2005-04-17 06:20:36 +08:00
|
|
|
/*
|
|
|
|
* fs/nfs/nfs4state.c
|
|
|
|
*
|
|
|
|
* Client-side XDR for NFSv4.
|
|
|
|
*
|
|
|
|
* Copyright (c) 2002 The Regents of the University of Michigan.
|
|
|
|
* All rights reserved.
|
|
|
|
*
|
|
|
|
* Kendrick Smith <kmsmith@umich.edu>
|
|
|
|
*
|
|
|
|
* Redistribution and use in source and binary forms, with or without
|
|
|
|
* modification, are permitted provided that the following conditions
|
|
|
|
* are met:
|
|
|
|
*
|
|
|
|
* 1. Redistributions of source code must retain the above copyright
|
|
|
|
* notice, this list of conditions and the following disclaimer.
|
|
|
|
* 2. Redistributions in binary form must reproduce the above copyright
|
|
|
|
* notice, this list of conditions and the following disclaimer in the
|
|
|
|
* documentation and/or other materials provided with the distribution.
|
|
|
|
* 3. Neither the name of the University nor the names of its
|
|
|
|
* contributors may be used to endorse or promote products derived
|
|
|
|
* from this software without specific prior written permission.
|
|
|
|
*
|
|
|
|
* THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
|
|
|
|
* WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
|
|
|
|
* MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
|
|
|
|
* DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
|
|
|
|
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
|
|
|
|
* CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
|
|
|
|
* SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
|
|
|
|
* BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
|
|
|
|
* LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
|
|
|
|
* NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
|
|
|
|
* SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
|
|
|
*
|
|
|
|
* Implementation of the NFSv4 state model. For the time being,
|
|
|
|
* this is minimal, but will be made much more complex in a
|
|
|
|
* subsequent patch.
|
|
|
|
*/
|
|
|
|
|
2007-07-09 04:49:11 +08:00
|
|
|
#include <linux/kernel.h>
|
2005-04-17 06:20:36 +08:00
|
|
|
#include <linux/slab.h>
|
2010-09-18 21:09:31 +08:00
|
|
|
#include <linux/fs.h>
|
2005-04-17 06:20:36 +08:00
|
|
|
#include <linux/nfs_fs.h>
|
|
|
|
#include <linux/nfs_idmap.h>
|
2006-01-03 16:55:23 +08:00
|
|
|
#include <linux/kthread.h>
|
|
|
|
#include <linux/module.h>
|
2007-07-03 01:58:33 +08:00
|
|
|
#include <linux/random.h>
|
2010-10-23 07:18:52 +08:00
|
|
|
#include <linux/ratelimit.h>
|
2005-04-17 06:20:36 +08:00
|
|
|
#include <linux/workqueue.h>
|
|
|
|
#include <linux/bitops.h>
|
NFS: Cache state owners after files are closed
Servers have a finite amount of memory to store NFSv4 open and lock
owners. Moreover, servers may have a difficult time determining when
they can reap their state owner table, thanks to gray areas in the
NFSv4 protocol specification. Thus clients should be careful to reuse
state owners when possible.
Currently Linux is not too careful. When a user has closed all her
files on one mount point, the state owner's reference count goes to
zero, and it is released. The next OPEN allocates a new one. A
workload that serially opens and closes files can run through a large
number of open owners this way.
When a state owner's reference count goes to zero, slap it onto a free
list for that nfs_server, with an expiry time. Garbage collect before
looking for a state owner. This makes state owners for active users
available for re-use.
Now that there can be unused state owners remaining at umount time,
purge the state owner free list when a server is destroyed. Also be
sure not to reclaim unused state owners during state recovery.
This change has benefits for the client as well. For some workloads,
this approach drops the number of OPEN_CONFIRM calls from the same as
the number of OPEN calls, down to just one. This reduces wire traffic
and thus open(2) latency. Before this patch, untarring a kernel
source tarball shows the OPEN_CONFIRM call counter steadily increasing
through the test. With the patch, the OPEN_CONFIRM count remains at 1
throughout the entire untar.
As long as the expiry time is kept short, I don't think garbage
collection should be terribly expensive, although it does bounce the
clp->cl_lock around a bit.
[ At some point we should rationalize the use of the nfs_server
->destroy method. ]
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
[Trond: Fixed a garbage collection race and a few efficiency issues]
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2011-12-07 05:13:48 +08:00
|
|
|
#include <linux/jiffies.h>
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2005-06-23 01:16:21 +08:00
|
|
|
#include "nfs4_fs.h"
|
2005-04-17 06:20:36 +08:00
|
|
|
#include "callback.h"
|
|
|
|
#include "delegation.h"
|
2006-08-23 08:06:10 +08:00
|
|
|
#include "internal.h"
|
2010-10-20 12:18:02 +08:00
|
|
|
#include "pnfs.h"
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
#define OPENOWNER_POOL_SIZE 8
|
|
|
|
|
2005-06-23 01:16:21 +08:00
|
|
|
const nfs4_stateid zero_stateid;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
static LIST_HEAD(nfs4_clientid_list);
|
|
|
|
|
2009-04-01 21:22:47 +08:00
|
|
|
int nfs4_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2011-04-25 02:28:18 +08:00
|
|
|
struct nfs4_setclientid_res clid = {
|
|
|
|
.clientid = clp->cl_clientid,
|
|
|
|
.confirm = clp->cl_confirm,
|
|
|
|
};
|
2009-03-19 08:48:06 +08:00
|
|
|
unsigned short port;
|
|
|
|
int status;
|
|
|
|
|
2011-04-25 02:28:18 +08:00
|
|
|
if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
|
|
|
|
goto do_confirm;
|
2009-03-19 08:48:06 +08:00
|
|
|
port = nfs_callback_tcpport;
|
|
|
|
if (clp->cl_addr.ss_family == AF_INET6)
|
|
|
|
port = nfs_callback_tcpport6;
|
|
|
|
|
2010-04-17 04:43:06 +08:00
|
|
|
status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
|
|
|
|
if (status != 0)
|
|
|
|
goto out;
|
2011-04-25 02:28:18 +08:00
|
|
|
clp->cl_clientid = clid.clientid;
|
|
|
|
clp->cl_confirm = clid.confirm;
|
|
|
|
set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
|
|
|
|
do_confirm:
|
2010-04-17 04:43:06 +08:00
|
|
|
status = nfs4_proc_setclientid_confirm(clp, &clid, cred);
|
|
|
|
if (status != 0)
|
|
|
|
goto out;
|
2011-04-25 02:28:18 +08:00
|
|
|
clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
|
2010-04-17 04:43:06 +08:00
|
|
|
nfs4_schedule_state_renewal(clp);
|
|
|
|
out:
|
2005-04-17 06:20:36 +08:00
|
|
|
return status;
|
|
|
|
}
|
|
|
|
|
2009-04-01 21:22:46 +08:00
|
|
|
struct rpc_cred *nfs4_get_machine_cred_locked(struct nfs_client *clp)
|
2008-04-09 04:02:17 +08:00
|
|
|
{
|
|
|
|
struct rpc_cred *cred = NULL;
|
|
|
|
|
|
|
|
if (clp->cl_machine_cred != NULL)
|
|
|
|
cred = get_rpccred(clp->cl_machine_cred);
|
|
|
|
return cred;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void nfs4_clear_machine_cred(struct nfs_client *clp)
|
|
|
|
{
|
|
|
|
struct rpc_cred *cred;
|
|
|
|
|
|
|
|
spin_lock(&clp->cl_lock);
|
|
|
|
cred = clp->cl_machine_cred;
|
|
|
|
clp->cl_machine_cred = NULL;
|
|
|
|
spin_unlock(&clp->cl_lock);
|
|
|
|
if (cred != NULL)
|
|
|
|
put_rpccred(cred);
|
|
|
|
}
|
|
|
|
|
2010-12-24 09:32:43 +08:00
|
|
|
static struct rpc_cred *
|
|
|
|
nfs4_get_renew_cred_server_locked(struct nfs_server *server)
|
2006-01-03 16:55:25 +08:00
|
|
|
{
|
2010-12-24 09:32:43 +08:00
|
|
|
struct rpc_cred *cred = NULL;
|
2006-01-03 16:55:25 +08:00
|
|
|
struct nfs4_state_owner *sp;
|
2007-07-03 01:58:33 +08:00
|
|
|
struct rb_node *pos;
|
2006-01-03 16:55:25 +08:00
|
|
|
|
2010-12-24 09:32:43 +08:00
|
|
|
for (pos = rb_first(&server->state_owners);
|
|
|
|
pos != NULL;
|
|
|
|
pos = rb_next(pos)) {
|
|
|
|
sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
|
2006-01-03 16:55:25 +08:00
|
|
|
if (list_empty(&sp->so_states))
|
|
|
|
continue;
|
|
|
|
cred = get_rpccred(sp->so_cred);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
return cred;
|
|
|
|
}
|
|
|
|
|
2010-12-24 09:32:43 +08:00
|
|
|
/**
|
|
|
|
* nfs4_get_renew_cred_locked - Acquire credential for a renew operation
|
|
|
|
* @clp: client state handle
|
|
|
|
*
|
|
|
|
* Returns an rpc_cred with reference count bumped, or NULL.
|
|
|
|
* Caller must hold clp->cl_lock.
|
|
|
|
*/
|
|
|
|
struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp)
|
|
|
|
{
|
|
|
|
struct rpc_cred *cred = NULL;
|
|
|
|
struct nfs_server *server;
|
|
|
|
|
|
|
|
rcu_read_lock();
|
|
|
|
list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
|
|
|
|
cred = nfs4_get_renew_cred_server_locked(server);
|
|
|
|
if (cred != NULL)
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
rcu_read_unlock();
|
|
|
|
return cred;
|
|
|
|
}
|
|
|
|
|
2009-04-01 21:22:49 +08:00
|
|
|
#if defined(CONFIG_NFS_V4_1)
|
|
|
|
|
2009-12-07 01:23:46 +08:00
|
|
|
static int nfs41_setup_state_renewal(struct nfs_client *clp)
|
|
|
|
{
|
|
|
|
int status;
|
|
|
|
struct nfs_fsinfo fsinfo;
|
|
|
|
|
2011-03-01 09:34:11 +08:00
|
|
|
if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) {
|
|
|
|
nfs4_schedule_state_renewal(clp);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2009-12-07 01:23:46 +08:00
|
|
|
status = nfs4_proc_get_lease_time(clp, &fsinfo);
|
|
|
|
if (status == 0) {
|
|
|
|
/* Update lease time and schedule renewal */
|
|
|
|
spin_lock(&clp->cl_lock);
|
|
|
|
clp->cl_lease_time = fsinfo.lease_time * HZ;
|
|
|
|
clp->cl_last_renewal = jiffies;
|
|
|
|
spin_unlock(&clp->cl_lock);
|
|
|
|
|
|
|
|
nfs4_schedule_state_renewal(clp);
|
|
|
|
}
|
|
|
|
|
|
|
|
return status;
|
|
|
|
}
|
|
|
|
|
2011-01-06 10:04:34 +08:00
|
|
|
/*
|
|
|
|
* Back channel returns NFS4ERR_DELAY for new requests when
|
|
|
|
* NFS4_SESSION_DRAINING is set so there is no work to be done when draining
|
|
|
|
* is ended.
|
|
|
|
*/
|
2009-12-15 13:27:58 +08:00
|
|
|
static void nfs4_end_drain_session(struct nfs_client *clp)
|
2009-12-07 01:57:34 +08:00
|
|
|
{
|
2009-12-15 13:27:58 +08:00
|
|
|
struct nfs4_session *ses = clp->cl_session;
|
2012-01-18 11:57:37 +08:00
|
|
|
struct nfs4_slot_table *tbl;
|
2009-12-15 13:27:56 +08:00
|
|
|
int max_slots;
|
|
|
|
|
2010-06-16 21:52:26 +08:00
|
|
|
if (ses == NULL)
|
|
|
|
return;
|
2012-01-18 11:57:37 +08:00
|
|
|
tbl = &ses->fc_slot_table;
|
2010-06-16 21:52:26 +08:00
|
|
|
if (test_and_clear_bit(NFS4_SESSION_DRAINING, &ses->session_state)) {
|
2012-01-18 11:57:37 +08:00
|
|
|
spin_lock(&tbl->slot_tbl_lock);
|
|
|
|
max_slots = tbl->max_slots;
|
2009-12-15 13:27:56 +08:00
|
|
|
while (max_slots--) {
|
2012-01-18 11:57:37 +08:00
|
|
|
if (rpc_wake_up_first(&tbl->slot_tbl_waitq,
|
|
|
|
nfs4_set_task_privileged,
|
|
|
|
NULL) == NULL)
|
2009-12-15 13:27:56 +08:00
|
|
|
break;
|
|
|
|
}
|
2012-01-18 11:57:37 +08:00
|
|
|
spin_unlock(&tbl->slot_tbl_lock);
|
2009-12-15 13:27:56 +08:00
|
|
|
}
|
2009-12-07 01:57:34 +08:00
|
|
|
}
|
|
|
|
|
2011-01-06 10:04:34 +08:00
|
|
|
static int nfs4_wait_on_slot_tbl(struct nfs4_slot_table *tbl)
|
2009-12-07 01:57:34 +08:00
|
|
|
{
|
|
|
|
spin_lock(&tbl->slot_tbl_lock);
|
2012-02-18 02:05:23 +08:00
|
|
|
if (tbl->highest_used_slotid != NFS4_NO_SLOT) {
|
2011-01-06 10:04:34 +08:00
|
|
|
INIT_COMPLETION(tbl->complete);
|
2009-12-07 01:57:34 +08:00
|
|
|
spin_unlock(&tbl->slot_tbl_lock);
|
2011-01-06 10:04:34 +08:00
|
|
|
return wait_for_completion_interruptible(&tbl->complete);
|
2009-12-07 01:57:34 +08:00
|
|
|
}
|
|
|
|
spin_unlock(&tbl->slot_tbl_lock);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2011-01-06 10:04:34 +08:00
|
|
|
static int nfs4_begin_drain_session(struct nfs_client *clp)
|
|
|
|
{
|
|
|
|
struct nfs4_session *ses = clp->cl_session;
|
|
|
|
int ret = 0;
|
|
|
|
|
|
|
|
set_bit(NFS4_SESSION_DRAINING, &ses->session_state);
|
|
|
|
/* back channel */
|
|
|
|
ret = nfs4_wait_on_slot_tbl(&ses->bc_slot_table);
|
|
|
|
if (ret)
|
|
|
|
return ret;
|
|
|
|
/* fore channel */
|
|
|
|
return nfs4_wait_on_slot_tbl(&ses->fc_slot_table);
|
|
|
|
}
|
|
|
|
|
2009-12-05 04:52:24 +08:00
|
|
|
int nfs41_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
|
|
|
|
{
|
|
|
|
int status;
|
|
|
|
|
2011-04-25 02:28:18 +08:00
|
|
|
if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
|
|
|
|
goto do_confirm;
|
2009-12-16 06:36:57 +08:00
|
|
|
nfs4_begin_drain_session(clp);
|
2009-12-05 04:52:24 +08:00
|
|
|
status = nfs4_proc_exchange_id(clp, cred);
|
2009-12-07 01:23:46 +08:00
|
|
|
if (status != 0)
|
|
|
|
goto out;
|
2011-04-25 02:28:18 +08:00
|
|
|
set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
|
|
|
|
do_confirm:
|
2009-12-07 01:23:46 +08:00
|
|
|
status = nfs4_proc_create_session(clp);
|
|
|
|
if (status != 0)
|
|
|
|
goto out;
|
2011-04-25 02:28:18 +08:00
|
|
|
clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
|
2009-12-07 01:23:46 +08:00
|
|
|
nfs41_setup_state_renewal(clp);
|
|
|
|
nfs_mark_client_ready(clp, NFS_CS_READY);
|
|
|
|
out:
|
2009-12-05 04:52:24 +08:00
|
|
|
return status;
|
|
|
|
}
|
|
|
|
|
2009-04-01 21:22:49 +08:00
|
|
|
struct rpc_cred *nfs4_get_exchange_id_cred(struct nfs_client *clp)
|
|
|
|
{
|
|
|
|
struct rpc_cred *cred;
|
|
|
|
|
|
|
|
spin_lock(&clp->cl_lock);
|
|
|
|
cred = nfs4_get_machine_cred_locked(clp);
|
|
|
|
spin_unlock(&clp->cl_lock);
|
|
|
|
return cred;
|
|
|
|
}
|
|
|
|
|
|
|
|
#endif /* CONFIG_NFS_V4_1 */
|
|
|
|
|
2010-12-24 09:32:43 +08:00
|
|
|
static struct rpc_cred *
|
|
|
|
nfs4_get_setclientid_cred_server(struct nfs_server *server)
|
2006-01-03 16:55:26 +08:00
|
|
|
{
|
2010-12-24 09:32:43 +08:00
|
|
|
struct nfs_client *clp = server->nfs_client;
|
|
|
|
struct rpc_cred *cred = NULL;
|
2006-01-03 16:55:26 +08:00
|
|
|
struct nfs4_state_owner *sp;
|
2007-07-03 01:58:33 +08:00
|
|
|
struct rb_node *pos;
|
2010-12-24 09:32:43 +08:00
|
|
|
|
|
|
|
spin_lock(&clp->cl_lock);
|
|
|
|
pos = rb_first(&server->state_owners);
|
|
|
|
if (pos != NULL) {
|
|
|
|
sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
|
|
|
|
cred = get_rpccred(sp->so_cred);
|
|
|
|
}
|
|
|
|
spin_unlock(&clp->cl_lock);
|
|
|
|
return cred;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* nfs4_get_setclientid_cred - Acquire credential for a setclientid operation
|
|
|
|
* @clp: client state handle
|
|
|
|
*
|
|
|
|
* Returns an rpc_cred with reference count bumped, or NULL.
|
|
|
|
*/
|
|
|
|
struct rpc_cred *nfs4_get_setclientid_cred(struct nfs_client *clp)
|
|
|
|
{
|
|
|
|
struct nfs_server *server;
|
2008-04-09 04:02:17 +08:00
|
|
|
struct rpc_cred *cred;
|
2006-01-03 16:55:26 +08:00
|
|
|
|
2008-12-24 04:21:41 +08:00
|
|
|
spin_lock(&clp->cl_lock);
|
|
|
|
cred = nfs4_get_machine_cred_locked(clp);
|
2010-12-24 09:32:43 +08:00
|
|
|
spin_unlock(&clp->cl_lock);
|
2008-04-09 04:02:17 +08:00
|
|
|
if (cred != NULL)
|
|
|
|
goto out;
|
2010-12-24 09:32:43 +08:00
|
|
|
|
|
|
|
rcu_read_lock();
|
|
|
|
list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
|
|
|
|
cred = nfs4_get_setclientid_cred_server(server);
|
|
|
|
if (cred != NULL)
|
|
|
|
break;
|
2006-01-03 16:55:26 +08:00
|
|
|
}
|
2010-12-24 09:32:43 +08:00
|
|
|
rcu_read_unlock();
|
|
|
|
|
2008-04-09 04:02:17 +08:00
|
|
|
out:
|
|
|
|
return cred;
|
2006-01-03 16:55:26 +08:00
|
|
|
}
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
static struct nfs4_state_owner *
|
2010-12-24 09:32:43 +08:00
|
|
|
nfs4_find_state_owner_locked(struct nfs_server *server, struct rpc_cred *cred)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2010-12-24 09:32:43 +08:00
|
|
|
struct rb_node **p = &server->state_owners.rb_node,
|
2007-07-03 01:58:33 +08:00
|
|
|
*parent = NULL;
|
2011-12-07 05:13:39 +08:00
|
|
|
struct nfs4_state_owner *sp;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2007-07-03 01:58:33 +08:00
|
|
|
while (*p != NULL) {
|
|
|
|
parent = *p;
|
2010-12-24 09:32:43 +08:00
|
|
|
sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
|
2007-07-03 01:58:33 +08:00
|
|
|
|
|
|
|
if (cred < sp->so_cred)
|
|
|
|
p = &parent->rb_left;
|
|
|
|
else if (cred > sp->so_cred)
|
|
|
|
p = &parent->rb_right;
|
|
|
|
else {
|
NFS: Cache state owners after files are closed
Servers have a finite amount of memory to store NFSv4 open and lock
owners. Moreover, servers may have a difficult time determining when
they can reap their state owner table, thanks to gray areas in the
NFSv4 protocol specification. Thus clients should be careful to reuse
state owners when possible.
Currently Linux is not too careful. When a user has closed all her
files on one mount point, the state owner's reference count goes to
zero, and it is released. The next OPEN allocates a new one. A
workload that serially opens and closes files can run through a large
number of open owners this way.
When a state owner's reference count goes to zero, slap it onto a free
list for that nfs_server, with an expiry time. Garbage collect before
looking for a state owner. This makes state owners for active users
available for re-use.
Now that there can be unused state owners remaining at umount time,
purge the state owner free list when a server is destroyed. Also be
sure not to reclaim unused state owners during state recovery.
This change has benefits for the client as well. For some workloads,
this approach drops the number of OPEN_CONFIRM calls from the same as
the number of OPEN calls, down to just one. This reduces wire traffic
and thus open(2) latency. Before this patch, untarring a kernel
source tarball shows the OPEN_CONFIRM call counter steadily increasing
through the test. With the patch, the OPEN_CONFIRM count remains at 1
throughout the entire untar.
As long as the expiry time is kept short, I don't think garbage
collection should be terribly expensive, although it does bounce the
clp->cl_lock around a bit.
[ At some point we should rationalize the use of the nfs_server
->destroy method. ]
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
[Trond: Fixed a garbage collection race and a few efficiency issues]
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2011-12-07 05:13:48 +08:00
|
|
|
if (!list_empty(&sp->so_lru))
|
|
|
|
list_del_init(&sp->so_lru);
|
2007-07-03 01:58:33 +08:00
|
|
|
atomic_inc(&sp->so_count);
|
2011-12-07 05:13:39 +08:00
|
|
|
return sp;
|
2007-07-03 01:58:33 +08:00
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
2011-12-07 05:13:39 +08:00
|
|
|
return NULL;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
2007-07-03 01:58:33 +08:00
|
|
|
static struct nfs4_state_owner *
|
2010-12-24 09:32:43 +08:00
|
|
|
nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
|
2007-07-03 01:58:33 +08:00
|
|
|
{
|
2010-12-24 09:32:43 +08:00
|
|
|
struct nfs_server *server = new->so_server;
|
|
|
|
struct rb_node **p = &server->state_owners.rb_node,
|
2007-07-03 01:58:33 +08:00
|
|
|
*parent = NULL;
|
|
|
|
struct nfs4_state_owner *sp;
|
2012-01-18 11:04:24 +08:00
|
|
|
int err;
|
2007-07-03 01:58:33 +08:00
|
|
|
|
|
|
|
while (*p != NULL) {
|
|
|
|
parent = *p;
|
2010-12-24 09:32:43 +08:00
|
|
|
sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
|
2007-07-03 01:58:33 +08:00
|
|
|
|
|
|
|
if (new->so_cred < sp->so_cred)
|
|
|
|
p = &parent->rb_left;
|
|
|
|
else if (new->so_cred > sp->so_cred)
|
|
|
|
p = &parent->rb_right;
|
|
|
|
else {
|
NFS: Cache state owners after files are closed
Servers have a finite amount of memory to store NFSv4 open and lock
owners. Moreover, servers may have a difficult time determining when
they can reap their state owner table, thanks to gray areas in the
NFSv4 protocol specification. Thus clients should be careful to reuse
state owners when possible.
Currently Linux is not too careful. When a user has closed all her
files on one mount point, the state owner's reference count goes to
zero, and it is released. The next OPEN allocates a new one. A
workload that serially opens and closes files can run through a large
number of open owners this way.
When a state owner's reference count goes to zero, slap it onto a free
list for that nfs_server, with an expiry time. Garbage collect before
looking for a state owner. This makes state owners for active users
available for re-use.
Now that there can be unused state owners remaining at umount time,
purge the state owner free list when a server is destroyed. Also be
sure not to reclaim unused state owners during state recovery.
This change has benefits for the client as well. For some workloads,
this approach drops the number of OPEN_CONFIRM calls from the same as
the number of OPEN calls, down to just one. This reduces wire traffic
and thus open(2) latency. Before this patch, untarring a kernel
source tarball shows the OPEN_CONFIRM call counter steadily increasing
through the test. With the patch, the OPEN_CONFIRM count remains at 1
throughout the entire untar.
As long as the expiry time is kept short, I don't think garbage
collection should be terribly expensive, although it does bounce the
clp->cl_lock around a bit.
[ At some point we should rationalize the use of the nfs_server
->destroy method. ]
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
[Trond: Fixed a garbage collection race and a few efficiency issues]
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2011-12-07 05:13:48 +08:00
|
|
|
if (!list_empty(&sp->so_lru))
|
|
|
|
list_del_init(&sp->so_lru);
|
2007-07-03 01:58:33 +08:00
|
|
|
atomic_inc(&sp->so_count);
|
|
|
|
return sp;
|
|
|
|
}
|
|
|
|
}
|
2012-01-18 11:04:25 +08:00
|
|
|
err = ida_get_new(&server->openowner_id, &new->so_seqid.owner_id);
|
2012-01-18 11:04:24 +08:00
|
|
|
if (err)
|
|
|
|
return ERR_PTR(err);
|
2010-12-24 09:32:43 +08:00
|
|
|
rb_link_node(&new->so_server_node, parent, p);
|
|
|
|
rb_insert_color(&new->so_server_node, &server->state_owners);
|
2007-07-03 01:58:33 +08:00
|
|
|
return new;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
2010-12-24 09:32:43 +08:00
|
|
|
nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
|
2007-07-03 01:58:33 +08:00
|
|
|
{
|
2010-12-24 09:32:43 +08:00
|
|
|
struct nfs_server *server = sp->so_server;
|
|
|
|
|
|
|
|
if (!RB_EMPTY_NODE(&sp->so_server_node))
|
|
|
|
rb_erase(&sp->so_server_node, &server->state_owners);
|
2012-01-18 11:04:25 +08:00
|
|
|
ida_remove(&server->openowner_id, sp->so_seqid.owner_id);
|
2007-07-03 01:58:33 +08:00
|
|
|
}
|
|
|
|
|
2012-01-18 11:04:25 +08:00
|
|
|
static void
|
|
|
|
nfs4_init_seqid_counter(struct nfs_seqid_counter *sc)
|
|
|
|
{
|
|
|
|
sc->flags = 0;
|
|
|
|
sc->counter = 0;
|
|
|
|
spin_lock_init(&sc->lock);
|
|
|
|
INIT_LIST_HEAD(&sc->list);
|
|
|
|
rpc_init_wait_queue(&sc->wait, "Seqid_waitqueue");
|
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
|
|
|
nfs4_destroy_seqid_counter(struct nfs_seqid_counter *sc)
|
|
|
|
{
|
|
|
|
rpc_destroy_wait_queue(&sc->wait);
|
|
|
|
}
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
/*
|
|
|
|
* nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
|
|
|
|
* create a new state_owner.
|
|
|
|
*
|
|
|
|
*/
|
|
|
|
static struct nfs4_state_owner *
|
2012-01-18 11:04:24 +08:00
|
|
|
nfs4_alloc_state_owner(struct nfs_server *server,
|
|
|
|
struct rpc_cred *cred,
|
|
|
|
gfp_t gfp_flags)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
|
|
|
struct nfs4_state_owner *sp;
|
|
|
|
|
2012-01-18 11:04:24 +08:00
|
|
|
sp = kzalloc(sizeof(*sp), gfp_flags);
|
2005-04-17 06:20:36 +08:00
|
|
|
if (!sp)
|
|
|
|
return NULL;
|
2012-01-18 11:04:24 +08:00
|
|
|
sp->so_server = server;
|
|
|
|
sp->so_cred = get_rpccred(cred);
|
2005-10-21 05:22:47 +08:00
|
|
|
spin_lock_init(&sp->so_lock);
|
2005-04-17 06:20:36 +08:00
|
|
|
INIT_LIST_HEAD(&sp->so_states);
|
2012-01-18 11:04:25 +08:00
|
|
|
nfs4_init_seqid_counter(&sp->so_seqid);
|
2005-04-17 06:20:36 +08:00
|
|
|
atomic_set(&sp->so_count, 1);
|
NFS: Cache state owners after files are closed
Servers have a finite amount of memory to store NFSv4 open and lock
owners. Moreover, servers may have a difficult time determining when
they can reap their state owner table, thanks to gray areas in the
NFSv4 protocol specification. Thus clients should be careful to reuse
state owners when possible.
Currently Linux is not too careful. When a user has closed all her
files on one mount point, the state owner's reference count goes to
zero, and it is released. The next OPEN allocates a new one. A
workload that serially opens and closes files can run through a large
number of open owners this way.
When a state owner's reference count goes to zero, slap it onto a free
list for that nfs_server, with an expiry time. Garbage collect before
looking for a state owner. This makes state owners for active users
available for re-use.
Now that there can be unused state owners remaining at umount time,
purge the state owner free list when a server is destroyed. Also be
sure not to reclaim unused state owners during state recovery.
This change has benefits for the client as well. For some workloads,
this approach drops the number of OPEN_CONFIRM calls from the same as
the number of OPEN calls, down to just one. This reduces wire traffic
and thus open(2) latency. Before this patch, untarring a kernel
source tarball shows the OPEN_CONFIRM call counter steadily increasing
through the test. With the patch, the OPEN_CONFIRM count remains at 1
throughout the entire untar.
As long as the expiry time is kept short, I don't think garbage
collection should be terribly expensive, although it does bounce the
clp->cl_lock around a bit.
[ At some point we should rationalize the use of the nfs_server
->destroy method. ]
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
[Trond: Fixed a garbage collection race and a few efficiency issues]
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2011-12-07 05:13:48 +08:00
|
|
|
INIT_LIST_HEAD(&sp->so_lru);
|
2005-04-17 06:20:36 +08:00
|
|
|
return sp;
|
|
|
|
}
|
|
|
|
|
2008-05-03 04:42:45 +08:00
|
|
|
static void
|
2005-04-17 06:20:36 +08:00
|
|
|
nfs4_drop_state_owner(struct nfs4_state_owner *sp)
|
|
|
|
{
|
2010-12-24 09:32:43 +08:00
|
|
|
if (!RB_EMPTY_NODE(&sp->so_server_node)) {
|
|
|
|
struct nfs_server *server = sp->so_server;
|
|
|
|
struct nfs_client *clp = server->nfs_client;
|
2007-07-03 01:58:33 +08:00
|
|
|
|
|
|
|
spin_lock(&clp->cl_lock);
|
2010-12-24 09:32:43 +08:00
|
|
|
rb_erase(&sp->so_server_node, &server->state_owners);
|
|
|
|
RB_CLEAR_NODE(&sp->so_server_node);
|
2007-07-03 01:58:33 +08:00
|
|
|
spin_unlock(&clp->cl_lock);
|
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
NFS: Cache state owners after files are closed
Servers have a finite amount of memory to store NFSv4 open and lock
owners. Moreover, servers may have a difficult time determining when
they can reap their state owner table, thanks to gray areas in the
NFSv4 protocol specification. Thus clients should be careful to reuse
state owners when possible.
Currently Linux is not too careful. When a user has closed all her
files on one mount point, the state owner's reference count goes to
zero, and it is released. The next OPEN allocates a new one. A
workload that serially opens and closes files can run through a large
number of open owners this way.
When a state owner's reference count goes to zero, slap it onto a free
list for that nfs_server, with an expiry time. Garbage collect before
looking for a state owner. This makes state owners for active users
available for re-use.
Now that there can be unused state owners remaining at umount time,
purge the state owner free list when a server is destroyed. Also be
sure not to reclaim unused state owners during state recovery.
This change has benefits for the client as well. For some workloads,
this approach drops the number of OPEN_CONFIRM calls from the same as
the number of OPEN calls, down to just one. This reduces wire traffic
and thus open(2) latency. Before this patch, untarring a kernel
source tarball shows the OPEN_CONFIRM call counter steadily increasing
through the test. With the patch, the OPEN_CONFIRM count remains at 1
throughout the entire untar.
As long as the expiry time is kept short, I don't think garbage
collection should be terribly expensive, although it does bounce the
clp->cl_lock around a bit.
[ At some point we should rationalize the use of the nfs_server
->destroy method. ]
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
[Trond: Fixed a garbage collection race and a few efficiency issues]
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2011-12-07 05:13:48 +08:00
|
|
|
static void nfs4_free_state_owner(struct nfs4_state_owner *sp)
|
|
|
|
{
|
2012-01-18 11:04:25 +08:00
|
|
|
nfs4_destroy_seqid_counter(&sp->so_seqid);
|
NFS: Cache state owners after files are closed
Servers have a finite amount of memory to store NFSv4 open and lock
owners. Moreover, servers may have a difficult time determining when
they can reap their state owner table, thanks to gray areas in the
NFSv4 protocol specification. Thus clients should be careful to reuse
state owners when possible.
Currently Linux is not too careful. When a user has closed all her
files on one mount point, the state owner's reference count goes to
zero, and it is released. The next OPEN allocates a new one. A
workload that serially opens and closes files can run through a large
number of open owners this way.
When a state owner's reference count goes to zero, slap it onto a free
list for that nfs_server, with an expiry time. Garbage collect before
looking for a state owner. This makes state owners for active users
available for re-use.
Now that there can be unused state owners remaining at umount time,
purge the state owner free list when a server is destroyed. Also be
sure not to reclaim unused state owners during state recovery.
This change has benefits for the client as well. For some workloads,
this approach drops the number of OPEN_CONFIRM calls from the same as
the number of OPEN calls, down to just one. This reduces wire traffic
and thus open(2) latency. Before this patch, untarring a kernel
source tarball shows the OPEN_CONFIRM call counter steadily increasing
through the test. With the patch, the OPEN_CONFIRM count remains at 1
throughout the entire untar.
As long as the expiry time is kept short, I don't think garbage
collection should be terribly expensive, although it does bounce the
clp->cl_lock around a bit.
[ At some point we should rationalize the use of the nfs_server
->destroy method. ]
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
[Trond: Fixed a garbage collection race and a few efficiency issues]
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2011-12-07 05:13:48 +08:00
|
|
|
put_rpccred(sp->so_cred);
|
|
|
|
kfree(sp);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void nfs4_gc_state_owners(struct nfs_server *server)
|
|
|
|
{
|
|
|
|
struct nfs_client *clp = server->nfs_client;
|
|
|
|
struct nfs4_state_owner *sp, *tmp;
|
|
|
|
unsigned long time_min, time_max;
|
|
|
|
LIST_HEAD(doomed);
|
|
|
|
|
|
|
|
spin_lock(&clp->cl_lock);
|
|
|
|
time_max = jiffies;
|
|
|
|
time_min = (long)time_max - (long)clp->cl_lease_time;
|
|
|
|
list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
|
|
|
|
/* NB: LRU is sorted so that oldest is at the head */
|
|
|
|
if (time_in_range(sp->so_expires, time_min, time_max))
|
|
|
|
break;
|
|
|
|
list_move(&sp->so_lru, &doomed);
|
|
|
|
nfs4_remove_state_owner_locked(sp);
|
|
|
|
}
|
|
|
|
spin_unlock(&clp->cl_lock);
|
|
|
|
|
|
|
|
list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
|
|
|
|
list_del(&sp->so_lru);
|
|
|
|
nfs4_free_state_owner(sp);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2010-12-24 09:32:43 +08:00
|
|
|
/**
|
|
|
|
* nfs4_get_state_owner - Look up a state owner given a credential
|
|
|
|
* @server: nfs_server to search
|
|
|
|
* @cred: RPC credential to match
|
|
|
|
*
|
|
|
|
* Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
|
|
|
|
*/
|
|
|
|
struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
|
2012-01-18 11:04:24 +08:00
|
|
|
struct rpc_cred *cred,
|
|
|
|
gfp_t gfp_flags)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2006-08-23 08:06:09 +08:00
|
|
|
struct nfs_client *clp = server->nfs_client;
|
2005-04-17 06:20:36 +08:00
|
|
|
struct nfs4_state_owner *sp, *new;
|
|
|
|
|
|
|
|
spin_lock(&clp->cl_lock);
|
2010-12-24 09:32:43 +08:00
|
|
|
sp = nfs4_find_state_owner_locked(server, cred);
|
2005-04-17 06:20:36 +08:00
|
|
|
spin_unlock(&clp->cl_lock);
|
|
|
|
if (sp != NULL)
|
NFS: Cache state owners after files are closed
Servers have a finite amount of memory to store NFSv4 open and lock
owners. Moreover, servers may have a difficult time determining when
they can reap their state owner table, thanks to gray areas in the
NFSv4 protocol specification. Thus clients should be careful to reuse
state owners when possible.
Currently Linux is not too careful. When a user has closed all her
files on one mount point, the state owner's reference count goes to
zero, and it is released. The next OPEN allocates a new one. A
workload that serially opens and closes files can run through a large
number of open owners this way.
When a state owner's reference count goes to zero, slap it onto a free
list for that nfs_server, with an expiry time. Garbage collect before
looking for a state owner. This makes state owners for active users
available for re-use.
Now that there can be unused state owners remaining at umount time,
purge the state owner free list when a server is destroyed. Also be
sure not to reclaim unused state owners during state recovery.
This change has benefits for the client as well. For some workloads,
this approach drops the number of OPEN_CONFIRM calls from the same as
the number of OPEN calls, down to just one. This reduces wire traffic
and thus open(2) latency. Before this patch, untarring a kernel
source tarball shows the OPEN_CONFIRM call counter steadily increasing
through the test. With the patch, the OPEN_CONFIRM count remains at 1
throughout the entire untar.
As long as the expiry time is kept short, I don't think garbage
collection should be terribly expensive, although it does bounce the
clp->cl_lock around a bit.
[ At some point we should rationalize the use of the nfs_server
->destroy method. ]
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
[Trond: Fixed a garbage collection race and a few efficiency issues]
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2011-12-07 05:13:48 +08:00
|
|
|
goto out;
|
2012-01-18 11:04:24 +08:00
|
|
|
new = nfs4_alloc_state_owner(server, cred, gfp_flags);
|
2007-07-03 01:58:33 +08:00
|
|
|
if (new == NULL)
|
NFS: Cache state owners after files are closed
Servers have a finite amount of memory to store NFSv4 open and lock
owners. Moreover, servers may have a difficult time determining when
they can reap their state owner table, thanks to gray areas in the
NFSv4 protocol specification. Thus clients should be careful to reuse
state owners when possible.
Currently Linux is not too careful. When a user has closed all her
files on one mount point, the state owner's reference count goes to
zero, and it is released. The next OPEN allocates a new one. A
workload that serially opens and closes files can run through a large
number of open owners this way.
When a state owner's reference count goes to zero, slap it onto a free
list for that nfs_server, with an expiry time. Garbage collect before
looking for a state owner. This makes state owners for active users
available for re-use.
Now that there can be unused state owners remaining at umount time,
purge the state owner free list when a server is destroyed. Also be
sure not to reclaim unused state owners during state recovery.
This change has benefits for the client as well. For some workloads,
this approach drops the number of OPEN_CONFIRM calls from the same as
the number of OPEN calls, down to just one. This reduces wire traffic
and thus open(2) latency. Before this patch, untarring a kernel
source tarball shows the OPEN_CONFIRM call counter steadily increasing
through the test. With the patch, the OPEN_CONFIRM count remains at 1
throughout the entire untar.
As long as the expiry time is kept short, I don't think garbage
collection should be terribly expensive, although it does bounce the
clp->cl_lock around a bit.
[ At some point we should rationalize the use of the nfs_server
->destroy method. ]
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
[Trond: Fixed a garbage collection race and a few efficiency issues]
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2011-12-07 05:13:48 +08:00
|
|
|
goto out;
|
2012-01-18 11:04:24 +08:00
|
|
|
do {
|
|
|
|
if (ida_pre_get(&server->openowner_id, gfp_flags) == 0)
|
|
|
|
break;
|
|
|
|
spin_lock(&clp->cl_lock);
|
|
|
|
sp = nfs4_insert_state_owner_locked(new);
|
|
|
|
spin_unlock(&clp->cl_lock);
|
|
|
|
} while (sp == ERR_PTR(-EAGAIN));
|
2012-01-18 11:04:24 +08:00
|
|
|
if (sp != new)
|
|
|
|
nfs4_free_state_owner(new);
|
NFS: Cache state owners after files are closed
Servers have a finite amount of memory to store NFSv4 open and lock
owners. Moreover, servers may have a difficult time determining when
they can reap their state owner table, thanks to gray areas in the
NFSv4 protocol specification. Thus clients should be careful to reuse
state owners when possible.
Currently Linux is not too careful. When a user has closed all her
files on one mount point, the state owner's reference count goes to
zero, and it is released. The next OPEN allocates a new one. A
workload that serially opens and closes files can run through a large
number of open owners this way.
When a state owner's reference count goes to zero, slap it onto a free
list for that nfs_server, with an expiry time. Garbage collect before
looking for a state owner. This makes state owners for active users
available for re-use.
Now that there can be unused state owners remaining at umount time,
purge the state owner free list when a server is destroyed. Also be
sure not to reclaim unused state owners during state recovery.
This change has benefits for the client as well. For some workloads,
this approach drops the number of OPEN_CONFIRM calls from the same as
the number of OPEN calls, down to just one. This reduces wire traffic
and thus open(2) latency. Before this patch, untarring a kernel
source tarball shows the OPEN_CONFIRM call counter steadily increasing
through the test. With the patch, the OPEN_CONFIRM count remains at 1
throughout the entire untar.
As long as the expiry time is kept short, I don't think garbage
collection should be terribly expensive, although it does bounce the
clp->cl_lock around a bit.
[ At some point we should rationalize the use of the nfs_server
->destroy method. ]
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
[Trond: Fixed a garbage collection race and a few efficiency issues]
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2011-12-07 05:13:48 +08:00
|
|
|
out:
|
|
|
|
nfs4_gc_state_owners(server);
|
2007-07-03 01:58:33 +08:00
|
|
|
return sp;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
2010-12-24 09:32:43 +08:00
|
|
|
/**
|
|
|
|
* nfs4_put_state_owner - Release a nfs4_state_owner
|
|
|
|
* @sp: state owner data to release
|
|
|
|
*/
|
2005-04-17 06:20:36 +08:00
|
|
|
void nfs4_put_state_owner(struct nfs4_state_owner *sp)
|
|
|
|
{
|
NFS: Cache state owners after files are closed
Servers have a finite amount of memory to store NFSv4 open and lock
owners. Moreover, servers may have a difficult time determining when
they can reap their state owner table, thanks to gray areas in the
NFSv4 protocol specification. Thus clients should be careful to reuse
state owners when possible.
Currently Linux is not too careful. When a user has closed all her
files on one mount point, the state owner's reference count goes to
zero, and it is released. The next OPEN allocates a new one. A
workload that serially opens and closes files can run through a large
number of open owners this way.
When a state owner's reference count goes to zero, slap it onto a free
list for that nfs_server, with an expiry time. Garbage collect before
looking for a state owner. This makes state owners for active users
available for re-use.
Now that there can be unused state owners remaining at umount time,
purge the state owner free list when a server is destroyed. Also be
sure not to reclaim unused state owners during state recovery.
This change has benefits for the client as well. For some workloads,
this approach drops the number of OPEN_CONFIRM calls from the same as
the number of OPEN calls, down to just one. This reduces wire traffic
and thus open(2) latency. Before this patch, untarring a kernel
source tarball shows the OPEN_CONFIRM call counter steadily increasing
through the test. With the patch, the OPEN_CONFIRM count remains at 1
throughout the entire untar.
As long as the expiry time is kept short, I don't think garbage
collection should be terribly expensive, although it does bounce the
clp->cl_lock around a bit.
[ At some point we should rationalize the use of the nfs_server
->destroy method. ]
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
[Trond: Fixed a garbage collection race and a few efficiency issues]
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2011-12-07 05:13:48 +08:00
|
|
|
struct nfs_server *server = sp->so_server;
|
|
|
|
struct nfs_client *clp = server->nfs_client;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
|
|
|
|
return;
|
NFS: Cache state owners after files are closed
Servers have a finite amount of memory to store NFSv4 open and lock
owners. Moreover, servers may have a difficult time determining when
they can reap their state owner table, thanks to gray areas in the
NFSv4 protocol specification. Thus clients should be careful to reuse
state owners when possible.
Currently Linux is not too careful. When a user has closed all her
files on one mount point, the state owner's reference count goes to
zero, and it is released. The next OPEN allocates a new one. A
workload that serially opens and closes files can run through a large
number of open owners this way.
When a state owner's reference count goes to zero, slap it onto a free
list for that nfs_server, with an expiry time. Garbage collect before
looking for a state owner. This makes state owners for active users
available for re-use.
Now that there can be unused state owners remaining at umount time,
purge the state owner free list when a server is destroyed. Also be
sure not to reclaim unused state owners during state recovery.
This change has benefits for the client as well. For some workloads,
this approach drops the number of OPEN_CONFIRM calls from the same as
the number of OPEN calls, down to just one. This reduces wire traffic
and thus open(2) latency. Before this patch, untarring a kernel
source tarball shows the OPEN_CONFIRM call counter steadily increasing
through the test. With the patch, the OPEN_CONFIRM count remains at 1
throughout the entire untar.
As long as the expiry time is kept short, I don't think garbage
collection should be terribly expensive, although it does bounce the
clp->cl_lock around a bit.
[ At some point we should rationalize the use of the nfs_server
->destroy method. ]
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
[Trond: Fixed a garbage collection race and a few efficiency issues]
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2011-12-07 05:13:48 +08:00
|
|
|
|
|
|
|
if (!RB_EMPTY_NODE(&sp->so_server_node)) {
|
|
|
|
sp->so_expires = jiffies;
|
|
|
|
list_add_tail(&sp->so_lru, &server->state_owners_lru);
|
|
|
|
spin_unlock(&clp->cl_lock);
|
|
|
|
} else {
|
|
|
|
nfs4_remove_state_owner_locked(sp);
|
|
|
|
spin_unlock(&clp->cl_lock);
|
|
|
|
nfs4_free_state_owner(sp);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* nfs4_purge_state_owners - Release all cached state owners
|
|
|
|
* @server: nfs_server with cached state owners to release
|
|
|
|
*
|
|
|
|
* Called at umount time. Remaining state owners will be on
|
|
|
|
* the LRU with ref count of zero.
|
|
|
|
*/
|
|
|
|
void nfs4_purge_state_owners(struct nfs_server *server)
|
|
|
|
{
|
|
|
|
struct nfs_client *clp = server->nfs_client;
|
|
|
|
struct nfs4_state_owner *sp, *tmp;
|
|
|
|
LIST_HEAD(doomed);
|
|
|
|
|
|
|
|
spin_lock(&clp->cl_lock);
|
|
|
|
list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
|
|
|
|
list_move(&sp->so_lru, &doomed);
|
|
|
|
nfs4_remove_state_owner_locked(sp);
|
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
spin_unlock(&clp->cl_lock);
|
NFS: Cache state owners after files are closed
Servers have a finite amount of memory to store NFSv4 open and lock
owners. Moreover, servers may have a difficult time determining when
they can reap their state owner table, thanks to gray areas in the
NFSv4 protocol specification. Thus clients should be careful to reuse
state owners when possible.
Currently Linux is not too careful. When a user has closed all her
files on one mount point, the state owner's reference count goes to
zero, and it is released. The next OPEN allocates a new one. A
workload that serially opens and closes files can run through a large
number of open owners this way.
When a state owner's reference count goes to zero, slap it onto a free
list for that nfs_server, with an expiry time. Garbage collect before
looking for a state owner. This makes state owners for active users
available for re-use.
Now that there can be unused state owners remaining at umount time,
purge the state owner free list when a server is destroyed. Also be
sure not to reclaim unused state owners during state recovery.
This change has benefits for the client as well. For some workloads,
this approach drops the number of OPEN_CONFIRM calls from the same as
the number of OPEN calls, down to just one. This reduces wire traffic
and thus open(2) latency. Before this patch, untarring a kernel
source tarball shows the OPEN_CONFIRM call counter steadily increasing
through the test. With the patch, the OPEN_CONFIRM count remains at 1
throughout the entire untar.
As long as the expiry time is kept short, I don't think garbage
collection should be terribly expensive, although it does bounce the
clp->cl_lock around a bit.
[ At some point we should rationalize the use of the nfs_server
->destroy method. ]
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
[Trond: Fixed a garbage collection race and a few efficiency issues]
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2011-12-07 05:13:48 +08:00
|
|
|
|
|
|
|
list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
|
|
|
|
list_del(&sp->so_lru);
|
|
|
|
nfs4_free_state_owner(sp);
|
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
static struct nfs4_state *
|
|
|
|
nfs4_alloc_open_state(void)
|
|
|
|
{
|
|
|
|
struct nfs4_state *state;
|
|
|
|
|
2010-05-14 00:51:01 +08:00
|
|
|
state = kzalloc(sizeof(*state), GFP_NOFS);
|
2005-04-17 06:20:36 +08:00
|
|
|
if (!state)
|
|
|
|
return NULL;
|
|
|
|
atomic_set(&state->count, 1);
|
|
|
|
INIT_LIST_HEAD(&state->lock_states);
|
2005-06-23 01:16:32 +08:00
|
|
|
spin_lock_init(&state->state_lock);
|
2007-07-09 22:45:42 +08:00
|
|
|
seqlock_init(&state->seqlock);
|
2005-04-17 06:20:36 +08:00
|
|
|
return state;
|
|
|
|
}
|
|
|
|
|
2005-11-05 04:32:58 +08:00
|
|
|
void
|
2008-12-24 04:21:56 +08:00
|
|
|
nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
|
2005-11-05 04:32:58 +08:00
|
|
|
{
|
2008-12-24 04:21:56 +08:00
|
|
|
if (state->state == fmode)
|
2005-11-05 04:32:58 +08:00
|
|
|
return;
|
|
|
|
/* NB! List reordering - see the reclaim code for why. */
|
2008-12-24 04:21:56 +08:00
|
|
|
if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
|
|
|
|
if (fmode & FMODE_WRITE)
|
2005-11-05 04:32:58 +08:00
|
|
|
list_move(&state->open_states, &state->owner->so_states);
|
|
|
|
else
|
|
|
|
list_move_tail(&state->open_states, &state->owner->so_states);
|
|
|
|
}
|
2008-12-24 04:21:56 +08:00
|
|
|
state->state = fmode;
|
2005-11-05 04:32:58 +08:00
|
|
|
}
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
static struct nfs4_state *
|
|
|
|
__nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
|
|
|
|
{
|
|
|
|
struct nfs_inode *nfsi = NFS_I(inode);
|
|
|
|
struct nfs4_state *state;
|
|
|
|
|
|
|
|
list_for_each_entry(state, &nfsi->open_states, inode_states) {
|
2007-07-04 02:41:19 +08:00
|
|
|
if (state->owner != owner)
|
2005-04-17 06:20:36 +08:00
|
|
|
continue;
|
2007-07-04 02:41:19 +08:00
|
|
|
if (atomic_inc_not_zero(&state->count))
|
2005-04-17 06:20:36 +08:00
|
|
|
return state;
|
|
|
|
}
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
|
|
|
nfs4_free_open_state(struct nfs4_state *state)
|
|
|
|
{
|
|
|
|
kfree(state);
|
|
|
|
}
|
|
|
|
|
|
|
|
struct nfs4_state *
|
|
|
|
nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
|
|
|
|
{
|
|
|
|
struct nfs4_state *state, *new;
|
|
|
|
struct nfs_inode *nfsi = NFS_I(inode);
|
|
|
|
|
|
|
|
spin_lock(&inode->i_lock);
|
|
|
|
state = __nfs4_find_state_byowner(inode, owner);
|
|
|
|
spin_unlock(&inode->i_lock);
|
|
|
|
if (state)
|
|
|
|
goto out;
|
|
|
|
new = nfs4_alloc_open_state();
|
2005-10-21 05:22:47 +08:00
|
|
|
spin_lock(&owner->so_lock);
|
2005-04-17 06:20:36 +08:00
|
|
|
spin_lock(&inode->i_lock);
|
|
|
|
state = __nfs4_find_state_byowner(inode, owner);
|
|
|
|
if (state == NULL && new != NULL) {
|
|
|
|
state = new;
|
|
|
|
state->owner = owner;
|
|
|
|
atomic_inc(&owner->so_count);
|
|
|
|
list_add(&state->inode_states, &nfsi->open_states);
|
2011-03-29 15:08:50 +08:00
|
|
|
ihold(inode);
|
|
|
|
state->inode = inode;
|
2005-04-17 06:20:36 +08:00
|
|
|
spin_unlock(&inode->i_lock);
|
2005-10-21 05:22:47 +08:00
|
|
|
/* Note: The reclaim code dictates that we add stateless
|
|
|
|
* and read-only stateids to the end of the list */
|
|
|
|
list_add_tail(&state->open_states, &owner->so_states);
|
|
|
|
spin_unlock(&owner->so_lock);
|
2005-04-17 06:20:36 +08:00
|
|
|
} else {
|
|
|
|
spin_unlock(&inode->i_lock);
|
2005-10-21 05:22:47 +08:00
|
|
|
spin_unlock(&owner->so_lock);
|
2005-04-17 06:20:36 +08:00
|
|
|
if (new)
|
|
|
|
nfs4_free_open_state(new);
|
|
|
|
}
|
|
|
|
out:
|
|
|
|
return state;
|
|
|
|
}
|
|
|
|
|
|
|
|
void nfs4_put_open_state(struct nfs4_state *state)
|
|
|
|
{
|
|
|
|
struct inode *inode = state->inode;
|
|
|
|
struct nfs4_state_owner *owner = state->owner;
|
|
|
|
|
2005-10-21 05:22:47 +08:00
|
|
|
if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
|
2005-04-17 06:20:36 +08:00
|
|
|
return;
|
2005-10-21 05:22:47 +08:00
|
|
|
spin_lock(&inode->i_lock);
|
2007-07-27 22:23:05 +08:00
|
|
|
list_del(&state->inode_states);
|
2005-04-17 06:20:36 +08:00
|
|
|
list_del(&state->open_states);
|
2005-10-21 05:22:47 +08:00
|
|
|
spin_unlock(&inode->i_lock);
|
|
|
|
spin_unlock(&owner->so_lock);
|
2005-04-17 06:20:36 +08:00
|
|
|
iput(inode);
|
|
|
|
nfs4_free_open_state(state);
|
|
|
|
nfs4_put_state_owner(owner);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
2005-10-19 05:20:13 +08:00
|
|
|
* Close the current file.
|
2005-04-17 06:20:36 +08:00
|
|
|
*/
|
2011-06-23 06:20:23 +08:00
|
|
|
static void __nfs4_close(struct nfs4_state *state,
|
2010-05-14 00:51:01 +08:00
|
|
|
fmode_t fmode, gfp_t gfp_mask, int wait)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
|
|
|
struct nfs4_state_owner *owner = state->owner;
|
2007-07-06 06:07:55 +08:00
|
|
|
int call_close = 0;
|
2008-12-24 04:21:56 +08:00
|
|
|
fmode_t newstate;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
atomic_inc(&owner->so_count);
|
|
|
|
/* Protect against nfs4_find_state() */
|
2005-10-21 05:22:47 +08:00
|
|
|
spin_lock(&owner->so_lock);
|
2008-12-24 04:21:56 +08:00
|
|
|
switch (fmode & (FMODE_READ | FMODE_WRITE)) {
|
2006-01-03 16:55:13 +08:00
|
|
|
case FMODE_READ:
|
|
|
|
state->n_rdonly--;
|
|
|
|
break;
|
|
|
|
case FMODE_WRITE:
|
|
|
|
state->n_wronly--;
|
|
|
|
break;
|
|
|
|
case FMODE_READ|FMODE_WRITE:
|
|
|
|
state->n_rdwr--;
|
|
|
|
}
|
2007-07-06 06:07:55 +08:00
|
|
|
newstate = FMODE_READ|FMODE_WRITE;
|
2006-01-03 16:55:13 +08:00
|
|
|
if (state->n_rdwr == 0) {
|
2007-07-06 06:07:55 +08:00
|
|
|
if (state->n_rdonly == 0) {
|
2006-01-03 16:55:13 +08:00
|
|
|
newstate &= ~FMODE_READ;
|
2007-07-06 06:07:55 +08:00
|
|
|
call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
|
|
|
|
call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
|
|
|
|
}
|
|
|
|
if (state->n_wronly == 0) {
|
2006-01-03 16:55:13 +08:00
|
|
|
newstate &= ~FMODE_WRITE;
|
2007-07-06 06:07:55 +08:00
|
|
|
call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
|
|
|
|
call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
|
|
|
|
}
|
|
|
|
if (newstate == 0)
|
|
|
|
clear_bit(NFS_DELEGATED_STATE, &state->flags);
|
2006-01-03 16:55:13 +08:00
|
|
|
}
|
2007-07-06 06:07:55 +08:00
|
|
|
nfs4_state_set_mode_locked(state, newstate);
|
2005-10-21 05:22:47 +08:00
|
|
|
spin_unlock(&owner->so_lock);
|
2005-11-05 04:32:58 +08:00
|
|
|
|
2007-07-06 06:07:55 +08:00
|
|
|
if (!call_close) {
|
2007-06-12 11:05:07 +08:00
|
|
|
nfs4_put_open_state(state);
|
|
|
|
nfs4_put_state_owner(owner);
|
2011-01-06 19:36:32 +08:00
|
|
|
} else {
|
|
|
|
bool roc = pnfs_roc(state->inode);
|
|
|
|
|
2011-06-23 06:20:23 +08:00
|
|
|
nfs4_do_close(state, gfp_mask, wait, roc);
|
2011-01-06 19:36:32 +08:00
|
|
|
}
|
2007-10-19 06:03:27 +08:00
|
|
|
}
|
|
|
|
|
2011-06-23 06:20:23 +08:00
|
|
|
void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
|
2007-10-19 06:03:27 +08:00
|
|
|
{
|
2011-06-23 06:20:23 +08:00
|
|
|
__nfs4_close(state, fmode, GFP_NOFS, 0);
|
2007-10-19 06:03:27 +08:00
|
|
|
}
|
|
|
|
|
2011-06-23 06:20:23 +08:00
|
|
|
void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
|
2007-10-19 06:03:27 +08:00
|
|
|
{
|
2011-06-23 06:20:23 +08:00
|
|
|
__nfs4_close(state, fmode, GFP_KERNEL, 1);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Search the state->lock_states for an existing lock_owner
|
|
|
|
* that is compatible with current->files
|
|
|
|
*/
|
|
|
|
static struct nfs4_lock_state *
|
2010-07-02 00:49:11 +08:00
|
|
|
__nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid, unsigned int type)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
|
|
|
struct nfs4_lock_state *pos;
|
|
|
|
list_for_each_entry(pos, &state->lock_states, ls_locks) {
|
2010-07-02 00:49:11 +08:00
|
|
|
if (type != NFS4_ANY_LOCK_TYPE && pos->ls_owner.lo_type != type)
|
2005-04-17 06:20:36 +08:00
|
|
|
continue;
|
2010-07-02 00:49:11 +08:00
|
|
|
switch (pos->ls_owner.lo_type) {
|
|
|
|
case NFS4_POSIX_LOCK_TYPE:
|
|
|
|
if (pos->ls_owner.lo_u.posix_owner != fl_owner)
|
|
|
|
continue;
|
|
|
|
break;
|
|
|
|
case NFS4_FLOCK_LOCK_TYPE:
|
|
|
|
if (pos->ls_owner.lo_u.flock_owner != fl_pid)
|
|
|
|
continue;
|
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
atomic_inc(&pos->ls_count);
|
|
|
|
return pos;
|
|
|
|
}
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Return a compatible lock_state. If no initialized lock_state structure
|
|
|
|
* exists, return an uninitialized one.
|
|
|
|
*
|
|
|
|
*/
|
2010-07-02 00:49:11 +08:00
|
|
|
static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid, unsigned int type)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
|
|
|
struct nfs4_lock_state *lsp;
|
2010-12-24 09:32:43 +08:00
|
|
|
struct nfs_server *server = state->owner->so_server;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2010-05-14 00:51:01 +08:00
|
|
|
lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
|
2005-04-17 06:20:36 +08:00
|
|
|
if (lsp == NULL)
|
|
|
|
return NULL;
|
2012-01-18 11:04:25 +08:00
|
|
|
nfs4_init_seqid_counter(&lsp->ls_seqid);
|
2005-04-17 06:20:36 +08:00
|
|
|
atomic_set(&lsp->ls_count, 1);
|
2009-07-22 04:47:46 +08:00
|
|
|
lsp->ls_state = state;
|
2010-07-02 00:49:11 +08:00
|
|
|
lsp->ls_owner.lo_type = type;
|
|
|
|
switch (lsp->ls_owner.lo_type) {
|
|
|
|
case NFS4_FLOCK_LOCK_TYPE:
|
|
|
|
lsp->ls_owner.lo_u.flock_owner = fl_pid;
|
|
|
|
break;
|
|
|
|
case NFS4_POSIX_LOCK_TYPE:
|
|
|
|
lsp->ls_owner.lo_u.posix_owner = fl_owner;
|
|
|
|
break;
|
|
|
|
default:
|
2012-01-18 11:04:25 +08:00
|
|
|
goto out_free;
|
2010-07-02 00:49:11 +08:00
|
|
|
}
|
2012-01-18 11:04:25 +08:00
|
|
|
lsp->ls_seqid.owner_id = ida_simple_get(&server->lockowner_id, 0, 0, GFP_NOFS);
|
|
|
|
if (lsp->ls_seqid.owner_id < 0)
|
2012-01-18 11:04:25 +08:00
|
|
|
goto out_free;
|
2005-06-23 01:16:32 +08:00
|
|
|
INIT_LIST_HEAD(&lsp->ls_locks);
|
2005-04-17 06:20:36 +08:00
|
|
|
return lsp;
|
2012-01-18 11:04:25 +08:00
|
|
|
out_free:
|
|
|
|
kfree(lsp);
|
|
|
|
return NULL;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
2007-07-03 01:58:33 +08:00
|
|
|
static void nfs4_free_lock_state(struct nfs4_lock_state *lsp)
|
|
|
|
{
|
2010-12-24 09:32:43 +08:00
|
|
|
struct nfs_server *server = lsp->ls_state->owner->so_server;
|
2007-07-03 01:58:33 +08:00
|
|
|
|
2012-01-18 11:04:25 +08:00
|
|
|
ida_simple_remove(&server->lockowner_id, lsp->ls_seqid.owner_id);
|
2012-01-18 11:04:25 +08:00
|
|
|
nfs4_destroy_seqid_counter(&lsp->ls_seqid);
|
2007-07-03 01:58:33 +08:00
|
|
|
kfree(lsp);
|
|
|
|
}
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
/*
|
|
|
|
* Return a compatible lock_state. If no initialized lock_state structure
|
|
|
|
* exists, return an uninitialized one.
|
|
|
|
*
|
|
|
|
*/
|
2010-07-02 00:49:11 +08:00
|
|
|
static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner, pid_t pid, unsigned int type)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2005-06-23 01:16:32 +08:00
|
|
|
struct nfs4_lock_state *lsp, *new = NULL;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2005-06-23 01:16:32 +08:00
|
|
|
for(;;) {
|
|
|
|
spin_lock(&state->state_lock);
|
2010-07-02 00:49:11 +08:00
|
|
|
lsp = __nfs4_find_lock_state(state, owner, pid, type);
|
2005-06-23 01:16:32 +08:00
|
|
|
if (lsp != NULL)
|
|
|
|
break;
|
|
|
|
if (new != NULL) {
|
|
|
|
list_add(&new->ls_locks, &state->lock_states);
|
|
|
|
set_bit(LK_STATE_IN_USE, &state->flags);
|
|
|
|
lsp = new;
|
|
|
|
new = NULL;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
spin_unlock(&state->state_lock);
|
2010-07-02 00:49:11 +08:00
|
|
|
new = nfs4_alloc_lock_state(state, owner, pid, type);
|
2005-06-23 01:16:32 +08:00
|
|
|
if (new == NULL)
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
spin_unlock(&state->state_lock);
|
2007-07-03 01:58:33 +08:00
|
|
|
if (new != NULL)
|
|
|
|
nfs4_free_lock_state(new);
|
2005-04-17 06:20:36 +08:00
|
|
|
return lsp;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
2005-06-23 01:16:32 +08:00
|
|
|
* Release reference to lock_state, and free it if we see that
|
|
|
|
* it is no longer in use
|
2005-04-17 06:20:36 +08:00
|
|
|
*/
|
2005-10-19 05:20:15 +08:00
|
|
|
void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2005-06-23 01:16:32 +08:00
|
|
|
struct nfs4_state *state;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2005-06-23 01:16:32 +08:00
|
|
|
if (lsp == NULL)
|
|
|
|
return;
|
|
|
|
state = lsp->ls_state;
|
|
|
|
if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
|
|
|
|
return;
|
|
|
|
list_del(&lsp->ls_locks);
|
|
|
|
if (list_empty(&state->lock_states))
|
|
|
|
clear_bit(LK_STATE_IN_USE, &state->flags);
|
|
|
|
spin_unlock(&state->state_lock);
|
2010-07-02 00:49:01 +08:00
|
|
|
if (lsp->ls_flags & NFS_LOCK_INITIALIZED)
|
|
|
|
nfs4_release_lockowner(lsp);
|
2007-07-03 01:58:33 +08:00
|
|
|
nfs4_free_lock_state(lsp);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
2005-06-23 01:16:32 +08:00
|
|
|
static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2005-06-23 01:16:32 +08:00
|
|
|
struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2005-06-23 01:16:32 +08:00
|
|
|
dst->fl_u.nfs4_fl.owner = lsp;
|
|
|
|
atomic_inc(&lsp->ls_count);
|
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2005-06-23 01:16:32 +08:00
|
|
|
static void nfs4_fl_release_lock(struct file_lock *fl)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2005-06-23 01:16:32 +08:00
|
|
|
nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
2009-09-22 08:01:11 +08:00
|
|
|
static const struct file_lock_operations nfs4_fl_lock_ops = {
|
2005-06-23 01:16:32 +08:00
|
|
|
.fl_copy_lock = nfs4_fl_copy_lock,
|
|
|
|
.fl_release_private = nfs4_fl_release_lock,
|
|
|
|
};
|
|
|
|
|
|
|
|
int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2005-06-23 01:16:32 +08:00
|
|
|
struct nfs4_lock_state *lsp;
|
|
|
|
|
|
|
|
if (fl->fl_ops != NULL)
|
|
|
|
return 0;
|
2010-07-02 00:49:11 +08:00
|
|
|
if (fl->fl_flags & FL_POSIX)
|
|
|
|
lsp = nfs4_get_lock_state(state, fl->fl_owner, 0, NFS4_POSIX_LOCK_TYPE);
|
|
|
|
else if (fl->fl_flags & FL_FLOCK)
|
|
|
|
lsp = nfs4_get_lock_state(state, 0, fl->fl_pid, NFS4_FLOCK_LOCK_TYPE);
|
|
|
|
else
|
|
|
|
return -EINVAL;
|
2005-06-23 01:16:32 +08:00
|
|
|
if (lsp == NULL)
|
|
|
|
return -ENOMEM;
|
|
|
|
fl->fl_u.nfs4_fl.owner = lsp;
|
|
|
|
fl->fl_ops = &nfs4_fl_lock_ops;
|
|
|
|
return 0;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
2005-06-23 01:16:32 +08:00
|
|
|
/*
|
|
|
|
* Byte-range lock aware utility to initialize the stateid of read/write
|
|
|
|
* requests.
|
2005-04-17 06:20:36 +08:00
|
|
|
*/
|
2010-07-02 00:49:11 +08:00
|
|
|
void nfs4_copy_stateid(nfs4_stateid *dst, struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2005-06-23 01:16:32 +08:00
|
|
|
struct nfs4_lock_state *lsp;
|
2007-07-09 22:45:42 +08:00
|
|
|
int seq;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2007-07-09 22:45:42 +08:00
|
|
|
do {
|
|
|
|
seq = read_seqbegin(&state->seqlock);
|
|
|
|
memcpy(dst, &state->stateid, sizeof(*dst));
|
|
|
|
} while (read_seqretry(&state->seqlock, seq));
|
2005-06-23 01:16:32 +08:00
|
|
|
if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
|
|
|
|
return;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2005-06-23 01:16:32 +08:00
|
|
|
spin_lock(&state->state_lock);
|
2010-07-02 00:49:11 +08:00
|
|
|
lsp = __nfs4_find_lock_state(state, fl_owner, fl_pid, NFS4_ANY_LOCK_TYPE);
|
2005-06-23 01:16:32 +08:00
|
|
|
if (lsp != NULL && (lsp->ls_flags & NFS_LOCK_INITIALIZED) != 0)
|
|
|
|
memcpy(dst, &lsp->ls_stateid, sizeof(*dst));
|
|
|
|
spin_unlock(&state->state_lock);
|
2005-04-17 06:20:36 +08:00
|
|
|
nfs4_put_lock_state(lsp);
|
|
|
|
}
|
|
|
|
|
2010-05-14 00:51:01 +08:00
|
|
|
struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
|
NFSv4: Add functions to order RPC calls
NFSv4 file state-changing functions such as OPEN, CLOSE, LOCK,... are all
labelled with "sequence identifiers" in order to prevent the server from
reordering RPC requests, as this could cause its file state to
become out of sync with the client.
Currently the NFS client code enforces this ordering locally using
semaphores to restrict access to structures until the RPC call is done.
This, of course, only works with synchronous RPC calls, since the
user process must first grab the semaphore.
By dropping semaphores, and instead teaching the RPC engine to hold
the RPC calls until they are ready to be sent, we can extend this
process to work nicely with asynchronous RPC calls too.
This patch adds a new list called "rpc_sequence" that defines the order
of the RPC calls to be sent. We add one such list for each state_owner.
When an RPC call is ready to be sent, it checks if it is top of the
rpc_sequence list. If so, it proceeds. If not, it goes back to sleep,
and loops until it hits top of the list.
Once the RPC call has completed, it can then bump the sequence id counter,
and remove itself from the rpc_sequence list, and then wake up the next
sleeper.
Note that the state_owner sequence ids and lock_owner sequence ids are
all indexed to the same rpc_sequence list, so OPEN, LOCK,... requests
are all ordered w.r.t. each other.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2005-10-19 05:20:12 +08:00
|
|
|
{
|
|
|
|
struct nfs_seqid *new;
|
|
|
|
|
2010-05-14 00:51:01 +08:00
|
|
|
new = kmalloc(sizeof(*new), gfp_mask);
|
NFSv4: Add functions to order RPC calls
NFSv4 file state-changing functions such as OPEN, CLOSE, LOCK,... are all
labelled with "sequence identifiers" in order to prevent the server from
reordering RPC requests, as this could cause its file state to
become out of sync with the client.
Currently the NFS client code enforces this ordering locally using
semaphores to restrict access to structures until the RPC call is done.
This, of course, only works with synchronous RPC calls, since the
user process must first grab the semaphore.
By dropping semaphores, and instead teaching the RPC engine to hold
the RPC calls until they are ready to be sent, we can extend this
process to work nicely with asynchronous RPC calls too.
This patch adds a new list called "rpc_sequence" that defines the order
of the RPC calls to be sent. We add one such list for each state_owner.
When an RPC call is ready to be sent, it checks if it is top of the
rpc_sequence list. If so, it proceeds. If not, it goes back to sleep,
and loops until it hits top of the list.
Once the RPC call has completed, it can then bump the sequence id counter,
and remove itself from the rpc_sequence list, and then wake up the next
sleeper.
Note that the state_owner sequence ids and lock_owner sequence ids are
all indexed to the same rpc_sequence list, so OPEN, LOCK,... requests
are all ordered w.r.t. each other.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2005-10-19 05:20:12 +08:00
|
|
|
if (new != NULL) {
|
|
|
|
new->sequence = counter;
|
2008-01-09 06:56:07 +08:00
|
|
|
INIT_LIST_HEAD(&new->list);
|
2012-01-21 07:47:05 +08:00
|
|
|
new->task = NULL;
|
NFSv4: Add functions to order RPC calls
NFSv4 file state-changing functions such as OPEN, CLOSE, LOCK,... are all
labelled with "sequence identifiers" in order to prevent the server from
reordering RPC requests, as this could cause its file state to
become out of sync with the client.
Currently the NFS client code enforces this ordering locally using
semaphores to restrict access to structures until the RPC call is done.
This, of course, only works with synchronous RPC calls, since the
user process must first grab the semaphore.
By dropping semaphores, and instead teaching the RPC engine to hold
the RPC calls until they are ready to be sent, we can extend this
process to work nicely with asynchronous RPC calls too.
This patch adds a new list called "rpc_sequence" that defines the order
of the RPC calls to be sent. We add one such list for each state_owner.
When an RPC call is ready to be sent, it checks if it is top of the
rpc_sequence list. If so, it proceeds. If not, it goes back to sleep,
and loops until it hits top of the list.
Once the RPC call has completed, it can then bump the sequence id counter,
and remove itself from the rpc_sequence list, and then wake up the next
sleeper.
Note that the state_owner sequence ids and lock_owner sequence ids are
all indexed to the same rpc_sequence list, so OPEN, LOCK,... requests
are all ordered w.r.t. each other.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2005-10-19 05:20:12 +08:00
|
|
|
}
|
|
|
|
return new;
|
|
|
|
}
|
|
|
|
|
2009-12-16 03:47:36 +08:00
|
|
|
void nfs_release_seqid(struct nfs_seqid *seqid)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2012-01-21 07:47:05 +08:00
|
|
|
struct nfs_seqid_counter *sequence;
|
NFSv4: Add functions to order RPC calls
NFSv4 file state-changing functions such as OPEN, CLOSE, LOCK,... are all
labelled with "sequence identifiers" in order to prevent the server from
reordering RPC requests, as this could cause its file state to
become out of sync with the client.
Currently the NFS client code enforces this ordering locally using
semaphores to restrict access to structures until the RPC call is done.
This, of course, only works with synchronous RPC calls, since the
user process must first grab the semaphore.
By dropping semaphores, and instead teaching the RPC engine to hold
the RPC calls until they are ready to be sent, we can extend this
process to work nicely with asynchronous RPC calls too.
This patch adds a new list called "rpc_sequence" that defines the order
of the RPC calls to be sent. We add one such list for each state_owner.
When an RPC call is ready to be sent, it checks if it is top of the
rpc_sequence list. If so, it proceeds. If not, it goes back to sleep,
and loops until it hits top of the list.
Once the RPC call has completed, it can then bump the sequence id counter,
and remove itself from the rpc_sequence list, and then wake up the next
sleeper.
Note that the state_owner sequence ids and lock_owner sequence ids are
all indexed to the same rpc_sequence list, so OPEN, LOCK,... requests
are all ordered w.r.t. each other.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2005-10-19 05:20:12 +08:00
|
|
|
|
2012-01-21 07:47:05 +08:00
|
|
|
if (list_empty(&seqid->list))
|
|
|
|
return;
|
|
|
|
sequence = seqid->sequence;
|
|
|
|
spin_lock(&sequence->lock);
|
|
|
|
list_del_init(&seqid->list);
|
|
|
|
if (!list_empty(&sequence->list)) {
|
|
|
|
struct nfs_seqid *next;
|
|
|
|
|
|
|
|
next = list_first_entry(&sequence->list,
|
|
|
|
struct nfs_seqid, list);
|
|
|
|
rpc_wake_up_queued_task(&sequence->wait, next->task);
|
2008-01-09 06:56:07 +08:00
|
|
|
}
|
2012-01-21 07:47:05 +08:00
|
|
|
spin_unlock(&sequence->lock);
|
2009-12-16 03:47:36 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
void nfs_free_seqid(struct nfs_seqid *seqid)
|
|
|
|
{
|
|
|
|
nfs_release_seqid(seqid);
|
NFSv4: Add functions to order RPC calls
NFSv4 file state-changing functions such as OPEN, CLOSE, LOCK,... are all
labelled with "sequence identifiers" in order to prevent the server from
reordering RPC requests, as this could cause its file state to
become out of sync with the client.
Currently the NFS client code enforces this ordering locally using
semaphores to restrict access to structures until the RPC call is done.
This, of course, only works with synchronous RPC calls, since the
user process must first grab the semaphore.
By dropping semaphores, and instead teaching the RPC engine to hold
the RPC calls until they are ready to be sent, we can extend this
process to work nicely with asynchronous RPC calls too.
This patch adds a new list called "rpc_sequence" that defines the order
of the RPC calls to be sent. We add one such list for each state_owner.
When an RPC call is ready to be sent, it checks if it is top of the
rpc_sequence list. If so, it proceeds. If not, it goes back to sleep,
and loops until it hits top of the list.
Once the RPC call has completed, it can then bump the sequence id counter,
and remove itself from the rpc_sequence list, and then wake up the next
sleeper.
Note that the state_owner sequence ids and lock_owner sequence ids are
all indexed to the same rpc_sequence list, so OPEN, LOCK,... requests
are all ordered w.r.t. each other.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2005-10-19 05:20:12 +08:00
|
|
|
kfree(seqid);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
NFSv4: Add functions to order RPC calls
NFSv4 file state-changing functions such as OPEN, CLOSE, LOCK,... are all
labelled with "sequence identifiers" in order to prevent the server from
reordering RPC requests, as this could cause its file state to
become out of sync with the client.
Currently the NFS client code enforces this ordering locally using
semaphores to restrict access to structures until the RPC call is done.
This, of course, only works with synchronous RPC calls, since the
user process must first grab the semaphore.
By dropping semaphores, and instead teaching the RPC engine to hold
the RPC calls until they are ready to be sent, we can extend this
process to work nicely with asynchronous RPC calls too.
This patch adds a new list called "rpc_sequence" that defines the order
of the RPC calls to be sent. We add one such list for each state_owner.
When an RPC call is ready to be sent, it checks if it is top of the
rpc_sequence list. If so, it proceeds. If not, it goes back to sleep,
and loops until it hits top of the list.
Once the RPC call has completed, it can then bump the sequence id counter,
and remove itself from the rpc_sequence list, and then wake up the next
sleeper.
Note that the state_owner sequence ids and lock_owner sequence ids are
all indexed to the same rpc_sequence list, so OPEN, LOCK,... requests
are all ordered w.r.t. each other.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2005-10-19 05:20:12 +08:00
|
|
|
* Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
|
|
|
|
* failed with a seqid incrementing error -
|
|
|
|
* see comments nfs_fs.h:seqid_mutating_error()
|
|
|
|
*/
|
2007-07-03 02:03:03 +08:00
|
|
|
static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
|
NFSv4: Add functions to order RPC calls
NFSv4 file state-changing functions such as OPEN, CLOSE, LOCK,... are all
labelled with "sequence identifiers" in order to prevent the server from
reordering RPC requests, as this could cause its file state to
become out of sync with the client.
Currently the NFS client code enforces this ordering locally using
semaphores to restrict access to structures until the RPC call is done.
This, of course, only works with synchronous RPC calls, since the
user process must first grab the semaphore.
By dropping semaphores, and instead teaching the RPC engine to hold
the RPC calls until they are ready to be sent, we can extend this
process to work nicely with asynchronous RPC calls too.
This patch adds a new list called "rpc_sequence" that defines the order
of the RPC calls to be sent. We add one such list for each state_owner.
When an RPC call is ready to be sent, it checks if it is top of the
rpc_sequence list. If so, it proceeds. If not, it goes back to sleep,
and loops until it hits top of the list.
Once the RPC call has completed, it can then bump the sequence id counter,
and remove itself from the rpc_sequence list, and then wake up the next
sleeper.
Note that the state_owner sequence ids and lock_owner sequence ids are
all indexed to the same rpc_sequence list, so OPEN, LOCK,... requests
are all ordered w.r.t. each other.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2005-10-19 05:20:12 +08:00
|
|
|
{
|
2012-01-18 11:04:25 +08:00
|
|
|
BUG_ON(list_first_entry(&seqid->sequence->list, struct nfs_seqid, list) != seqid);
|
NFSv4: Add functions to order RPC calls
NFSv4 file state-changing functions such as OPEN, CLOSE, LOCK,... are all
labelled with "sequence identifiers" in order to prevent the server from
reordering RPC requests, as this could cause its file state to
become out of sync with the client.
Currently the NFS client code enforces this ordering locally using
semaphores to restrict access to structures until the RPC call is done.
This, of course, only works with synchronous RPC calls, since the
user process must first grab the semaphore.
By dropping semaphores, and instead teaching the RPC engine to hold
the RPC calls until they are ready to be sent, we can extend this
process to work nicely with asynchronous RPC calls too.
This patch adds a new list called "rpc_sequence" that defines the order
of the RPC calls to be sent. We add one such list for each state_owner.
When an RPC call is ready to be sent, it checks if it is top of the
rpc_sequence list. If so, it proceeds. If not, it goes back to sleep,
and loops until it hits top of the list.
Once the RPC call has completed, it can then bump the sequence id counter,
and remove itself from the rpc_sequence list, and then wake up the next
sleeper.
Note that the state_owner sequence ids and lock_owner sequence ids are
all indexed to the same rpc_sequence list, so OPEN, LOCK,... requests
are all ordered w.r.t. each other.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2005-10-19 05:20:12 +08:00
|
|
|
switch (status) {
|
|
|
|
case 0:
|
|
|
|
break;
|
|
|
|
case -NFS4ERR_BAD_SEQID:
|
2007-07-09 04:49:11 +08:00
|
|
|
if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
|
|
|
|
return;
|
|
|
|
printk(KERN_WARNING "NFS: v4 server returned a bad"
|
2008-02-14 05:09:35 +08:00
|
|
|
" sequence-id error on an"
|
|
|
|
" unconfirmed sequence %p!\n",
|
2007-07-09 04:49:11 +08:00
|
|
|
seqid->sequence);
|
NFSv4: Add functions to order RPC calls
NFSv4 file state-changing functions such as OPEN, CLOSE, LOCK,... are all
labelled with "sequence identifiers" in order to prevent the server from
reordering RPC requests, as this could cause its file state to
become out of sync with the client.
Currently the NFS client code enforces this ordering locally using
semaphores to restrict access to structures until the RPC call is done.
This, of course, only works with synchronous RPC calls, since the
user process must first grab the semaphore.
By dropping semaphores, and instead teaching the RPC engine to hold
the RPC calls until they are ready to be sent, we can extend this
process to work nicely with asynchronous RPC calls too.
This patch adds a new list called "rpc_sequence" that defines the order
of the RPC calls to be sent. We add one such list for each state_owner.
When an RPC call is ready to be sent, it checks if it is top of the
rpc_sequence list. If so, it proceeds. If not, it goes back to sleep,
and loops until it hits top of the list.
Once the RPC call has completed, it can then bump the sequence id counter,
and remove itself from the rpc_sequence list, and then wake up the next
sleeper.
Note that the state_owner sequence ids and lock_owner sequence ids are
all indexed to the same rpc_sequence list, so OPEN, LOCK,... requests
are all ordered w.r.t. each other.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2005-10-19 05:20:12 +08:00
|
|
|
case -NFS4ERR_STALE_CLIENTID:
|
|
|
|
case -NFS4ERR_STALE_STATEID:
|
|
|
|
case -NFS4ERR_BAD_STATEID:
|
|
|
|
case -NFS4ERR_BADXDR:
|
|
|
|
case -NFS4ERR_RESOURCE:
|
|
|
|
case -NFS4ERR_NOFILEHANDLE:
|
|
|
|
/* Non-seqid mutating errors */
|
|
|
|
return;
|
|
|
|
};
|
|
|
|
/*
|
|
|
|
* Note: no locking needed as we are guaranteed to be first
|
|
|
|
* on the sequence list
|
|
|
|
*/
|
|
|
|
seqid->sequence->counter++;
|
|
|
|
}
|
|
|
|
|
|
|
|
void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
|
|
|
|
{
|
2009-04-01 21:22:52 +08:00
|
|
|
struct nfs4_state_owner *sp = container_of(seqid->sequence,
|
|
|
|
struct nfs4_state_owner, so_seqid);
|
|
|
|
struct nfs_server *server = sp->so_server;
|
|
|
|
|
|
|
|
if (status == -NFS4ERR_BAD_SEQID)
|
2005-04-17 06:20:36 +08:00
|
|
|
nfs4_drop_state_owner(sp);
|
2009-04-01 21:22:52 +08:00
|
|
|
if (!nfs4_has_session(server->nfs_client))
|
|
|
|
nfs_increment_seqid(status, seqid);
|
NFSv4: Add functions to order RPC calls
NFSv4 file state-changing functions such as OPEN, CLOSE, LOCK,... are all
labelled with "sequence identifiers" in order to prevent the server from
reordering RPC requests, as this could cause its file state to
become out of sync with the client.
Currently the NFS client code enforces this ordering locally using
semaphores to restrict access to structures until the RPC call is done.
This, of course, only works with synchronous RPC calls, since the
user process must first grab the semaphore.
By dropping semaphores, and instead teaching the RPC engine to hold
the RPC calls until they are ready to be sent, we can extend this
process to work nicely with asynchronous RPC calls too.
This patch adds a new list called "rpc_sequence" that defines the order
of the RPC calls to be sent. We add one such list for each state_owner.
When an RPC call is ready to be sent, it checks if it is top of the
rpc_sequence list. If so, it proceeds. If not, it goes back to sleep,
and loops until it hits top of the list.
Once the RPC call has completed, it can then bump the sequence id counter,
and remove itself from the rpc_sequence list, and then wake up the next
sleeper.
Note that the state_owner sequence ids and lock_owner sequence ids are
all indexed to the same rpc_sequence list, so OPEN, LOCK,... requests
are all ordered w.r.t. each other.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2005-10-19 05:20:12 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Increment the seqid if the LOCK/LOCKU succeeded, or
|
|
|
|
* failed with a seqid incrementing error -
|
|
|
|
* see comments nfs_fs.h:seqid_mutating_error()
|
|
|
|
*/
|
|
|
|
void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
|
|
|
|
{
|
2007-07-03 02:03:03 +08:00
|
|
|
nfs_increment_seqid(status, seqid);
|
NFSv4: Add functions to order RPC calls
NFSv4 file state-changing functions such as OPEN, CLOSE, LOCK,... are all
labelled with "sequence identifiers" in order to prevent the server from
reordering RPC requests, as this could cause its file state to
become out of sync with the client.
Currently the NFS client code enforces this ordering locally using
semaphores to restrict access to structures until the RPC call is done.
This, of course, only works with synchronous RPC calls, since the
user process must first grab the semaphore.
By dropping semaphores, and instead teaching the RPC engine to hold
the RPC calls until they are ready to be sent, we can extend this
process to work nicely with asynchronous RPC calls too.
This patch adds a new list called "rpc_sequence" that defines the order
of the RPC calls to be sent. We add one such list for each state_owner.
When an RPC call is ready to be sent, it checks if it is top of the
rpc_sequence list. If so, it proceeds. If not, it goes back to sleep,
and loops until it hits top of the list.
Once the RPC call has completed, it can then bump the sequence id counter,
and remove itself from the rpc_sequence list, and then wake up the next
sleeper.
Note that the state_owner sequence ids and lock_owner sequence ids are
all indexed to the same rpc_sequence list, so OPEN, LOCK,... requests
are all ordered w.r.t. each other.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2005-10-19 05:20:12 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
|
|
|
|
{
|
2012-01-18 11:04:25 +08:00
|
|
|
struct nfs_seqid_counter *sequence = seqid->sequence;
|
NFSv4: Add functions to order RPC calls
NFSv4 file state-changing functions such as OPEN, CLOSE, LOCK,... are all
labelled with "sequence identifiers" in order to prevent the server from
reordering RPC requests, as this could cause its file state to
become out of sync with the client.
Currently the NFS client code enforces this ordering locally using
semaphores to restrict access to structures until the RPC call is done.
This, of course, only works with synchronous RPC calls, since the
user process must first grab the semaphore.
By dropping semaphores, and instead teaching the RPC engine to hold
the RPC calls until they are ready to be sent, we can extend this
process to work nicely with asynchronous RPC calls too.
This patch adds a new list called "rpc_sequence" that defines the order
of the RPC calls to be sent. We add one such list for each state_owner.
When an RPC call is ready to be sent, it checks if it is top of the
rpc_sequence list. If so, it proceeds. If not, it goes back to sleep,
and loops until it hits top of the list.
Once the RPC call has completed, it can then bump the sequence id counter,
and remove itself from the rpc_sequence list, and then wake up the next
sleeper.
Note that the state_owner sequence ids and lock_owner sequence ids are
all indexed to the same rpc_sequence list, so OPEN, LOCK,... requests
are all ordered w.r.t. each other.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2005-10-19 05:20:12 +08:00
|
|
|
int status = 0;
|
|
|
|
|
|
|
|
spin_lock(&sequence->lock);
|
2012-01-21 07:47:05 +08:00
|
|
|
seqid->task = task;
|
2008-01-09 06:56:07 +08:00
|
|
|
if (list_empty(&seqid->list))
|
|
|
|
list_add_tail(&seqid->list, &sequence->list);
|
|
|
|
if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
|
|
|
|
goto unlock;
|
2008-02-23 05:34:17 +08:00
|
|
|
rpc_sleep_on(&sequence->wait, task, NULL);
|
2008-01-09 06:56:07 +08:00
|
|
|
status = -EAGAIN;
|
|
|
|
unlock:
|
NFSv4: Add functions to order RPC calls
NFSv4 file state-changing functions such as OPEN, CLOSE, LOCK,... are all
labelled with "sequence identifiers" in order to prevent the server from
reordering RPC requests, as this could cause its file state to
become out of sync with the client.
Currently the NFS client code enforces this ordering locally using
semaphores to restrict access to structures until the RPC call is done.
This, of course, only works with synchronous RPC calls, since the
user process must first grab the semaphore.
By dropping semaphores, and instead teaching the RPC engine to hold
the RPC calls until they are ready to be sent, we can extend this
process to work nicely with asynchronous RPC calls too.
This patch adds a new list called "rpc_sequence" that defines the order
of the RPC calls to be sent. We add one such list for each state_owner.
When an RPC call is ready to be sent, it checks if it is top of the
rpc_sequence list. If so, it proceeds. If not, it goes back to sleep,
and loops until it hits top of the list.
Once the RPC call has completed, it can then bump the sequence id counter,
and remove itself from the rpc_sequence list, and then wake up the next
sleeper.
Note that the state_owner sequence ids and lock_owner sequence ids are
all indexed to the same rpc_sequence list, so OPEN, LOCK,... requests
are all ordered w.r.t. each other.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2005-10-19 05:20:12 +08:00
|
|
|
spin_unlock(&sequence->lock);
|
|
|
|
return status;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
2008-12-24 04:21:48 +08:00
|
|
|
static int nfs4_run_state_manager(void *);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2008-12-24 04:21:48 +08:00
|
|
|
static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
|
2006-01-03 16:55:22 +08:00
|
|
|
{
|
|
|
|
smp_mb__before_clear_bit();
|
2008-12-24 04:21:48 +08:00
|
|
|
clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
|
2006-01-03 16:55:22 +08:00
|
|
|
smp_mb__after_clear_bit();
|
2008-12-24 04:21:48 +08:00
|
|
|
wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
|
2006-01-03 16:55:22 +08:00
|
|
|
rpc_wake_up(&clp->cl_rpcwaitq);
|
|
|
|
}
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
/*
|
2008-12-24 04:21:48 +08:00
|
|
|
* Schedule the nfs_client asynchronous state management routine
|
2005-04-17 06:20:36 +08:00
|
|
|
*/
|
2008-12-24 04:21:50 +08:00
|
|
|
void nfs4_schedule_state_manager(struct nfs_client *clp)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2006-01-03 16:55:23 +08:00
|
|
|
struct task_struct *task;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2008-12-24 04:21:48 +08:00
|
|
|
if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
|
|
|
|
return;
|
2006-01-03 16:55:23 +08:00
|
|
|
__module_get(THIS_MODULE);
|
|
|
|
atomic_inc(&clp->cl_count);
|
2008-12-24 04:21:48 +08:00
|
|
|
task = kthread_run(nfs4_run_state_manager, clp, "%s-manager",
|
2007-12-11 03:57:16 +08:00
|
|
|
rpc_peeraddr2str(clp->cl_rpcclient,
|
|
|
|
RPC_DISPLAY_ADDR));
|
2006-01-03 16:55:23 +08:00
|
|
|
if (!IS_ERR(task))
|
|
|
|
return;
|
2008-12-24 04:21:48 +08:00
|
|
|
nfs4_clear_state_manager_bit(clp);
|
2006-08-23 08:06:10 +08:00
|
|
|
nfs_put_client(clp);
|
2006-01-03 16:55:23 +08:00
|
|
|
module_put(THIS_MODULE);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
2011-03-10 05:00:53 +08:00
|
|
|
* Schedule a lease recovery attempt
|
2005-04-17 06:20:36 +08:00
|
|
|
*/
|
2011-03-10 05:00:53 +08:00
|
|
|
void nfs4_schedule_lease_recovery(struct nfs_client *clp)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
|
|
|
if (!clp)
|
|
|
|
return;
|
2008-12-24 04:21:42 +08:00
|
|
|
if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
|
|
|
|
set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
|
2008-12-24 04:21:48 +08:00
|
|
|
nfs4_schedule_state_manager(clp);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
2011-08-25 03:07:37 +08:00
|
|
|
void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
|
|
|
|
{
|
|
|
|
nfs_handle_cb_pathdown(clp);
|
|
|
|
nfs4_schedule_state_manager(clp);
|
|
|
|
}
|
|
|
|
|
2011-03-10 05:12:46 +08:00
|
|
|
static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
|
2008-12-24 04:21:41 +08:00
|
|
|
{
|
|
|
|
|
|
|
|
set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
|
|
|
|
/* Don't recover state that expired before the reboot */
|
|
|
|
if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
|
|
|
|
clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
|
|
|
|
return 0;
|
|
|
|
}
|
2008-12-24 04:21:43 +08:00
|
|
|
set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
|
2008-12-24 04:21:41 +08:00
|
|
|
set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
2011-03-10 05:12:46 +08:00
|
|
|
static int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
|
2008-12-24 04:21:41 +08:00
|
|
|
{
|
|
|
|
set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
|
|
|
|
clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
|
2008-12-24 04:21:43 +08:00
|
|
|
set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
|
2008-12-24 04:21:41 +08:00
|
|
|
set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
2011-03-10 05:00:53 +08:00
|
|
|
void nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
|
|
|
|
{
|
|
|
|
struct nfs_client *clp = server->nfs_client;
|
|
|
|
|
|
|
|
nfs4_state_mark_reclaim_nograce(clp, state);
|
|
|
|
nfs4_schedule_state_manager(clp);
|
|
|
|
}
|
|
|
|
|
2008-12-24 04:21:40 +08:00
|
|
|
static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
|
|
|
struct inode *inode = state->inode;
|
2008-12-24 04:21:44 +08:00
|
|
|
struct nfs_inode *nfsi = NFS_I(inode);
|
2005-04-17 06:20:36 +08:00
|
|
|
struct file_lock *fl;
|
|
|
|
int status = 0;
|
|
|
|
|
2009-06-18 04:23:00 +08:00
|
|
|
if (inode->i_flock == NULL)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
/* Guard against delegation returns and new lock/unlock calls */
|
2008-12-24 04:21:44 +08:00
|
|
|
down_write(&nfsi->rwsem);
|
2009-06-18 04:23:00 +08:00
|
|
|
/* Protect inode->i_flock using the BKL */
|
2010-09-18 21:09:31 +08:00
|
|
|
lock_flocks();
|
2008-02-21 05:03:05 +08:00
|
|
|
for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
|
2005-11-05 04:35:30 +08:00
|
|
|
if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
|
2005-04-17 06:20:36 +08:00
|
|
|
continue;
|
2007-08-11 05:44:32 +08:00
|
|
|
if (nfs_file_open_context(fl->fl_file)->state != state)
|
2005-04-17 06:20:36 +08:00
|
|
|
continue;
|
2010-09-18 21:09:31 +08:00
|
|
|
unlock_flocks();
|
2005-04-17 06:20:36 +08:00
|
|
|
status = ops->recover_lock(state, fl);
|
|
|
|
switch (status) {
|
2009-06-18 04:22:59 +08:00
|
|
|
case 0:
|
|
|
|
break;
|
|
|
|
case -ESTALE:
|
|
|
|
case -NFS4ERR_ADMIN_REVOKED:
|
|
|
|
case -NFS4ERR_STALE_STATEID:
|
|
|
|
case -NFS4ERR_BAD_STATEID:
|
|
|
|
case -NFS4ERR_EXPIRED:
|
|
|
|
case -NFS4ERR_NO_GRACE:
|
|
|
|
case -NFS4ERR_STALE_CLIENTID:
|
2009-12-04 04:58:56 +08:00
|
|
|
case -NFS4ERR_BADSESSION:
|
|
|
|
case -NFS4ERR_BADSLOT:
|
|
|
|
case -NFS4ERR_BAD_HIGH_SLOT:
|
|
|
|
case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
|
2009-06-18 04:22:59 +08:00
|
|
|
goto out;
|
2005-04-17 06:20:36 +08:00
|
|
|
default:
|
2012-01-27 02:32:23 +08:00
|
|
|
printk(KERN_ERR "NFS: %s: unhandled error %d. "
|
|
|
|
"Zeroing state\n", __func__, status);
|
2009-06-18 04:22:59 +08:00
|
|
|
case -ENOMEM:
|
|
|
|
case -NFS4ERR_DENIED:
|
2005-04-17 06:20:36 +08:00
|
|
|
case -NFS4ERR_RECLAIM_BAD:
|
|
|
|
case -NFS4ERR_RECLAIM_CONFLICT:
|
2005-11-05 04:35:30 +08:00
|
|
|
/* kill_proc(fl->fl_pid, SIGLOST, 1); */
|
2009-06-18 04:22:59 +08:00
|
|
|
status = 0;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
2010-09-18 21:09:31 +08:00
|
|
|
lock_flocks();
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
2010-09-18 21:09:31 +08:00
|
|
|
unlock_flocks();
|
2009-06-18 04:22:59 +08:00
|
|
|
out:
|
2008-12-24 04:21:44 +08:00
|
|
|
up_write(&nfsi->rwsem);
|
2005-04-17 06:20:36 +08:00
|
|
|
return status;
|
|
|
|
}
|
|
|
|
|
2008-12-24 04:21:40 +08:00
|
|
|
static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
|
|
|
struct nfs4_state *state;
|
|
|
|
struct nfs4_lock_state *lock;
|
|
|
|
int status = 0;
|
|
|
|
|
|
|
|
/* Note: we rely on the sp->so_states list being ordered
|
|
|
|
* so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
|
|
|
|
* states first.
|
|
|
|
* This is needed to ensure that the server won't give us any
|
|
|
|
* read delegations that we have to return if, say, we are
|
|
|
|
* recovering after a network partition or a reboot from a
|
|
|
|
* server that doesn't support a grace period.
|
|
|
|
*/
|
2008-12-24 04:21:43 +08:00
|
|
|
restart:
|
|
|
|
spin_lock(&sp->so_lock);
|
2005-04-17 06:20:36 +08:00
|
|
|
list_for_each_entry(state, &sp->so_states, open_states) {
|
2008-12-24 04:21:41 +08:00
|
|
|
if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
|
|
|
|
continue;
|
2005-04-17 06:20:36 +08:00
|
|
|
if (state->state == 0)
|
|
|
|
continue;
|
2008-12-24 04:21:43 +08:00
|
|
|
atomic_inc(&state->count);
|
|
|
|
spin_unlock(&sp->so_lock);
|
2005-04-17 06:20:36 +08:00
|
|
|
status = ops->recover_open(sp, state);
|
|
|
|
if (status >= 0) {
|
2008-12-24 04:21:40 +08:00
|
|
|
status = nfs4_reclaim_locks(state, ops);
|
|
|
|
if (status >= 0) {
|
2011-12-10 05:31:52 +08:00
|
|
|
spin_lock(&state->state_lock);
|
2008-12-24 04:21:40 +08:00
|
|
|
list_for_each_entry(lock, &state->lock_states, ls_locks) {
|
|
|
|
if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
|
2012-01-27 02:32:23 +08:00
|
|
|
printk("NFS: %s: Lock reclaim "
|
|
|
|
"failed!\n", __func__);
|
2008-12-24 04:21:40 +08:00
|
|
|
}
|
2011-12-10 05:31:52 +08:00
|
|
|
spin_unlock(&state->state_lock);
|
2008-12-24 04:21:43 +08:00
|
|
|
nfs4_put_open_state(state);
|
|
|
|
goto restart;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
switch (status) {
|
|
|
|
default:
|
2012-01-27 02:32:23 +08:00
|
|
|
printk(KERN_ERR "NFS: %s: unhandled error %d. "
|
|
|
|
"Zeroing state\n", __func__, status);
|
2005-04-17 06:20:36 +08:00
|
|
|
case -ENOENT:
|
2009-06-18 04:22:59 +08:00
|
|
|
case -ENOMEM:
|
2008-12-24 04:21:41 +08:00
|
|
|
case -ESTALE:
|
2005-04-17 06:20:36 +08:00
|
|
|
/*
|
|
|
|
* Open state on this file cannot be recovered
|
|
|
|
* All we can do is revert to using the zero stateid.
|
|
|
|
*/
|
|
|
|
memset(state->stateid.data, 0,
|
|
|
|
sizeof(state->stateid.data));
|
|
|
|
/* Mark the file as being 'closed' */
|
|
|
|
state->state = 0;
|
|
|
|
break;
|
2010-10-20 07:47:49 +08:00
|
|
|
case -EKEYEXPIRED:
|
|
|
|
/*
|
|
|
|
* User RPCSEC_GSS context has expired.
|
|
|
|
* We cannot recover this stateid now, so
|
|
|
|
* skip it and allow recovery thread to
|
|
|
|
* proceed.
|
|
|
|
*/
|
|
|
|
break;
|
2009-06-18 04:22:59 +08:00
|
|
|
case -NFS4ERR_ADMIN_REVOKED:
|
|
|
|
case -NFS4ERR_STALE_STATEID:
|
|
|
|
case -NFS4ERR_BAD_STATEID:
|
2008-12-24 04:21:41 +08:00
|
|
|
case -NFS4ERR_RECLAIM_BAD:
|
|
|
|
case -NFS4ERR_RECLAIM_CONFLICT:
|
2010-06-25 03:11:43 +08:00
|
|
|
nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
|
2008-12-24 04:21:41 +08:00
|
|
|
break;
|
2005-04-17 06:20:36 +08:00
|
|
|
case -NFS4ERR_EXPIRED:
|
|
|
|
case -NFS4ERR_NO_GRACE:
|
2010-06-25 03:11:43 +08:00
|
|
|
nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
|
2005-04-17 06:20:36 +08:00
|
|
|
case -NFS4ERR_STALE_CLIENTID:
|
2009-12-04 04:58:56 +08:00
|
|
|
case -NFS4ERR_BADSESSION:
|
|
|
|
case -NFS4ERR_BADSLOT:
|
|
|
|
case -NFS4ERR_BAD_HIGH_SLOT:
|
|
|
|
case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
|
2005-04-17 06:20:36 +08:00
|
|
|
goto out_err;
|
|
|
|
}
|
2008-12-24 04:21:43 +08:00
|
|
|
nfs4_put_open_state(state);
|
|
|
|
goto restart;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
2008-12-24 04:21:43 +08:00
|
|
|
spin_unlock(&sp->so_lock);
|
2005-04-17 06:20:36 +08:00
|
|
|
return 0;
|
|
|
|
out_err:
|
2008-12-24 04:21:43 +08:00
|
|
|
nfs4_put_open_state(state);
|
2005-04-17 06:20:36 +08:00
|
|
|
return status;
|
|
|
|
}
|
|
|
|
|
2008-12-24 04:21:41 +08:00
|
|
|
static void nfs4_clear_open_state(struct nfs4_state *state)
|
|
|
|
{
|
|
|
|
struct nfs4_lock_state *lock;
|
|
|
|
|
|
|
|
clear_bit(NFS_DELEGATED_STATE, &state->flags);
|
|
|
|
clear_bit(NFS_O_RDONLY_STATE, &state->flags);
|
|
|
|
clear_bit(NFS_O_WRONLY_STATE, &state->flags);
|
|
|
|
clear_bit(NFS_O_RDWR_STATE, &state->flags);
|
2011-12-10 05:31:52 +08:00
|
|
|
spin_lock(&state->state_lock);
|
2008-12-24 04:21:41 +08:00
|
|
|
list_for_each_entry(lock, &state->lock_states, ls_locks) {
|
|
|
|
lock->ls_seqid.flags = 0;
|
|
|
|
lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
|
|
|
|
}
|
2011-12-10 05:31:52 +08:00
|
|
|
spin_unlock(&state->state_lock);
|
2008-12-24 04:21:41 +08:00
|
|
|
}
|
|
|
|
|
2010-12-24 09:32:43 +08:00
|
|
|
static void nfs4_reset_seqids(struct nfs_server *server,
|
|
|
|
int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
|
NFSv4: Add functions to order RPC calls
NFSv4 file state-changing functions such as OPEN, CLOSE, LOCK,... are all
labelled with "sequence identifiers" in order to prevent the server from
reordering RPC requests, as this could cause its file state to
become out of sync with the client.
Currently the NFS client code enforces this ordering locally using
semaphores to restrict access to structures until the RPC call is done.
This, of course, only works with synchronous RPC calls, since the
user process must first grab the semaphore.
By dropping semaphores, and instead teaching the RPC engine to hold
the RPC calls until they are ready to be sent, we can extend this
process to work nicely with asynchronous RPC calls too.
This patch adds a new list called "rpc_sequence" that defines the order
of the RPC calls to be sent. We add one such list for each state_owner.
When an RPC call is ready to be sent, it checks if it is top of the
rpc_sequence list. If so, it proceeds. If not, it goes back to sleep,
and loops until it hits top of the list.
Once the RPC call has completed, it can then bump the sequence id counter,
and remove itself from the rpc_sequence list, and then wake up the next
sleeper.
Note that the state_owner sequence ids and lock_owner sequence ids are
all indexed to the same rpc_sequence list, so OPEN, LOCK,... requests
are all ordered w.r.t. each other.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2005-10-19 05:20:12 +08:00
|
|
|
{
|
2010-12-24 09:32:43 +08:00
|
|
|
struct nfs_client *clp = server->nfs_client;
|
NFSv4: Add functions to order RPC calls
NFSv4 file state-changing functions such as OPEN, CLOSE, LOCK,... are all
labelled with "sequence identifiers" in order to prevent the server from
reordering RPC requests, as this could cause its file state to
become out of sync with the client.
Currently the NFS client code enforces this ordering locally using
semaphores to restrict access to structures until the RPC call is done.
This, of course, only works with synchronous RPC calls, since the
user process must first grab the semaphore.
By dropping semaphores, and instead teaching the RPC engine to hold
the RPC calls until they are ready to be sent, we can extend this
process to work nicely with asynchronous RPC calls too.
This patch adds a new list called "rpc_sequence" that defines the order
of the RPC calls to be sent. We add one such list for each state_owner.
When an RPC call is ready to be sent, it checks if it is top of the
rpc_sequence list. If so, it proceeds. If not, it goes back to sleep,
and loops until it hits top of the list.
Once the RPC call has completed, it can then bump the sequence id counter,
and remove itself from the rpc_sequence list, and then wake up the next
sleeper.
Note that the state_owner sequence ids and lock_owner sequence ids are
all indexed to the same rpc_sequence list, so OPEN, LOCK,... requests
are all ordered w.r.t. each other.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2005-10-19 05:20:12 +08:00
|
|
|
struct nfs4_state_owner *sp;
|
2007-07-03 01:58:33 +08:00
|
|
|
struct rb_node *pos;
|
NFSv4: Add functions to order RPC calls
NFSv4 file state-changing functions such as OPEN, CLOSE, LOCK,... are all
labelled with "sequence identifiers" in order to prevent the server from
reordering RPC requests, as this could cause its file state to
become out of sync with the client.
Currently the NFS client code enforces this ordering locally using
semaphores to restrict access to structures until the RPC call is done.
This, of course, only works with synchronous RPC calls, since the
user process must first grab the semaphore.
By dropping semaphores, and instead teaching the RPC engine to hold
the RPC calls until they are ready to be sent, we can extend this
process to work nicely with asynchronous RPC calls too.
This patch adds a new list called "rpc_sequence" that defines the order
of the RPC calls to be sent. We add one such list for each state_owner.
When an RPC call is ready to be sent, it checks if it is top of the
rpc_sequence list. If so, it proceeds. If not, it goes back to sleep,
and loops until it hits top of the list.
Once the RPC call has completed, it can then bump the sequence id counter,
and remove itself from the rpc_sequence list, and then wake up the next
sleeper.
Note that the state_owner sequence ids and lock_owner sequence ids are
all indexed to the same rpc_sequence list, so OPEN, LOCK,... requests
are all ordered w.r.t. each other.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2005-10-19 05:20:12 +08:00
|
|
|
struct nfs4_state *state;
|
|
|
|
|
2010-12-24 09:32:43 +08:00
|
|
|
spin_lock(&clp->cl_lock);
|
|
|
|
for (pos = rb_first(&server->state_owners);
|
|
|
|
pos != NULL;
|
|
|
|
pos = rb_next(pos)) {
|
|
|
|
sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
|
NFSv4: Add functions to order RPC calls
NFSv4 file state-changing functions such as OPEN, CLOSE, LOCK,... are all
labelled with "sequence identifiers" in order to prevent the server from
reordering RPC requests, as this could cause its file state to
become out of sync with the client.
Currently the NFS client code enforces this ordering locally using
semaphores to restrict access to structures until the RPC call is done.
This, of course, only works with synchronous RPC calls, since the
user process must first grab the semaphore.
By dropping semaphores, and instead teaching the RPC engine to hold
the RPC calls until they are ready to be sent, we can extend this
process to work nicely with asynchronous RPC calls too.
This patch adds a new list called "rpc_sequence" that defines the order
of the RPC calls to be sent. We add one such list for each state_owner.
When an RPC call is ready to be sent, it checks if it is top of the
rpc_sequence list. If so, it proceeds. If not, it goes back to sleep,
and loops until it hits top of the list.
Once the RPC call has completed, it can then bump the sequence id counter,
and remove itself from the rpc_sequence list, and then wake up the next
sleeper.
Note that the state_owner sequence ids and lock_owner sequence ids are
all indexed to the same rpc_sequence list, so OPEN, LOCK,... requests
are all ordered w.r.t. each other.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2005-10-19 05:20:12 +08:00
|
|
|
sp->so_seqid.flags = 0;
|
2005-10-21 05:22:47 +08:00
|
|
|
spin_lock(&sp->so_lock);
|
NFSv4: Add functions to order RPC calls
NFSv4 file state-changing functions such as OPEN, CLOSE, LOCK,... are all
labelled with "sequence identifiers" in order to prevent the server from
reordering RPC requests, as this could cause its file state to
become out of sync with the client.
Currently the NFS client code enforces this ordering locally using
semaphores to restrict access to structures until the RPC call is done.
This, of course, only works with synchronous RPC calls, since the
user process must first grab the semaphore.
By dropping semaphores, and instead teaching the RPC engine to hold
the RPC calls until they are ready to be sent, we can extend this
process to work nicely with asynchronous RPC calls too.
This patch adds a new list called "rpc_sequence" that defines the order
of the RPC calls to be sent. We add one such list for each state_owner.
When an RPC call is ready to be sent, it checks if it is top of the
rpc_sequence list. If so, it proceeds. If not, it goes back to sleep,
and loops until it hits top of the list.
Once the RPC call has completed, it can then bump the sequence id counter,
and remove itself from the rpc_sequence list, and then wake up the next
sleeper.
Note that the state_owner sequence ids and lock_owner sequence ids are
all indexed to the same rpc_sequence list, so OPEN, LOCK,... requests
are all ordered w.r.t. each other.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2005-10-19 05:20:12 +08:00
|
|
|
list_for_each_entry(state, &sp->so_states, open_states) {
|
2008-12-24 04:21:41 +08:00
|
|
|
if (mark_reclaim(clp, state))
|
|
|
|
nfs4_clear_open_state(state);
|
NFSv4: Add functions to order RPC calls
NFSv4 file state-changing functions such as OPEN, CLOSE, LOCK,... are all
labelled with "sequence identifiers" in order to prevent the server from
reordering RPC requests, as this could cause its file state to
become out of sync with the client.
Currently the NFS client code enforces this ordering locally using
semaphores to restrict access to structures until the RPC call is done.
This, of course, only works with synchronous RPC calls, since the
user process must first grab the semaphore.
By dropping semaphores, and instead teaching the RPC engine to hold
the RPC calls until they are ready to be sent, we can extend this
process to work nicely with asynchronous RPC calls too.
This patch adds a new list called "rpc_sequence" that defines the order
of the RPC calls to be sent. We add one such list for each state_owner.
When an RPC call is ready to be sent, it checks if it is top of the
rpc_sequence list. If so, it proceeds. If not, it goes back to sleep,
and loops until it hits top of the list.
Once the RPC call has completed, it can then bump the sequence id counter,
and remove itself from the rpc_sequence list, and then wake up the next
sleeper.
Note that the state_owner sequence ids and lock_owner sequence ids are
all indexed to the same rpc_sequence list, so OPEN, LOCK,... requests
are all ordered w.r.t. each other.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2005-10-19 05:20:12 +08:00
|
|
|
}
|
2005-10-21 05:22:47 +08:00
|
|
|
spin_unlock(&sp->so_lock);
|
NFSv4: Add functions to order RPC calls
NFSv4 file state-changing functions such as OPEN, CLOSE, LOCK,... are all
labelled with "sequence identifiers" in order to prevent the server from
reordering RPC requests, as this could cause its file state to
become out of sync with the client.
Currently the NFS client code enforces this ordering locally using
semaphores to restrict access to structures until the RPC call is done.
This, of course, only works with synchronous RPC calls, since the
user process must first grab the semaphore.
By dropping semaphores, and instead teaching the RPC engine to hold
the RPC calls until they are ready to be sent, we can extend this
process to work nicely with asynchronous RPC calls too.
This patch adds a new list called "rpc_sequence" that defines the order
of the RPC calls to be sent. We add one such list for each state_owner.
When an RPC call is ready to be sent, it checks if it is top of the
rpc_sequence list. If so, it proceeds. If not, it goes back to sleep,
and loops until it hits top of the list.
Once the RPC call has completed, it can then bump the sequence id counter,
and remove itself from the rpc_sequence list, and then wake up the next
sleeper.
Note that the state_owner sequence ids and lock_owner sequence ids are
all indexed to the same rpc_sequence list, so OPEN, LOCK,... requests
are all ordered w.r.t. each other.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2005-10-19 05:20:12 +08:00
|
|
|
}
|
2010-12-24 09:32:43 +08:00
|
|
|
spin_unlock(&clp->cl_lock);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
|
|
|
|
int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
|
|
|
|
{
|
|
|
|
struct nfs_server *server;
|
|
|
|
|
|
|
|
rcu_read_lock();
|
|
|
|
list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
|
|
|
|
nfs4_reset_seqids(server, mark_reclaim);
|
|
|
|
rcu_read_unlock();
|
NFSv4: Add functions to order RPC calls
NFSv4 file state-changing functions such as OPEN, CLOSE, LOCK,... are all
labelled with "sequence identifiers" in order to prevent the server from
reordering RPC requests, as this could cause its file state to
become out of sync with the client.
Currently the NFS client code enforces this ordering locally using
semaphores to restrict access to structures until the RPC call is done.
This, of course, only works with synchronous RPC calls, since the
user process must first grab the semaphore.
By dropping semaphores, and instead teaching the RPC engine to hold
the RPC calls until they are ready to be sent, we can extend this
process to work nicely with asynchronous RPC calls too.
This patch adds a new list called "rpc_sequence" that defines the order
of the RPC calls to be sent. We add one such list for each state_owner.
When an RPC call is ready to be sent, it checks if it is top of the
rpc_sequence list. If so, it proceeds. If not, it goes back to sleep,
and loops until it hits top of the list.
Once the RPC call has completed, it can then bump the sequence id counter,
and remove itself from the rpc_sequence list, and then wake up the next
sleeper.
Note that the state_owner sequence ids and lock_owner sequence ids are
all indexed to the same rpc_sequence list, so OPEN, LOCK,... requests
are all ordered w.r.t. each other.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2005-10-19 05:20:12 +08:00
|
|
|
}
|
|
|
|
|
2008-12-24 04:21:41 +08:00
|
|
|
static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
|
|
|
|
{
|
|
|
|
/* Mark all delegations for reclaim */
|
|
|
|
nfs_delegation_mark_reclaim(clp);
|
|
|
|
nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
|
|
|
|
}
|
|
|
|
|
2009-12-06 05:08:41 +08:00
|
|
|
static void nfs4_reclaim_complete(struct nfs_client *clp,
|
|
|
|
const struct nfs4_state_recovery_ops *ops)
|
|
|
|
{
|
|
|
|
/* Notify the server we're done reclaiming our state */
|
|
|
|
if (ops->reclaim_complete)
|
|
|
|
(void)ops->reclaim_complete(clp);
|
|
|
|
}
|
|
|
|
|
2010-12-24 09:32:43 +08:00
|
|
|
static void nfs4_clear_reclaim_server(struct nfs_server *server)
|
2008-12-24 04:21:41 +08:00
|
|
|
{
|
2010-12-24 09:32:43 +08:00
|
|
|
struct nfs_client *clp = server->nfs_client;
|
2008-12-24 04:21:41 +08:00
|
|
|
struct nfs4_state_owner *sp;
|
|
|
|
struct rb_node *pos;
|
|
|
|
struct nfs4_state *state;
|
|
|
|
|
2010-12-24 09:32:43 +08:00
|
|
|
spin_lock(&clp->cl_lock);
|
|
|
|
for (pos = rb_first(&server->state_owners);
|
|
|
|
pos != NULL;
|
|
|
|
pos = rb_next(pos)) {
|
|
|
|
sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
|
2008-12-24 04:21:41 +08:00
|
|
|
spin_lock(&sp->so_lock);
|
|
|
|
list_for_each_entry(state, &sp->so_states, open_states) {
|
2010-12-24 09:32:43 +08:00
|
|
|
if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
|
|
|
|
&state->flags))
|
2008-12-24 04:21:41 +08:00
|
|
|
continue;
|
|
|
|
nfs4_state_mark_reclaim_nograce(clp, state);
|
|
|
|
}
|
|
|
|
spin_unlock(&sp->so_lock);
|
|
|
|
}
|
2010-12-24 09:32:43 +08:00
|
|
|
spin_unlock(&clp->cl_lock);
|
|
|
|
}
|
|
|
|
|
|
|
|
static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
|
|
|
|
{
|
|
|
|
struct nfs_server *server;
|
|
|
|
|
|
|
|
if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
rcu_read_lock();
|
|
|
|
list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
|
|
|
|
nfs4_clear_reclaim_server(server);
|
|
|
|
rcu_read_unlock();
|
2008-12-24 04:21:41 +08:00
|
|
|
|
|
|
|
nfs_delegation_reap_unclaimed(clp);
|
2010-10-05 05:59:08 +08:00
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
|
|
|
|
{
|
|
|
|
if (!nfs4_state_clear_reclaim_reboot(clp))
|
|
|
|
return;
|
|
|
|
nfs4_reclaim_complete(clp, clp->cl_mvops->reboot_recovery_ops);
|
2008-12-24 04:21:41 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
static void nfs_delegation_clear_all(struct nfs_client *clp)
|
|
|
|
{
|
|
|
|
nfs_delegation_mark_reclaim(clp);
|
|
|
|
nfs_delegation_reap_unclaimed(clp);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
|
|
|
|
{
|
|
|
|
nfs_delegation_clear_all(clp);
|
|
|
|
nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
|
|
|
|
}
|
|
|
|
|
2010-10-20 07:47:49 +08:00
|
|
|
static void nfs4_warn_keyexpired(const char *s)
|
|
|
|
{
|
|
|
|
printk_ratelimited(KERN_WARNING "Error: state manager"
|
|
|
|
" encountered RPCSEC_GSS session"
|
|
|
|
" expired against NFSv4 server %s.\n",
|
|
|
|
s);
|
|
|
|
}
|
|
|
|
|
2009-12-04 04:53:20 +08:00
|
|
|
static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
|
2008-12-24 04:21:42 +08:00
|
|
|
{
|
|
|
|
switch (error) {
|
2011-12-02 05:31:34 +08:00
|
|
|
case 0:
|
|
|
|
break;
|
2008-12-24 04:21:42 +08:00
|
|
|
case -NFS4ERR_CB_PATH_DOWN:
|
2008-12-24 04:21:47 +08:00
|
|
|
nfs_handle_cb_pathdown(clp);
|
2011-12-02 05:31:34 +08:00
|
|
|
break;
|
2009-12-04 04:53:21 +08:00
|
|
|
case -NFS4ERR_NO_GRACE:
|
|
|
|
nfs4_state_end_reclaim_reboot(clp);
|
2011-12-02 05:31:34 +08:00
|
|
|
break;
|
2008-12-24 04:21:42 +08:00
|
|
|
case -NFS4ERR_STALE_CLIENTID:
|
|
|
|
case -NFS4ERR_LEASE_MOVED:
|
|
|
|
set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
|
2010-10-05 05:59:08 +08:00
|
|
|
nfs4_state_clear_reclaim_reboot(clp);
|
2008-12-24 04:21:42 +08:00
|
|
|
nfs4_state_start_reclaim_reboot(clp);
|
|
|
|
break;
|
|
|
|
case -NFS4ERR_EXPIRED:
|
|
|
|
set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
|
|
|
|
nfs4_state_start_reclaim_nograce(clp);
|
2009-12-05 04:55:34 +08:00
|
|
|
break;
|
2009-04-01 21:22:39 +08:00
|
|
|
case -NFS4ERR_BADSESSION:
|
|
|
|
case -NFS4ERR_BADSLOT:
|
|
|
|
case -NFS4ERR_BAD_HIGH_SLOT:
|
|
|
|
case -NFS4ERR_DEADSESSION:
|
|
|
|
case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
|
|
|
|
case -NFS4ERR_SEQ_FALSE_RETRY:
|
|
|
|
case -NFS4ERR_SEQ_MISORDERED:
|
2009-12-05 04:55:05 +08:00
|
|
|
set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
|
2009-12-05 05:02:14 +08:00
|
|
|
/* Zero session reset errors */
|
2011-12-02 05:31:34 +08:00
|
|
|
break;
|
2010-10-20 07:47:49 +08:00
|
|
|
case -EKEYEXPIRED:
|
|
|
|
/* Nothing we can do */
|
|
|
|
nfs4_warn_keyexpired(clp->cl_hostname);
|
2011-12-02 05:31:34 +08:00
|
|
|
break;
|
|
|
|
default:
|
|
|
|
return error;
|
2008-12-24 04:21:42 +08:00
|
|
|
}
|
2011-12-02 05:31:34 +08:00
|
|
|
return 0;
|
2008-12-24 04:21:42 +08:00
|
|
|
}
|
|
|
|
|
2008-12-24 04:21:40 +08:00
|
|
|
static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2010-12-24 09:32:43 +08:00
|
|
|
struct nfs4_state_owner *sp;
|
|
|
|
struct nfs_server *server;
|
2007-07-03 01:58:33 +08:00
|
|
|
struct rb_node *pos;
|
2005-04-17 06:20:36 +08:00
|
|
|
int status = 0;
|
|
|
|
|
2008-12-24 04:21:43 +08:00
|
|
|
restart:
|
2010-12-24 09:32:43 +08:00
|
|
|
rcu_read_lock();
|
|
|
|
list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
|
NFS: Cache state owners after files are closed
Servers have a finite amount of memory to store NFSv4 open and lock
owners. Moreover, servers may have a difficult time determining when
they can reap their state owner table, thanks to gray areas in the
NFSv4 protocol specification. Thus clients should be careful to reuse
state owners when possible.
Currently Linux is not too careful. When a user has closed all her
files on one mount point, the state owner's reference count goes to
zero, and it is released. The next OPEN allocates a new one. A
workload that serially opens and closes files can run through a large
number of open owners this way.
When a state owner's reference count goes to zero, slap it onto a free
list for that nfs_server, with an expiry time. Garbage collect before
looking for a state owner. This makes state owners for active users
available for re-use.
Now that there can be unused state owners remaining at umount time,
purge the state owner free list when a server is destroyed. Also be
sure not to reclaim unused state owners during state recovery.
This change has benefits for the client as well. For some workloads,
this approach drops the number of OPEN_CONFIRM calls from the same as
the number of OPEN calls, down to just one. This reduces wire traffic
and thus open(2) latency. Before this patch, untarring a kernel
source tarball shows the OPEN_CONFIRM call counter steadily increasing
through the test. With the patch, the OPEN_CONFIRM count remains at 1
throughout the entire untar.
As long as the expiry time is kept short, I don't think garbage
collection should be terribly expensive, although it does bounce the
clp->cl_lock around a bit.
[ At some point we should rationalize the use of the nfs_server
->destroy method. ]
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
[Trond: Fixed a garbage collection race and a few efficiency issues]
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
2011-12-07 05:13:48 +08:00
|
|
|
nfs4_purge_state_owners(server);
|
2010-12-24 09:32:43 +08:00
|
|
|
spin_lock(&clp->cl_lock);
|
|
|
|
for (pos = rb_first(&server->state_owners);
|
|
|
|
pos != NULL;
|
|
|
|
pos = rb_next(pos)) {
|
|
|
|
sp = rb_entry(pos,
|
|
|
|
struct nfs4_state_owner, so_server_node);
|
|
|
|
if (!test_and_clear_bit(ops->owner_flag_bit,
|
|
|
|
&sp->so_flags))
|
|
|
|
continue;
|
|
|
|
atomic_inc(&sp->so_count);
|
|
|
|
spin_unlock(&clp->cl_lock);
|
|
|
|
rcu_read_unlock();
|
|
|
|
|
|
|
|
status = nfs4_reclaim_open_state(sp, ops);
|
|
|
|
if (status < 0) {
|
|
|
|
set_bit(ops->owner_flag_bit, &sp->so_flags);
|
|
|
|
nfs4_put_state_owner(sp);
|
|
|
|
return nfs4_recovery_handle_error(clp, status);
|
|
|
|
}
|
|
|
|
|
2008-12-24 04:21:43 +08:00
|
|
|
nfs4_put_state_owner(sp);
|
2010-12-24 09:32:43 +08:00
|
|
|
goto restart;
|
2008-12-24 04:21:43 +08:00
|
|
|
}
|
2010-12-24 09:32:43 +08:00
|
|
|
spin_unlock(&clp->cl_lock);
|
2008-12-24 04:21:40 +08:00
|
|
|
}
|
2010-12-24 09:32:43 +08:00
|
|
|
rcu_read_unlock();
|
2008-12-24 04:21:40 +08:00
|
|
|
return status;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int nfs4_check_lease(struct nfs_client *clp)
|
|
|
|
{
|
|
|
|
struct rpc_cred *cred;
|
2010-06-16 21:52:27 +08:00
|
|
|
const struct nfs4_state_maintenance_ops *ops =
|
|
|
|
clp->cl_mvops->state_renewal_ops;
|
2011-12-02 05:31:34 +08:00
|
|
|
int status;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2008-12-24 04:21:42 +08:00
|
|
|
/* Is the client already known to have an expired lease? */
|
|
|
|
if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
|
|
|
|
return 0;
|
2009-04-01 21:22:46 +08:00
|
|
|
spin_lock(&clp->cl_lock);
|
|
|
|
cred = ops->get_state_renewal_cred_locked(clp);
|
|
|
|
spin_unlock(&clp->cl_lock);
|
2008-12-24 04:21:42 +08:00
|
|
|
if (cred == NULL) {
|
|
|
|
cred = nfs4_get_setclientid_cred(clp);
|
2011-12-02 05:31:34 +08:00
|
|
|
status = -ENOKEY;
|
2008-12-24 04:21:42 +08:00
|
|
|
if (cred == NULL)
|
|
|
|
goto out;
|
2006-01-03 16:55:26 +08:00
|
|
|
}
|
2009-04-01 21:22:45 +08:00
|
|
|
status = ops->renew_lease(clp, cred);
|
2008-12-24 04:21:42 +08:00
|
|
|
put_rpccred(cred);
|
|
|
|
out:
|
2009-12-04 04:53:20 +08:00
|
|
|
return nfs4_recovery_handle_error(clp, status);
|
2008-12-24 04:21:40 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
static int nfs4_reclaim_lease(struct nfs_client *clp)
|
|
|
|
{
|
|
|
|
struct rpc_cred *cred;
|
2010-06-16 21:52:27 +08:00
|
|
|
const struct nfs4_state_recovery_ops *ops =
|
|
|
|
clp->cl_mvops->reboot_recovery_ops;
|
2008-12-24 04:21:40 +08:00
|
|
|
int status = -ENOENT;
|
|
|
|
|
2009-04-01 21:22:48 +08:00
|
|
|
cred = ops->get_clid_cred(clp);
|
2006-01-03 16:55:26 +08:00
|
|
|
if (cred != NULL) {
|
2009-04-01 21:22:47 +08:00
|
|
|
status = ops->establish_clid(clp, cred);
|
2006-01-03 16:55:26 +08:00
|
|
|
put_rpccred(cred);
|
2008-04-09 04:02:17 +08:00
|
|
|
/* Handle case where the user hasn't set up machine creds */
|
|
|
|
if (status == -EACCES && cred == clp->cl_machine_cred) {
|
|
|
|
nfs4_clear_machine_cred(clp);
|
2008-12-24 04:21:40 +08:00
|
|
|
status = -EAGAIN;
|
2008-04-09 04:02:17 +08:00
|
|
|
}
|
2009-04-01 21:21:51 +08:00
|
|
|
if (status == -NFS4ERR_MINOR_VERS_MISMATCH)
|
|
|
|
status = -EPROTONOSUPPORT;
|
2006-01-03 16:55:26 +08:00
|
|
|
}
|
2008-12-24 04:21:40 +08:00
|
|
|
return status;
|
|
|
|
}
|
|
|
|
|
2009-04-01 21:22:38 +08:00
|
|
|
#ifdef CONFIG_NFS_V4_1
|
2011-03-10 05:00:53 +08:00
|
|
|
void nfs4_schedule_session_recovery(struct nfs4_session *session)
|
|
|
|
{
|
2011-05-27 02:26:35 +08:00
|
|
|
struct nfs_client *clp = session->clp;
|
|
|
|
|
|
|
|
set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
|
|
|
|
nfs4_schedule_lease_recovery(clp);
|
2011-03-10 05:00:53 +08:00
|
|
|
}
|
2011-03-01 09:34:20 +08:00
|
|
|
EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
|
2011-03-10 05:00:53 +08:00
|
|
|
|
2010-01-21 05:06:27 +08:00
|
|
|
void nfs41_handle_recall_slot(struct nfs_client *clp)
|
|
|
|
{
|
|
|
|
set_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state);
|
2011-03-10 05:00:53 +08:00
|
|
|
nfs4_schedule_state_manager(clp);
|
2010-01-21 05:06:27 +08:00
|
|
|
}
|
|
|
|
|
2010-03-03 02:06:21 +08:00
|
|
|
static void nfs4_reset_all_state(struct nfs_client *clp)
|
|
|
|
{
|
|
|
|
if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
|
|
|
|
clp->cl_boot_time = CURRENT_TIME;
|
|
|
|
nfs4_state_start_reclaim_nograce(clp);
|
2011-03-10 05:00:53 +08:00
|
|
|
nfs4_schedule_state_manager(clp);
|
2010-03-03 02:06:21 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static void nfs41_handle_server_reboot(struct nfs_client *clp)
|
|
|
|
{
|
|
|
|
if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
|
|
|
|
nfs4_state_start_reclaim_reboot(clp);
|
2011-03-10 05:00:53 +08:00
|
|
|
nfs4_schedule_state_manager(clp);
|
2010-03-03 02:06:21 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static void nfs41_handle_state_revoked(struct nfs_client *clp)
|
|
|
|
{
|
|
|
|
/* Temporary */
|
|
|
|
nfs4_reset_all_state(clp);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
|
|
|
|
{
|
|
|
|
/* This will need to handle layouts too */
|
|
|
|
nfs_expire_all_delegations(clp);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void nfs41_handle_cb_path_down(struct nfs_client *clp)
|
|
|
|
{
|
|
|
|
nfs_expire_all_delegations(clp);
|
|
|
|
if (test_and_set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) == 0)
|
2011-03-10 05:00:53 +08:00
|
|
|
nfs4_schedule_state_manager(clp);
|
2010-03-03 02:06:21 +08:00
|
|
|
}
|
|
|
|
|
2009-12-06 02:46:14 +08:00
|
|
|
void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags)
|
|
|
|
{
|
|
|
|
if (!flags)
|
|
|
|
return;
|
2011-12-02 05:37:42 +08:00
|
|
|
if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
|
2010-03-03 02:06:21 +08:00
|
|
|
nfs41_handle_server_reboot(clp);
|
2011-12-02 05:37:42 +08:00
|
|
|
if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED |
|
2009-12-06 02:46:14 +08:00
|
|
|
SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
|
|
|
|
SEQ4_STATUS_ADMIN_STATE_REVOKED |
|
2010-03-03 02:06:21 +08:00
|
|
|
SEQ4_STATUS_LEASE_MOVED))
|
|
|
|
nfs41_handle_state_revoked(clp);
|
2011-12-02 05:37:42 +08:00
|
|
|
if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
|
2010-03-03 02:06:21 +08:00
|
|
|
nfs41_handle_recallable_state_revoked(clp);
|
2011-12-02 05:37:42 +08:00
|
|
|
if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
|
2009-12-06 02:46:14 +08:00
|
|
|
SEQ4_STATUS_BACKCHANNEL_FAULT |
|
|
|
|
SEQ4_STATUS_CB_PATH_DOWN_SESSION))
|
2010-03-03 02:06:21 +08:00
|
|
|
nfs41_handle_cb_path_down(clp);
|
2009-12-06 02:46:14 +08:00
|
|
|
}
|
|
|
|
|
2009-04-01 21:22:39 +08:00
|
|
|
static int nfs4_reset_session(struct nfs_client *clp)
|
|
|
|
{
|
|
|
|
int status;
|
|
|
|
|
2009-12-16 06:36:57 +08:00
|
|
|
nfs4_begin_drain_session(clp);
|
2009-04-01 21:22:39 +08:00
|
|
|
status = nfs4_proc_destroy_session(clp->cl_session);
|
|
|
|
if (status && status != -NFS4ERR_BADSESSION &&
|
|
|
|
status != -NFS4ERR_DEADSESSION) {
|
2009-12-07 22:16:09 +08:00
|
|
|
status = nfs4_recovery_handle_error(clp, status);
|
2009-04-01 21:22:39 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
|
2009-12-06 08:32:11 +08:00
|
|
|
status = nfs4_proc_create_session(clp);
|
2010-01-22 03:54:13 +08:00
|
|
|
if (status) {
|
2009-12-07 22:16:09 +08:00
|
|
|
status = nfs4_recovery_handle_error(clp, status);
|
2010-01-22 03:54:13 +08:00
|
|
|
goto out;
|
|
|
|
}
|
2011-05-27 02:26:35 +08:00
|
|
|
clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
|
2010-01-22 03:54:13 +08:00
|
|
|
/* create_session negotiated new slot table */
|
|
|
|
clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state);
|
2009-12-07 01:57:34 +08:00
|
|
|
|
2010-01-22 03:54:13 +08:00
|
|
|
/* Let the state manager reestablish state */
|
|
|
|
if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
|
2009-12-15 13:27:58 +08:00
|
|
|
nfs41_setup_state_renewal(clp);
|
2010-01-22 03:54:13 +08:00
|
|
|
out:
|
2009-04-01 21:22:39 +08:00
|
|
|
return status;
|
|
|
|
}
|
2009-04-01 21:22:38 +08:00
|
|
|
|
2010-01-21 05:06:27 +08:00
|
|
|
static int nfs4_recall_slot(struct nfs_client *clp)
|
|
|
|
{
|
|
|
|
struct nfs4_slot_table *fc_tbl = &clp->cl_session->fc_slot_table;
|
|
|
|
struct nfs4_channel_attrs *fc_attrs = &clp->cl_session->fc_attrs;
|
|
|
|
struct nfs4_slot *new, *old;
|
|
|
|
int i;
|
|
|
|
|
|
|
|
nfs4_begin_drain_session(clp);
|
|
|
|
new = kmalloc(fc_tbl->target_max_slots * sizeof(struct nfs4_slot),
|
2010-05-14 00:51:01 +08:00
|
|
|
GFP_NOFS);
|
2010-01-21 05:06:27 +08:00
|
|
|
if (!new)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
|
|
|
spin_lock(&fc_tbl->slot_tbl_lock);
|
|
|
|
for (i = 0; i < fc_tbl->target_max_slots; i++)
|
|
|
|
new[i].seq_nr = fc_tbl->slots[i].seq_nr;
|
|
|
|
old = fc_tbl->slots;
|
|
|
|
fc_tbl->slots = new;
|
|
|
|
fc_tbl->max_slots = fc_tbl->target_max_slots;
|
|
|
|
fc_tbl->target_max_slots = 0;
|
|
|
|
fc_attrs->max_reqs = fc_tbl->max_slots;
|
|
|
|
spin_unlock(&fc_tbl->slot_tbl_lock);
|
|
|
|
|
|
|
|
kfree(old);
|
|
|
|
nfs4_end_drain_session(clp);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2009-04-01 21:22:38 +08:00
|
|
|
#else /* CONFIG_NFS_V4_1 */
|
2009-04-01 21:22:39 +08:00
|
|
|
static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
|
2009-12-15 13:27:58 +08:00
|
|
|
static int nfs4_end_drain_session(struct nfs_client *clp) { return 0; }
|
2010-01-21 05:06:27 +08:00
|
|
|
static int nfs4_recall_slot(struct nfs_client *clp) { return 0; }
|
2009-04-01 21:22:38 +08:00
|
|
|
#endif /* CONFIG_NFS_V4_1 */
|
|
|
|
|
2009-06-17 02:43:15 +08:00
|
|
|
/* Set NFS4CLNT_LEASE_EXPIRED for all v4.0 errors and for recoverable errors
|
|
|
|
* on EXCHANGE_ID for v4.1
|
|
|
|
*/
|
|
|
|
static void nfs4_set_lease_expired(struct nfs_client *clp, int status)
|
|
|
|
{
|
2011-04-25 02:28:18 +08:00
|
|
|
switch (status) {
|
|
|
|
case -NFS4ERR_CLID_INUSE:
|
|
|
|
case -NFS4ERR_STALE_CLIENTID:
|
|
|
|
clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
|
|
|
|
break;
|
|
|
|
case -NFS4ERR_DELAY:
|
2011-04-25 02:29:33 +08:00
|
|
|
case -ETIMEDOUT:
|
2011-04-25 02:28:18 +08:00
|
|
|
case -EAGAIN:
|
|
|
|
ssleep(1);
|
|
|
|
break;
|
2009-06-17 02:43:15 +08:00
|
|
|
|
2011-04-25 02:28:18 +08:00
|
|
|
case -EKEYEXPIRED:
|
|
|
|
nfs4_warn_keyexpired(clp->cl_hostname);
|
|
|
|
case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
|
|
|
|
* in nfs4_exchange_id */
|
|
|
|
default:
|
|
|
|
return;
|
2009-06-17 02:43:15 +08:00
|
|
|
}
|
|
|
|
set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
|
|
|
|
}
|
|
|
|
|
2008-12-24 04:21:48 +08:00
|
|
|
static void nfs4_state_manager(struct nfs_client *clp)
|
2008-12-24 04:21:40 +08:00
|
|
|
{
|
|
|
|
int status = 0;
|
|
|
|
|
|
|
|
/* Ensure exclusive access to NFSv4 state */
|
2011-04-16 05:34:18 +08:00
|
|
|
do {
|
2008-12-24 04:21:41 +08:00
|
|
|
if (test_and_clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
|
|
|
|
/* We're going to have to re-establish a clientid */
|
|
|
|
status = nfs4_reclaim_lease(clp);
|
|
|
|
if (status) {
|
2009-06-17 02:43:15 +08:00
|
|
|
nfs4_set_lease_expired(clp, status);
|
2009-12-04 04:53:22 +08:00
|
|
|
if (test_bit(NFS4CLNT_LEASE_EXPIRED,
|
|
|
|
&clp->cl_state))
|
2008-12-24 04:21:41 +08:00
|
|
|
continue;
|
2009-04-01 21:22:38 +08:00
|
|
|
if (clp->cl_cons_state ==
|
|
|
|
NFS_CS_SESSION_INITING)
|
|
|
|
nfs_mark_client_ready(clp, status);
|
2008-12-24 04:21:41 +08:00
|
|
|
goto out_error;
|
|
|
|
}
|
2008-12-24 04:21:42 +08:00
|
|
|
clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
|
2011-06-01 07:05:47 +08:00
|
|
|
|
|
|
|
if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH,
|
|
|
|
&clp->cl_state))
|
|
|
|
nfs4_state_start_reclaim_nograce(clp);
|
|
|
|
else
|
|
|
|
set_bit(NFS4CLNT_RECLAIM_REBOOT,
|
|
|
|
&clp->cl_state);
|
|
|
|
|
2010-10-20 12:18:02 +08:00
|
|
|
pnfs_destroy_all_layouts(clp);
|
2008-12-24 04:21:42 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
|
|
|
|
status = nfs4_check_lease(clp);
|
2011-12-02 05:31:34 +08:00
|
|
|
if (status < 0)
|
|
|
|
goto out_error;
|
2009-12-04 04:53:22 +08:00
|
|
|
if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
|
2008-12-24 04:21:42 +08:00
|
|
|
continue;
|
2008-12-24 04:21:41 +08:00
|
|
|
}
|
2009-12-04 04:53:22 +08:00
|
|
|
|
2009-04-01 21:22:39 +08:00
|
|
|
/* Initialize or reset the session */
|
2009-12-05 04:55:05 +08:00
|
|
|
if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)
|
2009-08-25 07:21:29 +08:00
|
|
|
&& nfs4_has_session(clp)) {
|
2009-12-05 04:52:24 +08:00
|
|
|
status = nfs4_reset_session(clp);
|
2009-12-04 04:53:22 +08:00
|
|
|
if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
|
|
|
|
continue;
|
|
|
|
if (status < 0)
|
2009-04-01 21:22:38 +08:00
|
|
|
goto out_error;
|
|
|
|
}
|
2009-12-04 04:53:22 +08:00
|
|
|
|
2008-12-24 04:21:41 +08:00
|
|
|
/* First recover reboot state... */
|
2009-12-04 04:52:41 +08:00
|
|
|
if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
|
2009-04-01 21:22:47 +08:00
|
|
|
status = nfs4_do_reclaim(clp,
|
2010-06-16 21:52:27 +08:00
|
|
|
clp->cl_mvops->reboot_recovery_ops);
|
2009-12-04 04:53:22 +08:00
|
|
|
if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
|
2009-12-05 04:55:05 +08:00
|
|
|
test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
|
2009-04-01 21:22:39 +08:00
|
|
|
continue;
|
2008-12-24 04:21:41 +08:00
|
|
|
nfs4_state_end_reclaim_reboot(clp);
|
2009-12-04 04:53:22 +08:00
|
|
|
if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
|
|
|
|
continue;
|
|
|
|
if (status < 0)
|
|
|
|
goto out_error;
|
2008-12-24 04:21:40 +08:00
|
|
|
}
|
|
|
|
|
2008-12-24 04:21:41 +08:00
|
|
|
/* Now recover expired state... */
|
|
|
|
if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
|
2009-04-01 21:22:47 +08:00
|
|
|
status = nfs4_do_reclaim(clp,
|
2010-06-16 21:52:27 +08:00
|
|
|
clp->cl_mvops->nograce_recovery_ops);
|
2009-12-04 04:53:22 +08:00
|
|
|
if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
|
2009-12-05 04:55:05 +08:00
|
|
|
test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) ||
|
2009-12-04 04:53:22 +08:00
|
|
|
test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
|
|
|
|
continue;
|
|
|
|
if (status < 0)
|
2008-12-24 04:21:41 +08:00
|
|
|
goto out_error;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
2008-12-24 04:21:47 +08:00
|
|
|
|
2009-12-15 13:27:58 +08:00
|
|
|
nfs4_end_drain_session(clp);
|
2008-12-24 04:21:47 +08:00
|
|
|
if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
|
|
|
|
nfs_client_return_marked_delegations(clp);
|
|
|
|
continue;
|
|
|
|
}
|
2010-01-21 05:06:27 +08:00
|
|
|
/* Recall session slots */
|
|
|
|
if (test_and_clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state)
|
|
|
|
&& nfs4_has_session(clp)) {
|
|
|
|
status = nfs4_recall_slot(clp);
|
|
|
|
if (status < 0)
|
|
|
|
goto out_error;
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
2008-12-24 04:21:48 +08:00
|
|
|
|
|
|
|
nfs4_clear_state_manager_bit(clp);
|
2008-12-24 04:21:48 +08:00
|
|
|
/* Did we race with an attempt to give us more work? */
|
|
|
|
if (clp->cl_state == 0)
|
|
|
|
break;
|
|
|
|
if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
|
|
|
|
break;
|
2011-04-16 05:34:18 +08:00
|
|
|
} while (atomic_read(&clp->cl_count) > 1);
|
2008-12-24 04:21:48 +08:00
|
|
|
return;
|
2005-04-17 06:20:36 +08:00
|
|
|
out_error:
|
2012-01-27 02:32:22 +08:00
|
|
|
printk(KERN_WARNING "NFS: state manager failed on NFSv4 server %s"
|
2007-12-11 03:57:16 +08:00
|
|
|
" with error %d\n", clp->cl_hostname, -status);
|
2009-12-15 13:27:58 +08:00
|
|
|
nfs4_end_drain_session(clp);
|
2008-12-24 04:21:48 +08:00
|
|
|
nfs4_clear_state_manager_bit(clp);
|
|
|
|
}
|
|
|
|
|
|
|
|
static int nfs4_run_state_manager(void *ptr)
|
|
|
|
{
|
|
|
|
struct nfs_client *clp = ptr;
|
|
|
|
|
|
|
|
allow_signal(SIGKILL);
|
|
|
|
nfs4_state_manager(clp);
|
|
|
|
nfs_put_client(clp);
|
|
|
|
module_put_and_exit(0);
|
|
|
|
return 0;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Local variables:
|
|
|
|
* c-basic-offset: 8
|
|
|
|
* End:
|
|
|
|
*/
|