do exist
that are aliases of each other. Further, an organisation can, say, set up
a public cell and a private cell with overlapping, but not identical, sets
of servers. The difference is purely in the database attached to the VL
servers.
The current code will malfunction if it sees a server in two cells as it
assumes global address -> server record mappings and that each server is in
just one cell.
Further, each server may have multiple addresses - and may have addresses
of different families (IPv4 and IPv6, say).
To this end, the following structural changes are made:
(1) Server record management is overhauled:
(a) Server records are made independent of cell. The namespace keeps
track of them, volume records have lists of them and each vnode
has a server on which its callback interest currently resides.
(b) The cell record no longer keeps a list of servers known to be in
that cell.
(c) The server records are now kept in a flat list because there's no
single address to sort on.
(d) Server records are now keyed by their UUID within the namespace.
(e) The addresses for a server are obtained with the VL.GetAddrsU
rather than with VL.GetEntryByName, using the server's UUID as a
parameter.
(f) Cached server records are garbage collected after a period of
non-use and are counted out of existence before purging is allowed
to complete. This protects the work functions against rmmod.
(g) The servers list is now in /proc/fs/afs/servers.
(2) Volume record management is overhauled:
(a) An RCU-replaceable server list is introduced. This tracks both
servers and their coresponding callback interests.
(b) The superblock is now keyed on cell record and numeric volume ID.
(c) The volume record is now tied to the superblock which mounts it,
and is activated when mounted and deactivated when unmounted.
This makes it easier to handle the cache cookie without causing a
double-use in fscache.
(d) The volume record is loaded from the VLDB using VL.GetEntryByNameU
to get the server UUID list.
(e) The volume name is updated if it is seen to have changed when the
volume is updated (the update is keyed on the volume ID).
(3) The vlocation record is got rid of and VLDB records are no longer
cached. Sufficient information is stored in the volume record, though
an update to a volume record is now no longer shared between related
volumes (volumes come in bundles of three: R/W, R/O and backup).
and the following procedural changes are made:
(1) The fileserver cursor introduced previously is now fleshed out and
used to iterate over fileservers and their addresses.
(2) Volume status is checked during iteration, and the server list is
replaced if a change is detected.
(3) Server status is checked during iteration, and the address list is
replaced if a change is detected.
(4) The abort code is saved into the address list cursor and -ECONNABORTED
returned in afs_make_call() if a remote abort happened rather than
translating the abort into an error message. This allows actions to
be taken depending on the abort code more easily.
(a) If a VMOVED abort is seen then this is handled by rechecking the
volume and restarting the iteration.
(b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is
handled by sleeping for a short period and retrying and/or trying
other servers that might serve that volume. A message is also
displayed once until the condition has cleared.
(c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the
moment.
(d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to
see if it has been deleted; if not, the fileserver is probably
indicating that the volume couldn't be attached and needs
salvaging.
(e) If statfs() sees one of these aborts, it does not sleep, but
rather returns an error, so as not to block the umount program.
(5) The fileserver iteration functions in vnode.c are now merged into
their callers and more heavily macroised around the cursor. vnode.c
is removed.
(6) Operations on a particular vnode are serialised on that vnode because
the server will lock that vnode whilst it operates on it, so a second
op sent will just have to wait.
(7) Fileservers are probed with FS.GetCapabilities before being used.
This is where service upgrade will be done.
(8) A callback interest on a fileserver is set up before an FS operation
is performed and passed through to afs_make_call() so that it can be
set on the vnode if the operation returns a callback. The callback
interest is passed through to afs_iget() also so that it can be set
there too.
In general, record updating is done on an as-needed basis when we try to
access servers, volumes or vnodes rather than offloading it to work items
and special threads.
Notes:
(1) Pre AFS-3.4 servers are no longer supported, though this can be added
back if necessary (AFS-3.4 was released in 1998).
(2) VBUSY is retried forever for the moment at intervals of 1s.
(3) /proc/fs/afs/<cell>/servers no longer exists.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 23:27:50 +08:00
|
|
|
inode = afs_iget(fc->vnode->vfs_inode.i_sb, fc->key,
|
|
|
|
newfid, newstatus, newcb, fc->cbi);
|
|
|
|
if (IS_ERR(inode)) {
|
|
|
|
/* ENOMEM or EINTR at a really inconvenient time - just abandon
|
|
|
|
* the new directory on the server.
|
|
|
|
*/
|
|
|
|
fc->ac.error = PTR_ERR(inode);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
2017-11-21 07:04:08 +08:00
|
|
|
d_add(new_dentry, inode);
|
afs: Overhaul volume and server record caching and fileserver rotation
The current code assumes that volumes and servers are per-cell and are
never shared, but this is not enforced, and, indeed, public cells do exist
that are aliases of each other. Further, an organisation can, say, set up
a public cell and a private cell with overlapping, but not identical, sets
of servers. The difference is purely in the database attached to the VL
servers.
The current code will malfunction if it sees a server in two cells as it
assumes global address -> server record mappings and that each server is in
just one cell.
Further, each server may have multiple addresses - and may have addresses
of different families (IPv4 and IPv6, say).
To this end, the following structural changes are made:
(1) Server record management is overhauled:
(a) Server records are made independent of cell. The namespace keeps
track of them, volume records have lists of them and each vnode
has a server on which its callback interest currently resides.
(b) The cell record no longer keeps a list of servers known to be in
that cell.
(c) The server records are now kept in a flat list because there's no
single address to sort on.
(d) Server records are now keyed by their UUID within the namespace.
(e) The addresses for a server are obtained with the VL.GetAddrsU
rather than with VL.GetEntryByName, using the server's UUID as a
parameter.
(f) Cached server records are garbage collected after a period of
non-use and are counted out of existence before purging is allowed
to complete. This protects the work functions against rmmod.
(g) The servers list is now in /proc/fs/afs/servers.
(2) Volume record management is overhauled:
(a) An RCU-replaceable server list is introduced. This tracks both
servers and their coresponding callback interests.
(b) The superblock is now keyed on cell record and numeric volume ID.
(c) The volume record is now tied to the superblock which mounts it,
and is activated when mounted and deactivated when unmounted.
This makes it easier to handle the cache cookie without causing a
double-use in fscache.
(d) The volume record is loaded from the VLDB using VL.GetEntryByNameU
to get the server UUID list.
(e) The volume name is updated if it is seen to have changed when the
volume is updated (the update is keyed on the volume ID).
(3) The vlocation record is got rid of and VLDB records are no longer
cached. Sufficient information is stored in the volume record, though
an update to a volume record is now no longer shared between related
volumes (volumes come in bundles of three: R/W, R/O and backup).
and the following procedural changes are made:
(1) The fileserver cursor introduced previously is now fleshed out and
used to iterate over fileservers and their addresses.
(2) Volume status is checked during iteration, and the server list is
replaced if a change is detected.
(3) Server status is checked during iteration, and the address list is
replaced if a change is detected.
(4) The abort code is saved into the address list cursor and -ECONNABORTED
returned in afs_make_call() if a remote abort happened rather than
translating the abort into an error message. This allows actions to
be taken depending on the abort code more easily.
(a) If a VMOVED abort is seen then this is handled by rechecking the
volume and restarting the iteration.
(b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is
handled by sleeping for a short period and retrying and/or trying
other servers that might serve that volume. A message is also
displayed once until the condition has cleared.
(c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the
moment.
(d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to
see if it has been deleted; if not, the fileserver is probably
indicating that the volume couldn't be attached and needs
salvaging.
(e) If statfs() sees one of these aborts, it does not sleep, but
rather returns an error, so as not to block the umount program.
(5) The fileserver iteration functions in vnode.c are now merged into
their callers and more heavily macroised around the cursor. vnode.c
is removed.
(6) Operations on a particular vnode are serialised on that vnode because
the server will lock that vnode whilst it operates on it, so a second
op sent will just have to wait.
(7) Fileservers are probed with FS.GetCapabilities before being used.
This is where service upgrade will be done.
(8) A callback interest on a fileserver is set up before an FS operation
is performed and passed through to afs_make_call() so that it can be
set on the vnode if the operation returns a callback. The callback
interest is passed through to afs_iget() also so that it can be set
there too.
In general, record updating is done on an as-needed basis when we try to
access servers, volumes or vnodes rather than offloading it to work items
and special threads.
Notes:
(1) Pre AFS-3.4 servers are no longer supported, though this can be added
back if necessary (AFS-3.4 was released in 1998).
(2) VBUSY is retried forever for the moment at intervals of 1s.
(3) /proc/fs/afs/<cell>/servers no longer exists.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 23:27:50 +08:00
|
|
|
}
|
|
|
|
|
2007-04-27 06:59:35 +08:00
|
|
|
/*
|
|
|
|
* create a directory on an AFS filesystem
|
|
|
|
*/
|
2011-07-26 13:41:39 +08:00
|
|
|
static int afs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
|
2007-04-27 06:59:35 +08:00
|
|
|
{
|
afs: Overhaul volume and server record caching and fileserver rotation
The current code assumes that volumes and servers are per-cell and are
never shared, but this is not enforced, and, indeed, public cells do exist
that are aliases of each other. Further, an organisation can, say, set up
a public cell and a private cell with overlapping, but not identical, sets
of servers. The difference is purely in the database attached to the VL
servers.
The current code will malfunction if it sees a server in two cells as it
assumes global address -> server record mappings and that each server is in
just one cell.
Further, each server may have multiple addresses - and may have addresses
of different families (IPv4 and IPv6, say).
To this end, the following structural changes are made:
(1) Server record management is overhauled:
(a) Server records are made independent of cell. The namespace keeps
track of them, volume records have lists of them and each vnode
has a server on which its callback interest currently resides.
(b) The cell record no longer keeps a list of servers known to be in
that cell.
(c) The server records are now kept in a flat list because there's no
single address to sort on.
(d) Server records are now keyed by their UUID within the namespace.
(e) The addresses for a server are obtained with the VL.GetAddrsU
rather than with VL.GetEntryByName, using the server's UUID as a
parameter.
(f) Cached server records are garbage collected after a period of
non-use and are counted out of existence before purging is allowed
to complete. This protects the work functions against rmmod.
(g) The servers list is now in /proc/fs/afs/servers.
(2) Volume record management is overhauled:
(a) An RCU-replaceable server list is introduced. This tracks both
servers and their coresponding callback interests.
(b) The superblock is now keyed on cell record and numeric volume ID.
(c) The volume record is now tied to the superblock which mounts it,
and is activated when mounted and deactivated when unmounted.
This makes it easier to handle the cache cookie without causing a
double-use in fscache.
(d) The volume record is loaded from the VLDB using VL.GetEntryByNameU
to get the server UUID list.
(e) The volume name is updated if it is seen to have changed when the
volume is updated (the update is keyed on the volume ID).
(3) The vlocation record is got rid of and VLDB records are no longer
cached. Sufficient information is stored in the volume record, though
an update to a volume record is now no longer shared between related
volumes (volumes come in bundles of three: R/W, R/O and backup).
and the following procedural changes are made:
(1) The fileserver cursor introduced previously is now fleshed out and
used to iterate over fileservers and their addresses.
(2) Volume status is checked during iteration, and the server list is
replaced if a change is detected.
(3) Server status is checked during iteration, and the address list is
replaced if a change is detected.
(4) The abort code is saved into the address list cursor and -ECONNABORTED
returned in afs_make_call() if a remote abort happened rather than
translating the abort into an error message. This allows actions to
be taken depending on the abort code more easily.
(a) If a VMOVED abort is seen then this is handled by rechecking the
volume and restarting the iteration.
(b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is
handled by sleeping for a short period and retrying and/or trying
other servers that might serve that volume. A message is also
displayed once until the condition has cleared.
(c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the
moment.
(d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to
see if it has been deleted; if not, the fileserver is probably
indicating that the volume couldn't be attached and needs
salvaging.
(e) If statfs() sees one of these aborts, it does not sleep, but
rather returns an error, so as not to block the umount program.
(5) The fileserver iteration functions in vnode.c are now merged into
their callers and more heavily macroised around the cursor. vnode.c
is removed.
(6) Operations on a particular vnode are serialised on that vnode because
the server will lock that vnode whilst it operates on it, so a second
op sent will just have to wait.
(7) Fileservers are probed with FS.GetCapabilities before being used.
This is where service upgrade will be done.
(8) A callback interest on a fileserver is set up before an FS operation
is performed and passed through to afs_make_call() so that it can be
set on the vnode if the operation returns a callback. The callback
interest is passed through to afs_iget() also so that it can be set
there too.
In general, record updating is done on an as-needed basis when we try to
access servers, volumes or vnodes rather than offloading it to work items
and special threads.
Notes:
(1) Pre AFS-3.4 servers are no longer supported, though this can be added
back if necessary (AFS-3.4 was released in 1998).
(2) VBUSY is retried forever for the moment at intervals of 1s.
(3) /proc/fs/afs/<cell>/servers no longer exists.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 23:27:50 +08:00
|
|
|
struct afs_file_status newstatus;
|
|
|
|
struct afs_fs_cursor fc;
|
|
|
|
struct afs_callback newcb;
|
|
|
|
struct afs_vnode *dvnode = AFS_FS_I(dir);
|
|
|
|
struct afs_fid newfid;
|
2007-04-27 06:59:35 +08:00
|
|
|
struct key *key;
|
|
|
|
int ret;
|
|
|
|
|
afs: Overhaul volume and server record caching and fileserver rotation
The current code assumes that volumes and servers are per-cell and are
never shared, but this is not enforced, and, indeed, public cells do exist
that are aliases of each other. Further, an organisation can, say, set up
a public cell and a private cell with overlapping, but not identical, sets
of servers. The difference is purely in the database attached to the VL
servers.
The current code will malfunction if it sees a server in two cells as it
assumes global address -> server record mappings and that each server is in
just one cell.
Further, each server may have multiple addresses - and may have addresses
of different families (IPv4 and IPv6, say).
To this end, the following structural changes are made:
(1) Server record management is overhauled:
(a) Server records are made independent of cell. The namespace keeps
track of them, volume records have lists of them and each vnode
has a server on which its callback interest currently resides.
(b) The cell record no longer keeps a list of servers known to be in
that cell.
(c) The server records are now kept in a flat list because there's no
single address to sort on.
(d) Server records are now keyed by their UUID within the namespace.
(e) The addresses for a server are obtained with the VL.GetAddrsU
rather than with VL.GetEntryByName, using the server's UUID as a
parameter.
(f) Cached server records are garbage collected after a period of
non-use and are counted out of existence before purging is allowed
to complete. This protects the work functions against rmmod.
(g) The servers list is now in /proc/fs/afs/servers.
(2) Volume record management is overhauled:
(a) An RCU-replaceable server list is introduced. This tracks both
servers and their coresponding callback interests.
(b) The superblock is now keyed on cell record and numeric volume ID.
(c) The volume record is now tied to the superblock which mounts it,
and is activated when mounted and deactivated when unmounted.
This makes it easier to handle the cache cookie without causing a
double-use in fscache.
(d) The volume record is loaded from the VLDB using VL.GetEntryByNameU
to get the server UUID list.
(e) The volume name is updated if it is seen to have changed when the
volume is updated (the update is keyed on the volume ID).
(3) The vlocation record is got rid of and VLDB records are no longer
cached. Sufficient information is stored in the volume record, though
an update to a volume record is now no longer shared between related
volumes (volumes come in bundles of three: R/W, R/O and backup).
and the following procedural changes are made:
(1) The fileserver cursor introduced previously is now fleshed out and
used to iterate over fileservers and their addresses.
(2) Volume status is checked during iteration, and the server list is
replaced if a change is detected.
(3) Server status is checked during iteration, and the address list is
replaced if a change is detected.
(4) The abort code is saved into the address list cursor and -ECONNABORTED
returned in afs_make_call() if a remote abort happened rather than
translating the abort into an error message. This allows actions to
be taken depending on the abort code more easily.
(a) If a VMOVED abort is seen then this is handled by rechecking the
volume and restarting the iteration.
(b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is
handled by sleeping for a short period and retrying and/or trying
other servers that might serve that volume. A message is also
displayed once until the condition has cleared.
(c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the
moment.
(d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to
see if it has been deleted; if not, the fileserver is probably
indicating that the volume couldn't be attached and needs
salvaging.
(e) If statfs() sees one of these aborts, it does not sleep, but
rather returns an error, so as not to block the umount program.
(5) The fileserver iteration functions in vnode.c are now merged into
their callers and more heavily macroised around the cursor. vnode.c
is removed.
(6) Operations on a particular vnode are serialised on that vnode because
the server will lock that vnode whilst it operates on it, so a second
op sent will just have to wait.
(7) Fileservers are probed with FS.GetCapabilities before being used.
This is where service upgrade will be done.
(8) A callback interest on a fileserver is set up before an FS operation
is performed and passed through to afs_make_call() so that it can be
set on the vnode if the operation returns a callback. The callback
interest is passed through to afs_iget() also so that it can be set
there too.
In general, record updating is done on an as-needed basis when we try to
access servers, volumes or vnodes rather than offloading it to work items
and special threads.
Notes:
(1) Pre AFS-3.4 servers are no longer supported, though this can be added
back if necessary (AFS-3.4 was released in 1998).
(2) VBUSY is retried forever for the moment at intervals of 1s.
(3) /proc/fs/afs/<cell>/servers no longer exists.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 23:27:50 +08:00
|
|
|
mode |= S_IFDIR;
|
2007-04-27 06:59:35 +08:00
|
|
|
|
2014-10-22 08:11:25 +08:00
|
|
|
_enter("{%x:%u},{%pd},%ho",
|
|
|
|
dvnode->fid.vid, dvnode->fid.vnode, dentry, mode);
|
2007-04-27 06:59:35 +08:00
|
|
|
|
|
|
|
key = afs_request_key(dvnode->volume->cell);
|
|
|
|
if (IS_ERR(key)) {
|
|
|
|
ret = PTR_ERR(key);
|
|
|
|
goto error;
|
|
|
|
}
|
|
|
|
|
afs: Overhaul volume and server record caching and fileserver rotation
The current code assumes that volumes and servers are per-cell and are
never shared, but this is not enforced, and, indeed, public cells do exist
that are aliases of each other. Further, an organisation can, say, set up
a public cell and a private cell with overlapping, but not identical, sets
of servers. The difference is purely in the database attached to the VL
servers.
The current code will malfunction if it sees a server in two cells as it
assumes global address -> server record mappings and that each server is in
just one cell.
Further, each server may have multiple addresses - and may have addresses
of different families (IPv4 and IPv6, say).
To this end, the following structural changes are made:
(1) Server record management is overhauled:
(a) Server records are made independent of cell. The namespace keeps
track of them, volume records have lists of them and each vnode
has a server on which its callback interest currently resides.
(b) The cell record no longer keeps a list of servers known to be in
that cell.
(c) The server records are now kept in a flat list because there's no
single address to sort on.
(d) Server records are now keyed by their UUID within the namespace.
(e) The addresses for a server are obtained with the VL.GetAddrsU
rather than with VL.GetEntryByName, using the server's UUID as a
parameter.
(f) Cached server records are garbage collected after a period of
non-use and are counted out of existence before purging is allowed
to complete. This protects the work functions against rmmod.
(g) The servers list is now in /proc/fs/afs/servers.
(2) Volume record management is overhauled:
(a) An RCU-replaceable server list is introduced. This tracks both
servers and their coresponding callback interests.
(b) The superblock is now keyed on cell record and numeric volume ID.
(c) The volume record is now tied to the superblock which mounts it,
and is activated when mounted and deactivated when unmounted.
This makes it easier to handle the cache cookie without causing a
double-use in fscache.
(d) The volume record is loaded from the VLDB using VL.GetEntryByNameU
to get the server UUID list.
(e) The volume name is updated if it is seen to have changed when the
volume is updated (the update is keyed on the volume ID).
(3) The vlocation record is got rid of and VLDB records are no longer
cached. Sufficient information is stored in the volume record, though
an update to a volume record is now no longer shared between related
volumes (volumes come in bundles of three: R/W, R/O and backup).
and the following procedural changes are made:
(1) The fileserver cursor introduced previously is now fleshed out and
used to iterate over fileservers and their addresses.
(2) Volume status is checked during iteration, and the server list is
replaced if a change is detected.
(3) Server status is checked during iteration, and the address list is
replaced if a change is detected.
(4) The abort code is saved into the address list cursor and -ECONNABORTED
returned in afs_make_call() if a remote abort happened rather than
translating the abort into an error message. This allows actions to
be taken depending on the abort code more easily.
(a) If a VMOVED abort is seen then this is handled by rechecking the
volume and restarting the iteration.
(b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is
handled by sleeping for a short period and retrying and/or trying
other servers that might serve that volume. A message is also
displayed once until the condition has cleared.
(c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the
moment.
(d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to
see if it has been deleted; if not, the fileserver is probably
indicating that the volume couldn't be attached and needs
salvaging.
(e) If statfs() sees one of these aborts, it does not sleep, but
rather returns an error, so as not to block the umount program.
(5) The fileserver iteration functions in vnode.c are now merged into
their callers and more heavily macroised around the cursor. vnode.c
is removed.
(6) Operations on a particular vnode are serialised on that vnode because
the server will lock that vnode whilst it operates on it, so a second
op sent will just have to wait.
(7) Fileservers are probed with FS.GetCapabilities before being used.
This is where service upgrade will be done.
(8) A callback interest on a fileserver is set up before an FS operation
is performed and passed through to afs_make_call() so that it can be
set on the vnode if the operation returns a callback. The callback
interest is passed through to afs_iget() also so that it can be set
there too.
In general, record updating is done on an as-needed basis when we try to
access servers, volumes or vnodes rather than offloading it to work items
and special threads.
Notes:
(1) Pre AFS-3.4 servers are no longer supported, though this can be added
back if necessary (AFS-3.4 was released in 1998).
(2) VBUSY is retried forever for the moment at intervals of 1s.
(3) /proc/fs/afs/<cell>/servers no longer exists.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 23:27:50 +08:00
|
|
|
ret = -ERESTARTSYS;
|
|
|
|
if (afs_begin_vnode_operation(&fc, dvnode, key)) {
|
|
|
|
while (afs_select_fileserver(&fc)) {
|
|
|
|
fc.cb_break = dvnode->cb_break + dvnode->cb_s_break;
|
|
|
|
afs_fs_create(&fc, dentry->d_name.name, mode,
|
|
|
|
&newfid, &newstatus, &newcb);
|
|
|
|
}
|
2007-04-27 06:59:35 +08:00
|
|
|
|
afs: Overhaul volume and server record caching and fileserver rotation
The current code assumes that volumes and servers are per-cell and are
never shared, but this is not enforced, and, indeed, public cells do exist
that are aliases of each other. Further, an organisation can, say, set up
a public cell and a private cell with overlapping, but not identical, sets
of servers. The difference is purely in the database attached to the VL
servers.
The current code will malfunction if it sees a server in two cells as it
assumes global address -> server record mappings and that each server is in
just one cell.
Further, each server may have multiple addresses - and may have addresses
of different families (IPv4 and IPv6, say).
To this end, the following structural changes are made:
(1) Server record management is overhauled:
(a) Server records are made independent of cell. The namespace keeps
track of them, volume records have lists of them and each vnode
has a server on which its callback interest currently resides.
(b) The cell record no longer keeps a list of servers known to be in
that cell.
(c) The server records are now kept in a flat list because there's no
single address to sort on.
(d) Server records are now keyed by their UUID within the namespace.
(e) The addresses for a server are obtained with the VL.GetAddrsU
rather than with VL.GetEntryByName, using the server's UUID as a
parameter.
(f) Cached server records are garbage collected after a period of
non-use and are counted out of existence before purging is allowed
to complete. This protects the work functions against rmmod.
(g) The servers list is now in /proc/fs/afs/servers.
(2) Volume record management is overhauled:
(a) An RCU-replaceable server list is introduced. This tracks both
servers and their coresponding callback interests.
(b) The superblock is now keyed on cell record and numeric volume ID.
(c) The volume record is now tied to the superblock which mounts it,
and is activated when mounted and deactivated when unmounted.
This makes it easier to handle the cache cookie without causing a
double-use in fscache.
(d) The volume record is loaded from the VLDB using VL.GetEntryByNameU
to get the server UUID list.
(e) The volume name is updated if it is seen to have changed when the
volume is updated (the update is keyed on the volume ID).
(3) The vlocation record is got rid of and VLDB records are no longer
cached. Sufficient information is stored in the volume record, though
an update to a volume record is now no longer shared between related
volumes (volumes come in bundles of three: R/W, R/O and backup).
and the following procedural changes are made:
(1) The fileserver cursor introduced previously is now fleshed out and
used to iterate over fileservers and their addresses.
(2) Volume status is checked during iteration, and the server list is
replaced if a change is detected.
(3) Server status is checked during iteration, and the address list is
replaced if a change is detected.
(4) The abort code is saved into the address list cursor and -ECONNABORTED
returned in afs_make_call() if a remote abort happened rather than
translating the abort into an error message. This allows actions to
be taken depending on the abort code more easily.
(a) If a VMOVED abort is seen then this is handled by rechecking the
volume and restarting the iteration.
(b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is
handled by sleeping for a short period and retrying and/or trying
other servers that might serve that volume. A message is also
displayed once until the condition has cleared.
(c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the
moment.
(d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to
see if it has been deleted; if not, the fileserver is probably
indicating that the volume couldn't be attached and needs
salvaging.
(e) If statfs() sees one of these aborts, it does not sleep, but
rather returns an error, so as not to block the umount program.
(5) The fileserver iteration functions in vnode.c are now merged into
their callers and more heavily macroised around the cursor. vnode.c
is removed.
(6) Operations on a particular vnode are serialised on that vnode because
the server will lock that vnode whilst it operates on it, so a second
op sent will just have to wait.
(7) Fileservers are probed with FS.GetCapabilities before being used.
This is where service upgrade will be done.
(8) A callback interest on a fileserver is set up before an FS operation
is performed and passed through to afs_make_call() so that it can be
set on the vnode if the operation returns a callback. The callback
interest is passed through to afs_iget() also so that it can be set
there too.
In general, record updating is done on an as-needed basis when we try to
access servers, volumes or vnodes rather than offloading it to work items
and special threads.
Notes:
(1) Pre AFS-3.4 servers are no longer supported, though this can be added
back if necessary (AFS-3.4 was released in 1998).
(2) VBUSY is retried forever for the moment at intervals of 1s.
(3) /proc/fs/afs/<cell>/servers no longer exists.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 23:27:50 +08:00
|
|
|
afs_check_for_remote_deletion(&fc, fc.vnode);
|
|
|
|
afs_vnode_commit_status(&fc, dvnode, fc.cb_break);
|
|
|
|
afs_vnode_new_inode(&fc, dentry, &newfid, &newstatus, &newcb);
|
|
|
|
ret = afs_end_vnode_operation(&fc);
|
|
|
|
if (ret < 0)
|
|
|
|
goto error_key;
|
2017-11-21 06:41:00 +08:00
|
|
|
} else {
|
|
|
|
goto error_key;
|
2007-04-27 06:59:35 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
key_put(key);
|
|
|
|
_leave(" = 0");
|
|
|
|
return 0;
|
|
|
|
|
afs: Overhaul volume and server record caching and fileserver rotation
The current code assumes that volumes and servers are per-cell and are
never shared, but this is not enforced, and, indeed, public cells do exist
that are aliases of each other. Further, an organisation can, say, set up
a public cell and a private cell with overlapping, but not identical, sets
of servers. The difference is purely in the database attached to the VL
servers.
The current code will malfunction if it sees a server in two cells as it
assumes global address -> server record mappings and that each server is in
just one cell.
Further, each server may have multiple addresses - and may have addresses
of different families (IPv4 and IPv6, say).
To this end, the following structural changes are made:
(1) Server record management is overhauled:
(a) Server records are made independent of cell. The namespace keeps
track of them, volume records have lists of them and each vnode
has a server on which its callback interest currently resides.
(b) The cell record no longer keeps a list of servers known to be in
that cell.
(c) The server records are now kept in a flat list because there's no
single address to sort on.
(d) Server records are now keyed by their UUID within the namespace.
(e) The addresses for a server are obtained with the VL.GetAddrsU
rather than with VL.GetEntryByName, using the server's UUID as a
parameter.
(f) Cached server records are garbage collected after a period of
non-use and are counted out of existence before purging is allowed
to complete. This protects the work functions against rmmod.
(g) The servers list is now in /proc/fs/afs/servers.
(2) Volume record management is overhauled:
(a) An RCU-replaceable server list is introduced. This tracks both
servers and their coresponding callback interests.
(b) The superblock is now keyed on cell record and numeric volume ID.
(c) The volume record is now tied to the superblock which mounts it,
and is activated when mounted and deactivated when unmounted.
This makes it easier to handle the cache cookie without causing a
double-use in fscache.
(d) The volume record is loaded from the VLDB using VL.GetEntryByNameU
to get the server UUID list.
(e) The volume name is updated if it is seen to have changed when the
volume is updated (the update is keyed on the volume ID).
(3) The vlocation record is got rid of and VLDB records are no longer
cached. Sufficient information is stored in the volume record, though
an update to a volume record is now no longer shared between related
volumes (volumes come in bundles of three: R/W, R/O and backup).
and the following procedural changes are made:
(1) The fileserver cursor introduced previously is now fleshed out and
used to iterate over fileservers and their addresses.
(2) Volume status is checked during iteration, and the server list is
replaced if a change is detected.
(3) Server status is checked during iteration, and the address list is
replaced if a change is detected.
(4) The abort code is saved into the address list cursor and -ECONNABORTED
returned in afs_make_call() if a remote abort happened rather than
translating the abort into an error message. This allows actions to
be taken depending on the abort code more easily.
(a) If a VMOVED abort is seen then this is handled by rechecking the
volume and restarting the iteration.
(b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is
handled by sleeping for a short period and retrying and/or trying
other servers that might serve that volume. A message is also
displayed once until the condition has cleared.
(c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the
moment.
(d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to
see if it has been deleted; if not, the fileserver is probably
indicating that the volume couldn't be attached and needs
salvaging.
(e) If statfs() sees one of these aborts, it does not sleep, but
rather returns an error, so as not to block the umount program.
(5) The fileserver iteration functions in vnode.c are now merged into
their callers and more heavily macroised around the cursor. vnode.c
is removed.
(6) Operations on a particular vnode are serialised on that vnode because
the server will lock that vnode whilst it operates on it, so a second
op sent will just have to wait.
(7) Fileservers are probed with FS.GetCapabilities before being used.
This is where service upgrade will be done.
(8) A callback interest on a fileserver is set up before an FS operation
is performed and passed through to afs_make_call() so that it can be
set on the vnode if the operation returns a callback. The callback
interest is passed through to afs_iget() also so that it can be set
there too.
In general, record updating is done on an as-needed basis when we try to
access servers, volumes or vnodes rather than offloading it to work items
and special threads.
Notes:
(1) Pre AFS-3.4 servers are no longer supported, though this can be added
back if necessary (AFS-3.4 was released in 1998).
(2) VBUSY is retried forever for the moment at intervals of 1s.
(3) /proc/fs/afs/<cell>/servers no longer exists.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 23:27:50 +08:00
|
|
|
error_key:
|
2007-04-27 06:59:35 +08:00
|
|
|
key_put(key);
|
|
|
|
error:
|
|
|
|
d_drop(dentry);
|
|
|
|
_leave(" = %d", ret);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
afs: Overhaul volume and server record caching and fileserver rotation
The current code assumes that volumes and servers are per-cell and are
never shared, but this is not enforced, and, indeed, public cells do exist
that are aliases of each other. Further, an organisation can, say, set up
a public cell and a private cell with overlapping, but not identical, sets
of servers. The difference is purely in the database attached to the VL
servers.
The current code will malfunction if it sees a server in two cells as it
assumes global address -> server record mappings and that each server is in
just one cell.
Further, each server may have multiple addresses - and may have addresses
of different families (IPv4 and IPv6, say).
To this end, the following structural changes are made:
(1) Server record management is overhauled:
(a) Server records are made independent of cell. The namespace keeps
track of them, volume records have lists of them and each vnode
has a server on which its callback interest currently resides.
(b) The cell record no longer keeps a list of servers known to be in
that cell.
(c) The server records are now kept in a flat list because there's no
single address to sort on.
(d) Server records are now keyed by their UUID within the namespace.
(e) The addresses for a server are obtained with the VL.GetAddrsU
rather than with VL.GetEntryByName, using the server's UUID as a
parameter.
(f) Cached server records are garbage collected after a period of
non-use and are counted out of existence before purging is allowed
to complete. This protects the work functions against rmmod.
(g) The servers list is now in /proc/fs/afs/servers.
(2) Volume record management is overhauled:
(a) An RCU-replaceable server list is introduced. This tracks both
servers and their coresponding callback interests.
(b) The superblock is now keyed on cell record and numeric volume ID.
(c) The volume record is now tied to the superblock which mounts it,
and is activated when mounted and deactivated when unmounted.
This makes it easier to handle the cache cookie without causing a
double-use in fscache.
(d) The volume record is loaded from the VLDB using VL.GetEntryByNameU
to get the server UUID list.
(e) The volume name is updated if it is seen to have changed when the
volume is updated (the update is keyed on the volume ID).
(3) The vlocation record is got rid of and VLDB records are no longer
cached. Sufficient information is stored in the volume record, though
an update to a volume record is now no longer shared between related
volumes (volumes come in bundles of three: R/W, R/O and backup).
and the following procedural changes are made:
(1) The fileserver cursor introduced previously is now fleshed out and
used to iterate over fileservers and their addresses.
(2) Volume status is checked during iteration, and the server list is
replaced if a change is detected.
(3) Server status is checked during iteration, and the address list is
replaced if a change is detected.
(4) The abort code is saved into the address list cursor and -ECONNABORTED
returned in afs_make_call() if a remote abort happened rather than
translating the abort into an error message. This allows actions to
be taken depending on the abort code more easily.
(a) If a VMOVED abort is seen then this is handled by rechecking the
volume and restarting the iteration.
(b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is
handled by sleeping for a short period and retrying and/or trying
other servers that might serve that volume. A message is also
displayed once until the condition has cleared.
(c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the
moment.
(d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to
see if it has been deleted; if not, the fileserver is probably
indicating that the volume couldn't be attached and needs
salvaging.
(e) If statfs() sees one of these aborts, it does not sleep, but
rather returns an error, so as not to block the umount program.
(5) The fileserver iteration functions in vnode.c are now merged into
their callers and more heavily macroised around the cursor. vnode.c
is removed.
(6) Operations on a particular vnode are serialised on that vnode because
the server will lock that vnode whilst it operates on it, so a second
op sent will just have to wait.
(7) Fileservers are probed with FS.GetCapabilities before being used.
This is where service upgrade will be done.
(8) A callback interest on a fileserver is set up before an FS operation
is performed and passed through to afs_make_call() so that it can be
set on the vnode if the operation returns a callback. The callback
interest is passed through to afs_iget() also so that it can be set
there too.
In general, record updating is done on an as-needed basis when we try to
access servers, volumes or vnodes rather than offloading it to work items
and special threads.
Notes:
(1) Pre AFS-3.4 servers are no longer supported, though this can be added
back if necessary (AFS-3.4 was released in 1998).
(2) VBUSY is retried forever for the moment at intervals of 1s.
(3) /proc/fs/afs/<cell>/servers no longer exists.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 23:27:50 +08:00
|
|
|
/*
|
|
|
|
* Remove a subdir from a directory.
|
|
|
|
*/
|
|
|
|
static void afs_dir_remove_subdir(struct dentry *dentry)
|
|
|
|
{
|
|
|
|
if (d_really_is_positive(dentry)) {
|
|
|
|
struct afs_vnode *vnode = AFS_FS_I(d_inode(dentry));
|
|
|
|
|
|
|
|
clear_nlink(&vnode->vfs_inode);
|
|
|
|
set_bit(AFS_VNODE_DELETED, &vnode->flags);
|
|
|
|
clear_bit(AFS_VNODE_CB_PROMISED, &vnode->flags);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2007-04-27 06:59:35 +08:00
|
|
|
/*
|
|
|
|
* remove a directory from an AFS filesystem
|
|
|
|
*/
|
|
|
|
static int afs_rmdir(struct inode *dir, struct dentry *dentry)
|
|
|
|
{
|
afs: Overhaul volume and server record caching and fileserver rotation
The current code assumes that volumes and servers are per-cell and are
never shared, but this is not enforced, and, indeed, public cells do exist
that are aliases of each other. Further, an organisation can, say, set up
a public cell and a private cell with overlapping, but not identical, sets
of servers. The difference is purely in the database attached to the VL
servers.
The current code will malfunction if it sees a server in two cells as it
assumes global address -> server record mappings and that each server is in
just one cell.
Further, each server may have multiple addresses - and may have addresses
of different families (IPv4 and IPv6, say).
To this end, the following structural changes are made:
(1) Server record management is overhauled:
(a) Server records are made independent of cell. The namespace keeps
track of them, volume records have lists of them and each vnode
has a server on which its callback interest currently resides.
(b) The cell record no longer keeps a list of servers known to be in
that cell.
(c) The server records are now kept in a flat list because there's no
single address to sort on.
(d) Server records are now keyed by their UUID within the namespace.
(e) The addresses for a server are obtained with the VL.GetAddrsU
rather than with VL.GetEntryByName, using the server's UUID as a
parameter.
(f) Cached server records are garbage collected after a period of
non-use and are counted out of existence before purging is allowed
to complete. This protects the work functions against rmmod.
(g) The servers list is now in /proc/fs/afs/servers.
(2) Volume record management is overhauled:
(a) An RCU-replaceable server list is introduced. This tracks both
servers and their coresponding callback interests.
(b) The superblock is now keyed on cell record and numeric volume ID.
(c) The volume record is now tied to the superblock which mounts it,
and is activated when mounted and deactivated when unmounted.
This makes it easier to handle the cache cookie without causing a
double-use in fscache.
(d) The volume record is loaded from the VLDB using VL.GetEntryByNameU
to get the server UUID list.
(e) The volume name is updated if it is seen to have changed when the
volume is updated (the update is keyed on the volume ID).
(3) The vlocation record is got rid of and VLDB records are no longer
cached. Sufficient information is stored in the volume record, though
an update to a volume record is now no longer shared between related
volumes (volumes come in bundles of three: R/W, R/O and backup).
and the following procedural changes are made:
(1) The fileserver cursor introduced previously is now fleshed out and
used to iterate over fileservers and their addresses.
(2) Volume status is checked during iteration, and the server list is
replaced if a change is detected.
(3) Server status is checked during iteration, and the address list is
replaced if a change is detected.
(4) The abort code is saved into the address list cursor and -ECONNABORTED
returned in afs_make_call() if a remote abort happened rather than
translating the abort into an error message. This allows actions to
be taken depending on the abort code more easily.
(a) If a VMOVED abort is seen then this is handled by rechecking the
volume and restarting the iteration.
(b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is
handled by sleeping for a short period and retrying and/or trying
other servers that might serve that volume. A message is also
displayed once until the condition has cleared.
(c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the
moment.
(d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to
see if it has been deleted; if not, the fileserver is probably
indicating that the volume couldn't be attached and needs
salvaging.
(e) If statfs() sees one of these aborts, it does not sleep, but
rather returns an error, so as not to block the umount program.
(5) The fileserver iteration functions in vnode.c are now merged into
their callers and more heavily macroised around the cursor. vnode.c
is removed.
(6) Operations on a particular vnode are serialised on that vnode because
the server will lock that vnode whilst it operates on it, so a second
op sent will just have to wait.
(7) Fileservers are probed with FS.GetCapabilities before being used.
This is where service upgrade will be done.
(8) A callback interest on a fileserver is set up before an FS operation
is performed and passed through to afs_make_call() so that it can be
set on the vnode if the operation returns a callback. The callback
interest is passed through to afs_iget() also so that it can be set
there too.
In general, record updating is done on an as-needed basis when we try to
access servers, volumes or vnodes rather than offloading it to work items
and special threads.
Notes:
(1) Pre AFS-3.4 servers are no longer supported, though this can be added
back if necessary (AFS-3.4 was released in 1998).
(2) VBUSY is retried forever for the moment at intervals of 1s.
(3) /proc/fs/afs/<cell>/servers no longer exists.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 23:27:50 +08:00
|
|
|
struct afs_fs_cursor fc;
|
|
|
|
struct afs_vnode *dvnode = AFS_FS_I(dir);
|
2007-04-27 06:59:35 +08:00
|
|
|
struct key *key;
|
|
|
|
int ret;
|
|
|
|
|
2014-10-22 08:11:25 +08:00
|
|
|
_enter("{%x:%u},{%pd}",
|
|
|
|
dvnode->fid.vid, dvnode->fid.vnode, dentry);
|
2007-04-27 06:59:35 +08:00
|
|
|
|
|
|
|
key = afs_request_key(dvnode->volume->cell);
|
|
|
|
if (IS_ERR(key)) {
|
|
|
|
ret = PTR_ERR(key);
|
|
|
|
goto error;
|
|
|
|
}
|
|
|
|
|
afs: Overhaul volume and server record caching and fileserver rotation
The current code assumes that volumes and servers are per-cell and are
never shared, but this is not enforced, and, indeed, public cells do exist
that are aliases of each other. Further, an organisation can, say, set up
a public cell and a private cell with overlapping, but not identical, sets
of servers. The difference is purely in the database attached to the VL
servers.
The current code will malfunction if it sees a server in two cells as it
assumes global address -> server record mappings and that each server is in
just one cell.
Further, each server may have multiple addresses - and may have addresses
of different families (IPv4 and IPv6, say).
To this end, the following structural changes are made:
(1) Server record management is overhauled:
(a) Server records are made independent of cell. The namespace keeps
track of them, volume records have lists of them and each vnode
has a server on which its callback interest currently resides.
(b) The cell record no longer keeps a list of servers known to be in
that cell.
(c) The server records are now kept in a flat list because there's no
single address to sort on.
(d) Server records are now keyed by their UUID within the namespace.
(e) The addresses for a server are obtained with the VL.GetAddrsU
rather than with VL.GetEntryByName, using the server's UUID as a
parameter.
(f) Cached server records are garbage collected after a period of
non-use and are counted out of existence before purging is allowed
to complete. This protects the work functions against rmmod.
(g) The servers list is now in /proc/fs/afs/servers.
(2) Volume record management is overhauled:
(a) An RCU-replaceable server list is introduced. This tracks both
servers and their coresponding callback interests.
(b) The superblock is now keyed on cell record and numeric volume ID.
(c) The volume record is now tied to the superblock which mounts it,
and is activated when mounted and deactivated when unmounted.
This makes it easier to handle the cache cookie without causing a
double-use in fscache.
(d) The volume record is loaded from the VLDB using VL.GetEntryByNameU
to get the server UUID list.
(e) The volume name is updated if it is seen to have changed when the
volume is updated (the update is keyed on the volume ID).
(3) The vlocation record is got rid of and VLDB records are no longer
cached. Sufficient information is stored in the volume record, though
an update to a volume record is now no longer shared between related
volumes (volumes come in bundles of three: R/W, R/O and backup).
and the following procedural changes are made:
(1) The fileserver cursor introduced previously is now fleshed out and
used to iterate over fileservers and their addresses.
(2) Volume status is checked during iteration, and the server list is
replaced if a change is detected.
(3) Server status is checked during iteration, and the address list is
replaced if a change is detected.
(4) The abort code is saved into the address list cursor and -ECONNABORTED
returned in afs_make_call() if a remote abort happened rather than
translating the abort into an error message. This allows actions to
be taken depending on the abort code more easily.
(a) If a VMOVED abort is seen then this is handled by rechecking the
volume and restarting the iteration.
(b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is
handled by sleeping for a short period and retrying and/or trying
other servers that might serve that volume. A message is also
displayed once until the condition has cleared.
(c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the
moment.
(d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to
see if it has been deleted; if not, the fileserver is probably
indicating that the volume couldn't be attached and needs
salvaging.
(e) If statfs() sees one of these aborts, it does not sleep, but
rather returns an error, so as not to block the umount program.
(5) The fileserver iteration functions in vnode.c are now merged into
their callers and more heavily macroised around the cursor. vnode.c
is removed.
(6) Operations on a particular vnode are serialised on that vnode because
the server will lock that vnode whilst it operates on it, so a second
op sent will just have to wait.
(7) Fileservers are probed with FS.GetCapabilities before being used.
This is where service upgrade will be done.
(8) A callback interest on a fileserver is set up before an FS operation
is performed and passed through to afs_make_call() so that it can be
set on the vnode if the operation returns a callback. The callback
interest is passed through to afs_iget() also so that it can be set
there too.
In general, record updating is done on an as-needed basis when we try to
access servers, volumes or vnodes rather than offloading it to work items
and special threads.
Notes:
(1) Pre AFS-3.4 servers are no longer supported, though this can be added
back if necessary (AFS-3.4 was released in 1998).
(2) VBUSY is retried forever for the moment at intervals of 1s.
(3) /proc/fs/afs/<cell>/servers no longer exists.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 23:27:50 +08:00
|
|
|
ret = -ERESTARTSYS;
|
|
|
|
if (afs_begin_vnode_operation(&fc, dvnode, key)) {
|
|
|
|
while (afs_select_fileserver(&fc)) {
|
|
|
|
fc.cb_break = dvnode->cb_break + dvnode->cb_s_break;
|
|
|
|
afs_fs_remove(&fc, dentry->d_name.name, true);
|
|
|
|
}
|
2007-04-27 06:59:35 +08:00
|
|
|
|
afs: Overhaul volume and server record caching and fileserver rotation
The current code assumes that volumes and servers are per-cell and are
never shared, but this is not enforced, and, indeed, public cells do exist
that are aliases of each other. Further, an organisation can, say, set up
a public cell and a private cell with overlapping, but not identical, sets
of servers. The difference is purely in the database attached to the VL
servers.
The current code will malfunction if it sees a server in two cells as it
assumes global address -> server record mappings and that each server is in
just one cell.
Further, each server may have multiple addresses - and may have addresses
of different families (IPv4 and IPv6, say).
To this end, the following structural changes are made:
(1) Server record management is overhauled:
(a) Server records are made independent of cell. The namespace keeps
track of them, volume records have lists of them and each vnode
has a server on which its callback interest currently resides.
(b) The cell record no longer keeps a list of servers known to be in
that cell.
(c) The server records are now kept in a flat list because there's no
single address to sort on.
(d) Server records are now keyed by their UUID within the namespace.
(e) The addresses for a server are obtained with the VL.GetAddrsU
rather than with VL.GetEntryByName, using the server's UUID as a
parameter.
(f) Cached server records are garbage collected after a period of
non-use and are counted out of existence before purging is allowed
to complete. This protects the work functions against rmmod.
(g) The servers list is now in /proc/fs/afs/servers.
(2) Volume record management is overhauled:
(a) An RCU-replaceable server list is introduced. This tracks both
servers and their coresponding callback interests.
(b) The superblock is now keyed on cell record and numeric volume ID.
(c) The volume record is now tied to the superblock which mounts it,
and is activated when mounted and deactivated when unmounted.
This makes it easier to handle the cache cookie without causing a
double-use in fscache.
(d) The volume record is loaded from the VLDB using VL.GetEntryByNameU
to get the server UUID list.
(e) The volume name is updated if it is seen to have changed when the
volume is updated (the update is keyed on the volume ID).
(3) The vlocation record is got rid of and VLDB records are no longer
cached. Sufficient information is stored in the volume record, though
an update to a volume record is now no longer shared between related
volumes (volumes come in bundles of three: R/W, R/O and backup).
and the following procedural changes are made:
(1) The fileserver cursor introduced previously is now fleshed out and
used to iterate over fileservers and their addresses.
(2) Volume status is checked during iteration, and the server list is
replaced if a change is detected.
(3) Server status is checked during iteration, and the address list is
replaced if a change is detected.
(4) The abort code is saved into the address list cursor and -ECONNABORTED
returned in afs_make_call() if a remote abort happened rather than
translating the abort into an error message. This allows actions to
be taken depending on the abort code more easily.
(a) If a VMOVED abort is seen then this is handled by rechecking the
volume and restarting the iteration.
(b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is
handled by sleeping for a short period and retrying and/or trying
other servers that might serve that volume. A message is also
displayed once until the condition has cleared.
(c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the
moment.
(d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to
see if it has been deleted; if not, the fileserver is probably
indicating that the volume couldn't be attached and needs
salvaging.
(e) If statfs() sees one of these aborts, it does not sleep, but
rather returns an error, so as not to block the umount program.
(5) The fileserver iteration functions in vnode.c are now merged into
their callers and more heavily macroised around the cursor. vnode.c
is removed.
(6) Operations on a particular vnode are serialised on that vnode because
the server will lock that vnode whilst it operates on it, so a second
op sent will just have to wait.
(7) Fileservers are probed with FS.GetCapabilities before being used.
This is where service upgrade will be done.
(8) A callback interest on a fileserver is set up before an FS operation
is performed and passed through to afs_make_call() so that it can be
set on the vnode if the operation returns a callback. The callback
interest is passed through to afs_iget() also so that it can be set
there too.
In general, record updating is done on an as-needed basis when we try to
access servers, volumes or vnodes rather than offloading it to work items
and special threads.
Notes:
(1) Pre AFS-3.4 servers are no longer supported, though this can be added
back if necessary (AFS-3.4 was released in 1998).
(2) VBUSY is retried forever for the moment at intervals of 1s.
(3) /proc/fs/afs/<cell>/servers no longer exists.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 23:27:50 +08:00
|
|
|
afs_vnode_commit_status(&fc, dvnode, fc.cb_break);
|
|
|
|
ret = afs_end_vnode_operation(&fc);
|
|
|
|
if (ret == 0)
|
|
|
|
afs_dir_remove_subdir(dentry);
|
2007-04-27 06:59:35 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
key_put(key);
|
|
|
|
error:
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
afs: Overhaul volume and server record caching and fileserver rotation
The current code assumes that volumes and servers are per-cell and are
never shared, but this is not enforced, and, indeed, public cells do exist
that are aliases of each other. Further, an organisation can, say, set up
a public cell and a private cell with overlapping, but not identical, sets
of servers. The difference is purely in the database attached to the VL
servers.
The current code will malfunction if it sees a server in two cells as it
assumes global address -> server record mappings and that each server is in
just one cell.
Further, each server may have multiple addresses - and may have addresses
of different families (IPv4 and IPv6, say).
To this end, the following structural changes are made:
(1) Server record management is overhauled:
(a) Server records are made independent of cell. The namespace keeps
track of them, volume records have lists of them and each vnode
has a server on which its callback interest currently resides.
(b) The cell record no longer keeps a list of servers known to be in
that cell.
(c) The server records are now kept in a flat list because there's no
single address to sort on.
(d) Server records are now keyed by their UUID within the namespace.
(e) The addresses for a server are obtained with the VL.GetAddrsU
rather than with VL.GetEntryByName, using the server's UUID as a
parameter.
(f) Cached server records are garbage collected after a period of
non-use and are counted out of existence before purging is allowed
to complete. This protects the work functions against rmmod.
(g) The servers list is now in /proc/fs/afs/servers.
(2) Volume record management is overhauled:
(a) An RCU-replaceable server list is introduced. This tracks both
servers and their coresponding callback interests.
(b) The superblock is now keyed on cell record and numeric volume ID.
(c) The volume record is now tied to the superblock which mounts it,
and is activated when mounted and deactivated when unmounted.
This makes it easier to handle the cache cookie without causing a
double-use in fscache.
(d) The volume record is loaded from the VLDB using VL.GetEntryByNameU
to get the server UUID list.
(e) The volume name is updated if it is seen to have changed when the
volume is updated (the update is keyed on the volume ID).
(3) The vlocation record is got rid of and VLDB records are no longer
cached. Sufficient information is stored in the volume record, though
an update to a volume record is now no longer shared between related
volumes (volumes come in bundles of three: R/W, R/O and backup).
and the following procedural changes are made:
(1) The fileserver cursor introduced previously is now fleshed out and
used to iterate over fileservers and their addresses.
(2) Volume status is checked during iteration, and the server list is
replaced if a change is detected.
(3) Server status is checked during iteration, and the address list is
replaced if a change is detected.
(4) The abort code is saved into the address list cursor and -ECONNABORTED
returned in afs_make_call() if a remote abort happened rather than
translating the abort into an error message. This allows actions to
be taken depending on the abort code more easily.
(a) If a VMOVED abort is seen then this is handled by rechecking the
volume and restarting the iteration.
(b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is
handled by sleeping for a short period and retrying and/or trying
other servers that might serve that volume. A message is also
displayed once until the condition has cleared.
(c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the
moment.
(d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to
see if it has been deleted; if not, the fileserver is probably
indicating that the volume couldn't be attached and needs
salvaging.
(e) If statfs() sees one of these aborts, it does not sleep, but
rather returns an error, so as not to block the umount program.
(5) The fileserver iteration functions in vnode.c are now merged into
their callers and more heavily macroised around the cursor. vnode.c
is removed.
(6) Operations on a particular vnode are serialised on that vnode because
the server will lock that vnode whilst it operates on it, so a second
op sent will just have to wait.
(7) Fileservers are probed with FS.GetCapabilities before being used.
This is where service upgrade will be done.
(8) A callback interest on a fileserver is set up before an FS operation
is performed and passed through to afs_make_call() so that it can be
set on the vnode if the operation returns a callback. The callback
interest is passed through to afs_iget() also so that it can be set
there too.
In general, record updating is done on an as-needed basis when we try to
access servers, volumes or vnodes rather than offloading it to work items
and special threads.
Notes:
(1) Pre AFS-3.4 servers are no longer supported, though this can be added
back if necessary (AFS-3.4 was released in 1998).
(2) VBUSY is retried forever for the moment at intervals of 1s.
(3) /proc/fs/afs/<cell>/servers no longer exists.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 23:27:50 +08:00
|
|
|
* Remove a link to a file or symlink from a directory.
|
|
|
|
*
|
|
|
|
* If the file was not deleted due to excess hard links, the fileserver will
|
|
|
|
* break the callback promise on the file - if it had one - before it returns
|
|
|
|
* to us, and if it was deleted, it won't
|
|
|
|
*
|
|
|
|
* However, if we didn't have a callback promise outstanding, or it was
|
|
|
|
* outstanding on a different server, then it won't break it either...
|
|
|
|
*/
|
2018-01-02 18:02:19 +08:00
|
|
|
static int afs_dir_remove_link(struct dentry *dentry, struct key *key,
|
|
|
|
unsigned long d_version_before,
|
|
|
|
unsigned long d_version_after)
|
afs: Overhaul volume and server record caching and fileserver rotation
The current code assumes that volumes and servers are per-cell and are
never shared, but this is not enforced, and, indeed, public cells do exist
that are aliases of each other. Further, an organisation can, say, set up
a public cell and a private cell with overlapping, but not identical, sets
of servers. The difference is purely in the database attached to the VL
servers.
The current code will malfunction if it sees a server in two cells as it
assumes global address -> server record mappings and that each server is in
just one cell.
Further, each server may have multiple addresses - and may have addresses
of different families (IPv4 and IPv6, say).
To this end, the following structural changes are made:
(1) Server record management is overhauled:
(a) Server records are made independent of cell. The namespace keeps
track of them, volume records have lists of them and each vnode
has a server on which its callback interest currently resides.
(b) The cell record no longer keeps a list of servers known to be in
that cell.
(c) The server records are now kept in a flat list because there's no
single address to sort on.
(d) Server records are now keyed by their UUID within the namespace.
(e) The addresses for a server are obtained with the VL.GetAddrsU
rather than with VL.GetEntryByName, using the server's UUID as a
parameter.
(f) Cached server records are garbage collected after a period of
non-use and are counted out of existence before purging is allowed
to complete. This protects the work functions against rmmod.
(g) The servers list is now in /proc/fs/afs/servers.
(2) Volume record management is overhauled:
(a) An RCU-replaceable server list is introduced. This tracks both
servers and their coresponding callback interests.
(b) The superblock is now keyed on cell record and numeric volume ID.
(c) The volume record is now tied to the superblock which mounts it,
and is activated when mounted and deactivated when unmounted.
This makes it easier to handle the cache cookie without causing a
double-use in fscache.
(d) The volume record is loaded from the VLDB using VL.GetEntryByNameU
to get the server UUID list.
(e) The volume name is updated if it is seen to have changed when the
volume is updated (the update is keyed on the volume ID).
(3) The vlocation record is got rid of and VLDB records are no longer
cached. Sufficient information is stored in the volume record, though
an update to a volume record is now no longer shared between related
volumes (volumes come in bundles of three: R/W, R/O and backup).
and the following procedural changes are made:
(1) The fileserver cursor introduced previously is now fleshed out and
used to iterate over fileservers and their addresses.
(2) Volume status is checked during iteration, and the server list is
replaced if a change is detected.
(3) Server status is checked during iteration, and the address list is
replaced if a change is detected.
(4) The abort code is saved into the address list cursor and -ECONNABORTED
returned in afs_make_call() if a remote abort happened rather than
translating the abort into an error message. This allows actions to
be taken depending on the abort code more easily.
(a) If a VMOVED abort is seen then this is handled by rechecking the
volume and restarting the iteration.
(b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is
handled by sleeping for a short period and retrying and/or trying
other servers that might serve that volume. A message is also
displayed once until the condition has cleared.
(c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the
moment.
(d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to
see if it has been deleted; if not, the fileserver is probably
indicating that the volume couldn't be attached and needs
salvaging.
(e) If statfs() sees one of these aborts, it does not sleep, but
rather returns an error, so as not to block the umount program.
(5) The fileserver iteration functions in vnode.c are now merged into
their callers and more heavily macroised around the cursor. vnode.c
is removed.
(6) Operations on a particular vnode are serialised on that vnode because
the server will lock that vnode whilst it operates on it, so a second
op sent will just have to wait.
(7) Fileservers are probed with FS.GetCapabilities before being used.
This is where service upgrade will be done.
(8) A callback interest on a fileserver is set up before an FS operation
is performed and passed through to afs_make_call() so that it can be
set on the vnode if the operation returns a callback. The callback
interest is passed through to afs_iget() also so that it can be set
there too.
In general, record updating is done on an as-needed basis when we try to
access servers, volumes or vnodes rather than offloading it to work items
and special threads.
Notes:
(1) Pre AFS-3.4 servers are no longer supported, though this can be added
back if necessary (AFS-3.4 was released in 1998).
(2) VBUSY is retried forever for the moment at intervals of 1s.
(3) /proc/fs/afs/<cell>/servers no longer exists.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 23:27:50 +08:00
|
|
|
{
|
2018-01-02 18:02:19 +08:00
|
|
|
bool dir_valid;
|
afs: Overhaul volume and server record caching and fileserver rotation
The current code assumes that volumes and servers are per-cell and are
never shared, but this is not enforced, and, indeed, public cells do exist
that are aliases of each other. Further, an organisation can, say, set up
a public cell and a private cell with overlapping, but not identical, sets
of servers. The difference is purely in the database attached to the VL
servers.
The current code will malfunction if it sees a server in two cells as it
assumes global address -> server record mappings and that each server is in
just one cell.
Further, each server may have multiple addresses - and may have addresses
of different families (IPv4 and IPv6, say).
To this end, the following structural changes are made:
(1) Server record management is overhauled:
(a) Server records are made independent of cell. The namespace keeps
track of them, volume records have lists of them and each vnode
has a server on which its callback interest currently resides.
(b) The cell record no longer keeps a list of servers known to be in
that cell.
(c) The server records are now kept in a flat list because there's no
single address to sort on.
(d) Server records are now keyed by their UUID within the namespace.
(e) The addresses for a server are obtained with the VL.GetAddrsU
rather than with VL.GetEntryByName, using the server's UUID as a
parameter.
(f) Cached server records are garbage collected after a period of
non-use and are counted out of existence before purging is allowed
to complete. This protects the work functions against rmmod.
(g) The servers list is now in /proc/fs/afs/servers.
(2) Volume record management is overhauled:
(a) An RCU-replaceable server list is introduced. This tracks both
servers and their coresponding callback interests.
(b) The superblock is now keyed on cell record and numeric volume ID.
(c) The volume record is now tied to the superblock which mounts it,
and is activated when mounted and deactivated when unmounted.
This makes it easier to handle the cache cookie without causing a
double-use in fscache.
(d) The volume record is loaded from the VLDB using VL.GetEntryByNameU
to get the server UUID list.
(e) The volume name is updated if it is seen to have changed when the
volume is updated (the update is keyed on the volume ID).
(3) The vlocation record is got rid of and VLDB records are no longer
cached. Sufficient information is stored in the volume record, though
an update to a volume record is now no longer shared between related
volumes (volumes come in bundles of three: R/W, R/O and backup).
and the following procedural changes are made:
(1) The fileserver cursor introduced previously is now fleshed out and
used to iterate over fileservers and their addresses.
(2) Volume status is checked during iteration, and the server list is
replaced if a change is detected.
(3) Server status is checked during iteration, and the address list is
replaced if a change is detected.
(4) The abort code is saved into the address list cursor and -ECONNABORTED
returned in afs_make_call() if a remote abort happened rather than
translating the abort into an error message. This allows actions to
be taken depending on the abort code more easily.
(a) If a VMOVED abort is seen then this is handled by rechecking the
volume and restarting the iteration.
(b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is
handled by sleeping for a short period and retrying and/or trying
other servers that might serve that volume. A message is also
displayed once until the condition has cleared.
(c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the
moment.
(d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to
see if it has been deleted; if not, the fileserver is probably
indicating that the volume couldn't be attached and needs
salvaging.
(e) If statfs() sees one of these aborts, it does not sleep, but
rather returns an error, so as not to block the umount program.
(5) The fileserver iteration functions in vnode.c are now merged into
their callers and more heavily macroised around the cursor. vnode.c
is removed.
(6) Operations on a particular vnode are serialised on that vnode because
the server will lock that vnode whilst it operates on it, so a second
op sent will just have to wait.
(7) Fileservers are probed with FS.GetCapabilities before being used.
This is where service upgrade will be done.
(8) A callback interest on a fileserver is set up before an FS operation
is performed and passed through to afs_make_call() so that it can be
set on the vnode if the operation returns a callback. The callback
interest is passed through to afs_iget() also so that it can be set
there too.
In general, record updating is done on an as-needed basis when we try to
access servers, volumes or vnodes rather than offloading it to work items
and special threads.
Notes:
(1) Pre AFS-3.4 servers are no longer supported, though this can be added
back if necessary (AFS-3.4 was released in 1998).
(2) VBUSY is retried forever for the moment at intervals of 1s.
(3) /proc/fs/afs/<cell>/servers no longer exists.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 23:27:50 +08:00
|
|
|
int ret = 0;
|
|
|
|
|
2018-01-02 18:02:19 +08:00
|
|
|
/* There were no intervening changes on the server if the version
|
|
|
|
* number we got back was incremented by exactly 1.
|
|
|
|
*/
|
|
|
|
dir_valid = (d_version_after == d_version_before + 1);
|
|
|
|
|
afs: Overhaul volume and server record caching and fileserver rotation
The current code assumes that volumes and servers are per-cell and are
never shared, but this is not enforced, and, indeed, public cells do exist
that are aliases of each other. Further, an organisation can, say, set up
a public cell and a private cell with overlapping, but not identical, sets
of servers. The difference is purely in the database attached to the VL
servers.
The current code will malfunction if it sees a server in two cells as it
assumes global address -> server record mappings and that each server is in
just one cell.
Further, each server may have multiple addresses - and may have addresses
of different families (IPv4 and IPv6, say).
To this end, the following structural changes are made:
(1) Server record management is overhauled:
(a) Server records are made independent of cell. The namespace keeps
track of them, volume records have lists of them and each vnode
has a server on which its callback interest currently resides.
(b) The cell record no longer keeps a list of servers known to be in
that cell.
(c) The server records are now kept in a flat list because there's no
single address to sort on.
(d) Server records are now keyed by their UUID within the namespace.
(e) The addresses for a server are obtained with the VL.GetAddrsU
rather than with VL.GetEntryByName, using the server's UUID as a
parameter.
(f) Cached server records are garbage collected after a period of
non-use and are counted out of existence before purging is allowed
to complete. This protects the work functions against rmmod.
(g) The servers list is now in /proc/fs/afs/servers.
(2) Volume record management is overhauled:
(a) An RCU-replaceable server list is introduced. This tracks both
servers and their coresponding callback interests.
(b) The superblock is now keyed on cell record and numeric volume ID.
(c) The volume record is now tied to the superblock which mounts it,
and is activated when mounted and deactivated when unmounted.
This makes it easier to handle the cache cookie without causing a
double-use in fscache.
(d) The volume record is loaded from the VLDB using VL.GetEntryByNameU
to get the server UUID list.
(e) The volume name is updated if it is seen to have changed when the
volume is updated (the update is keyed on the volume ID).
(3) The vlocation record is got rid of and VLDB records are no longer
cached. Sufficient information is stored in the volume record, though
an update to a volume record is now no longer shared between related
volumes (volumes come in bundles of three: R/W, R/O and backup).
and the following procedural changes are made:
(1) The fileserver cursor introduced previously is now fleshed out and
used to iterate over fileservers and their addresses.
(2) Volume status is checked during iteration, and the server list is
replaced if a change is detected.
(3) Server status is checked during iteration, and the address list is
replaced if a change is detected.
(4) The abort code is saved into the address list cursor and -ECONNABORTED
returned in afs_make_call() if a remote abort happened rather than
translating the abort into an error message. This allows actions to
be taken depending on the abort code more easily.
(a) If a VMOVED abort is seen then this is handled by rechecking the
volume and restarting the iteration.
(b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is
handled by sleeping for a short period and retrying and/or trying
other servers that might serve that volume. A message is also
displayed once until the condition has cleared.
(c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the
moment.
(d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to
see if it has been deleted; if not, the fileserver is probably
indicating that the volume couldn't be attached and needs
salvaging.
(e) If statfs() sees one of these aborts, it does not sleep, but
rather returns an error, so as not to block the umount program.
(5) The fileserver iteration functions in vnode.c are now merged into
their callers and more heavily macroised around the cursor. vnode.c
is removed.
(6) Operations on a particular vnode are serialised on that vnode because
the server will lock that vnode whilst it operates on it, so a second
op sent will just have to wait.
(7) Fileservers are probed with FS.GetCapabilities before being used.
This is where service upgrade will be done.
(8) A callback interest on a fileserver is set up before an FS operation
is performed and passed through to afs_make_call() so that it can be
set on the vnode if the operation returns a callback. The callback
interest is passed through to afs_iget() also so that it can be set
there too.
In general, record updating is done on an as-needed basis when we try to
access servers, volumes or vnodes rather than offloading it to work items
and special threads.
Notes:
(1) Pre AFS-3.4 servers are no longer supported, though this can be added
back if necessary (AFS-3.4 was released in 1998).
(2) VBUSY is retried forever for the moment at intervals of 1s.
(3) /proc/fs/afs/<cell>/servers no longer exists.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 23:27:50 +08:00
|
|
|
if (d_really_is_positive(dentry)) {
|
|
|
|
struct afs_vnode *vnode = AFS_FS_I(d_inode(dentry));
|
|
|
|
|
2018-01-02 18:02:19 +08:00
|
|
|
if (dir_valid) {
|
|
|
|
drop_nlink(&vnode->vfs_inode);
|
|
|
|
if (vnode->vfs_inode.i_nlink == 0) {
|
|
|
|
set_bit(AFS_VNODE_DELETED, &vnode->flags);
|
|
|
|
clear_bit(AFS_VNODE_CB_PROMISED, &vnode->flags);
|
|
|
|
}
|
afs: Overhaul volume and server record caching and fileserver rotation
The current code assumes that volumes and servers are per-cell and are
never shared, but this is not enforced, and, indeed, public cells do exist
that are aliases of each other. Further, an organisation can, say, set up
a public cell and a private cell with overlapping, but not identical, sets
of servers. The difference is purely in the database attached to the VL
servers.
The current code will malfunction if it sees a server in two cells as it
assumes global address -> server record mappings and that each server is in
just one cell.
Further, each server may have multiple addresses - and may have addresses
of different families (IPv4 and IPv6, say).
To this end, the following structural changes are made:
(1) Server record management is overhauled:
(a) Server records are made independent of cell. The namespace keeps
track of them, volume records have lists of them and each vnode
has a server on which its callback interest currently resides.
(b) The cell record no longer keeps a list of servers known to be in
that cell.
(c) The server records are now kept in a flat list because there's no
single address to sort on.
(d) Server records are now keyed by their UUID within the namespace.
(e) The addresses for a server are obtained with the VL.GetAddrsU
rather than with VL.GetEntryByName, using the server's UUID as a
parameter.
(f) Cached server records are garbage collected after a period of
non-use and are counted out of existence before purging is allowed
to complete. This protects the work functions against rmmod.
(g) The servers list is now in /proc/fs/afs/servers.
(2) Volume record management is overhauled:
(a) An RCU-replaceable server list is introduced. This tracks both
servers and their coresponding callback interests.
(b) The superblock is now keyed on cell record and numeric volume ID.
(c) The volume record is now tied to the superblock which mounts it,
and is activated when mounted and deactivated when unmounted.
This makes it easier to handle the cache cookie without causing a
double-use in fscache.
(d) The volume record is loaded from the VLDB using VL.GetEntryByNameU
to get the server UUID list.
(e) The volume name is updated if it is seen to have changed when the
volume is updated (the update is keyed on the volume ID).
(3) The vlocation record is got rid of and VLDB records are no longer
cached. Sufficient information is stored in the volume record, though
an update to a volume record is now no longer shared between related
volumes (volumes come in bundles of three: R/W, R/O and backup).
and the following procedural changes are made:
(1) The fileserver cursor introduced previously is now fleshed out and
used to iterate over fileservers and their addresses.
(2) Volume status is checked during iteration, and the server list is
replaced if a change is detected.
(3) Server status is checked during iteration, and the address list is
replaced if a change is detected.
(4) The abort code is saved into the address list cursor and -ECONNABORTED
returned in afs_make_call() if a remote abort happened rather than
translating the abort into an error message. This allows actions to
be taken depending on the abort code more easily.
(a) If a VMOVED abort is seen then this is handled by rechecking the
volume and restarting the iteration.
(b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is
handled by sleeping for a short period and retrying and/or trying
other servers that might serve that volume. A message is also
displayed once until the condition has cleared.
(c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the
moment.
(d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to
see if it has been deleted; if not, the fileserver is probably
indicating that the volume couldn't be attached and needs
salvaging.
(e) If statfs() sees one of these aborts, it does not sleep, but
rather returns an error, so as not to block the umount program.
(5) The fileserver iteration functions in vnode.c are now merged into
their callers and more heavily macroised around the cursor. vnode.c
is removed.
(6) Operations on a particular vnode are serialised on that vnode because
the server will lock that vnode whilst it operates on it, so a second
op sent will just have to wait.
(7) Fileservers are probed with FS.GetCapabilities before being used.
This is where service upgrade will be done.
(8) A callback interest on a fileserver is set up before an FS operation
is performed and passed through to afs_make_call() so that it can be
set on the vnode if the operation returns a callback. The callback
interest is passed through to afs_iget() also so that it can be set
there too.
In general, record updating is done on an as-needed basis when we try to
access servers, volumes or vnodes rather than offloading it to work items
and special threads.
Notes:
(1) Pre AFS-3.4 servers are no longer supported, though this can be added
back if necessary (AFS-3.4 was released in 1998).
(2) VBUSY is retried forever for the moment at intervals of 1s.
(3) /proc/fs/afs/<cell>/servers no longer exists.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 23:27:50 +08:00
|
|
|
ret = 0;
|
2018-01-02 18:02:19 +08:00
|
|
|
} else {
|
|
|
|
clear_bit(AFS_VNODE_CB_PROMISED, &vnode->flags);
|
|
|
|
|
|
|
|
if (test_bit(AFS_VNODE_DELETED, &vnode->flags))
|
|
|
|
kdebug("AFS_VNODE_DELETED");
|
|
|
|
|
|
|
|
ret = afs_validate(vnode, key);
|
|
|
|
if (ret == -ESTALE)
|
|
|
|
ret = 0;
|
|
|
|
}
|
afs: Overhaul volume and server record caching and fileserver rotation
The current code assumes that volumes and servers are per-cell and are
never shared, but this is not enforced, and, indeed, public cells do exist
that are aliases of each other. Further, an organisation can, say, set up
a public cell and a private cell with overlapping, but not identical, sets
of servers. The difference is purely in the database attached to the VL
servers.
The current code will malfunction if it sees a server in two cells as it
assumes global address -> server record mappings and that each server is in
just one cell.
Further, each server may have multiple addresses - and may have addresses
of different families (IPv4 and IPv6, say).
To this end, the following structural changes are made:
(1) Server record management is overhauled:
(a) Server records are made independent of cell. The namespace keeps
track of them, volume records have lists of them and each vnode
has a server on which its callback interest currently resides.
(b) The cell record no longer keeps a list of servers known to be in
that cell.
(c) The server records are now kept in a flat list because there's no
single address to sort on.
(d) Server records are now keyed by their UUID within the namespace.
(e) The addresses for a server are obtained with the VL.GetAddrsU
rather than with VL.GetEntryByName, using the server's UUID as a
parameter.
(f) Cached server records are garbage collected after a period of
non-use and are counted out of existence before purging is allowed
to complete. This protects the work functions against rmmod.
(g) The servers list is now in /proc/fs/afs/servers.
(2) Volume record management is overhauled:
(a) An RCU-replaceable server list is introduced. This tracks both
servers and their coresponding callback interests.
(b) The superblock is now keyed on cell record and numeric volume ID.
(c) The volume record is now tied to the superblock which mounts it,
and is activated when mounted and deactivated when unmounted.
This makes it easier to handle the cache cookie without causing a
double-use in fscache.
(d) The volume record is loaded from the VLDB using VL.GetEntryByNameU
to get the server UUID list.
(e) The volume name is updated if it is seen to have changed when the
volume is updated (the update is keyed on the volume ID).
(3) The vlocation record is got rid of and VLDB records are no longer
cached. Sufficient information is stored in the volume record, though
an update to a volume record is now no longer shared between related
volumes (volumes come in bundles of three: R/W, R/O and backup).
and the following procedural changes are made:
(1) The fileserver cursor introduced previously is now fleshed out and
used to iterate over fileservers and their addresses.
(2) Volume status is checked during iteration, and the server list is
replaced if a change is detected.
(3) Server status is checked during iteration, and the address list is
replaced if a change is detected.
(4) The abort code is saved into the address list cursor and -ECONNABORTED
returned in afs_make_call() if a remote abort happened rather than
translating the abort into an error message. This allows actions to
be taken depending on the abort code more easily.
(a) If a VMOVED abort is seen then this is handled by rechecking the
volume and restarting the iteration.
(b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is
handled by sleeping for a short period and retrying and/or trying
other servers that might serve that volume. A message is also
displayed once until the condition has cleared.
(c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the
moment.
(d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to
see if it has been deleted; if not, the fileserver is probably
indicating that the volume couldn't be attached and needs
salvaging.
(e) If statfs() sees one of these aborts, it does not sleep, but
rather returns an error, so as not to block the umount program.
(5) The fileserver iteration functions in vnode.c are now merged into
their callers and more heavily macroised around the cursor. vnode.c
is removed.
(6) Operations on a particular vnode are serialised on that vnode because
the server will lock that vnode whilst it operates on it, so a second
op sent will just have to wait.
(7) Fileservers are probed with FS.GetCapabilities before being used.
This is where service upgrade will be done.
(8) A callback interest on a fileserver is set up before an FS operation
is performed and passed through to afs_make_call() so that it can be
set on the vnode if the operation returns a callback. The callback
interest is passed through to afs_iget() also so that it can be set
there too.
In general, record updating is done on an as-needed basis when we try to
access servers, volumes or vnodes rather than offloading it to work items
and special threads.
Notes:
(1) Pre AFS-3.4 servers are no longer supported, though this can be added
back if necessary (AFS-3.4 was released in 1998).
(2) VBUSY is retried forever for the moment at intervals of 1s.
(3) /proc/fs/afs/<cell>/servers no longer exists.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 23:27:50 +08:00
|
|
|
_debug("nlink %d [val %d]", vnode->vfs_inode.i_nlink, ret);
|
|
|
|
}
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Remove a file or symlink from an AFS filesystem.
|
2007-04-27 06:59:35 +08:00
|
|
|
*/
|
|
|
|
static int afs_unlink(struct inode *dir, struct dentry *dentry)
|
|
|
|
{
|
afs: Overhaul volume and server record caching and fileserver rotation
The current code assumes that volumes and servers are per-cell and are
never shared, but this is not enforced, and, indeed, public cells do exist
that are aliases of each other. Further, an organisation can, say, set up
a public cell and a private cell with overlapping, but not identical, sets
of servers. The difference is purely in the database attached to the VL
servers.
The current code will malfunction if it sees a server in two cells as it
assumes global address -> server record mappings and that each server is in
just one cell.
Further, each server may have multiple addresses - and may have addresses
of different families (IPv4 and IPv6, say).
To this end, the following structural changes are made:
(1) Server record management is overhauled:
(a) Server records are made independent of cell. The namespace keeps
track of them, volume records have lists of them and each vnode
has a server on which its callback interest currently resides.
(b) The cell record no longer keeps a list of servers known to be in
that cell.
(c) The server records are now kept in a flat list because there's no
single address to sort on.
(d) Server records are now keyed by their UUID within the namespace.
(e) The addresses for a server are obtained with the VL.GetAddrsU
rather than with VL.GetEntryByName, using the server's UUID as a
parameter.
(f) Cached server records are garbage collected after a period of
non-use and are counted out of existence before purging is allowed
to complete. This protects the work functions against rmmod.
(g) The servers list is now in /proc/fs/afs/servers.
(2) Volume record management is overhauled:
(a) An RCU-replaceable server list is introduced. This tracks both
servers and their coresponding callback interests.
(b) The superblock is now keyed on cell record and numeric volume ID.
(c) The volume record is now tied to the superblock which mounts it,
and is activated when mounted and deactivated when unmounted.
This makes it easier to handle the cache cookie without causing a
double-use in fscache.
(d) The volume record is loaded from the VLDB using VL.GetEntryByNameU
to get the server UUID list.
(e) The volume name is updated if it is seen to have changed when the
volume is updated (the update is keyed on the volume ID).
(3) The vlocation record is got rid of and VLDB records are no longer
cached. Sufficient information is stored in the volume record, though
an update to a volume record is now no longer shared between related
volumes (volumes come in bundles of three: R/W, R/O and backup).
and the following procedural changes are made:
(1) The fileserver cursor introduced previously is now fleshed out and
used to iterate over fileservers and their addresses.
(2) Volume status is checked during iteration, and the server list is
replaced if a change is detected.
(3) Server status is checked during iteration, and the address list is
replaced if a change is detected.
(4) The abort code is saved into the address list cursor and -ECONNABORTED
returned in afs_make_call() if a remote abort happened rather than
translating the abort into an error message. This allows actions to
be taken depending on the abort code more easily.
(a) If a VMOVED abort is seen then this is handled by rechecking the
volume and restarting the iteration.
(b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is
handled by sleeping for a short period and retrying and/or trying
other servers that might serve that volume. A message is also
displayed once until the condition has cleared.
(c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the
moment.
(d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to
see if it has been deleted; if not, the fileserver is probably
indicating that the volume couldn't be attached and needs
salvaging.
(e) If statfs() sees one of these aborts, it does not sleep, but
rather returns an error, so as not to block the umount program.
(5) The fileserver iteration functions in vnode.c are now merged into
their callers and more heavily macroised around the cursor. vnode.c
is removed.
(6) Operations on a particular vnode are serialised on that vnode because
the server will lock that vnode whilst it operates on it, so a second
op sent will just have to wait.
(7) Fileservers are probed with FS.GetCapabilities before being used.
This is where service upgrade will be done.
(8) A callback interest on a fileserver is set up before an FS operation
is performed and passed through to afs_make_call() so that it can be
set on the vnode if the operation returns a callback. The callback
interest is passed through to afs_iget() also so that it can be set
there too.
In general, record updating is done on an as-needed basis when we try to
access servers, volumes or vnodes rather than offloading it to work items
and special threads.
Notes:
(1) Pre AFS-3.4 servers are no longer supported, though this can be added
back if necessary (AFS-3.4 was released in 1998).
(2) VBUSY is retried forever for the moment at intervals of 1s.
(3) /proc/fs/afs/<cell>/servers no longer exists.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 23:27:50 +08:00
|
|
|
struct afs_fs_cursor fc;
|
|
|
|
struct afs_vnode *dvnode = AFS_FS_I(dir), *vnode;
|
2007-04-27 06:59:35 +08:00
|
|
|
struct key *key;
|
2018-01-02 18:02:19 +08:00
|
|
|
unsigned long d_version = (unsigned long)dentry->d_fsdata;
|
2007-04-27 06:59:35 +08:00
|
|
|
int ret;
|
|
|
|
|
2014-10-22 08:11:25 +08:00
|
|
|
_enter("{%x:%u},{%pd}",
|
|
|
|
dvnode->fid.vid, dvnode->fid.vnode, dentry);
|
2007-04-27 06:59:35 +08:00
|
|
|
|
2007-05-11 13:22:20 +08:00
|
|
|
if (dentry->d_name.len >= AFSNAMEMAX)
|
afs: Overhaul volume and server record caching and fileserver rotation
The current code assumes that volumes and servers are per-cell and are
never shared, but this is not enforced, and, indeed, public cells do exist
that are aliases of each other. Further, an organisation can, say, set up
a public cell and a private cell with overlapping, but not identical, sets
of servers. The difference is purely in the database attached to the VL
servers.
The current code will malfunction if it sees a server in two cells as it
assumes global address -> server record mappings and that each server is in
just one cell.
Further, each server may have multiple addresses - and may have addresses
of different families (IPv4 and IPv6, say).
To this end, the following structural changes are made:
(1) Server record management is overhauled:
(a) Server records are made independent of cell. The namespace keeps
track of them, volume records have lists of them and each vnode
has a server on which its callback interest currently resides.
(b) The cell record no longer keeps a list of servers known to be in
that cell.
(c) The server records are now kept in a flat list because there's no
single address to sort on.
(d) Server records are now keyed by their UUID within the namespace.
(e) The addresses for a server are obtained with the VL.GetAddrsU
rather than with VL.GetEntryByName, using the server's UUID as a
parameter.
(f) Cached server records are garbage collected after a period of
non-use and are counted out of existence before purging is allowed
to complete. This protects the work functions against rmmod.
(g) The servers list is now in /proc/fs/afs/servers.
(2) Volume record management is overhauled:
(a) An RCU-replaceable server list is introduced. This tracks both
servers and their coresponding callback interests.
(b) The superblock is now keyed on cell record and numeric volume ID.
(c) The volume record is now tied to the superblock which mounts it,
and is activated when mounted and deactivated when unmounted.
This makes it easier to handle the cache cookie without causing a
double-use in fscache.
(d) The volume record is loaded from the VLDB using VL.GetEntryByNameU
to get the server UUID list.
(e) The volume name is updated if it is seen to have changed when the
volume is updated (the update is keyed on the volume ID).
(3) The vlocation record is got rid of and VLDB records are no longer
cached. Sufficient information is stored in the volume record, though
an update to a volume record is now no longer shared between related
volumes (volumes come in bundles of three: R/W, R/O and backup).
and the following procedural changes are made:
(1) The fileserver cursor introduced previously is now fleshed out and
used to iterate over fileservers and their addresses.
(2) Volume status is checked during iteration, and the server list is
replaced if a change is detected.
(3) Server status is checked during iteration, and the address list is
replaced if a change is detected.
(4) The abort code is saved into the address list cursor and -ECONNABORTED
returned in afs_make_call() if a remote abort happened rather than
translating the abort into an error message. This allows actions to
be taken depending on the abort code more easily.
(a) If a VMOVED abort is seen then this is handled by rechecking the
volume and restarting the iteration.
(b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is
handled by sleeping for a short period and retrying and/or trying
other servers that might serve that volume. A message is also
displayed once until the condition has cleared.
(c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the
moment.
(d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to
see if it has been deleted; if not, the fileserver is probably
indicating that the volume couldn't be attached and needs
salvaging.
(e) If statfs() sees one of these aborts, it does not sleep, but
rather returns an error, so as not to block the umount program.
(5) The fileserver iteration functions in vnode.c are now merged into
their callers and more heavily macroised around the cursor. vnode.c
is removed.
(6) Operations on a particular vnode are serialised on that vnode because
the server will lock that vnode whilst it operates on it, so a second
op sent will just have to wait.
(7) Fileservers are probed with FS.GetCapabilities before being used.
This is where service upgrade will be done.
(8) A callback interest on a fileserver is set up before an FS operation
is performed and passed through to afs_make_call() so that it can be
set on the vnode if the operation returns a callback. The callback
interest is passed through to afs_iget() also so that it can be set
there too.
In general, record updating is done on an as-needed basis when we try to
access servers, volumes or vnodes rather than offloading it to work items
and special threads.
Notes:
(1) Pre AFS-3.4 servers are no longer supported, though this can be added
back if necessary (AFS-3.4 was released in 1998).
(2) VBUSY is retried forever for the moment at intervals of 1s.
(3) /proc/fs/afs/<cell>/servers no longer exists.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 23:27:50 +08:00
|
|
|
return -ENAMETOOLONG;
|
2007-04-27 06:59:35 +08:00
|
|
|
|
|
|
|
key = afs_request_key(dvnode->volume->cell);
|
|
|
|
if (IS_ERR(key)) {
|
|
|
|
ret = PTR_ERR(key);
|
|
|
|
goto error;
|
|
|
|
}
|
|
|
|
|
afs: Overhaul volume and server record caching and fileserver rotation
The current code assumes that volumes and servers are per-cell and are
never shared, but this is not enforced, and, indeed, public cells do exist
that are aliases of each other. Further, an organisation can, say, set up
a public cell and a private cell with overlapping, but not identical, sets
of servers. The difference is purely in the database attached to the VL
servers.
The current code will malfunction if it sees a server in two cells as it
assumes global address -> server record mappings and that each server is in
just one cell.
Further, each server may have multiple addresses - and may have addresses
of different families (IPv4 and IPv6, say).
To this end, the following structural changes are made:
(1) Server record management is overhauled:
(a) Server records are made independent of cell. The namespace keeps
track of them, volume records have lists of them and each vnode
has a server on which its callback interest currently resides.
(b) The cell record no longer keeps a list of servers known to be in
that cell.
(c) The server records are now kept in a flat list because there's no
single address to sort on.
(d) Server records are now keyed by their UUID within the namespace.
(e) The addresses for a server are obtained with the VL.GetAddrsU
rather than with VL.GetEntryByName, using the server's UUID as a
parameter.
(f) Cached server records are garbage collected after a period of
non-use and are counted out of existence before purging is allowed
to complete. This protects the work functions against rmmod.
(g) The servers list is now in /proc/fs/afs/servers.
(2) Volume record management is overhauled:
(a) An RCU-replaceable server list is introduced. This tracks both
servers and their coresponding callback interests.
(b) The superblock is now keyed on cell record and numeric volume ID.
(c) The volume record is now tied to the superblock which mounts it,
and is activated when mounted and deactivated when unmounted.
This makes it easier to handle the cache cookie without causing a
double-use in fscache.
(d) The volume record is loaded from the VLDB using VL.GetEntryByNameU
to get the server UUID list.
(e) The volume name is updated if it is seen to have changed when the
volume is updated (the update is keyed on the volume ID).
(3) The vlocation record is got rid of and VLDB records are no longer
cached. Sufficient information is stored in the volume record, though
an update to a volume record is now no longer shared between related
volumes (volumes come in bundles of three: R/W, R/O and backup).
and the following procedural changes are made:
(1) The fileserver cursor introduced previously is now fleshed out and
used to iterate over fileservers and their addresses.
(2) Volume status is checked during iteration, and the server list is
replaced if a change is detected.
(3) Server status is checked during iteration, and the address list is
replaced if a change is detected.
(4) The abort code is saved into the address list cursor and -ECONNABORTED
returned in afs_make_call() if a remote abort happened rather than
translating the abort into an error message. This allows actions to
be taken depending on the abort code more easily.
(a) If a VMOVED abort is seen then this is handled by rechecking the
volume and restarting the iteration.
(b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is
handled by sleeping for a short period and retrying and/or trying
other servers that might serve that volume. A message is also
displayed once until the condition has cleared.
(c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the
moment.
(d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to
see if it has been deleted; if not, the fileserver is probably
indicating that the volume couldn't be attached and needs
salvaging.
(e) If statfs() sees one of these aborts, it does not sleep, but
rather returns an error, so as not to block the umount program.
(5) The fileserver iteration functions in vnode.c are now merged into
their callers and more heavily macroised around the cursor. vnode.c
is removed.
(6) Operations on a particular vnode are serialised on that vnode because
the server will lock that vnode whilst it operates on it, so a second
op sent will just have to wait.
(7) Fileservers are probed with FS.GetCapabilities before being used.
This is where service upgrade will be done.
(8) A callback interest on a fileserver is set up before an FS operation
is performed and passed through to afs_make_call() so that it can be
set on the vnode if the operation returns a callback. The callback
interest is passed through to afs_iget() also so that it can be set
there too.
In general, record updating is done on an as-needed basis when we try to
access servers, volumes or vnodes rather than offloading it to work items
and special threads.
Notes:
(1) Pre AFS-3.4 servers are no longer supported, though this can be added
back if necessary (AFS-3.4 was released in 1998).
(2) VBUSY is retried forever for the moment at intervals of 1s.
(3) /proc/fs/afs/<cell>/servers no longer exists.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 23:27:50 +08:00
|
|
|
/* Try to make sure we have a callback promise on the victim. */
|
2015-03-18 06:25:59 +08:00
|
|
|
if (d_really_is_positive(dentry)) {
|
|
|
|
vnode = AFS_FS_I(d_inode(dentry));
|
2007-04-27 06:59:35 +08:00
|
|
|
ret = afs_validate(vnode, key);
|
|
|
|
if (ret < 0)
|
afs: Overhaul volume and server record caching and fileserver rotation
The current code assumes that volumes and servers are per-cell and are
never shared, but this is not enforced, and, indeed, public cells do exist
that are aliases of each other. Further, an organisation can, say, set up
a public cell and a private cell with overlapping, but not identical, sets
of servers. The difference is purely in the database attached to the VL
servers.
The current code will malfunction if it sees a server in two cells as it
assumes global address -> server record mappings and that each server is in
just one cell.
Further, each server may have multiple addresses - and may have addresses
of different families (IPv4 and IPv6, say).
To this end, the following structural changes are made:
(1) Server record management is overhauled:
(a) Server records are made independent of cell. The namespace keeps
track of them, volume records have lists of them and each vnode
has a server on which its callback interest currently resides.
(b) The cell record no longer keeps a list of servers known to be in
that cell.
(c) The server records are now kept in a flat list because there's no
single address to sort on.
(d) Server records are now keyed by their UUID within the namespace.
(e) The addresses for a server are obtained with the VL.GetAddrsU
rather than with VL.GetEntryByName, using the server's UUID as a
parameter.
(f) Cached server records are garbage collected after a period of
non-use and are counted out of existence before purging is allowed
to complete. This protects the work functions against rmmod.
(g) The servers list is now in /proc/fs/afs/servers.
(2) Volume record management is overhauled:
(a) An RCU-replaceable server list is introduced. This tracks both
servers and their coresponding callback interests.
(b) The superblock is now keyed on cell record and numeric volume ID.
(c) The volume record is now tied to the superblock which mounts it,
and is activated when mounted and deactivated when unmounted.
This makes it easier to handle the cache cookie without causing a
double-use in fscache.
(d) The volume record is loaded from the VLDB using VL.GetEntryByNameU
to get the server UUID list.
(e) The volume name is updated if it is seen to have changed when the
volume is updated (the update is keyed on the volume ID).
(3) The vlocation record is got rid of and VLDB records are no longer
cached. Sufficient information is stored in the volume record, though
an update to a volume record is now no longer shared between related
volumes (volumes come in bundles of three: R/W, R/O and backup).
and the following procedural changes are made:
(1) The fileserver cursor introduced previously is now fleshed out and
used to iterate over fileservers and their addresses.
(2) Volume status is checked during iteration, and the server list is
replaced if a change is detected.
(3) Server status is checked during iteration, and the address list is
replaced if a change is detected.
(4) The abort code is saved into the address list cursor and -ECONNABORTED
returned in afs_make_call() if a remote abort happened rather than
translating the abort into an error message. This allows actions to
be taken depending on the abort code more easily.
(a) If a VMOVED abort is seen then this is handled by rechecking the
volume and restarting the iteration.
(b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is
handled by sleeping for a short period and retrying and/or trying
other servers that might serve that volume. A message is also
displayed once until the condition has cleared.
(c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the
moment.
(d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to
see if it has been deleted; if not, the fileserver is probably
indicating that the volume couldn't be attached and needs
salvaging.
(e) If statfs() sees one of these aborts, it does not sleep, but
rather returns an error, so as not to block the umount program.
(5) The fileserver iteration functions in vnode.c are now merged into
their callers and more heavily macroised around the cursor. vnode.c
is removed.
(6) Operations on a particular vnode are serialised on that vnode because
the server will lock that vnode whilst it operates on it, so a second
op sent will just have to wait.
(7) Fileservers are probed with FS.GetCapabilities before being used.
This is where service upgrade will be done.
(8) A callback interest on a fileserver is set up before an FS operation
is performed and passed through to afs_make_call() so that it can be
set on the vnode if the operation returns a callback. The callback
interest is passed through to afs_iget() also so that it can be set
there too.
In general, record updating is done on an as-needed basis when we try to
access servers, volumes or vnodes rather than offloading it to work items
and special threads.
Notes:
(1) Pre AFS-3.4 servers are no longer supported, though this can be added
back if necessary (AFS-3.4 was released in 1998).
(2) VBUSY is retried forever for the moment at intervals of 1s.
(3) /proc/fs/afs/<cell>/servers no longer exists.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 23:27:50 +08:00
|
|
|
goto error_key;
|
2007-04-27 06:59:35 +08:00
|
|
|
}
|
|
|
|
|
afs: Overhaul volume and server record caching and fileserver rotation
The current code assumes that volumes and servers are per-cell and are
never shared, but this is not enforced, and, indeed, public cells do exist
that are aliases of each other. Further, an organisation can, say, set up
a public cell and a private cell with overlapping, but not identical, sets
of servers. The difference is purely in the database attached to the VL
servers.
The current code will malfunction if it sees a server in two cells as it
assumes global address -> server record mappings and that each server is in
just one cell.
Further, each server may have multiple addresses - and may have addresses
of different families (IPv4 and IPv6, say).
To this end, the following structural changes are made:
(1) Server record management is overhauled:
(a) Server records are made independent of cell. The namespace keeps
track of them, volume records have lists of them and each vnode
has a server on which its callback interest currently resides.
(b) The cell record no longer keeps a list of servers known to be in
that cell.
(c) The server records are now kept in a flat list because there's no
single address to sort on.
(d) Server records are now keyed by their UUID within the namespace.
(e) The addresses for a server are obtained with the VL.GetAddrsU
rather than with VL.GetEntryByName, using the server's UUID as a
parameter.
(f) Cached server records are garbage collected after a period of
non-use and are counted out of existence before purging is allowed
to complete. This protects the work functions against rmmod.
(g) The servers list is now in /proc/fs/afs/servers.
(2) Volume record management is overhauled:
(a) An RCU-replaceable server list is introduced. This tracks both
servers and their coresponding callback interests.
(b) The superblock is now keyed on cell record and numeric volume ID.
(c) The volume record is now tied to the superblock which mounts it,
and is activated when mounted and deactivated when unmounted.
This makes it easier to handle the cache cookie without causing a
double-use in fscache.
(d) The volume record is loaded from the VLDB using VL.GetEntryByNameU
to get the server UUID list.
(e) The volume name is updated if it is seen to have changed when the
volume is updated (the update is keyed on the volume ID).
(3) The vlocation record is got rid of and VLDB records are no longer
cached. Sufficient information is stored in the volume record, though
an update to a volume record is now no longer shared between related
volumes (volumes come in bundles of three: R/W, R/O and backup).
and the following procedural changes are made:
(1) The fileserver cursor introduced previously is now fleshed out and
used to iterate over fileservers and their addresses.
(2) Volume status is checked during iteration, and the server list is
replaced if a change is detected.
(3) Server status is checked during iteration, and the address list is
replaced if a change is detected.
(4) The abort code is saved into the address list cursor and -ECONNABORTED
returned in afs_make_call() if a remote abort happened rather than
translating the abort into an error message. This allows actions to
be taken depending on the abort code more easily.
(a) If a VMOVED abort is seen then this is handled by rechecking the
volume and restarting the iteration.
(b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is
handled by sleeping for a short period and retrying and/or trying
other servers that might serve that volume. A message is also
displayed once until the condition has cleared.
(c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the
moment.
(d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to
see if it has been deleted; if not, the fileserver is probably
indicating that the volume couldn't be attached and needs
salvaging.
(e) If statfs() sees one of these aborts, it does not sleep, but
rather returns an error, so as not to block the umount program.
(5) The fileserver iteration functions in vnode.c are now merged into
their callers and more heavily macroised around the cursor. vnode.c
is removed.
(6) Operations on a particular vnode are serialised on that vnode because
the server will lock that vnode whilst it operates on it, so a second
op sent will just have to wait.
(7) Fileservers are probed with FS.GetCapabilities before being used.
This is where service upgrade will be done.
(8) A callback interest on a fileserver is set up before an FS operation
is performed and passed through to afs_make_call() so that it can be
set on the vnode if the operation returns a callback. The callback
interest is passed through to afs_iget() also so that it can be set
there too.
In general, record updating is done on an as-needed basis when we try to
access servers, volumes or vnodes rather than offloading it to work items
and special threads.
Notes:
(1) Pre AFS-3.4 servers are no longer supported, though this can be added
back if necessary (AFS-3.4 was released in 1998).
(2) VBUSY is retried forever for the moment at intervals of 1s.
(3) /proc/fs/afs/<cell>/servers no longer exists.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 23:27:50 +08:00
|
|
|
ret = -ERESTARTSYS;
|
|
|
|
if (afs_begin_vnode_operation(&fc, dvnode, key)) {
|
|
|
|
while (afs_select_fileserver(&fc)) {
|
|
|
|
fc.cb_break = dvnode->cb_break + dvnode->cb_s_break;
|
|
|
|
afs_fs_remove(&fc, dentry->d_name.name, false);
|
|
|
|
}
|
2007-04-27 06:59:35 +08:00
|
|
|
|
afs: Overhaul volume and server record caching and fileserver rotation
The current code assumes that volumes and servers are per-cell and are
never shared, but this is not enforced, and, indeed, public cells do exist
that are aliases of each other. Further, an organisation can, say, set up
a public cell and a private cell with overlapping, but not identical, sets
of servers. The difference is purely in the database attached to the VL
servers.
The current code will malfunction if it sees a server in two cells as it
assumes global address -> server record mappings and that each server is in
just one cell.
Further, each server may have multiple addresses - and may have addresses
of different families (IPv4 and IPv6, say).
To this end, the following structural changes are made:
(1) Server record management is overhauled:
(a) Server records are made independent of cell. The namespace keeps
track of them, volume records have lists of them and each vnode
has a server on which its callback interest currently resides.
(b) The cell record no longer keeps a list of servers known to be in
that cell.
(c) The server records are now kept in a flat list because there's no
single address to sort on.
(d) Server records are now keyed by their UUID within the namespace.
(e) The addresses for a server are obtained with the VL.GetAddrsU
rather than with VL.GetEntryByName, using the server's UUID as a
parameter.
(f) Cached server records are garbage collected after a period of
non-use and are counted out of existence before purging is allowed
to complete. This protects the work functions against rmmod.
(g) The servers list is now in /proc/fs/afs/servers.
(2) Volume record management is overhauled:
(a) An RCU-replaceable server list is introduced. This tracks both
servers and their coresponding callback interests.
(b) The superblock is now keyed on cell record and numeric volume ID.
(c) The volume record is now tied to the superblock which mounts it,
and is activated when mounted and deactivated when unmounted.
This makes it easier to handle the cache cookie without causing a
double-use in fscache.
(d) The volume record is loaded from the VLDB using VL.GetEntryByNameU
to get the server UUID list.
(e) The volume name is updated if it is seen to have changed when the
volume is updated (the update is keyed on the volume ID).
(3) The vlocation record is got rid of and VLDB records are no longer
cached. Sufficient information is stored in the volume record, though
an update to a volume record is now no longer shared between related
volumes (volumes come in bundles of three: R/W, R/O and backup).
and the following procedural changes are made:
(1) The fileserver cursor introduced previously is now fleshed out and
used to iterate over fileservers and their addresses.
(2) Volume status is checked during iteration, and the server list is
replaced if a change is detected.
(3) Server status is checked during iteration, and the address list is
replaced if a change is detected.
(4) The abort code is saved into the address list cursor and -ECONNABORTED
returned in afs_make_call() if a remote abort happened rather than
translating the abort into an error message. This allows actions to
be taken depending on the abort code more easily.
(a) If a VMOVED abort is seen then this is handled by rechecking the
volume and restarting the iteration.
(b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is
handled by sleeping for a short period and retrying and/or trying
other servers that might serve that volume. A message is also
displayed once until the condition has cleared.
(c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the
moment.
(d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to
see if it has been deleted; if not, the fileserver is probably
indicating that the volume couldn't be attached and needs
salvaging.
(e) If statfs() sees one of these aborts, it does not sleep, but
rather returns an error, so as not to block the umount program.
(5) The fileserver iteration functions in vnode.c are now merged into
their callers and more heavily macroised around the cursor. vnode.c
is removed.
(6) Operations on a particular vnode are serialised on that vnode because
the server will lock that vnode whilst it operates on it, so a second
op sent will just have to wait.
(7) Fileservers are probed with FS.GetCapabilities before being used.
This is where service upgrade will be done.
(8) A callback interest on a fileserver is set up before an FS operation
is performed and passed through to afs_make_call() so that it can be
set on the vnode if the operation returns a callback. The callback
interest is passed through to afs_iget() also so that it can be set
there too.
In general, record updating is done on an as-needed basis when we try to
access servers, volumes or vnodes rather than offloading it to work items
and special threads.
Notes:
(1) Pre AFS-3.4 servers are no longer supported, though this can be added
back if necessary (AFS-3.4 was released in 1998).
(2) VBUSY is retried forever for the moment at intervals of 1s.
(3) /proc/fs/afs/<cell>/servers no longer exists.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 23:27:50 +08:00
|
|
|
afs_vnode_commit_status(&fc, dvnode, fc.cb_break);
|
|
|
|
ret = afs_end_vnode_operation(&fc);
|
|
|
|
if (ret == 0)
|
2018-01-02 18:02:19 +08:00
|
|
|
ret = afs_dir_remove_link(
|
|
|
|
dentry, key, d_version,
|
|
|
|
(unsigned long)dvnode->status.data_version);
|
2007-04-27 06:59:35 +08:00
|
|
|
}
|
|
|
|
|
afs: Overhaul volume and server record caching and fileserver rotation
The current code assumes that volumes and servers are per-cell and are
never shared, but this is not enforced, and, indeed, public cells do exist
that are aliases of each other. Further, an organisation can, say, set up
a public cell and a private cell with overlapping, but not identical, sets
of servers. The difference is purely in the database attached to the VL
servers.
The current code will malfunction if it sees a server in two cells as it
assumes global address -> server record mappings and that each server is in
just one cell.
Further, each server may have multiple addresses - and may have addresses
of different families (IPv4 and IPv6, say).
To this end, the following structural changes are made:
(1) Server record management is overhauled:
(a) Server records are made independent of cell. The namespace keeps
track of them, volume records have lists of them and each vnode
has a server on which its callback interest currently resides.
(b) The cell record no longer keeps a list of servers known to be in
that cell.
(c) The server records are now kept in a flat list because there's no
single address to sort on.
(d) Server records are now keyed by their UUID within the namespace.
(e) The addresses for a server are obtained with the VL.GetAddrsU
rather than with VL.GetEntryByName, using the server's UUID as a
parameter.
(f) Cached server records are garbage collected after a period of
non-use and are counted out of existence before purging is allowed
to complete. This protects the work functions against rmmod.
(g) The servers list is now in /proc/fs/afs/servers.
(2) Volume record management is overhauled:
(a) An RCU-replaceable server list is introduced. This tracks both
servers and their coresponding callback interests.
(b) The superblock is now keyed on cell record and numeric volume ID.
(c) The volume record is now tied to the superblock which mounts it,
and is activated when mounted and deactivated when unmounted.
This makes it easier to handle the cache cookie without causing a
double-use in fscache.
(d) The volume record is loaded from the VLDB using VL.GetEntryByNameU
to get the server UUID list.
(e) The volume name is updated if it is seen to have changed when the
volume is updated (the update is keyed on the volume ID).
(3) The vlocation record is got rid of and VLDB records are no longer
cached. Sufficient information is stored in the volume record, though
an update to a volume record is now no longer shared between related
volumes (volumes come in bundles of three: R/W, R/O and backup).
and the following procedural changes are made:
(1) The fileserver cursor introduced previously is now fleshed out and
used to iterate over fileservers and their addresses.
(2) Volume status is checked during iteration, and the server list is
replaced if a change is detected.
(3) Server status is checked during iteration, and the address list is
replaced if a change is detected.
(4) The abort code is saved into the address list cursor and -ECONNABORTED
returned in afs_make_call() if a remote abort happened rather than
translating the abort into an error message. This allows actions to
be taken depending on the abort code more easily.
(a) If a VMOVED abort is seen then this is handled by rechecking the
volume and restarting the iteration.
(b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is
handled by sleeping for a short period and retrying and/or trying
other servers that might serve that volume. A message is also
displayed once until the condition has cleared.
(c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the
moment.
(d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to
see if it has been deleted; if not, the fileserver is probably
indicating that the volume couldn't be attached and needs
salvaging.
(e) If statfs() sees one of these aborts, it does not sleep, but
rather returns an error, so as not to block the umount program.
(5) The fileserver iteration functions in vnode.c are now merged into
their callers and more heavily macroised around the cursor. vnode.c
is removed.
(6) Operations on a particular vnode are serialised on that vnode because
the server will lock that vnode whilst it operates on it, so a second
op sent will just have to wait.
(7) Fileservers are probed with FS.GetCapabilities before being used.
This is where service upgrade will be done.
(8) A callback interest on a fileserver is set up before an FS operation
is performed and passed through to afs_make_call() so that it can be
set on the vnode if the operation returns a callback. The callback
interest is passed through to afs_iget() also so that it can be set
there too.
In general, record updating is done on an as-needed basis when we try to
access servers, volumes or vnodes rather than offloading it to work items
and special threads.
Notes:
(1) Pre AFS-3.4 servers are no longer supported, though this can be added
back if necessary (AFS-3.4 was released in 1998).
(2) VBUSY is retried forever for the moment at intervals of 1s.
(3) /proc/fs/afs/<cell>/servers no longer exists.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 23:27:50 +08:00
|
|
|
error_key:
|
2007-04-27 06:59:35 +08:00
|
|
|
key_put(key);
|
|
|
|
error:
|
|
|
|
_leave(" = %d", ret);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* create a regular file on an AFS filesystem
|
|
|
|
*/
|
2011-07-26 13:42:34 +08:00
|
|
|
static int afs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
|
2012-06-11 06:05:36 +08:00
|
|
|
bool excl)
|
2007-04-27 06:59:35 +08:00
|
|
|
{
|
afs: Overhaul volume and server record caching and fileserver rotation
The current code assumes that volumes and servers are per-cell and are
never shared, but this is not enforced, and, indeed, public cells do exist
that are aliases of each other. Further, an organisation can, say, set up
a public cell and a private cell with overlapping, but not identical, sets
of servers. The difference is purely in the database attached to the VL
servers.
The current code will malfunction if it sees a server in two cells as it
assumes global address -> server record mappings and that each server is in
just one cell.
Further, each server may have multiple addresses - and may have addresses
of different families (IPv4 and IPv6, say).
To this end, the following structural changes are made:
(1) Server record management is overhauled:
(a) Server records are made independent of cell. The namespace keeps
track of them, volume records have lists of them and each vnode
has a server on which its callback interest currently resides.
(b) The cell record no longer keeps a list of servers known to be in
that cell.
(c) The server records are now kept in a flat list because there's no
single address to sort on.
(d) Server records are now keyed by their UUID within the namespace.
(e) The addresses for a server are obtained with the VL.GetAddrsU
rather than with VL.GetEntryByName, using the server's UUID as a
parameter.
(f) Cached server records are garbage collected after a period of
non-use and are counted out of existence before purging is allowed
to complete. This protects the work functions against rmmod.
(g) The servers list is now in /proc/fs/afs/servers.
(2) Volume record management is overhauled:
(a) An RCU-replaceable server list is introduced. This tracks both
servers and their coresponding callback interests.
(b) The superblock is now keyed on cell record and numeric volume ID.
(c) The volume record is now tied to the superblock which mounts it,
and is activated when mounted and deactivated when unmounted.
This makes it easier to handle the cache cookie without causing a
double-use in fscache.
(d) The volume record is loaded from the VLDB using VL.GetEntryByNameU
to get the server UUID list.
(e) The volume name is updated if it is seen to have changed when the
volume is updated (the update is keyed on the volume ID).
(3) The vlocation record is got rid of and VLDB records are no longer
cached. Sufficient information is stored in the volume record, though
an update to a volume record is now no longer shared between related
volumes (volumes come in bundles of three: R/W, R/O and backup).
and the following procedural changes are made:
(1) The fileserver cursor introduced previously is now fleshed out and
used to iterate over fileservers and their addresses.
(2) Volume status is checked during iteration, and the server list is
replaced if a change is detected.
(3) Server status is checked during iteration, and the address list is
replaced if a change is detected.
(4) The abort code is saved into the address list cursor and -ECONNABORTED
returned in afs_make_call() if a remote abort happened rather than
translating the abort into an error message. This allows actions to
be taken depending on the abort code more easily.
(a) If a VMOVED abort is seen then this is handled by rechecking the
volume and restarting the iteration.
(b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is
handled by sleeping for a short period and retrying and/or trying
other servers that might serve that volume. A message is also
displayed once until the condition has cleared.
(c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the
moment.
(d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to
see if it has been deleted; if not, the fileserver is probably
indicating that the volume couldn't be attached and needs
salvaging.
(e) If statfs() sees one of these aborts, it does not sleep, but
rather returns an error, so as not to block the umount program.
(5) The fileserver iteration functions in vnode.c are now merged into
their callers and more heavily macroised around the cursor. vnode.c
is removed.
(6) Operations on a particular vnode are serialised on that vnode because
the server will lock that vnode whilst it operates on it, so a second
op sent will just have to wait.
(7) Fileservers are probed with FS.GetCapabilities before being used.
This is where service upgrade will be done.
(8) A callback interest on a fileserver is set up before an FS operation
is performed and passed through to afs_make_call() so that it can be
set on the vnode if the operation returns a callback. The callback
interest is passed through to afs_iget() also so that it can be set
there too.
In general, record updating is done on an as-needed basis when we try to
access servers, volumes or vnodes rather than offloading it to work items
and special threads.
Notes:
(1) Pre AFS-3.4 servers are no longer supported, though this can be added
back if necessary (AFS-3.4 was released in 1998).
(2) VBUSY is retried forever for the moment at intervals of 1s.
(3) /proc/fs/afs/<cell>/servers no longer exists.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 23:27:50 +08:00
|
|
|
struct afs_fs_cursor fc;
|
|
|
|
struct afs_file_status newstatus;
|
|
|
|
struct afs_callback newcb;
|
2017-11-20 21:58:20 +08:00
|
|
|
struct afs_vnode *dvnode = AFS_FS_I(dir);
|
afs: Overhaul volume and server record caching and fileserver rotation
The current code assumes that volumes and servers are per-cell and are
never shared, but this is not enforced, and, indeed, public cells do exist
that are aliases of each other. Further, an organisation can, say, set up
a public cell and a private cell with overlapping, but not identical, sets
of servers. The difference is purely in the database attached to the VL
servers.
The current code will malfunction if it sees a server in two cells as it
assumes global address -> server record mappings and that each server is in
just one cell.
Further, each server may have multiple addresses - and may have addresses
of different families (IPv4 and IPv6, say).
To this end, the following structural changes are made:
(1) Server record management is overhauled:
(a) Server records are made independent of cell. The namespace keeps
track of them, volume records have lists of them and each vnode
has a server on which its callback interest currently resides.
(b) The cell record no longer keeps a list of servers known to be in
that cell.
(c) The server records are now kept in a flat list because there's no
single address to sort on.
(d) Server records are now keyed by their UUID within the namespace.
(e) The addresses for a server are obtained with the VL.GetAddrsU
rather than with VL.GetEntryByName, using the server's UUID as a
parameter.
(f) Cached server records are garbage collected after a period of
non-use and are counted out of existence before purging is allowed
to complete. This protects the work functions against rmmod.
(g) The servers list is now in /proc/fs/afs/servers.
(2) Volume record management is overhauled:
(a) An RCU-replaceable server list is introduced. This tracks both
servers and their coresponding callback interests.
(b) The superblock is now keyed on cell record and numeric volume ID.
(c) The volume record is now tied to the superblock which mounts it,
and is activated when mounted and deactivated when unmounted.
This makes it easier to handle the cache cookie without causing a
double-use in fscache.
(d) The volume record is loaded from the VLDB using VL.GetEntryByNameU
to get the server UUID list.
(e) The volume name is updated if it is seen to have changed when the
volume is updated (the update is keyed on the volume ID).
(3) The vlocation record is got rid of and VLDB records are no longer
cached. Sufficient information is stored in the volume record, though
an update to a volume record is now no longer shared between related
volumes (volumes come in bundles of three: R/W, R/O and backup).
and the following procedural changes are made:
(1) The fileserver cursor introduced previously is now fleshed out and
used to iterate over fileservers and their addresses.
(2) Volume status is checked during iteration, and the server list is
replaced if a change is detected.
(3) Server status is checked during iteration, and the address list is
replaced if a change is detected.
(4) The abort code is saved into the address list cursor and -ECONNABORTED
returned in afs_make_call() if a remote abort happened rather than
translating the abort into an error message. This allows actions to
be taken depending on the abort code more easily.
(a) If a VMOVED abort is seen then this is handled by rechecking the
volume and restarting the iteration.
(b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is
handled by sleeping for a short period and retrying and/or trying
other servers that might serve that volume. A message is also
displayed once until the condition has cleared.
(c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the
moment.
(d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to
see if it has been deleted; if not, the fileserver is probably
indicating that the volume couldn't be attached and needs
salvaging.
(e) If statfs() sees one of these aborts, it does not sleep, but
rather returns an error, so as not to block the umount program.
(5) The fileserver iteration functions in vnode.c are now merged into
their callers and more heavily macroised around the cursor. vnode.c
is removed.
(6) Operations on a particular vnode are serialised on that vnode because
the server will lock that vnode whilst it operates on it, so a second
op sent will just have to wait.
(7) Fileservers are probed with FS.GetCapabilities before being used.
This is where service upgrade will be done.
(8) A callback interest on a fileserver is set up before an FS operation
is performed and passed through to afs_make_call() so that it can be
set on the vnode if the operation returns a callback. The callback
interest is passed through to afs_iget() also so that it can be set
there too.
In general, record updating is done on an as-needed basis when we try to
access servers, volumes or vnodes rather than offloading it to work items
and special threads.
Notes:
(1) Pre AFS-3.4 servers are no longer supported, though this can be added
back if necessary (AFS-3.4 was released in 1998).
(2) VBUSY is retried forever for the moment at intervals of 1s.
(3) /proc/fs/afs/<cell>/servers no longer exists.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 23:27:50 +08:00
|
|
|
struct afs_fid newfid;
|
2007-04-27 06:59:35 +08:00
|
|
|
struct key *key;
|
|
|
|
int ret;
|
|
|
|
|
afs: Overhaul volume and server record caching and fileserver rotation
The current code assumes that volumes and servers are per-cell and are
never shared, but this is not enforced, and, indeed, public cells do exist
that are aliases of each other. Further, an organisation can, say, set up
a public cell and a private cell with overlapping, but not identical, sets
of servers. The difference is purely in the database attached to the VL
servers.
The current code will malfunction if it sees a server in two cells as it
assumes global address -> server record mappings and that each server is in
just one cell.
Further, each server may have multiple addresses - and may have addresses
of different families (IPv4 and IPv6, say).
To this end, the following structural changes are made:
(1) Server record management is overhauled:
(a) Server records are made independent of cell. The namespace keeps
track of them, volume records have lists of them and each vnode
has a server on which its callback interest currently resides.
(b) The cell record no longer keeps a list of servers known to be in
that cell.
(c) The server records are now kept in a flat list because there's no
single address to sort on.
(d) Server records are now keyed by their UUID within the namespace.
(e) The addresses for a server are obtained with the VL.GetAddrsU
rather than with VL.GetEntryByName, using the server's UUID as a
parameter.
(f) Cached server records are garbage collected after a period of
non-use and are counted out of existence before purging is allowed
to complete. This protects the work functions against rmmod.
(g) The servers list is now in /proc/fs/afs/servers.
(2) Volume record management is overhauled:
(a) An RCU-replaceable server list is introduced. This tracks both
servers and their coresponding callback interests.
(b) The superblock is now keyed on cell record and numeric volume ID.
(c) The volume record is now tied to the superblock which mounts it,
and is activated when mounted and deactivated when unmounted.
This makes it easier to handle the cache cookie without causing a
double-use in fscache.
(d) The volume record is loaded from the VLDB using VL.GetEntryByNameU
to get the server UUID list.
(e) The volume name is updated if it is seen to have changed when the
volume is updated (the update is keyed on the volume ID).
(3) The vlocation record is got rid of and VLDB records are no longer
cached. Sufficient information is stored in the volume record, though
an update to a volume record is now no longer shared between related
volumes (volumes come in bundles of three: R/W, R/O and backup).
and the following procedural changes are made:
(1) The fileserver cursor introduced previously is now fleshed out and
used to iterate over fileservers and their addresses.
(2) Volume status is checked during iteration, and the server list is
replaced if a change is detected.
(3) Server status is checked during iteration, and the address list is
replaced if a change is detected.
(4) The abort code is saved into the address list cursor and -ECONNABORTED
returned in afs_make_call() if a remote abort happened rather than
translating the abort into an error message. This allows actions to
be taken depending on the abort code more easily.
(a) If a VMOVED abort is seen then this is handled by rechecking the
volume and restarting the iteration.
(b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is
handled by sleeping for a short period and retrying and/or trying
other servers that might serve that volume. A message is also
displayed once until the condition has cleared.
(c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the
moment.
(d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to
see if it has been deleted; if not, the fileserver is probably
indicating that the volume couldn't be attached and needs
salvaging.
(e) If statfs() sees one of these aborts, it does not sleep, but
rather returns an error, so as not to block the umount program.
(5) The fileserver iteration functions in vnode.c are now merged into
their callers and more heavily macroised around the cursor. vnode.c
is removed.
(6) Operations on a particular vnode are serialised on that vnode because
the server will lock that vnode whilst it operates on it, so a second
op sent will just have to wait.
(7) Fileservers are probed with FS.GetCapabilities before being used.
This is where service upgrade will be done.
(8) A callback interest on a fileserver is set up before an FS operation
is performed and passed through to afs_make_call() so that it can be
set on the vnode if the operation returns a callback. The callback
interest is passed through to afs_iget() also so that it can be set
there too.
In general, record updating is done on an as-needed basis when we try to
access servers, volumes or vnodes rather than offloading it to work items
and special threads.
Notes:
(1) Pre AFS-3.4 servers are no longer supported, though this can be added
back if necessary (AFS-3.4 was released in 1998).
(2) VBUSY is retried forever for the moment at intervals of 1s.
(3) /proc/fs/afs/<cell>/servers no longer exists.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 23:27:50 +08:00
|
|
|
mode |= S_IFREG;
|
2007-04-27 06:59:35 +08:00
|
|
|
|
2014-10-22 08:11:25 +08:00
|
|
|
_enter("{%x:%u},{%pd},%ho,",
|
|
|
|
dvnode->fid.vid, dvnode->fid.vnode, dentry, mode);
|
2007-04-27 06:59:35 +08:00
|
|
|
|
afs: Overhaul volume and server record caching and fileserver rotation
The current code assumes that volumes and servers are per-cell and are
never shared, but this is not enforced, and, indeed, public cells do exist
that are aliases of each other. Further, an organisation can, say, set up
a public cell and a private cell with overlapping, but not identical, sets
of servers. The difference is purely in the database attached to the VL
servers.
The current code will malfunction if it sees a server in two cells as it
assumes global address -> server record mappings and that each server is in
just one cell.
Further, each server may have multiple addresses - and may have addresses
of different families (IPv4 and IPv6, say).
To this end, the following structural changes are made:
(1) Server record management is overhauled:
(a) Server records are made independent of cell. The namespace keeps
track of them, volume records have lists of them and each vnode
has a server on which its callback interest currently resides.
(b) The cell record no longer keeps a list of servers known to be in
that cell.
(c) The server records are now kept in a flat list because there's no
single address to sort on.
(d) Server records are now keyed by their UUID within the namespace.
(e) The addresses for a server are obtained with the VL.GetAddrsU
rather than with VL.GetEntryByName, using the server's UUID as a
parameter.
(f) Cached server records are garbage collected after a period of
non-use and are counted out of existence before purging is allowed
to complete. This protects the work functions against rmmod.
(g) The servers list is now in /proc/fs/afs/servers.
(2) Volume record management is overhauled:
(a) An RCU-replaceable server list is introduced. This tracks both
servers and their coresponding callback interests.
(b) The superblock is now keyed on cell record and numeric volume ID.
(c) The volume record is now tied to the superblock which mounts it,
and is activated when mounted and deactivated when unmounted.
This makes it easier to handle the cache cookie without causing a
double-use in fscache.
(d) The volume record is loaded from the VLDB using VL.GetEntryByNameU
to get the server UUID list.
(e) The volume name is updated if it is seen to have changed when the
volume is updated (the update is keyed on the volume ID).
(3) The vlocation record is got rid of and VLDB records are no longer
cached. Sufficient information is stored in the volume record, though
an update to a volume record is now no longer shared between related
volumes (volumes come in bundles of three: R/W, R/O and backup).
and the following procedural changes are made:
(1) The fileserver cursor introduced previously is now fleshed out and
used to iterate over fileservers and their addresses.
(2) Volume status is checked during iteration, and the server list is
replaced if a change is detected.
(3) Server status is checked during iteration, and the address list is
replaced if a change is detected.
(4) The abort code is saved into the address list cursor and -ECONNABORTED
returned in afs_make_call() if a remote abort happened rather than
translating the abort into an error message. This allows actions to
be taken depending on the abort code more easily.
(a) If a VMOVED abort is seen then this is handled by rechecking the
volume and restarting the iteration.
(b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is
handled by sleeping for a short period and retrying and/or trying
other servers that might serve that volume. A message is also
displayed once until the condition has cleared.
(c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the
moment.
(d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to
see if it has been deleted; if not, the fileserver is probably
indicating that the volume couldn't be attached and needs
salvaging.
(e) If statfs() sees one of these aborts, it does not sleep, but
rather returns an error, so as not to block the umount program.
(5) The fileserver iteration functions in vnode.c are now merged into
their callers and more heavily macroised around the cursor. vnode.c
is removed.
(6) Operations on a particular vnode are serialised on that vnode because
the server will lock that vnode whilst it operates on it, so a second
op sent will just have to wait.
(7) Fileservers are probed with FS.GetCapabilities before being used.
This is where service upgrade will be done.
(8) A callback interest on a fileserver is set up before an FS operation
is performed and passed through to afs_make_call() so that it can be
set on the vnode if the operation returns a callback. The callback
interest is passed through to afs_iget() also so that it can be set
there too.
In general, record updating is done on an as-needed basis when we try to
access servers, volumes or vnodes rather than offloading it to work items
and special threads.
Notes:
(1) Pre AFS-3.4 servers are no longer supported, though this can be added
back if necessary (AFS-3.4 was released in 1998).
(2) VBUSY is retried forever for the moment at intervals of 1s.
(3) /proc/fs/afs/<cell>/servers no longer exists.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 23:27:50 +08:00
|
|
|
ret = -ENAMETOOLONG;
|
|
|
|
if (dentry->d_name.len >= AFSNAMEMAX)
|
|
|
|
goto error;
|
|
|
|
|
2007-04-27 06:59:35 +08:00
|
|
|
key = afs_request_key(dvnode->volume->cell);
|
|
|
|
if (IS_ERR(key)) {
|
|
|
|
ret = PTR_ERR(key);
|
|
|
|
goto error;
|
|
|
|
}
|
|
|
|
|
afs: Overhaul volume and server record caching and fileserver rotation
The current code assumes that volumes and servers are per-cell and are
never shared, but this is not enforced, and, indeed, public cells do exist
that are aliases of each other. Further, an organisation can, say, set up
a public cell and a private cell with overlapping, but not identical, sets
of servers. The difference is purely in the database attached to the VL
servers.
The current code will malfunction if it sees a server in two cells as it
assumes global address -> server record mappings and that each server is in
just one cell.
Further, each server may have multiple addresses - and may have addresses
of different families (IPv4 and IPv6, say).
To this end, the following structural changes are made:
(1) Server record management is overhauled:
(a) Server records are made independent of cell. The namespace keeps
track of them, volume records have lists of them and each vnode
has a server on which its callback interest currently resides.
(b) The cell record no longer keeps a list of servers known to be in
that cell.
(c) The server records are now kept in a flat list because there's no
single address to sort on.
(d) Server records are now keyed by their UUID within the namespace.
(e) The addresses for a server are obtained with the VL.GetAddrsU
rather than with VL.GetEntryByName, using the server's UUID as a
parameter.
(f) Cached server records are garbage collected after a period of
non-use and are counted out of existence before purging is allowed
to complete. This protects the work functions against rmmod.
(g) The servers list is now in /proc/fs/afs/servers.
(2) Volume record management is overhauled:
(a) An RCU-replaceable server list is introduced. This tracks both
servers and their coresponding callback interests.
(b) The superblock is now keyed on cell record and numeric volume ID.
(c) The volume record is now tied to the superblock which mounts it,
and is activated when mounted and deactivated when unmounted.
This makes it easier to handle the cache cookie without causing a
double-use in fscache.
(d) The volume record is loaded from the VLDB using VL.GetEntryByNameU
to get the server UUID list.
(e) The volume name is updated if it is seen to have changed when the
volume is updated (the update is keyed on the volume ID).
(3) The vlocation record is got rid of and VLDB records are no longer
cached. Sufficient information is stored in the volume record, though
an update to a volume record is now no longer shared between related
volumes (volumes come in bundles of three: R/W, R/O and backup).
and the following procedural changes are made:
(1) The fileserver cursor introduced previously is now fleshed out and
used to iterate over fileservers and their addresses.
(2) Volume status is checked during iteration, and the server list is
replaced if a change is detected.
(3) Server status is checked during iteration, and the address list is
replaced if a change is detected.
(4) The abort code is saved into the address list cursor and -ECONNABORTED
returned in afs_make_call() if a remote abort happened rather than
translating the abort into an error message. This allows actions to
be taken depending on the abort code more easily.
(a) If a VMOVED abort is seen then this is handled by rechecking the
volume and restarting the iteration.
(b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is
handled by sleeping for a short period and retrying and/or trying
other servers that might serve that volume. A message is also
displayed once until the condition has cleared.
(c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the
moment.
(d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to
see if it has been deleted; if not, the fileserver is probably
indicating that the volume couldn't be attached and needs
salvaging.
(e) If statfs() sees one of these aborts, it does not sleep, but
rather returns an error, so as not to block the umount program.
(5) The fileserver iteration functions in vnode.c are now merged into
their callers and more heavily macroised around the cursor. vnode.c
is removed.
(6) Operations on a particular vnode are serialised on that vnode because
the server will lock that vnode whilst it operates on it, so a second
op sent will just have to wait.
(7) Fileservers are probed with FS.GetCapabilities before being used.
This is where service upgrade will be done.
(8) A callback interest on a fileserver is set up before an FS operation
is performed and passed through to afs_make_call() so that it can be
set on the vnode if the operation returns a callback. The callback
interest is passed through to afs_iget() also so that it can be set
there too.
In general, record updating is done on an as-needed basis when we try to
access servers, volumes or vnodes rather than offloading it to work items
and special threads.
Notes:
(1) Pre AFS-3.4 servers are no longer supported, though this can be added
back if necessary (AFS-3.4 was released in 1998).
(2) VBUSY is retried forever for the moment at intervals of 1s.
(3) /proc/fs/afs/<cell>/servers no longer exists.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 23:27:50 +08:00
|
|
|
ret = -ERESTARTSYS;
|
|
|
|
if (afs_begin_vnode_operation(&fc, dvnode, key)) {
|
|
|
|
while (afs_select_fileserver(&fc)) {
|
|
|
|
fc.cb_break = dvnode->cb_break + dvnode->cb_s_break;
|
|
|
|
afs_fs_create(&fc, dentry->d_name.name, mode,
|
|
|
|
&newfid, &newstatus, &newcb);
|
|
|
|
}
|
2007-04-27 06:59:35 +08:00
|
|
|
|
afs: Overhaul volume and server record caching and fileserver rotation
The current code assumes that volumes and servers are per-cell and are
never shared, but this is not enforced, and, indeed, public cells do exist
that are aliases of each other. Further, an organisation can, say, set up
a public cell and a private cell with overlapping, but not identical, sets
of servers. The difference is purely in the database attached to the VL
servers.
The current code will malfunction if it sees a server in two cells as it
assumes global address -> server record mappings and that each server is in
just one cell.
Further, each server may have multiple addresses - and may have addresses
of different families (IPv4 and IPv6, say).
To this end, the following structural changes are made:
(1) Server record management is overhauled:
(a) Server records are made independent of cell. The namespace keeps
track of them, volume records have lists of them and each vnode
has a server on which its callback interest currently resides.
(b) The cell record no longer keeps a list of servers known to be in
that cell.
(c) The server records are now kept in a flat list because there's no
single address to sort on.
(d) Server records are now keyed by their UUID within the namespace.
(e) The addresses for a server are obtained with the VL.GetAddrsU
rather than with VL.GetEntryByName, using the server's UUID as a
parameter.
(f) Cached server records are garbage collected after a period of
non-use and are counted out of existence before purging is allowed
to complete. This protects the work functions against rmmod.
(g) The servers list is now in /proc/fs/afs/servers.
(2) Volume record management is overhauled:
(a) An RCU-replaceable server list is introduced. This tracks both
servers and their coresponding callback interests.
(b) The superblock is now keyed on cell record and numeric volume ID.
(c) The volume record is now tied to the superblock which mounts it,
and is activated when mounted and deactivated when unmounted.
This makes it easier to handle the cache cookie without causing a
double-use in fscache.
(d) The volume record is loaded from the VLDB using VL.GetEntryByNameU
to get the server UUID list.
(e) The volume name is updated if it is seen to have changed when the
volume is updated (the update is keyed on the volume ID).
(3) The vlocation record is got rid of and VLDB records are no longer
cached. Sufficient information is stored in the volume record, though
an update to a volume record is now no longer shared between related
volumes (volumes come in bundles of three: R/W, R/O and backup).
and the following procedural changes are made:
(1) The fileserver cursor introduced previously is now fleshed out and
used to iterate over fileservers and their addresses.
(2) Volume status is checked during iteration, and the server list is
replaced if a change is detected.
(3) Server status is checked during iteration, and the address list is
replaced if a change is detected.
(4) The abort code is saved into the address list cursor and -ECONNABORTED
returned in afs_make_call() if a remote abort happened rather than
translating the abort into an error message. This allows actions to
be taken depending on the abort code more easily.
(a) If a VMOVED abort is seen then this is handled by rechecking the
volume and restarting the iteration.
(b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is
handled by sleeping for a short period and retrying and/or trying
other servers that might serve that volume. A message is also
displayed once until the condition has cleared.
(c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the
moment.
(d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to
see if it has been deleted; if not, the fileserver is probably
indicating that the volume couldn't be attached and needs
salvaging.
(e) If statfs() sees one of these aborts, it does not sleep, but
rather returns an error, so as not to block the umount program.
(5) The fileserver iteration functions in vnode.c are now merged into
their callers and more heavily macroised around the cursor. vnode.c
is removed.
(6) Operations on a particular vnode are serialised on that vnode because
the server will lock that vnode whilst it operates on it, so a second
op sent will just have to wait.
(7) Fileservers are probed with FS.GetCapabilities before being used.
This is where service upgrade will be done.
(8) A callback interest on a fileserver is set up before an FS operation
is performed and passed through to afs_make_call() so that it can be
set on the vnode if the operation returns a callback. The callback
interest is passed through to afs_iget() also so that it can be set
there too.
In general, record updating is done on an as-needed basis when we try to
access servers, volumes or vnodes rather than offloading it to work items
and special threads.
Notes:
(1) Pre AFS-3.4 servers are no longer supported, though this can be added
back if necessary (AFS-3.4 was released in 1998).
(2) VBUSY is retried forever for the moment at intervals of 1s.
(3) /proc/fs/afs/<cell>/servers no longer exists.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 23:27:50 +08:00
|
|
|
afs_check_for_remote_deletion(&fc, fc.vnode);
|
|
|
|
afs_vnode_commit_status(&fc, dvnode, fc.cb_break);
|
|
|
|
afs_vnode_new_inode(&fc, dentry, &newfid, &newstatus, &newcb);
|
|
|
|
ret = afs_end_vnode_operation(&fc);
|
|
|
|
if (ret < 0)
|
|
|
|
goto error_key;
|
2017-11-21 06:41:00 +08:00
|
|
|
} else {
|
|
|
|
goto error_key;
|
2007-04-27 06:59:35 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
key_put(key);
|
|
|
|
_leave(" = 0");
|
|
|
|
return 0;
|
|
|
|
|
afs: Overhaul volume and server record caching and fileserver rotation
The current code assumes that volumes and servers are per-cell and are
never shared, but this is not enforced, and, indeed, public cells do exist
that are aliases of each other. Further, an organisation can, say, set up
a public cell and a private cell with overlapping, but not identical, sets
of servers. The difference is purely in the database attached to the VL
servers.
The current code will malfunction if it sees a server in two cells as it
assumes global address -> server record mappings and that each server is in
just one cell.
Further, each server may have multiple addresses - and may have addresses
of different families (IPv4 and IPv6, say).
To this end, the following structural changes are made:
(1) Server record management is overhauled:
(a) Server records are made independent of cell. The namespace keeps
track of them, volume records have lists of them and each vnode
has a server on which its callback interest currently resides.
(b) The cell record no longer keeps a list of servers known to be in
that cell.
(c) The server records are now kept in a flat list because there's no
single address to sort on.
(d) Server records are now keyed by their UUID within the namespace.
(e) The addresses for a server are obtained with the VL.GetAddrsU
rather than with VL.GetEntryByName, using the server's UUID as a
parameter.
(f) Cached server records are garbage collected after a period of
non-use and are counted out of existence before purging is allowed
to complete. This protects the work functions against rmmod.
(g) The servers list is now in /proc/fs/afs/servers.
(2) Volume record management is overhauled:
(a) An RCU-replaceable server list is introduced. This tracks both
servers and their coresponding callback interests.
(b) The superblock is now keyed on cell record and numeric volume ID.
(c) The volume record is now tied to the superblock which mounts it,
and is activated when mounted and deactivated when unmounted.
This makes it easier to handle the cache cookie without causing a
double-use in fscache.
(d) The volume record is loaded from the VLDB using VL.GetEntryByNameU
to get the server UUID list.
(e) The volume name is updated if it is seen to have changed when the
volume is updated (the update is keyed on the volume ID).
(3) The vlocation record is got rid of and VLDB records are no longer
cached. Sufficient information is stored in the volume record, though
an update to a volume record is now no longer shared between related
volumes (volumes come in bundles of three: R/W, R/O and backup).
and the following procedural changes are made:
(1) The fileserver cursor introduced previously is now fleshed out and
used to iterate over fileservers and their addresses.
(2) Volume status is checked during iteration, and the server list is
replaced if a change is detected.
(3) Server status is checked during iteration, and the address list is
replaced if a change is detected.
(4) The abort code is saved into the address list cursor and -ECONNABORTED
returned in afs_make_call() if a remote abort happened rather than
translating the abort into an error message. This allows actions to
be taken depending on the abort code more easily.
(a) If a VMOVED abort is seen then this is handled by rechecking the
volume and restarting the iteration.
(b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is
handled by sleeping for a short period and retrying and/or trying
other servers that might serve that volume. A message is also
displayed once until the condition has cleared.
(c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the
moment.
(d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to
see if it has been deleted; if not, the fileserver is probably
indicating that the volume couldn't be attached and needs
salvaging.
(e) If statfs() sees one of these aborts, it does not sleep, but
rather returns an error, so as not to block the umount program.
(5) The fileserver iteration functions in vnode.c are now merged into
their callers and more heavily macroised around the cursor. vnode.c
is removed.
(6) Operations on a particular vnode are serialised on that vnode because
the server will lock that vnode whilst it operates on it, so a second
op sent will just have to wait.
(7) Fileservers are probed with FS.GetCapabilities before being used.
This is where service upgrade will be done.
(8) A callback interest on a fileserver is set up before an FS operation
is performed and passed through to afs_make_call() so that it can be
set on the vnode if the operation returns a callback. The callback
interest is passed through to afs_iget() also so that it can be set
there too.
In general, record updating is done on an as-needed basis when we try to
access servers, volumes or vnodes rather than offloading it to work items
and special threads.
Notes:
(1) Pre AFS-3.4 servers are no longer supported, though this can be added
back if necessary (AFS-3.4 was released in 1998).
(2) VBUSY is retried forever for the moment at intervals of 1s.
(3) /proc/fs/afs/<cell>/servers no longer exists.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 23:27:50 +08:00
|
|
|
error_key:
|
2007-04-27 06:59:35 +08:00
|
|
|
key_put(key);
|
|
|
|
error:
|
|
|
|
d_drop(dentry);
|
|
|
|
_leave(" = %d", ret);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* create a hard link between files in an AFS filesystem
|
|
|
|
*/
|
|
|
|
static int afs_link(struct dentry *from, struct inode *dir,
|
|
|
|
struct dentry *dentry)
|
|
|
|
{
|
afs: Overhaul volume and server record caching and fileserver rotation
The current code assumes that volumes and servers are per-cell and are
never shared, but this is not enforced, and, indeed, public cells do exist
that are aliases of each other. Further, an organisation can, say, set up
a public cell and a private cell with overlapping, but not identical, sets
of servers. The difference is purely in the database attached to the VL
servers.
The current code will malfunction if it sees a server in two cells as it
assumes global address -> server record mappings and that each server is in
just one cell.
Further, each server may have multiple addresses - and may have addresses
of different families (IPv4 and IPv6, say).
To this end, the following structural changes are made:
(1) Server record management is overhauled:
(a) Server records are made independent of cell. The namespace keeps
track of them, volume records have lists of them and each vnode
has a server on which its callback interest currently resides.
(b) The cell record no longer keeps a list of servers known to be in
that cell.
(c) The server records are now kept in a flat list because there's no
single address to sort on.
(d) Server records are now keyed by their UUID within the namespace.
(e) The addresses for a server are obtained with the VL.GetAddrsU
rather than with VL.GetEntryByName, using the server's UUID as a
parameter.
(f) Cached server records are garbage collected after a period of
non-use and are counted out of existence before purging is allowed
to complete. This protects the work functions against rmmod.
(g) The servers list is now in /proc/fs/afs/servers.
(2) Volume record management is overhauled:
(a) An RCU-replaceable server list is introduced. This tracks both
servers and their coresponding callback interests.
(b) The superblock is now keyed on cell record and numeric volume ID.
(c) The volume record is now tied to the superblock which mounts it,
and is activated when mounted and deactivated when unmounted.
This makes it easier to handle the cache cookie without causing a
double-use in fscache.
(d) The volume record is loaded from the VLDB using VL.GetEntryByNameU
to get the server UUID list.
(e) The volume name is updated if it is seen to have changed when the
volume is updated (the update is keyed on the volume ID).
(3) The vlocation record is got rid of and VLDB records are no longer
cached. Sufficient information is stored in the volume record, though
an update to a volume record is now no longer shared between related
volumes (volumes come in bundles of three: R/W, R/O and backup).
and the following procedural changes are made:
(1) The fileserver cursor introduced previously is now fleshed out and
used to iterate over fileservers and their addresses.
(2) Volume status is checked during iteration, and the server list is
replaced if a change is detected.
(3) Server status is checked during iteration, and the address list is
replaced if a change is detected.
(4) The abort code is saved into the address list cursor and -ECONNABORTED
returned in afs_make_call() if a remote abort happened rather than
translating the abort into an error message. This allows actions to
be taken depending on the abort code more easily.
(a) If a VMOVED abort is seen then this is handled by rechecking the
volume and restarting the iteration.
(b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is
handled by sleeping for a short period and retrying and/or trying
other servers that might serve that volume. A message is also
displayed once until the condition has cleared.
(c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the
moment.
(d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to
see if it has been deleted; if not, the fileserver is probably
indicating that the volume couldn't be attached and needs
salvaging.
(e) If statfs() sees one of these aborts, it does not sleep, but
rather returns an error, so as not to block the umount program.
(5) The fileserver iteration functions in vnode.c are now merged into
their callers and more heavily macroised around the cursor. vnode.c
is removed.
(6) Operations on a particular vnode are serialised on that vnode because
the server will lock that vnode whilst it operates on it, so a second
op sent will just have to wait.
(7) Fileservers are probed with FS.GetCapabilities before being used.
This is where service upgrade will be done.
(8) A callback interest on a fileserver is set up before an FS operation
is performed and passed through to afs_make_call() so that it can be
set on the vnode if the operation returns a callback. The callback
interest is passed through to afs_iget() also so that it can be set
there too.
In general, record updating is done on an as-needed basis when we try to
access servers, volumes or vnodes rather than offloading it to work items
and special threads.
Notes:
(1) Pre AFS-3.4 servers are no longer supported, though this can be added
back if necessary (AFS-3.4 was released in 1998).
(2) VBUSY is retried forever for the moment at intervals of 1s.
(3) /proc/fs/afs/<cell>/servers no longer exists.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 23:27:50 +08:00
|
|
|
struct afs_fs_cursor fc;
|
2007-04-27 06:59:35 +08:00
|
|
|
struct afs_vnode *dvnode, *vnode;
|
|
|
|
struct key *key;
|
|
|
|
int ret;
|
|
|
|
|
2015-03-18 06:25:59 +08:00
|
|
|
vnode = AFS_FS_I(d_inode(from));
|
2007-04-27 06:59:35 +08:00
|
|
|
dvnode = AFS_FS_I(dir);
|
|
|
|
|
2014-10-22 08:11:25 +08:00
|
|
|
_enter("{%x:%u},{%x:%u},{%pd}",
|
2007-04-27 06:59:35 +08:00
|
|
|
vnode->fid.vid, vnode->fid.vnode,
|
|
|
|
dvnode->fid.vid, dvnode->fid.vnode,
|
2014-10-22 08:11:25 +08:00
|
|
|
dentry);
|
2007-04-27 06:59:35 +08:00
|
|
|
|
afs: Overhaul volume and server record caching and fileserver rotation
The current code assumes that volumes and servers are per-cell and are
never shared, but this is not enforced, and, indeed, public cells do exist
that are aliases of each other. Further, an organisation can, say, set up
a public cell and a private cell with overlapping, but not identical, sets
of servers. The difference is purely in the database attached to the VL
servers.
The current code will malfunction if it sees a server in two cells as it
assumes global address -> server record mappings and that each server is in
just one cell.
Further, each server may have multiple addresses - and may have addresses
of different families (IPv4 and IPv6, say).
To this end, the following structural changes are made:
(1) Server record management is overhauled:
(a) Server records are made independent of cell. The namespace keeps
track of them, volume records have lists of them and each vnode
has a server on which its callback interest currently resides.
(b) The cell record no longer keeps a list of servers known to be in
that cell.
(c) The server records are now kept in a flat list because there's no
single address to sort on.
(d) Server records are now keyed by their UUID within the namespace.
(e) The addresses for a server are obtained with the VL.GetAddrsU
rather than with VL.GetEntryByName, using the server's UUID as a
parameter.
(f) Cached server records are garbage collected after a period of
non-use and are counted out of existence before purging is allowed
to complete. This protects the work functions against rmmod.
(g) The servers list is now in /proc/fs/afs/servers.
(2) Volume record management is overhauled:
(a) An RCU-replaceable server list is introduced. This tracks both
servers and their coresponding callback interests.
(b) The superblock is now keyed on cell record and numeric volume ID.
(c) The volume record is now tied to the superblock which mounts it,
and is activated when mounted and deactivated when unmounted.
This makes it easier to handle the cache cookie without causing a
double-use in fscache.
(d) The volume record is loaded from the VLDB using VL.GetEntryByNameU
to get the server UUID list.
(e) The volume name is updated if it is seen to have changed when the
volume is updated (the update is keyed on the volume ID).
(3) The vlocation record is got rid of and VLDB records are no longer
cached. Sufficient information is stored in the volume record, though
an update to a volume record is now no longer shared between related
volumes (volumes come in bundles of three: R/W, R/O and backup).
and the following procedural changes are made:
(1) The fileserver cursor introduced previously is now fleshed out and
used to iterate over fileservers and their addresses.
(2) Volume status is checked during iteration, and the server list is
replaced if a change is detected.
(3) Server status is checked during iteration, and the address list is
replaced if a change is detected.
(4) The abort code is saved into the address list cursor and -ECONNABORTED
returned in afs_make_call() if a remote abort happened rather than
translating the abort into an error message. This allows actions to
be taken depending on the abort code more easily.
(a) If a VMOVED abort is seen then this is handled by rechecking the
volume and restarting the iteration.
(b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is
handled by sleeping for a short period and retrying and/or trying
other servers that might serve that volume. A message is also
displayed once until the condition has cleared.
(c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the
moment.
(d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to
see if it has been deleted; if not, the fileserver is probably
indicating that the volume couldn't be attached and needs
salvaging.
(e) If statfs() sees one of these aborts, it does not sleep, but
rather returns an error, so as not to block the umount program.
(5) The fileserver iteration functions in vnode.c are now merged into
their callers and more heavily macroised around the cursor. vnode.c
is removed.
(6) Operations on a particular vnode are serialised on that vnode because
the server will lock that vnode whilst it operates on it, so a second
op sent will just have to wait.
(7) Fileservers are probed with FS.GetCapabilities before being used.
This is where service upgrade will be done.
(8) A callback interest on a fileserver is set up before an FS operation
is performed and passed through to afs_make_call() so that it can be
set on the vnode if the operation returns a callback. The callback
interest is passed through to afs_iget() also so that it can be set
there too.
In general, record updating is done on an as-needed basis when we try to
access servers, volumes or vnodes rather than offloading it to work items
and special threads.
Notes:
(1) Pre AFS-3.4 servers are no longer supported, though this can be added
back if necessary (AFS-3.4 was released in 1998).
(2) VBUSY is retried forever for the moment at intervals of 1s.
(3) /proc/fs/afs/<cell>/servers no longer exists.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 23:27:50 +08:00
|
|
|
ret = -ENAMETOOLONG;
|
|
|
|
if (dentry->d_name.len >= AFSNAMEMAX)
|
|
|
|
goto error;
|
|
|
|
|
2007-04-27 06:59:35 +08:00
|
|
|
key = afs_request_key(dvnode->volume->cell);
|
|
|
|
if (IS_ERR(key)) {
|
|
|
|
ret = PTR_ERR(key);
|
|
|
|
goto error;
|
|
|
|
}
|
|
|
|
|
afs: Overhaul volume and server record caching and fileserver rotation
The current code assumes that volumes and servers are per-cell and are
never shared, but this is not enforced, and, indeed, public cells do exist
that are aliases of each other. Further, an organisation can, say, set up
a public cell and a private cell with overlapping, but not identical, sets
of servers. The difference is purely in the database attached to the VL
servers.
The current code will malfunction if it sees a server in two cells as it
assumes global address -> server record mappings and that each server is in
just one cell.
Further, each server may have multiple addresses - and may have addresses
of different families (IPv4 and IPv6, say).
To this end, the following structural changes are made:
(1) Server record management is overhauled:
(a) Server records are made independent of cell. The namespace keeps
track of them, volume records have lists of them and each vnode
has a server on which its callback interest currently resides.
(b) The cell record no longer keeps a list of servers known to be in
that cell.
(c) The server records are now kept in a flat list because there's no
single address to sort on.
(d) Server records are now keyed by their UUID within the namespace.
(e) The addresses for a server are obtained with the VL.GetAddrsU
rather than with VL.GetEntryByName, using the server's UUID as a
parameter.
(f) Cached server records are garbage collected after a period of
non-use and are counted out of existence before purging is allowed
to complete. This protects the work functions against rmmod.
(g) The servers list is now in /proc/fs/afs/servers.
(2) Volume record management is overhauled:
(a) An RCU-replaceable server list is introduced. This tracks both
servers and their coresponding callback interests.
(b) The superblock is now keyed on cell record and numeric volume ID.
(c) The volume record is now tied to the superblock which mounts it,
and is activated when mounted and deactivated when unmounted.
This makes it easier to handle the cache cookie without causing a
double-use in fscache.
(d) The volume record is loaded from the VLDB using VL.GetEntryByNameU
to get the server UUID list.
(e) The volume name is updated if it is seen to have changed when the
volume is updated (the update is keyed on the volume ID).
(3) The vlocation record is got rid of and VLDB records are no longer
cached. Sufficient information is stored in the volume record, though
an update to a volume record is now no longer shared between related
volumes (volumes come in bundles of three: R/W, R/O and backup).
and the following procedural changes are made:
(1) The fileserver cursor introduced previously is now fleshed out and
used to iterate over fileservers and their addresses.
(2) Volume status is checked during iteration, and the server list is
replaced if a change is detected.
(3) Server status is checked during iteration, and the address list is
replaced if a change is detected.
(4) The abort code is saved into the address list cursor and -ECONNABORTED
returned in afs_make_call() if a remote abort happened rather than
translating the abort into an error message. This allows actions to
be taken depending on the abort code more easily.
(a) If a VMOVED abort is seen then this is handled by rechecking the
volume and restarting the iteration.
(b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is
handled by sleeping for a short period and retrying and/or trying
other servers that might serve that volume. A message is also
displayed once until the condition has cleared.
(c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the
moment.
(d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to
see if it has been deleted; if not, the fileserver is probably
indicating that the volume couldn't be attached and needs
salvaging.
(e) If statfs() sees one of these aborts, it does not sleep, but
rather returns an error, so as not to block the umount program.
(5) The fileserver iteration functions in vnode.c are now merged into
their callers and more heavily macroised around the cursor. vnode.c
is removed.
(6) Operations on a particular vnode are serialised on that vnode because
the server will lock that vnode whilst it operates on it, so a second
op sent will just have to wait.
(7) Fileservers are probed with FS.GetCapabilities before being used.
This is where service upgrade will be done.
(8) A callback interest on a fileserver is set up before an FS operation
is performed and passed through to afs_make_call() so that it can be
set on the vnode if the operation returns a callback. The callback
interest is passed through to afs_iget() also so that it can be set
there too.
In general, record updating is done on an as-needed basis when we try to
access servers, volumes or vnodes rather than offloading it to work items
and special threads.
Notes:
(1) Pre AFS-3.4 servers are no longer supported, though this can be added
back if necessary (AFS-3.4 was released in 1998).
(2) VBUSY is retried forever for the moment at intervals of 1s.
(3) /proc/fs/afs/<cell>/servers no longer exists.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 23:27:50 +08:00
|
|
|
ret = -ERESTARTSYS;
|
|
|
|
if (afs_begin_vnode_operation(&fc, dvnode, key)) {
|
|
|
|
if (mutex_lock_interruptible_nested(&vnode->io_lock, 1) < 0) {
|
|
|
|
afs_end_vnode_operation(&fc);
|
2017-11-21 07:04:08 +08:00
|
|
|
goto error_key;
|
afs: Overhaul volume and server record caching and fileserver rotation
The current code assumes that volumes and servers are per-cell and are
never shared, but this is not enforced, and, indeed, public cells do exist
that are aliases of each other. Further, an organisation can, say, set up
a public cell and a private cell with overlapping, but not identical, sets
of servers. The difference is purely in the database attached to the VL
servers.
The current code will malfunction if it sees a server in two cells as it
assumes global address -> server record mappings and that each server is in
just one cell.
Further, each server may have multiple addresses - and may have addresses
of different families (IPv4 and IPv6, say).
To this end, the following structural changes are made:
(1) Server record management is overhauled:
(a) Server records are made independent of cell. The namespace keeps
track of them, volume records have lists of them and each vnode
has a server on which its callback interest currently resides.
(b) The cell record no longer keeps a list of servers known to be in
that cell.
(c) The server records are now kept in a flat list because there's no
single address to sort on.
(d) Server records are now keyed by their UUID within the namespace.
(e) The addresses for a server are obtained with the VL.GetAddrsU
rather than with VL.GetEntryByName, using the server's UUID as a
parameter.
(f) Cached server records are garbage collected after a period of
non-use and are counted out of existence before purging is allowed
to complete. This protects the work functions against rmmod.
(g) The servers list is now in /proc/fs/afs/servers.
(2) Volume record management is overhauled:
(a) An RCU-replaceable server list is introduced. This tracks both
servers and their coresponding callback interests.
(b) The superblock is now keyed on cell record and numeric volume ID.
(c) The volume record is now tied to the superblock which mounts it,
and is activated when mounted and deactivated when unmounted.
This makes it easier to handle the cache cookie without causing a
double-use in fscache.
(d) The volume record is loaded from the VLDB using VL.GetEntryByNameU
to get the server UUID list.
(e) The volume name is updated if it is seen to have changed when the
volume is updated (the update is keyed on the volume ID).
(3) The vlocation record is got rid of and VLDB records are no longer
cached. Sufficient information is stored in the volume record, though
an update to a volume record is now no longer shared between related
volumes (volumes come in bundles of three: R/W, R/O and backup).
and the following procedural changes are made:
(1) The fileserver cursor introduced previously is now fleshed out and
used to iterate over fileservers and their addresses.
(2) Volume status is checked during iteration, and the server list is
replaced if a change is detected.
(3) Server status is checked during iteration, and the address list is
replaced if a change is detected.
(4) The abort code is saved into the address list cursor and -ECONNABORTED
returned in afs_make_call() if a remote abort happened rather than
translating the abort into an error message. This allows actions to
be taken depending on the abort code more easily.
(a) If a VMOVED abort is seen then this is handled by rechecking the
volume and restarting the iteration.
(b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is
handled by sleeping for a short period and retrying and/or trying
other servers that might serve that volume. A message is also
displayed once until the condition has cleared.
(c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the
moment.
(d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to
see if it has been deleted; if not, the fileserver is probably
indicating that the volume couldn't be attached and needs
salvaging.
(e) If statfs() sees one of these aborts, it does not sleep, but
rather returns an error, so as not to block the umount program.
(5) The fileserver iteration functions in vnode.c are now merged into
their callers and more heavily macroised around the cursor. vnode.c
is removed.
(6) Operations on a particular vnode are serialised on that vnode because
the server will lock that vnode whilst it operates on it, so a second
op sent will just have to wait.
(7) Fileservers are probed with FS.GetCapabilities before being used.
This is where service upgrade will be done.
(8) A callback interest on a fileserver is set up before an FS operation
is performed and passed through to afs_make_call() so that it can be
set on the vnode if the operation returns a callback. The callback
interest is passed through to afs_iget() also so that it can be set
there too.
In general, record updating is done on an as-needed basis when we try to
access servers, volumes or vnodes rather than offloading it to work items
and special threads.
Notes:
(1) Pre AFS-3.4 servers are no longer supported, though this can be added
back if necessary (AFS-3.4 was released in 1998).
(2) VBUSY is retried forever for the moment at intervals of 1s.
(3) /proc/fs/afs/<cell>/servers no longer exists.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 23:27:50 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
while (afs_select_fileserver(&fc)) {
|
|
|
|
fc.cb_break = dvnode->cb_break + dvnode->cb_s_break;
|
|
|
|
fc.cb_break_2 = vnode->cb_break + vnode->cb_s_break;
|
|
|
|
afs_fs_link(&fc, vnode, dentry->d_name.name);
|
|
|
|
}
|
|
|
|
|
|
|
|
afs_vnode_commit_status(&fc, dvnode, fc.cb_break);
|
|
|
|
afs_vnode_commit_status(&fc, vnode, fc.cb_break_2);
|
|
|
|
ihold(&vnode->vfs_inode);
|
|
|
|
d_instantiate(dentry, &vnode->vfs_inode);
|
|
|
|
|
|
|
|
mutex_unlock(&vnode->io_lock);
|
|
|
|
ret = afs_end_vnode_operation(&fc);
|
|
|
|
if (ret < 0)
|
|
|
|
goto error_key;
|
2017-11-21 06:41:00 +08:00
|
|
|
} else {
|
|
|
|
goto error_key;
|
afs: Overhaul volume and server record caching and fileserver rotation
The current code assumes that volumes and servers are per-cell and are
never shared, but this is not enforced, and, indeed, public cells do exist
that are aliases of each other. Further, an organisation can, say, set up
a public cell and a private cell with overlapping, but not identical, sets
of servers. The difference is purely in the database attached to the VL
servers.
The current code will malfunction if it sees a server in two cells as it
assumes global address -> server record mappings and that each server is in
just one cell.
Further, each server may have multiple addresses - and may have addresses
of different families (IPv4 and IPv6, say).
To this end, the following structural changes are made:
(1) Server record management is overhauled:
(a) Server records are made independent of cell. The namespace keeps
track of them, volume records have lists of them and each vnode
has a server on which its callback interest currently resides.
(b) The cell record no longer keeps a list of servers known to be in
that cell.
(c) The server records are now kept in a flat list because there's no
single address to sort on.
(d) Server records are now keyed by their UUID within the namespace.
(e) The addresses for a server are obtained with the VL.GetAddrsU
rather than with VL.GetEntryByName, using the server's UUID as a
parameter.
(f) Cached server records are garbage collected after a period of
non-use and are counted out of existence before purging is allowed
to complete. This protects the work functions against rmmod.
(g) The servers list is now in /proc/fs/afs/servers.
(2) Volume record management is overhauled:
(a) An RCU-replaceable server list is introduced. This tracks both
servers and their coresponding callback interests.
(b) The superblock is now keyed on cell record and numeric volume ID.
(c) The volume record is now tied to the superblock which mounts it,
and is activated when mounted and deactivated when unmounted.
This makes it easier to handle the cache cookie without causing a
double-use in fscache.
(d) The volume record is loaded from the VLDB using VL.GetEntryByNameU
to get the server UUID list.
(e) The volume name is updated if it is seen to have changed when the
volume is updated (the update is keyed on the volume ID).
(3) The vlocation record is got rid of and VLDB records are no longer
cached. Sufficient information is stored in the volume record, though
an update to a volume record is now no longer shared between related
volumes (volumes come in bundles of three: R/W, R/O and backup).
and the following procedural changes are made:
(1) The fileserver cursor introduced previously is now fleshed out and
used to iterate over fileservers and their addresses.
(2) Volume status is checked during iteration, and the server list is
replaced if a change is detected.
(3) Server status is checked during iteration, and the address list is
replaced if a change is detected.
(4) The abort code is saved into the address list cursor and -ECONNABORTED
returned in afs_make_call() if a remote abort happened rather than
translating the abort into an error message. This allows actions to
be taken depending on the abort code more easily.
(a) If a VMOVED abort is seen then this is handled by rechecking the
volume and restarting the iteration.
(b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is
handled by sleeping for a short period and retrying and/or trying
other servers that might serve that volume. A message is also
displayed once until the condition has cleared.
(c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the
moment.
(d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to
see if it has been deleted; if not, the fileserver is probably
indicating that the volume couldn't be attached and needs
salvaging.
(e) If statfs() sees one of these aborts, it does not sleep, but
rather returns an error, so as not to block the umount program.
(5) The fileserver iteration functions in vnode.c are now merged into
their callers and more heavily macroised around the cursor. vnode.c
is removed.
(6) Operations on a particular vnode are serialised on that vnode because
the server will lock that vnode whilst it operates on it, so a second
op sent will just have to wait.
(7) Fileservers are probed with FS.GetCapabilities before being used.
This is where service upgrade will be done.
(8) A callback interest on a fileserver is set up before an FS operation
is performed and passed through to afs_make_call() so that it can be
set on the vnode if the operation returns a callback. The callback
interest is passed through to afs_iget() also so that it can be set
there too.
In general, record updating is done on an as-needed basis when we try to
access servers, volumes or vnodes rather than offloading it to work items
and special threads.
Notes:
(1) Pre AFS-3.4 servers are no longer supported, though this can be added
back if necessary (AFS-3.4 was released in 1998).
(2) VBUSY is retried forever for the moment at intervals of 1s.
(3) /proc/fs/afs/<cell>/servers no longer exists.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 23:27:50 +08:00
|
|
|
}
|
2007-04-27 06:59:35 +08:00
|
|
|
|
|
|
|
key_put(key);
|
|
|
|
_leave(" = 0");
|
|
|
|
return 0;
|
|
|
|
|
afs: Overhaul volume and server record caching and fileserver rotation
The current code assumes that volumes and servers are per-cell and are
never shared, but this is not enforced, and, indeed, public cells do exist
that are aliases of each other. Further, an organisation can, say, set up
a public cell and a private cell with overlapping, but not identical, sets
of servers. The difference is purely in the database attached to the VL
servers.
The current code will malfunction if it sees a server in two cells as it
assumes global address -> server record mappings and that each server is in
just one cell.
Further, each server may have multiple addresses - and may have addresses
of different families (IPv4 and IPv6, say).
To this end, the following structural changes are made:
(1) Server record management is overhauled:
(a) Server records are made independent of cell. The namespace keeps
track of them, volume records have lists of them and each vnode
has a server on which its callback interest currently resides.
(b) The cell record no longer keeps a list of servers known to be in
that cell.
(c) The server records are now kept in a flat list because there's no
single address to sort on.
(d) Server records are now keyed by their UUID within the namespace.
(e) The addresses for a server are obtained with the VL.GetAddrsU
rather than with VL.GetEntryByName, using the server's UUID as a
parameter.
(f) Cached server records are garbage collected after a period of
non-use and are counted out of existence before purging is allowed
to complete. This protects the work functions against rmmod.
(g) The servers list is now in /proc/fs/afs/servers.
(2) Volume record management is overhauled:
(a) An RCU-replaceable server list is introduced. This tracks both
servers and their coresponding callback interests.
(b) The superblock is now keyed on cell record and numeric volume ID.
(c) The volume record is now tied to the superblock which mounts it,
and is activated when mounted and deactivated when unmounted.
This makes it easier to handle the cache cookie without causing a
double-use in fscache.
(d) The volume record is loaded from the VLDB using VL.GetEntryByNameU
to get the server UUID list.
(e) The volume name is updated if it is seen to have changed when the
volume is updated (the update is keyed on the volume ID).
(3) The vlocation record is got rid of and VLDB records are no longer
cached. Sufficient information is stored in the volume record, though
an update to a volume record is now no longer shared between related
volumes (volumes come in bundles of three: R/W, R/O and backup).
and the following procedural changes are made:
(1) The fileserver cursor introduced previously is now fleshed out and
used to iterate over fileservers and their addresses.
(2) Volume status is checked during iteration, and the server list is
replaced if a change is detected.
(3) Server status is checked during iteration, and the address list is
replaced if a change is detected.
(4) The abort code is saved into the address list cursor and -ECONNABORTED
returned in afs_make_call() if a remote abort happened rather than
translating the abort into an error message. This allows actions to
be taken depending on the abort code more easily.
(a) If a VMOVED abort is seen then this is handled by rechecking the
volume and restarting the iteration.
(b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is
handled by sleeping for a short period and retrying and/or trying
other servers that might serve that volume. A message is also
displayed once until the condition has cleared.
(c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the
moment.
(d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to
see if it has been deleted; if not, the fileserver is probably
indicating that the volume couldn't be attached and needs
salvaging.
(e) If statfs() sees one of these aborts, it does not sleep, but
rather returns an error, so as not to block the umount program.
(5) The fileserver iteration functions in vnode.c are now merged into
their callers and more heavily macroised around the cursor. vnode.c
is removed.
(6) Operations on a particular vnode are serialised on that vnode because
the server will lock that vnode whilst it operates on it, so a second
op sent will just have to wait.
(7) Fileservers are probed with FS.GetCapabilities before being used.
This is where service upgrade will be done.
(8) A callback interest on a fileserver is set up before an FS operation
is performed and passed through to afs_make_call() so that it can be
set on the vnode if the operation returns a callback. The callback
interest is passed through to afs_iget() also so that it can be set
there too.
In general, record updating is done on an as-needed basis when we try to
access servers, volumes or vnodes rather than offloading it to work items
and special threads.
Notes:
(1) Pre AFS-3.4 servers are no longer supported, though this can be added
back if necessary (AFS-3.4 was released in 1998).
(2) VBUSY is retried forever for the moment at intervals of 1s.
(3) /proc/fs/afs/<cell>/servers no longer exists.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 23:27:50 +08:00
|
|
|
error_key:
|
2007-04-27 06:59:35 +08:00
|
|
|
key_put(key);
|
|
|
|
error:
|
|
|
|
d_drop(dentry);
|
|
|
|
_leave(" = %d", ret);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* create a symlink in an AFS filesystem
|
|
|
|
*/
|
|
|
|
static int afs_symlink(struct inode *dir, struct dentry *dentry,
|
|
|
|
const char *content)
|
|
|
|
{
|
afs: Overhaul volume and server record caching and fileserver rotation
The current code assumes that volumes and servers are per-cell and are
never shared, but this is not enforced, and, indeed, public cells do exist
that are aliases of each other. Further, an organisation can, say, set up
a public cell and a private cell with overlapping, but not identical, sets
of servers. The difference is purely in the database attached to the VL
servers.
The current code will malfunction if it sees a server in two cells as it
assumes global address -> server record mappings and that each server is in
just one cell.
Further, each server may have multiple addresses - and may have addresses
of different families (IPv4 and IPv6, say).
To this end, the following structural changes are made:
(1) Server record management is overhauled:
(a) Server records are made independent of cell. The namespace keeps
track of them, volume records have lists of them and each vnode
has a server on which its callback interest currently resides.
(b) The cell record no longer keeps a list of servers known to be in
that cell.
(c) The server records are now kept in a flat list because there's no
single address to sort on.
(d) Server records are now keyed by their UUID within the namespace.
(e) The addresses for a server are obtained with the VL.GetAddrsU
rather than with VL.GetEntryByName, using the server's UUID as a
parameter.
(f) Cached server records are garbage collected after a period of
non-use and are counted out of existence before purging is allowed
to complete. This protects the work functions against rmmod.
(g) The servers list is now in /proc/fs/afs/servers.
(2) Volume record management is overhauled:
(a) An RCU-replaceable server list is introduced. This tracks both
servers and their coresponding callback interests.
(b) The superblock is now keyed on cell record and numeric volume ID.
(c) The volume record is now tied to the superblock which mounts it,
and is activated when mounted and deactivated when unmounted.
This makes it easier to handle the cache cookie without causing a
double-use in fscache.
(d) The volume record is loaded from the VLDB using VL.GetEntryByNameU
to get the server UUID list.
(e) The volume name is updated if it is seen to have changed when the
volume is updated (the update is keyed on the volume ID).
(3) The vlocation record is got rid of and VLDB records are no longer
cached. Sufficient information is stored in the volume record, though
an update to a volume record is now no longer shared between related
volumes (volumes come in bundles of three: R/W, R/O and backup).
and the following procedural changes are made:
(1) The fileserver cursor introduced previously is now fleshed out and
used to iterate over fileservers and their addresses.
(2) Volume status is checked during iteration, and the server list is
replaced if a change is detected.
(3) Server status is checked during iteration, and the address list is
replaced if a change is detected.
(4) The abort code is saved into the address list cursor and -ECONNABORTED
returned in afs_make_call() if a remote abort happened rather than
translating the abort into an error message. This allows actions to
be taken depending on the abort code more easily.
(a) If a VMOVED abort is seen then this is handled by rechecking the
volume and restarting the iteration.
(b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is
handled by sleeping for a short period and retrying and/or trying
other servers that might serve that volume. A message is also
displayed once until the condition has cleared.
(c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the
moment.
(d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to
see if it has been deleted; if not, the fileserver is probably
indicating that the volume couldn't be attached and needs
salvaging.
(e) If statfs() sees one of these aborts, it does not sleep, but
rather returns an error, so as not to block the umount program.
(5) The fileserver iteration functions in vnode.c are now merged into
their callers and more heavily macroised around the cursor. vnode.c
is removed.
(6) Operations on a particular vnode are serialised on that vnode because
the server will lock that vnode whilst it operates on it, so a second
op sent will just have to wait.
(7) Fileservers are probed with FS.GetCapabilities before being used.
This is where service upgrade will be done.
(8) A callback interest on a fileserver is set up before an FS operation
is performed and passed through to afs_make_call() so that it can be
set on the vnode if the operation returns a callback. The callback
interest is passed through to afs_iget() also so that it can be set
there too.
In general, record updating is done on an as-needed basis when we try to
access servers, volumes or vnodes rather than offloading it to work items
and special threads.
Notes:
(1) Pre AFS-3.4 servers are no longer supported, though this can be added
back if necessary (AFS-3.4 was released in 1998).
(2) VBUSY is retried forever for the moment at intervals of 1s.
(3) /proc/fs/afs/<cell>/servers no longer exists.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 23:27:50 +08:00
|
|
|
struct afs_fs_cursor fc;
|
|
|
|
struct afs_file_status newstatus;
|
|
|
|
struct afs_vnode *dvnode = AFS_FS_I(dir);
|
|
|
|
struct afs_fid newfid;
|
2007-04-27 06:59:35 +08:00
|
|
|
struct key *key;
|
|
|
|
int ret;
|
|
|
|
|
2014-10-22 08:11:25 +08:00
|
|
|
_enter("{%x:%u},{%pd},%s",
|
|
|
|
dvnode->fid.vid, dvnode->fid.vnode, dentry,
|
2007-04-27 06:59:35 +08:00
|
|
|
content);
|
|
|
|
|
afs: Overhaul volume and server record caching and fileserver rotation
The current code assumes that volumes and servers are per-cell and are
never shared, but this is not enforced, and, indeed, public cells do exist
that are aliases of each other. Further, an organisation can, say, set up
a public cell and a private cell with overlapping, but not identical, sets
of servers. The difference is purely in the database attached to the VL
servers.
The current code will malfunction if it sees a server in two cells as it
assumes global address -> server record mappings and that each server is in
just one cell.
Further, each server may have multiple addresses - and may have addresses
of different families (IPv4 and IPv6, say).
To this end, the following structural changes are made:
(1) Server record management is overhauled:
(a) Server records are made independent of cell. The namespace keeps
track of them, volume records have lists of them and each vnode
has a server on which its callback interest currently resides.
(b) The cell record no longer keeps a list of servers known to be in
that cell.
(c) The server records are now kept in a flat list because there's no
single address to sort on.
(d) Server records are now keyed by their UUID within the namespace.
(e) The addresses for a server are obtained with the VL.GetAddrsU
rather than with VL.GetEntryByName, using the server's UUID as a
parameter.
(f) Cached server records are garbage collected after a period of
non-use and are counted out of existence before purging is allowed
to complete. This protects the work functions against rmmod.
(g) The servers list is now in /proc/fs/afs/servers.
(2) Volume record management is overhauled:
(a) An RCU-replaceable server list is introduced. This tracks both
servers and their coresponding callback interests.
(b) The superblock is now keyed on cell record and numeric volume ID.
(c) The volume record is now tied to the superblock which mounts it,
and is activated when mounted and deactivated when unmounted.
This makes it easier to handle the cache cookie without causing a
double-use in fscache.
(d) The volume record is loaded from the VLDB using VL.GetEntryByNameU
to get the server UUID list.
(e) The volume name is updated if it is seen to have changed when the
volume is updated (the update is keyed on the volume ID).
(3) The vlocation record is got rid of and VLDB records are no longer
cached. Sufficient information is stored in the volume record, though
an update to a volume record is now no longer shared between related
volumes (volumes come in bundles of three: R/W, R/O and backup).
and the following procedural changes are made:
(1) The fileserver cursor introduced previously is now fleshed out and
used to iterate over fileservers and their addresses.
(2) Volume status is checked during iteration, and the server list is
replaced if a change is detected.
(3) Server status is checked during iteration, and the address list is
replaced if a change is detected.
(4) The abort code is saved into the address list cursor and -ECONNABORTED
returned in afs_make_call() if a remote abort happened rather than
translating the abort into an error message. This allows actions to
be taken depending on the abort code more easily.
(a) If a VMOVED abort is seen then this is handled by rechecking the
volume and restarting the iteration.
(b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is
handled by sleeping for a short period and retrying and/or trying
other servers that might serve that volume. A message is also
displayed once until the condition has cleared.
(c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the
moment.
(d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to
see if it has been deleted; if not, the fileserver is probably
indicating that the volume couldn't be attached and needs
salvaging.
(e) If statfs() sees one of these aborts, it does not sleep, but
rather returns an error, so as not to block the umount program.
(5) The fileserver iteration functions in vnode.c are now merged into
their callers and more heavily macroised around the cursor. vnode.c
is removed.
(6) Operations on a particular vnode are serialised on that vnode because
the server will lock that vnode whilst it operates on it, so a second
op sent will just have to wait.
(7) Fileservers are probed with FS.GetCapabilities before being used.
This is where service upgrade will be done.
(8) A callback interest on a fileserver is set up before an FS operation
is performed and passed through to afs_make_call() so that it can be
set on the vnode if the operation returns a callback. The callback
interest is passed through to afs_iget() also so that it can be set
there too.
In general, record updating is done on an as-needed basis when we try to
access servers, volumes or vnodes rather than offloading it to work items
and special threads.
Notes:
(1) Pre AFS-3.4 servers are no longer supported, though this can be added
back if necessary (AFS-3.4 was released in 1998).
(2) VBUSY is retried forever for the moment at intervals of 1s.
(3) /proc/fs/afs/<cell>/servers no longer exists.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 23:27:50 +08:00
|
|
|
ret = -ENAMETOOLONG;
|
|
|
|
if (dentry->d_name.len >= AFSNAMEMAX)
|
|
|
|
goto error;
|
|
|
|
|
2007-04-27 06:59:35 +08:00
|
|
|
ret = -EINVAL;
|
2007-05-11 13:22:20 +08:00
|
|
|
if (strlen(content) >= AFSPATHMAX)
|
2007-04-27 06:59:35 +08:00
|
|
|
goto error;
|
|
|
|
|
|
|
|
key = afs_request_key(dvnode->volume->cell);
|
|
|
|
if (IS_ERR(key)) {
|
|
|
|
ret = PTR_ERR(key);
|
|
|
|
goto error;
|
|
|
|
}
|
|
|
|
|
afs: Overhaul volume and server record caching and fileserver rotation
The current code assumes that volumes and servers are per-cell and are
never shared, but this is not enforced, and, indeed, public cells do exist
that are aliases of each other. Further, an organisation can, say, set up
a public cell and a private cell with overlapping, but not identical, sets
of servers. The difference is purely in the database attached to the VL
servers.
The current code will malfunction if it sees a server in two cells as it
assumes global address -> server record mappings and that each server is in
just one cell.
Further, each server may have multiple addresses - and may have addresses
of different families (IPv4 and IPv6, say).
To this end, the following structural changes are made:
(1) Server record management is overhauled:
(a) Server records are made independent of cell. The namespace keeps
track of them, volume records have lists of them and each vnode
has a server on which its callback interest currently resides.
(b) The cell record no longer keeps a list of servers known to be in
that cell.
(c) The server records are now kept in a flat list because there's no
single address to sort on.
(d) Server records are now keyed by their UUID within the namespace.
(e) The addresses for a server are obtained with the VL.GetAddrsU
rather than with VL.GetEntryByName, using the server's UUID as a
parameter.
(f) Cached server records are garbage collected after a period of
non-use and are counted out of existence before purging is allowed
to complete. This protects the work functions against rmmod.
(g) The servers list is now in /proc/fs/afs/servers.
(2) Volume record management is overhauled:
(a) An RCU-replaceable server list is introduced. This tracks both
servers and their coresponding callback interests.
(b) The superblock is now keyed on cell record and numeric volume ID.
(c) The volume record is now tied to the superblock which mounts it,
and is activated when mounted and deactivated when unmounted.
This makes it easier to handle the cache cookie without causing a
double-use in fscache.
(d) The volume record is loaded from the VLDB using VL.GetEntryByNameU
to get the server UUID list.
(e) The volume name is updated if it is seen to have changed when the
volume is updated (the update is keyed on the volume ID).
(3) The vlocation record is got rid of and VLDB records are no longer
cached. Sufficient information is stored in the volume record, though
an update to a volume record is now no longer shared between related
volumes (volumes come in bundles of three: R/W, R/O and backup).
and the following procedural changes are made:
(1) The fileserver cursor introduced previously is now fleshed out and
used to iterate over fileservers and their addresses.
(2) Volume status is checked during iteration, and the server list is
replaced if a change is detected.
(3) Server status is checked during iteration, and the address list is
replaced if a change is detected.
(4) The abort code is saved into the address list cursor and -ECONNABORTED
returned in afs_make_call() if a remote abort happened rather than
translating the abort into an error message. This allows actions to
be taken depending on the abort code more easily.
(a) If a VMOVED abort is seen then this is handled by rechecking the
volume and restarting the iteration.
(b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is
handled by sleeping for a short period and retrying and/or trying
other servers that might serve that volume. A message is also
displayed once until the condition has cleared.
(c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the
moment.
(d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to
see if it has been deleted; if not, the fileserver is probably
indicating that the volume couldn't be attached and needs
salvaging.
(e) If statfs() sees one of these aborts, it does not sleep, but
rather returns an error, so as not to block the umount program.
(5) The fileserver iteration functions in vnode.c are now merged into
their callers and more heavily macroised around the cursor. vnode.c
is removed.
(6) Operations on a particular vnode are serialised on that vnode because
the server will lock that vnode whilst it operates on it, so a second
op sent will just have to wait.
(7) Fileservers are probed with FS.GetCapabilities before being used.
This is where service upgrade will be done.
(8) A callback interest on a fileserver is set up before an FS operation
is performed and passed through to afs_make_call() so that it can be
set on the vnode if the operation returns a callback. The callback
interest is passed through to afs_iget() also so that it can be set
there too.
In general, record updating is done on an as-needed basis when we try to
access servers, volumes or vnodes rather than offloading it to work items
and special threads.
Notes:
(1) Pre AFS-3.4 servers are no longer supported, though this can be added
back if necessary (AFS-3.4 was released in 1998).
(2) VBUSY is retried forever for the moment at intervals of 1s.
(3) /proc/fs/afs/<cell>/servers no longer exists.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 23:27:50 +08:00
|
|
|
ret = -ERESTARTSYS;
|
|
|
|
if (afs_begin_vnode_operation(&fc, dvnode, key)) {
|
|
|
|
while (afs_select_fileserver(&fc)) {
|
|
|
|
fc.cb_break = dvnode->cb_break + dvnode->cb_s_break;
|
|
|
|
afs_fs_symlink(&fc, dentry->d_name.name, content,
|
|
|
|
&newfid, &newstatus);
|
|
|
|
}
|
2007-04-27 06:59:35 +08:00
|
|
|
|
afs: Overhaul volume and server record caching and fileserver rotation
The current code assumes that volumes and servers are per-cell and are
never shared, but this is not enforced, and, indeed, public cells do exist
that are aliases of each other. Further, an organisation can, say, set up
a public cell and a private cell with overlapping, but not identical, sets
of servers. The difference is purely in the database attached to the VL
servers.
The current code will malfunction if it sees a server in two cells as it
assumes global address -> server record mappings and that each server is in
just one cell.
Further, each server may have multiple addresses - and may have addresses
of different families (IPv4 and IPv6, say).
To this end, the following structural changes are made:
(1) Server record management is overhauled:
(a) Server records are made independent of cell. The namespace keeps
track of them, volume records have lists of them and each vnode
has a server on which its callback interest currently resides.
(b) The cell record no longer keeps a list of servers known to be in
that cell.
(c) The server records are now kept in a flat list because there's no
single address to sort on.
(d) Server records are now keyed by their UUID within the namespace.
(e) The addresses for a server are obtained with the VL.GetAddrsU
rather than with VL.GetEntryByName, using the server's UUID as a
parameter.
(f) Cached server records are garbage collected after a period of
non-use and are counted out of existence before purging is allowed
to complete. This protects the work functions against rmmod.
(g) The servers list is now in /proc/fs/afs/servers.
(2) Volume record management is overhauled:
(a) An RCU-replaceable server list is introduced. This tracks both
servers and their coresponding callback interests.
(b) The superblock is now keyed on cell record and numeric volume ID.
(c) The volume record is now tied to the superblock which mounts it,
and is activated when mounted and deactivated when unmounted.
This makes it easier to handle the cache cookie without causing a
double-use in fscache.
(d) The volume record is loaded from the VLDB using VL.GetEntryByNameU
to get the server UUID list.
(e) The volume name is updated if it is seen to have changed when the
volume is updated (the update is keyed on the volume ID).
(3) The vlocation record is got rid of and VLDB records are no longer
cached. Sufficient information is stored in the volume record, though
an update to a volume record is now no longer shared between related
volumes (volumes come in bundles of three: R/W, R/O and backup).
and the following procedural changes are made:
(1) The fileserver cursor introduced previously is now fleshed out and
used to iterate over fileservers and their addresses.
(2) Volume status is checked during iteration, and the server list is
replaced if a change is detected.
(3) Server status is checked during iteration, and the address list is
replaced if a change is detected.
(4) The abort code is saved into the address list cursor and -ECONNABORTED
returned in afs_make_call() if a remote abort happened rather than
translating the abort into an error message. This allows actions to
be taken depending on the abort code more easily.
(a) If a VMOVED abort is seen then this is handled by rechecking the
volume and restarting the iteration.
(b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is
handled by sleeping for a short period and retrying and/or trying
other servers that might serve that volume. A message is also
displayed once until the condition has cleared.
(c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the
moment.
(d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to
see if it has been deleted; if not, the fileserver is probably
indicating that the volume couldn't be attached and needs
salvaging.
(e) If statfs() sees one of these aborts, it does not sleep, but
rather returns an error, so as not to block the umount program.
(5) The fileserver iteration functions in vnode.c are now merged into
their callers and more heavily macroised around the cursor. vnode.c
is removed.
(6) Operations on a particular vnode are serialised on that vnode because
the server will lock that vnode whilst it operates on it, so a second
op sent will just have to wait.
(7) Fileservers are probed with FS.GetCapabilities before being used.
This is where service upgrade will be done.
(8) A callback interest on a fileserver is set up before an FS operation
is performed and passed through to afs_make_call() so that it can be
set on the vnode if the operation returns a callback. The callback
interest is passed through to afs_iget() also so that it can be set
there too.
In general, record updating is done on an as-needed basis when we try to
access servers, volumes or vnodes rather than offloading it to work items
and special threads.
Notes:
(1) Pre AFS-3.4 servers are no longer supported, though this can be added
back if necessary (AFS-3.4 was released in 1998).
(2) VBUSY is retried forever for the moment at intervals of 1s.
(3) /proc/fs/afs/<cell>/servers no longer exists.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 23:27:50 +08:00
|
|
|
afs_check_for_remote_deletion(&fc, fc.vnode);
|
|
|
|
afs_vnode_commit_status(&fc, dvnode, fc.cb_break);
|
|
|
|
afs_vnode_new_inode(&fc, dentry, &newfid, &newstatus, NULL);
|
|
|
|
ret = afs_end_vnode_operation(&fc);
|
|
|
|
if (ret < 0)
|
|
|
|
goto error_key;
|
2017-11-21 06:41:00 +08:00
|
|
|
} else {
|
|
|
|
goto error_key;
|
2007-04-27 06:59:35 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
key_put(key);
|
|
|
|
_leave(" = 0");
|
|
|
|
return 0;
|
|
|
|
|
afs: Overhaul volume and server record caching and fileserver rotation
The current code assumes that volumes and servers are per-cell and are
never shared, but this is not enforced, and, indeed, public cells do exist
that are aliases of each other. Further, an organisation can, say, set up
a public cell and a private cell with overlapping, but not identical, sets
of servers. The difference is purely in the database attached to the VL
servers.
The current code will malfunction if it sees a server in two cells as it
assumes global address -> server record mappings and that each server is in
just one cell.
Further, each server may have multiple addresses - and may have addresses
of different families (IPv4 and IPv6, say).
To this end, the following structural changes are made:
(1) Server record management is overhauled:
(a) Server records are made independent of cell. The namespace keeps
track of them, volume records have lists of them and each vnode
has a server on which its callback interest currently resides.
(b) The cell record no longer keeps a list of servers known to be in
that cell.
(c) The server records are now kept in a flat list because there's no
single address to sort on.
(d) Server records are now keyed by their UUID within the namespace.
(e) The addresses for a server are obtained with the VL.GetAddrsU
rather than with VL.GetEntryByName, using the server's UUID as a
parameter.
(f) Cached server records are garbage collected after a period of
non-use and are counted out of existence before purging is allowed
to complete. This protects the work functions against rmmod.
(g) The servers list is now in /proc/fs/afs/servers.
(2) Volume record management is overhauled:
(a) An RCU-replaceable server list is introduced. This tracks both
servers and their coresponding callback interests.
(b) The superblock is now keyed on cell record and numeric volume ID.
(c) The volume record is now tied to the superblock which mounts it,
and is activated when mounted and deactivated when unmounted.
This makes it easier to handle the cache cookie without causing a
double-use in fscache.
(d) The volume record is loaded from the VLDB using VL.GetEntryByNameU
to get the server UUID list.
(e) The volume name is updated if it is seen to have changed when the
volume is updated (the update is keyed on the volume ID).
(3) The vlocation record is got rid of and VLDB records are no longer
cached. Sufficient information is stored in the volume record, though
an update to a volume record is now no longer shared between related
volumes (volumes come in bundles of three: R/W, R/O and backup).
and the following procedural changes are made:
(1) The fileserver cursor introduced previously is now fleshed out and
used to iterate over fileservers and their addresses.
(2) Volume status is checked during iteration, and the server list is
replaced if a change is detected.
(3) Server status is checked during iteration, and the address list is
replaced if a change is detected.
(4) The abort code is saved into the address list cursor and -ECONNABORTED
returned in afs_make_call() if a remote abort happened rather than
translating the abort into an error message. This allows actions to
be taken depending on the abort code more easily.
(a) If a VMOVED abort is seen then this is handled by rechecking the
volume and restarting the iteration.
(b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is
handled by sleeping for a short period and retrying and/or trying
other servers that might serve that volume. A message is also
displayed once until the condition has cleared.
(c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the
moment.
(d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to
see if it has been deleted; if not, the fileserver is probably
indicating that the volume couldn't be attached and needs
salvaging.
(e) If statfs() sees one of these aborts, it does not sleep, but
rather returns an error, so as not to block the umount program.
(5) The fileserver iteration functions in vnode.c are now merged into
their callers and more heavily macroised around the cursor. vnode.c
is removed.
(6) Operations on a particular vnode are serialised on that vnode because
the server will lock that vnode whilst it operates on it, so a second
op sent will just have to wait.
(7) Fileservers are probed with FS.GetCapabilities before being used.
This is where service upgrade will be done.
(8) A callback interest on a fileserver is set up before an FS operation
is performed and passed through to afs_make_call() so that it can be
set on the vnode if the operation returns a callback. The callback
interest is passed through to afs_iget() also so that it can be set
there too.
In general, record updating is done on an as-needed basis when we try to
access servers, volumes or vnodes rather than offloading it to work items
and special threads.
Notes:
(1) Pre AFS-3.4 servers are no longer supported, though this can be added
back if necessary (AFS-3.4 was released in 1998).
(2) VBUSY is retried forever for the moment at intervals of 1s.
(3) /proc/fs/afs/<cell>/servers no longer exists.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 23:27:50 +08:00
|
|
|
error_key:
|
2007-04-27 06:59:35 +08:00
|
|
|
key_put(key);
|
|
|
|
error:
|
|
|
|
d_drop(dentry);
|
|
|
|
_leave(" = %d", ret);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* rename a file in an AFS filesystem and/or move it between directories
|
|
|
|
*/
|
|
|
|
static int afs_rename(struct inode *old_dir, struct dentry *old_dentry,
|
fs: make remaining filesystems use .rename2
This is trivial to do:
- add flags argument to foo_rename()
- check if flags is zero
- assign foo_rename() to .rename2 instead of .rename
This doesn't mean it's impossible to support RENAME_NOREPLACE for these
filesystems, but it is not trivial, like for local filesystems.
RENAME_NOREPLACE must guarantee atomicity (i.e. it shouldn't be possible
for a file to be created on one host while it is overwritten by rename on
another host).
Filesystems converted:
9p, afs, ceph, coda, ecryptfs, kernfs, lustre, ncpfs, nfs, ocfs2, orangefs.
After this, we can get rid of the duplicate interfaces for rename.
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Acked-by: David Howells <dhowells@redhat.com> [AFS]
Acked-by: Mike Marshall <hubcap@omnibond.com>
Cc: Eric Van Hensbergen <ericvh@gmail.com>
Cc: Ilya Dryomov <idryomov@gmail.com>
Cc: Jan Harkes <jaharkes@cs.cmu.edu>
Cc: Tyler Hicks <tyhicks@canonical.com>
Cc: Oleg Drokin <oleg.drokin@intel.com>
Cc: Trond Myklebust <trond.myklebust@primarydata.com>
Cc: Mark Fasheh <mfasheh@suse.com>
2016-09-27 17:03:58 +08:00
|
|
|
struct inode *new_dir, struct dentry *new_dentry,
|
|
|
|
unsigned int flags)
|
2007-04-27 06:59:35 +08:00
|
|
|
{
|
afs: Overhaul volume and server record caching and fileserver rotation
The current code assumes that volumes and servers are per-cell and are
never shared, but this is not enforced, and, indeed, public cells do exist
that are aliases of each other. Further, an organisation can, say, set up
a public cell and a private cell with overlapping, but not identical, sets
of servers. The difference is purely in the database attached to the VL
servers.
The current code will malfunction if it sees a server in two cells as it
assumes global address -> server record mappings and that each server is in
just one cell.
Further, each server may have multiple addresses - and may have addresses
of different families (IPv4 and IPv6, say).
To this end, the following structural changes are made:
(1) Server record management is overhauled:
(a) Server records are made independent of cell. The namespace keeps
track of them, volume records have lists of them and each vnode
has a server on which its callback interest currently resides.
(b) The cell record no longer keeps a list of servers known to be in
that cell.
(c) The server records are now kept in a flat list because there's no
single address to sort on.
(d) Server records are now keyed by their UUID within the namespace.
(e) The addresses for a server are obtained with the VL.GetAddrsU
rather than with VL.GetEntryByName, using the server's UUID as a
parameter.
(f) Cached server records are garbage collected after a period of
non-use and are counted out of existence before purging is allowed
to complete. This protects the work functions against rmmod.
(g) The servers list is now in /proc/fs/afs/servers.
(2) Volume record management is overhauled:
(a) An RCU-replaceable server list is introduced. This tracks both
servers and their coresponding callback interests.
(b) The superblock is now keyed on cell record and numeric volume ID.
(c) The volume record is now tied to the superblock which mounts it,
and is activated when mounted and deactivated when unmounted.
This makes it easier to handle the cache cookie without causing a
double-use in fscache.
(d) The volume record is loaded from the VLDB using VL.GetEntryByNameU
to get the server UUID list.
(e) The volume name is updated if it is seen to have changed when the
volume is updated (the update is keyed on the volume ID).
(3) The vlocation record is got rid of and VLDB records are no longer
cached. Sufficient information is stored in the volume record, though
an update to a volume record is now no longer shared between related
volumes (volumes come in bundles of three: R/W, R/O and backup).
and the following procedural changes are made:
(1) The fileserver cursor introduced previously is now fleshed out and
used to iterate over fileservers and their addresses.
(2) Volume status is checked during iteration, and the server list is
replaced if a change is detected.
(3) Server status is checked during iteration, and the address list is
replaced if a change is detected.
(4) The abort code is saved into the address list cursor and -ECONNABORTED
returned in afs_make_call() if a remote abort happened rather than
translating the abort into an error message. This allows actions to
be taken depending on the abort code more easily.
(a) If a VMOVED abort is seen then this is handled by rechecking the
volume and restarting the iteration.
(b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is
handled by sleeping for a short period and retrying and/or trying
other servers that might serve that volume. A message is also
displayed once until the condition has cleared.
(c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the
moment.
(d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to
see if it has been deleted; if not, the fileserver is probably
indicating that the volume couldn't be attached and needs
salvaging.
(e) If statfs() sees one of these aborts, it does not sleep, but
rather returns an error, so as not to block the umount program.
(5) The fileserver iteration functions in vnode.c are now merged into
their callers and more heavily macroised around the cursor. vnode.c
is removed.
(6) Operations on a particular vnode are serialised on that vnode because
the server will lock that vnode whilst it operates on it, so a second
op sent will just have to wait.
(7) Fileservers are probed with FS.GetCapabilities before being used.
This is where service upgrade will be done.
(8) A callback interest on a fileserver is set up before an FS operation
is performed and passed through to afs_make_call() so that it can be
set on the vnode if the operation returns a callback. The callback
interest is passed through to afs_iget() also so that it can be set
there too.
In general, record updating is done on an as-needed basis when we try to
access servers, volumes or vnodes rather than offloading it to work items
and special threads.
Notes:
(1) Pre AFS-3.4 servers are no longer supported, though this can be added
back if necessary (AFS-3.4 was released in 1998).
(2) VBUSY is retried forever for the moment at intervals of 1s.
(3) /proc/fs/afs/<cell>/servers no longer exists.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 23:27:50 +08:00
|
|
|
struct afs_fs_cursor fc;
|
2007-04-27 06:59:35 +08:00
|
|
|
struct afs_vnode *orig_dvnode, *new_dvnode, *vnode;
|
|
|
|
struct key *key;
|
|
|
|
int ret;
|
|
|
|
|
fs: make remaining filesystems use .rename2
This is trivial to do:
- add flags argument to foo_rename()
- check if flags is zero
- assign foo_rename() to .rename2 instead of .rename
This doesn't mean it's impossible to support RENAME_NOREPLACE for these
filesystems, but it is not trivial, like for local filesystems.
RENAME_NOREPLACE must guarantee atomicity (i.e. it shouldn't be possible
for a file to be created on one host while it is overwritten by rename on
another host).
Filesystems converted:
9p, afs, ceph, coda, ecryptfs, kernfs, lustre, ncpfs, nfs, ocfs2, orangefs.
After this, we can get rid of the duplicate interfaces for rename.
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Acked-by: David Howells <dhowells@redhat.com> [AFS]
Acked-by: Mike Marshall <hubcap@omnibond.com>
Cc: Eric Van Hensbergen <ericvh@gmail.com>
Cc: Ilya Dryomov <idryomov@gmail.com>
Cc: Jan Harkes <jaharkes@cs.cmu.edu>
Cc: Tyler Hicks <tyhicks@canonical.com>
Cc: Oleg Drokin <oleg.drokin@intel.com>
Cc: Trond Myklebust <trond.myklebust@primarydata.com>
Cc: Mark Fasheh <mfasheh@suse.com>
2016-09-27 17:03:58 +08:00
|
|
|
if (flags)
|
|
|
|
return -EINVAL;
|
|
|
|
|
2015-03-18 06:25:59 +08:00
|
|
|
vnode = AFS_FS_I(d_inode(old_dentry));
|
2007-04-27 06:59:35 +08:00
|
|
|
orig_dvnode = AFS_FS_I(old_dir);
|
|
|
|
new_dvnode = AFS_FS_I(new_dir);
|
|
|
|
|
2014-10-22 08:11:25 +08:00
|
|
|
_enter("{%x:%u},{%x:%u},{%x:%u},{%pd}",
|
2007-04-27 06:59:35 +08:00
|
|
|
orig_dvnode->fid.vid, orig_dvnode->fid.vnode,
|
|
|
|
vnode->fid.vid, vnode->fid.vnode,
|
|
|
|
new_dvnode->fid.vid, new_dvnode->fid.vnode,
|
2014-10-22 08:11:25 +08:00
|
|
|
new_dentry);
|
2007-04-27 06:59:35 +08:00
|
|
|
|
|
|
|
key = afs_request_key(orig_dvnode->volume->cell);
|
|
|
|
if (IS_ERR(key)) {
|
|
|
|
ret = PTR_ERR(key);
|
|
|
|
goto error;
|
|
|
|
}
|
|
|
|
|
afs: Overhaul volume and server record caching and fileserver rotation
The current code assumes that volumes and servers are per-cell and are
never shared, but this is not enforced, and, indeed, public cells do exist
that are aliases of each other. Further, an organisation can, say, set up
a public cell and a private cell with overlapping, but not identical, sets
of servers. The difference is purely in the database attached to the VL
servers.
The current code will malfunction if it sees a server in two cells as it
assumes global address -> server record mappings and that each server is in
just one cell.
Further, each server may have multiple addresses - and may have addresses
of different families (IPv4 and IPv6, say).
To this end, the following structural changes are made:
(1) Server record management is overhauled:
(a) Server records are made independent of cell. The namespace keeps
track of them, volume records have lists of them and each vnode
has a server on which its callback interest currently resides.
(b) The cell record no longer keeps a list of servers known to be in
that cell.
(c) The server records are now kept in a flat list because there's no
single address to sort on.
(d) Server records are now keyed by their UUID within the namespace.
(e) The addresses for a server are obtained with the VL.GetAddrsU
rather than with VL.GetEntryByName, using the server's UUID as a
parameter.
(f) Cached server records are garbage collected after a period of
non-use and are counted out of existence before purging is allowed
to complete. This protects the work functions against rmmod.
(g) The servers list is now in /proc/fs/afs/servers.
(2) Volume record management is overhauled:
(a) An RCU-replaceable server list is introduced. This tracks both
servers and their coresponding callback interests.
(b) The superblock is now keyed on cell record and numeric volume ID.
(c) The volume record is now tied to the superblock which mounts it,
and is activated when mounted and deactivated when unmounted.
This makes it easier to handle the cache cookie without causing a
double-use in fscache.
(d) The volume record is loaded from the VLDB using VL.GetEntryByNameU
to get the server UUID list.
(e) The volume name is updated if it is seen to have changed when the
volume is updated (the update is keyed on the volume ID).
(3) The vlocation record is got rid of and VLDB records are no longer
cached. Sufficient information is stored in the volume record, though
an update to a volume record is now no longer shared between related
volumes (volumes come in bundles of three: R/W, R/O and backup).
and the following procedural changes are made:
(1) The fileserver cursor introduced previously is now fleshed out and
used to iterate over fileservers and their addresses.
(2) Volume status is checked during iteration, and the server list is
replaced if a change is detected.
(3) Server status is checked during iteration, and the address list is
replaced if a change is detected.
(4) The abort code is saved into the address list cursor and -ECONNABORTED
returned in afs_make_call() if a remote abort happened rather than
translating the abort into an error message. This allows actions to
be taken depending on the abort code more easily.
(a) If a VMOVED abort is seen then this is handled by rechecking the
volume and restarting the iteration.
(b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is
handled by sleeping for a short period and retrying and/or trying
other servers that might serve that volume. A message is also
displayed once until the condition has cleared.
(c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the
moment.
(d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to
see if it has been deleted; if not, the fileserver is probably
indicating that the volume couldn't be attached and needs
salvaging.
(e) If statfs() sees one of these aborts, it does not sleep, but
rather returns an error, so as not to block the umount program.
(5) The fileserver iteration functions in vnode.c are now merged into
their callers and more heavily macroised around the cursor. vnode.c
is removed.
(6) Operations on a particular vnode are serialised on that vnode because
the server will lock that vnode whilst it operates on it, so a second
op sent will just have to wait.
(7) Fileservers are probed with FS.GetCapabilities before being used.
This is where service upgrade will be done.
(8) A callback interest on a fileserver is set up before an FS operation
is performed and passed through to afs_make_call() so that it can be
set on the vnode if the operation returns a callback. The callback
interest is passed through to afs_iget() also so that it can be set
there too.
In general, record updating is done on an as-needed basis when we try to
access servers, volumes or vnodes rather than offloading it to work items
and special threads.
Notes:
(1) Pre AFS-3.4 servers are no longer supported, though this can be added
back if necessary (AFS-3.4 was released in 1998).
(2) VBUSY is retried forever for the moment at intervals of 1s.
(3) /proc/fs/afs/<cell>/servers no longer exists.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 23:27:50 +08:00
|
|
|
ret = -ERESTARTSYS;
|
|
|
|
if (afs_begin_vnode_operation(&fc, orig_dvnode, key)) {
|
|
|
|
if (orig_dvnode != new_dvnode) {
|
|
|
|
if (mutex_lock_interruptible_nested(&new_dvnode->io_lock, 1) < 0) {
|
|
|
|
afs_end_vnode_operation(&fc);
|
2017-11-21 07:04:08 +08:00
|
|
|
goto error_key;
|
afs: Overhaul volume and server record caching and fileserver rotation
The current code assumes that volumes and servers are per-cell and are
never shared, but this is not enforced, and, indeed, public cells do exist
that are aliases of each other. Further, an organisation can, say, set up
a public cell and a private cell with overlapping, but not identical, sets
of servers. The difference is purely in the database attached to the VL
servers.
The current code will malfunction if it sees a server in two cells as it
assumes global address -> server record mappings and that each server is in
just one cell.
Further, each server may have multiple addresses - and may have addresses
of different families (IPv4 and IPv6, say).
To this end, the following structural changes are made:
(1) Server record management is overhauled:
(a) Server records are made independent of cell. The namespace keeps
track of them, volume records have lists of them and each vnode
has a server on which its callback interest currently resides.
(b) The cell record no longer keeps a list of servers known to be in
that cell.
(c) The server records are now kept in a flat list because there's no
single address to sort on.
(d) Server records are now keyed by their UUID within the namespace.
(e) The addresses for a server are obtained with the VL.GetAddrsU
rather than with VL.GetEntryByName, using the server's UUID as a
parameter.
(f) Cached server records are garbage collected after a period of
non-use and are counted out of existence before purging is allowed
to complete. This protects the work functions against rmmod.
(g) The servers list is now in /proc/fs/afs/servers.
(2) Volume record management is overhauled:
(a) An RCU-replaceable server list is introduced. This tracks both
servers and their coresponding callback interests.
(b) The superblock is now keyed on cell record and numeric volume ID.
(c) The volume record is now tied to the superblock which mounts it,
and is activated when mounted and deactivated when unmounted.
This makes it easier to handle the cache cookie without causing a
double-use in fscache.
(d) The volume record is loaded from the VLDB using VL.GetEntryByNameU
to get the server UUID list.
(e) The volume name is updated if it is seen to have changed when the
volume is updated (the update is keyed on the volume ID).
(3) The vlocation record is got rid of and VLDB records are no longer
cached. Sufficient information is stored in the volume record, though
an update to a volume record is now no longer shared between related
volumes (volumes come in bundles of three: R/W, R/O and backup).
and the following procedural changes are made:
(1) The fileserver cursor introduced previously is now fleshed out and
used to iterate over fileservers and their addresses.
(2) Volume status is checked during iteration, and the server list is
replaced if a change is detected.
(3) Server status is checked during iteration, and the address list is
replaced if a change is detected.
(4) The abort code is saved into the address list cursor and -ECONNABORTED
returned in afs_make_call() if a remote abort happened rather than
translating the abort into an error message. This allows actions to
be taken depending on the abort code more easily.
(a) If a VMOVED abort is seen then this is handled by rechecking the
volume and restarting the iteration.
(b) If a VBUSY, VRESTARTING or VSALVAGING abort is seen then this is
handled by sleeping for a short period and retrying and/or trying
other servers that might serve that volume. A message is also
displayed once until the condition has cleared.
(c) If a VOFFLINE abort is seen, then this is handled as VBUSY for the
moment.
(d) If a VNOVOL abort is seen, the volume is rechecked in the VLDB to
see if it has been deleted; if not, the fileserver is probably
indicating that the volume couldn't be attached and needs
salvaging.
(e) If statfs() sees one of these aborts, it does not sleep, but
rather returns an error, so as not to block the umount program.
(5) The fileserver iteration functions in vnode.c are now merged into
their callers and more heavily macroised around the cursor. vnode.c
is removed.
(6) Operations on a particular vnode are serialised on that vnode because
the server will lock that vnode whilst it operates on it, so a second
op sent will just have to wait.
(7) Fileservers are probed with FS.GetCapabilities before being used.
This is where service upgrade will be done.
(8) A callback interest on a fileserver is set up before an FS operation
is performed and passed through to afs_make_call() so that it can be
set on the vnode if the operation returns a callback. The callback
interest is passed through to afs_iget() also so that it can be set
there too.
In general, record updating is done on an as-needed basis when we try to
access servers, volumes or vnodes rather than offloading it to work items
and special threads.
Notes:
(1) Pre AFS-3.4 servers are no longer supported, though this can be added
back if necessary (AFS-3.4 was released in 1998).
(2) VBUSY is retried forever for the moment at intervals of 1s.
(3) /proc/fs/afs/<cell>/servers no longer exists.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 23:27:50 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
while (afs_select_fileserver(&fc)) {
|
|
|
|
fc.cb_break = orig_dvnode->cb_break + orig_dvnode->cb_s_break;
|
|
|
|
fc.cb_break_2 = new_dvnode->cb_break + new_dvnode->cb_s_break;
|
|
|
|
afs_fs_rename(&fc, old_dentry->d_name.name,
|
|
|
|
new_dvnode, new_dentry->d_name.name);
|
|
|
|
}
|
|
|
|
|
|
|
|
afs_vnode_commit_status(&fc, orig_dvnode, fc.cb_break);
|
|
|
|
afs_vnode_commit_status(&fc, new_dvnode, fc.cb_break_2);
|
|
|
|
if (orig_dvnode != new_dvnode)
|
|
|
|
mutex_unlock(&new_dvnode->io_lock);
|
|
|
|
ret = afs_end_vnode_operation(&fc);
|
|
|
|
if (ret < 0)
|
|
|
|
goto error_key;
|
|
|
|
}
|
|
|
|
|
|
|
|
error_key:
|
2007-04-27 06:59:35 +08:00
|
|
|
key_put(key);
|
|
|
|
error:
|
|
|
|
_leave(" = %d", ret);
|
|
|
|
return ret;
|
|
|
|
}
|