The BKL is completely out of the picture in the lockd and sunrpc code
these days. Update the antiquated comments that refer to it.
Signed-off-by: Jeff Layton <jlayton@primarydata.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
Testing has shown that the pool->sp_lock can be a bottleneck on a busy
server. Every time data is received on a socket, the server must take
that lock in order to dequeue a thread from the sp_threads list.
Address this problem by eliminating the sp_threads list (which contains
threads that are currently idle) and replacing it with a RQ_BUSY flag in
svc_rqst. This allows us to walk the sp_all_threads list under the
rcu_read_lock and find a suitable thread for the xprt by doing a
test_and_set_bit.
Note that we do still have a potential atomicity problem however with
this approach. We don't want svc_xprt_do_enqueue to set the
rqst->rq_xprt pointer unless a test_and_set_bit of RQ_BUSY returned
zero (which indicates that the thread was idle). But, by the time we
check that, the bit could be flipped by a waking thread.
To address this, we acquire a new per-rqst spinlock (rq_lock) and take
that before doing the test_and_set_bit. If that returns false, then we
can set rq_xprt and drop the spinlock. Then, when the thread wakes up,
it must set the bit under the same spinlock and can trust that if it was
already set then the rq_xprt is also properly set.
With this scheme, the case where we have an idle thread no longer needs
to take the highly contended pool->sp_lock at all, and that removes the
bottleneck.
That still leaves one issue: What of the case where we walk the whole
sp_all_threads list and don't find an idle thread? Because the search is
lockess, it's possible for the queueing to race with a thread that is
going to sleep. To address that, we queue the xprt and then search again.
If we find an idle thread at that point, we can't attach the xprt to it
directly since that might race with a different thread waking up and
finding it. All we can do is wake the idle thread back up and let it
attempt to find the now-queued xprt.
Signed-off-by: Jeff Layton <jlayton@primarydata.com>
Tested-by: Chris Worley <chris.worley@primarydata.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
In a later patch, we'll be removing some spinlocking around the socket
and thread queueing code in order to fix some contention problems. At
that point, the stats counters will no longer be protected by the
sp_lock.
Change the counters to atomic_long_t fields, except for the
"sockets_queued" counter which will still be manipulated under a
spinlock.
Signed-off-by: Jeff Layton <jlayton@primarydata.com>
Tested-by: Chris Worley <chris.worley@primarydata.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
...also make the manipulation of sp_all_threads list use RCU-friendly
functions.
Signed-off-by: Jeff Layton <jlayton@primarydata.com>
Tested-by: Chris Worley <chris.worley@primarydata.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
In a later patch, we'll want to be able to handle this flag without
holding the sp_lock. Change this field to an unsigned long flags
field, and declare a new flag in it that can be managed with atomic
bitops.
Signed-off-by: Jeff Layton <jlayton@primarydata.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
There are a couple of holes in the svc_rqst field on x86_64. Move the
rq_cachetype to a different location to eliminate both of them.
Signed-off-by: Jeff Layton <jlayton@primarydata.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
In a later patch, we're going to need some atomic bit flags. Since that
field will need to be an unsigned long, we mitigate that space
consumption by migrating some other bitflags to the new field. Start
with the rq_secure flag.
Signed-off-by: Jeff Layton <jlayton@primarydata.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
We're always _only_ waking up tasks from within the sp_threads list, so
we know that they are enqueued and alive. The rq_wait waitqueue is just
a distraction with extra atomic semantics.
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
rq_usedeferral and rq_splice_ok are used as 0 and 1, just defined to bool.
Signed-off-by: Kinglong Mee <kinglongmee@gmail.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
RPC_MAX_AUTH_SIZE is scattered around several places. Better to set it
once in the auth code, where this kind of estimate should be made. And
while we're at it we can leave it zero when we're not using krb5i or
krb5p.
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
After this we can handle for example getattr of very large ACLs.
Read, readdir, readlink are still special cases with their own limits.
Also we can't handle a new operation starting close to the end of a
page.
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
If an incoming NFS request is coming from the local host, then
nfsd will need to perform some special handling. So detect that
possibility and make the source visible in rq_local.
Signed-off-by: NeilBrown <neilb@suse.de>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
NFSv4 clients can contact port 2049 directly instead of needing the
portmapper.
Therefore a failure to register to the portmapper when starting an
NFSv4-only server isn't really a problem.
But Gareth Williams reports that an attempt to start an NFSv4-only
server without starting portmap fails:
#rpc.nfsd -N 2 -N 3
rpc.nfsd: writing fd to kernel failed: errno 111 (Connection refused)
rpc.nfsd: unable to set any sockets for nfsd
Add a flag to svc_version to tell the rpc layer it can safely ignore an
rpcbind failure in the NFSv4-only case.
Reported-by: Gareth Williams <gareth@garethwilliams.me.uk>
Reviewed-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Kinglong Mee <kinglongmee@gmail.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
I forgot to remove this in
afc59400d6 "nfsd4: cleanup: replace
rq_resused count by rq_next_page pointer".
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
There is a race in enqueueing thread to a pool and
waking up a thread.
lockd doesn't wake up on reception of lock granted callback
if svc_wake_up() is called before lockd's thread is added
to a pool.
Signed-off-by: Andriy Skulysh <Andriy_Skulysh@xyratex.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
This is a cleanup patch - makes code looks simplier.
It replaces widely used rqstp->rq_xprt->xpt_net by introduced SVC_NET(rqstp).
Signed-off-by: Stanislav Kinsbursky <skinsbursky@parallels.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
Move the rq_flavor into struct svc_cred, and use it in setclientid and
exchange_id comparisons as well.
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
This new routine is responsible for service registration in a specified
network context.
The idea is to separate service creation from per-net operations.
Note also: since registering service with svc_bind() can fail, the
service will be destroyed and during destruction it will try to
unregister itself from rpcbind. In this case unregistration has to be
skipped.
Signed-off-by: Stanislav Kinsbursky <skinsbursky@parallels.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
This patch introduces per-net Lockd initialization and destruction routines.
The logic is the same as in global Lockd up and down routines. Probably the
solution is not the best one. But at least it looks clear.
So per-net "up" routine are called only in case of lockd is running already. If
per-net resources are not allocated yet, then service is being registered with
local portmapper and lockd sockets created.
Per-net "down" routine is called on every lockd_down() call in case of global
users counter is not zero.
Signed-off-by: Stanislav Kinsbursky <skinsbursky@parallels.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
On service shutdown we can be sure, that no more users of it left except
current. Thus it looks like using current network namespace context is safe in
this case.
Signed-off-by: Stanislav Kinsbursky <skinsbursky@parallels.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Lockd and NFSd services will handle requests from and to many network
nsamespaces. And thus have to be registered and unregistered per network
namespace.
Signed-off-by: Stanislav Kinsbursky <skinsbursky@parallels.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
We have to call svc_rpcb_cleanup() explicitly from nfsd_last_thread() since
this function is registered as service shutdown callback and thus nobody else
will done it for us.
Signed-off-by: Stanislav Kinsbursky <skinsbursky@parallels.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
For IPv6 local address, lockd can not callback to client for
missing scope id when binding address at inet6_bind:
324 if (addr_type & IPV6_ADDR_LINKLOCAL) {
325 if (addr_len >= sizeof(struct sockaddr_in6) &&
326 addr->sin6_scope_id) {
327 /* Override any existing binding, if another one
328 * is supplied by user.
329 */
330 sk->sk_bound_dev_if = addr->sin6_scope_id;
331 }
332
333 /* Binding to link-local address requires an interface */
334 if (!sk->sk_bound_dev_if) {
335 err = -EINVAL;
336 goto out_unlock;
337 }
Replacing svc_addr_u by sockaddr_storage, let rqstp->rq_daddr contains more info
besides address.
Reviewed-by: Jeff Layton <jlayton@redhat.com>
Reviewed-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Mi Jinlong <mijinlong@cn.fujitsu.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
Use NUMA aware allocations to reduce latencies and increase throughput.
sunrpc kthreads can use kthread_create_on_node() if pool_mode is
"percpu" or "pernode", and svc_prepare_thread()/svc_init_buffer() can
also take into account NUMA node affinity for memory allocations.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
CC: "J. Bruce Fields" <bfields@fieldses.org>
CC: Neil Brown <neilb@suse.de>
CC: David Miller <davem@davemloft.net>
Reviewed-by: Greg Banks <gnb@fastmail.fm>
[bfields@redhat.com: fix up caller nfs41_callback_up]
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
* 'nfs-for-3.1' of git://git.linux-nfs.org/projects/trondmy/linux-nfs: (44 commits)
NFSv4: Don't use the delegation->inode in nfs_mark_return_delegation()
nfs: don't use d_move in nfs_async_rename_done
RDMA: Increasing RPCRDMA_MAX_DATA_SEGS
SUNRPC: Replace xprt->resend and xprt->sending with a priority queue
SUNRPC: Allow caller of rpc_sleep_on() to select priority levels
SUNRPC: Support dynamic slot allocation for TCP connections
SUNRPC: Clean up the slot table allocation
SUNRPC: Initalise the struct xprt upon allocation
SUNRPC: Ensure that we grab the XPRT_LOCK before calling xprt_alloc_slot
pnfs: simplify pnfs files module autoloading
nfs: document nfsv4 sillyrename issues
NFS: Convert nfs4_set_ds_client to EXPORT_SYMBOL_GPL
SUNRPC: Convert the backchannel exports to EXPORT_SYMBOL_GPL
SUNRPC: sunrpc should not explicitly depend on NFS config options
NFS: Clean up - simplify the switch to read/write-through-MDS
NFS: Move the pnfs write code into pnfs.c
NFS: Move the pnfs read code into pnfs.c
NFS: Allow the nfs_pageio_descriptor to signal that a re-coalesce is needed
NFS: Use the nfs_pageio_descriptor->pg_bsize in the read/write request
NFS: Cache rpc_ops in struct nfs_pageio_descriptor
...
It's sort of ridiculous that we've never had a working reply cache for
NFSv4.
On the other hand, we may still not: our current reply cache is likely
not very good, especially in the TCP case (which is the only case that
matters for v4). What we really need here is some serious testing.
Anyway, here's a start.
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
Change explicit references to CONFIG_NFS_V4_1 to implicit ones
Get rid of the unnecessary defines in backchannel_rqst.c and
bc_svc.c: the Makefile takes care of those dependency.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
* 'for-2.6.38' of git://linux-nfs.org/~bfields/linux: (62 commits)
nfsd4: fix callback restarting
nfsd: break lease on unlink, link, and rename
nfsd4: break lease on nfsd setattr
nfsd: don't support msnfs export option
nfsd4: initialize cb_per_client
nfsd4: allow restarting callbacks
nfsd4: simplify nfsd4_cb_prepare
nfsd4: give out delegations more quickly in 4.1 case
nfsd4: add helper function to run callbacks
nfsd4: make sure sequence flags are set after destroy_session
nfsd4: re-probe callback on connection loss
nfsd4: set sequence flag when backchannel is down
nfsd4: keep finer-grained callback status
rpc: allow xprt_class->setup to return a preexisting xprt
rpc: keep backchannel xprt as long as server connection
rpc: move sk_bc_xprt to svc_xprt
nfsd4: allow backchannel recovery
nfsd4: support BIND_CONN_TO_SESSION
nfsd4: modify session list under cl_lock
Documentation: fl_mylease no longer exists
...
Fix up conflicts in fs/nfsd/vfs.c with the vfs-scale work. The
vfs-scale work touched some msnfs cases, and this merge removes support
for that entirely, so the conflict was trivial to resolve.
Differentiate from server backchannel
Signed-off-by: Andy Adamson <andros@netapp.com>
Acked-by: Bruce Fields <bfields@redhat.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Currently we use -EAGAIN returns to determine when to drop a deferred
request. On its own, that is error-prone, as it makes us treat -EAGAIN
returns from other functions specially to prevent inadvertent dropping.
So, use a flag on the request instead.
Returning an error on request deferral is still required, to prevent
further processing, but we no longer need worry that an error return on
its own could result in a drop.
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
This reverts commit 59a252ff8c.
This helps in an entirely cached workload but not necessarily in
workloads that require waiting on disk.
Conflicts:
include/linux/sunrpc/svc.h
net/sunrpc/svc_xprt.c
Reported-by: Simon Kirby <sim@hostway.ca>
Tested-by: Jesper Krogh <jesper@krogh.cc>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
The version 4.1 DRC memory limit and tracking variables are server wide and
session specific. Replace struct svc_serv fields with globals.
Stop using the svc_serv sv_lock.
Add a spinlock to serialize access to the DRC limit management variables which
change on session creation and deletion (usage counter) or (future)
administrative action to adjust the total DRC memory limit.
Signed-off-by: Andy Adamson <andros@netapp.com>
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
This svc_xprt is passed on to the callback service thread to be later used
to processes incoming svc_rqst's
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
Implement the NFSv4.1 backchannel service. Invokes the common callback
processing logic svc_process_common() to authenticate the call and
dispatch the appropriate NFSv4.1 XDR decoder and operation procedure.
It then invokes bc_send() to send the reply over the same connection.
bc_send() is implemented in a separate patch.
At this time there is no slot validation or reply cache handling.
[nfs41: Preallocate rpc_rqst receive buffer for handling callbacks]
Signed-off-by: Ricardo Labiaga <Ricardo.Labiaga@netapp.com>
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
[Move bc_svc_process() declaration to correct patch]
Signed-off-by: Ricardo Labiaga <Ricardo.Labiaga@netapp.com>
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
Adds new list of rpc_xprt structures, and a readers/writers lock to
protect the list. The list is used to preallocate resources for
the backchannel during backchannel requests. Callbacks are not
expected to cause significant latency, so only one callback will
be allowed at this time.
It also adds a pointer to the NFS callback service so that
requests can be directed to it for processing.
New callback members added to svc_serv. The NFSv4.1 callback service will
sleep on the svc_serv->svc_cb_waitq until new callback requests arrive.
The request will be queued in svc_serv->svc_cb_list. This patch adds this
list, the sleep queue and spinlock to svc_serv.
[nfs41: NFSv4.1 callback support]
Signed-off-by: Ricardo Labiaga <ricardo.labiaga@netapp.com>
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
* 'for-2.6.30' of git://linux-nfs.org/~bfields/linux: (81 commits)
nfsd41: define nfsd4_set_statp as noop for !CONFIG_NFSD_V4
nfsd41: define NFSD_DRC_SIZE_SHIFT in set_max_drc
nfsd41: Documentation/filesystems/nfs41-server.txt
nfsd41: CREATE_EXCLUSIVE4_1
nfsd41: SUPPATTR_EXCLCREAT attribute
nfsd41: support for 3-word long attribute bitmask
nfsd: dynamically skip encoded fattr bitmap in _nfsd4_verify
nfsd41: pass writable attrs mask to nfsd4_decode_fattr
nfsd41: provide support for minor version 1 at rpc level
nfsd41: control nfsv4.1 svc via /proc/fs/nfsd/versions
nfsd41: add OPEN4_SHARE_ACCESS_WANT nfs4_stateid bmap
nfsd41: access_valid
nfsd41: clientid handling
nfsd41: check encode size for sessions maxresponse cached
nfsd41: stateid handling
nfsd: pass nfsd4_compound_state* to nfs4_preprocess_{state,seq}id_op
nfsd41: destroy_session operation
nfsd41: non-page DRC for solo sequence responses
nfsd41: Add a create session replay cache
nfsd41: create_session operation
...
Use no more than 1/128th of the number of free pages at nfsd startup for the
v4.1 DRC.
This is an arbitrary default which should probably end up under the control
of an administrator.
Signed-off-by: Andy Adamson <andros@netapp.com>
[moved added fields in struct svc_serv under CONFIG_NFSD_V4_1]
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
[fix set_max_drc calculation of sv_drc_max_pages]
[moved NFSD_DRC_SIZE_SHIFT's declaration up in header file]
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
On an NFSv4.1 server cache miss that causes an upcall, NFS4ERR_DELAY will be
returned. It is up to the NFSv4.1 client to resend only the operations that
have not been processed.
Initialize rq_usedeferral to 1 in svc_process(). It sill be turned off in
nfsd4_proc_compound() only when NFSv4.1 Sessions are used.
Note: this isn't an adequate solution on its own. It's acceptable as a way
to get some minimal 4.1 up and working, but we're going to have to find a
way to avoid returning DELAY in all common cases before 4.1 can really be
considered ready.
Signed-off-by: Andy Adamson <andros@netapp.com>
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
[nfsd41: reverse rq_nodeferral negative logic]
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
[sunrpc: initialize rq_usedeferral]
Signed-off-by: Andy Adamson <andros@netapp.com>
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Since an RPC service listener's protocol family is specified now via
svc_create_xprt(), it no longer needs to be passed to svc_create() or
svc_create_pooled(). Remove that argument from the synopsis of those
functions, and remove the sv_family field from the svc_serv struct.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
The sv_family field is going away. Instead of using sv_family, have
the svc_register() function take a protocol family argument.
Since this argument represents a protocol family, and not an address
family, this argument takes an int, as this is what is passed to
sock_create_kern(). Also make sure svc_register's helpers are
checking for PF_FOO instead of AF_FOO. The value of [AP]F_FOO are
equivalent; this is simply a symbolic change to reflect the semantics
of the value stored in that variable.
sock_create_kern() should return EPFNOSUPPORT if the passed-in
protocol family isn't supported, but it uses EAFNOSUPPORT for this
case. We will stick with that tradition here, as svc_register()
is called by the RPC server in the same path as sock_create_kern().
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Add /proc/fs/nfsd/pool_stats to export to userspace various
statistics about the operation of rpc server thread pools.
This patch is based on a forward-ported version of
knfsd-add-pool-thread-stats which has been shipping in the SGI
"Enhanced NFS" product since 2006 and which was previously
posted:
http://article.gmane.org/gmane.linux.nfs/10375
It has also been updated thus:
* moved EXPORT_SYMBOL() to near the function it exports
* made the new struct struct seq_operations const
* used SEQ_START_TOKEN instead of ((void *)1)
* merged fix from SGI PV 990526 "sunrpc: use dprintk instead of
printk in svc_pool_stats_*()" by Harshula Jayasuriya.
* merged fix from SGI PV 964001 "Crash reading pool_stats before
nfsds are started".
Signed-off-by: Greg Banks <gnb@sgi.com>
Signed-off-by: Harshula Jayasuriya <harshula@sgi.com>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Avoid overloading the CPU scheduler with enormous load averages
when handling high call-rate NFS loads. When the knfsd bottom half
is made aware of an incoming call by the socket layer, it tries to
choose an nfsd thread and wake it up. As long as there are idle
threads, one will be woken up.
If there are lot of nfsd threads (a sensible configuration when
the server is disk-bound or is running an HSM), there will be many
more nfsd threads than CPUs to run them. Under a high call-rate
low service-time workload, the result is that almost every nfsd is
runnable, but only a handful are actually able to run. This situation
causes two significant problems:
1. The CPU scheduler takes over 10% of each CPU, which is robbing
the nfsd threads of valuable CPU time.
2. At a high enough load, the nfsd threads starve userspace threads
of CPU time, to the point where daemons like portmap and rpc.mountd
do not schedule for tens of seconds at a time. Clients attempting
to mount an NFS filesystem timeout at the very first step (opening
a TCP connection to portmap) because portmap cannot wake up from
select() and call accept() in time.
Disclaimer: these effects were observed on a SLES9 kernel, modern
kernels' schedulers may behave more gracefully.
The solution is simple: keep in each svc_pool a counter of the number
of threads which have been woken but have not yet run, and do not wake
any more if that count reaches an arbitrary small threshold.
Testing was on a 4 CPU 4 NIC Altix using 4 IRIX clients, each with 16
synthetic client threads simulating an rsync (i.e. recursive directory
listing) workload reading from an i386 RH9 install image (161480
regular files in 10841 directories) on the server. That tree is small
enough to fill in the server's RAM so no disk traffic was involved.
This setup gives a sustained call rate in excess of 60000 calls/sec
before being CPU-bound on the server. The server was running 128 nfsds.
Profiling showed schedule() taking 6.7% of every CPU, and __wake_up()
taking 5.2%. This patch drops those contributions to 3.0% and 2.2%.
Load average was over 120 before the patch, and 20.9 after.
This patch is a forward-ported version of knfsd-avoid-nfsd-overload
which has been shipping in the SGI "Enhanced NFS" product since 2006.
It has been posted before:
http://article.gmane.org/gmane.linux.nfs/10374
Signed-off-by: Greg Banks <gnb@sgi.com>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Jeff Layton