2005-04-17 00:20:36 +02:00
|
|
|
/*
|
|
|
|
* 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.
|
|
|
|
*/
|
|
|
|
|
2007-07-08 22:49:11 +02:00
|
|
|
#include <linux/kernel.h>
|
2005-04-17 00:20:36 +02:00
|
|
|
#include <linux/slab.h>
|
|
|
|
#include <linux/smp_lock.h>
|
|
|
|
#include <linux/nfs_fs.h>
|
|
|
|
#include <linux/nfs_idmap.h>
|
2006-01-03 09:55:23 +01:00
|
|
|
#include <linux/kthread.h>
|
|
|
|
#include <linux/module.h>
|
2007-07-02 19:58:33 +02:00
|
|
|
#include <linux/random.h>
|
2005-04-17 00:20:36 +02:00
|
|
|
#include <linux/workqueue.h>
|
|
|
|
#include <linux/bitops.h>
|
|
|
|
|
2005-06-22 19:16:21 +02:00
|
|
|
#include "nfs4_fs.h"
|
2005-04-17 00:20:36 +02:00
|
|
|
#include "callback.h"
|
|
|
|
#include "delegation.h"
|
2006-08-23 02:06:10 +02:00
|
|
|
#include "internal.h"
|
2005-04-17 00:20:36 +02:00
|
|
|
|
|
|
|
#define OPENOWNER_POOL_SIZE 8
|
|
|
|
|
2005-06-22 19:16:21 +02:00
|
|
|
const nfs4_stateid zero_stateid;
|
2005-04-17 00:20:36 +02:00
|
|
|
|
|
|
|
static LIST_HEAD(nfs4_clientid_list);
|
|
|
|
|
2006-08-23 02:06:08 +02:00
|
|
|
static int nfs4_init_client(struct nfs_client *clp, struct rpc_cred *cred)
|
2005-04-17 00:20:36 +02:00
|
|
|
{
|
2006-01-03 09:55:26 +01:00
|
|
|
int status = nfs4_proc_setclientid(clp, NFS4_CALLBACK,
|
|
|
|
nfs_callback_tcpport, cred);
|
2005-04-17 00:20:36 +02:00
|
|
|
if (status == 0)
|
2006-01-03 09:55:26 +01:00
|
|
|
status = nfs4_proc_setclientid_confirm(clp, cred);
|
2005-04-17 00:20:36 +02:00
|
|
|
if (status == 0)
|
|
|
|
nfs4_schedule_state_renewal(clp);
|
|
|
|
return status;
|
|
|
|
}
|
|
|
|
|
2006-08-23 02:06:08 +02:00
|
|
|
struct rpc_cred *nfs4_get_renew_cred(struct nfs_client *clp)
|
2006-01-03 09:55:25 +01:00
|
|
|
{
|
|
|
|
struct nfs4_state_owner *sp;
|
2007-07-02 19:58:33 +02:00
|
|
|
struct rb_node *pos;
|
2006-01-03 09:55:25 +01:00
|
|
|
struct rpc_cred *cred = NULL;
|
|
|
|
|
2007-07-02 19:58:33 +02:00
|
|
|
for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
|
|
|
|
sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
|
2006-01-03 09:55:25 +01:00
|
|
|
if (list_empty(&sp->so_states))
|
|
|
|
continue;
|
|
|
|
cred = get_rpccred(sp->so_cred);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
return cred;
|
|
|
|
}
|
|
|
|
|
2007-05-14 23:16:04 +02:00
|
|
|
static struct rpc_cred *nfs4_get_setclientid_cred(struct nfs_client *clp)
|
2006-01-03 09:55:26 +01:00
|
|
|
{
|
|
|
|
struct nfs4_state_owner *sp;
|
2007-07-02 19:58:33 +02:00
|
|
|
struct rb_node *pos;
|
2006-01-03 09:55:26 +01:00
|
|
|
|
2007-07-02 19:58:33 +02:00
|
|
|
pos = rb_first(&clp->cl_state_owners);
|
|
|
|
if (pos != NULL) {
|
|
|
|
sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
|
2006-01-03 09:55:26 +01:00
|
|
|
return get_rpccred(sp->so_cred);
|
|
|
|
}
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
2007-07-02 19:58:33 +02:00
|
|
|
static void nfs_alloc_unique_id(struct rb_root *root, struct nfs_unique_id *new,
|
|
|
|
__u64 minval, int maxbits)
|
|
|
|
{
|
|
|
|
struct rb_node **p, *parent;
|
|
|
|
struct nfs_unique_id *pos;
|
|
|
|
__u64 mask = ~0ULL;
|
|
|
|
|
|
|
|
if (maxbits < 64)
|
|
|
|
mask = (1ULL << maxbits) - 1ULL;
|
|
|
|
|
|
|
|
/* Ensure distribution is more or less flat */
|
|
|
|
get_random_bytes(&new->id, sizeof(new->id));
|
|
|
|
new->id &= mask;
|
|
|
|
if (new->id < minval)
|
|
|
|
new->id += minval;
|
|
|
|
retry:
|
|
|
|
p = &root->rb_node;
|
|
|
|
parent = NULL;
|
|
|
|
|
|
|
|
while (*p != NULL) {
|
|
|
|
parent = *p;
|
|
|
|
pos = rb_entry(parent, struct nfs_unique_id, rb_node);
|
|
|
|
|
|
|
|
if (new->id < pos->id)
|
|
|
|
p = &(*p)->rb_left;
|
|
|
|
else if (new->id > pos->id)
|
|
|
|
p = &(*p)->rb_right;
|
|
|
|
else
|
|
|
|
goto id_exists;
|
|
|
|
}
|
|
|
|
rb_link_node(&new->rb_node, parent, p);
|
|
|
|
rb_insert_color(&new->rb_node, root);
|
|
|
|
return;
|
|
|
|
id_exists:
|
|
|
|
for (;;) {
|
|
|
|
new->id++;
|
|
|
|
if (new->id < minval || (new->id & mask) != new->id) {
|
|
|
|
new->id = minval;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
parent = rb_next(parent);
|
|
|
|
if (parent == NULL)
|
|
|
|
break;
|
|
|
|
pos = rb_entry(parent, struct nfs_unique_id, rb_node);
|
|
|
|
if (new->id < pos->id)
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
goto retry;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void nfs_free_unique_id(struct rb_root *root, struct nfs_unique_id *id)
|
|
|
|
{
|
|
|
|
rb_erase(&id->rb_node, root);
|
|
|
|
}
|
|
|
|
|
2005-04-17 00:20:36 +02:00
|
|
|
static struct nfs4_state_owner *
|
2007-07-06 16:53:21 +02:00
|
|
|
nfs4_find_state_owner(struct nfs_server *server, struct rpc_cred *cred)
|
2005-04-17 00:20:36 +02:00
|
|
|
{
|
2007-07-06 16:53:21 +02:00
|
|
|
struct nfs_client *clp = server->nfs_client;
|
2007-07-02 19:58:33 +02:00
|
|
|
struct rb_node **p = &clp->cl_state_owners.rb_node,
|
|
|
|
*parent = NULL;
|
2005-04-17 00:20:36 +02:00
|
|
|
struct nfs4_state_owner *sp, *res = NULL;
|
|
|
|
|
2007-07-02 19:58:33 +02:00
|
|
|
while (*p != NULL) {
|
|
|
|
parent = *p;
|
|
|
|
sp = rb_entry(parent, struct nfs4_state_owner, so_client_node);
|
|
|
|
|
2007-07-06 16:53:21 +02:00
|
|
|
if (server < sp->so_server) {
|
|
|
|
p = &parent->rb_left;
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
if (server > sp->so_server) {
|
|
|
|
p = &parent->rb_right;
|
|
|
|
continue;
|
|
|
|
}
|
2007-07-02 19:58:33 +02:00
|
|
|
if (cred < sp->so_cred)
|
|
|
|
p = &parent->rb_left;
|
|
|
|
else if (cred > sp->so_cred)
|
|
|
|
p = &parent->rb_right;
|
|
|
|
else {
|
|
|
|
atomic_inc(&sp->so_count);
|
|
|
|
res = sp;
|
|
|
|
break;
|
|
|
|
}
|
2005-04-17 00:20:36 +02:00
|
|
|
}
|
|
|
|
return res;
|
|
|
|
}
|
|
|
|
|
2007-07-02 19:58:33 +02:00
|
|
|
static struct nfs4_state_owner *
|
|
|
|
nfs4_insert_state_owner(struct nfs_client *clp, struct nfs4_state_owner *new)
|
|
|
|
{
|
|
|
|
struct rb_node **p = &clp->cl_state_owners.rb_node,
|
|
|
|
*parent = NULL;
|
|
|
|
struct nfs4_state_owner *sp;
|
|
|
|
|
|
|
|
while (*p != NULL) {
|
|
|
|
parent = *p;
|
|
|
|
sp = rb_entry(parent, struct nfs4_state_owner, so_client_node);
|
|
|
|
|
2007-07-06 16:53:21 +02:00
|
|
|
if (new->so_server < sp->so_server) {
|
|
|
|
p = &parent->rb_left;
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
if (new->so_server > sp->so_server) {
|
|
|
|
p = &parent->rb_right;
|
|
|
|
continue;
|
|
|
|
}
|
2007-07-02 19:58:33 +02:00
|
|
|
if (new->so_cred < sp->so_cred)
|
|
|
|
p = &parent->rb_left;
|
|
|
|
else if (new->so_cred > sp->so_cred)
|
|
|
|
p = &parent->rb_right;
|
|
|
|
else {
|
|
|
|
atomic_inc(&sp->so_count);
|
|
|
|
return sp;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
nfs_alloc_unique_id(&clp->cl_openowner_id, &new->so_owner_id, 1, 64);
|
|
|
|
rb_link_node(&new->so_client_node, parent, p);
|
|
|
|
rb_insert_color(&new->so_client_node, &clp->cl_state_owners);
|
|
|
|
return new;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
|
|
|
nfs4_remove_state_owner(struct nfs_client *clp, struct nfs4_state_owner *sp)
|
|
|
|
{
|
|
|
|
if (!RB_EMPTY_NODE(&sp->so_client_node))
|
|
|
|
rb_erase(&sp->so_client_node, &clp->cl_state_owners);
|
|
|
|
nfs_free_unique_id(&clp->cl_openowner_id, &sp->so_owner_id);
|
|
|
|
}
|
|
|
|
|
2005-04-17 00:20:36 +02:00
|
|
|
/*
|
|
|
|
* nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
|
|
|
|
* create a new state_owner.
|
|
|
|
*
|
|
|
|
*/
|
|
|
|
static struct nfs4_state_owner *
|
|
|
|
nfs4_alloc_state_owner(void)
|
|
|
|
{
|
|
|
|
struct nfs4_state_owner *sp;
|
|
|
|
|
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>
2005-10-18 23:20:12 +02:00
|
|
|
sp = kzalloc(sizeof(*sp),GFP_KERNEL);
|
2005-04-17 00:20:36 +02:00
|
|
|
if (!sp)
|
|
|
|
return NULL;
|
2005-10-20 23:22:47 +02:00
|
|
|
spin_lock_init(&sp->so_lock);
|
2005-04-17 00:20:36 +02:00
|
|
|
INIT_LIST_HEAD(&sp->so_states);
|
|
|
|
INIT_LIST_HEAD(&sp->so_delegations);
|
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>
2005-10-18 23:20:12 +02:00
|
|
|
rpc_init_wait_queue(&sp->so_sequence.wait, "Seqid_waitqueue");
|
|
|
|
sp->so_seqid.sequence = &sp->so_sequence;
|
|
|
|
spin_lock_init(&sp->so_sequence.lock);
|
|
|
|
INIT_LIST_HEAD(&sp->so_sequence.list);
|
2005-04-17 00:20:36 +02:00
|
|
|
atomic_set(&sp->so_count, 1);
|
|
|
|
return sp;
|
|
|
|
}
|
|
|
|
|
|
|
|
void
|
|
|
|
nfs4_drop_state_owner(struct nfs4_state_owner *sp)
|
|
|
|
{
|
2007-07-02 19:58:33 +02:00
|
|
|
if (!RB_EMPTY_NODE(&sp->so_client_node)) {
|
|
|
|
struct nfs_client *clp = sp->so_client;
|
|
|
|
|
|
|
|
spin_lock(&clp->cl_lock);
|
|
|
|
rb_erase(&sp->so_client_node, &clp->cl_state_owners);
|
|
|
|
RB_CLEAR_NODE(&sp->so_client_node);
|
|
|
|
spin_unlock(&clp->cl_lock);
|
|
|
|
}
|
2005-04-17 00:20:36 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Note: must be called with clp->cl_sem held in order to prevent races
|
|
|
|
* with reboot recovery!
|
|
|
|
*/
|
|
|
|
struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server, struct rpc_cred *cred)
|
|
|
|
{
|
2006-08-23 02:06:09 +02:00
|
|
|
struct nfs_client *clp = server->nfs_client;
|
2005-04-17 00:20:36 +02:00
|
|
|
struct nfs4_state_owner *sp, *new;
|
|
|
|
|
|
|
|
spin_lock(&clp->cl_lock);
|
2007-07-06 16:53:21 +02:00
|
|
|
sp = nfs4_find_state_owner(server, cred);
|
2005-04-17 00:20:36 +02:00
|
|
|
spin_unlock(&clp->cl_lock);
|
|
|
|
if (sp != NULL)
|
|
|
|
return sp;
|
2007-07-02 19:58:33 +02:00
|
|
|
new = nfs4_alloc_state_owner();
|
|
|
|
if (new == NULL)
|
|
|
|
return NULL;
|
|
|
|
new->so_client = clp;
|
2007-07-06 16:53:21 +02:00
|
|
|
new->so_server = server;
|
2007-07-02 19:58:33 +02:00
|
|
|
new->so_cred = cred;
|
|
|
|
spin_lock(&clp->cl_lock);
|
|
|
|
sp = nfs4_insert_state_owner(clp, new);
|
|
|
|
spin_unlock(&clp->cl_lock);
|
|
|
|
if (sp == new)
|
|
|
|
get_rpccred(cred);
|
|
|
|
else
|
|
|
|
kfree(new);
|
|
|
|
return sp;
|
2005-04-17 00:20:36 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Must be called with clp->cl_sem held in order to avoid races
|
|
|
|
* with state recovery...
|
|
|
|
*/
|
|
|
|
void nfs4_put_state_owner(struct nfs4_state_owner *sp)
|
|
|
|
{
|
2006-08-23 02:06:08 +02:00
|
|
|
struct nfs_client *clp = sp->so_client;
|
2005-04-17 00:20:36 +02:00
|
|
|
struct rpc_cred *cred = sp->so_cred;
|
|
|
|
|
|
|
|
if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
|
|
|
|
return;
|
2007-07-02 19:58:33 +02:00
|
|
|
nfs4_remove_state_owner(clp, sp);
|
2005-04-17 00:20:36 +02:00
|
|
|
spin_unlock(&clp->cl_lock);
|
|
|
|
put_rpccred(cred);
|
|
|
|
kfree(sp);
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct nfs4_state *
|
|
|
|
nfs4_alloc_open_state(void)
|
|
|
|
{
|
|
|
|
struct nfs4_state *state;
|
|
|
|
|
2006-01-03 09:55:13 +01:00
|
|
|
state = kzalloc(sizeof(*state), GFP_KERNEL);
|
2005-04-17 00:20:36 +02:00
|
|
|
if (!state)
|
|
|
|
return NULL;
|
|
|
|
atomic_set(&state->count, 1);
|
|
|
|
INIT_LIST_HEAD(&state->lock_states);
|
2005-06-22 19:16:32 +02:00
|
|
|
spin_lock_init(&state->state_lock);
|
2007-07-09 16:45:42 +02:00
|
|
|
seqlock_init(&state->seqlock);
|
2005-04-17 00:20:36 +02:00
|
|
|
return state;
|
|
|
|
}
|
|
|
|
|
2005-11-04 21:32:58 +01:00
|
|
|
void
|
|
|
|
nfs4_state_set_mode_locked(struct nfs4_state *state, mode_t mode)
|
|
|
|
{
|
|
|
|
if (state->state == mode)
|
|
|
|
return;
|
|
|
|
/* NB! List reordering - see the reclaim code for why. */
|
|
|
|
if ((mode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
|
|
|
|
if (mode & FMODE_WRITE)
|
|
|
|
list_move(&state->open_states, &state->owner->so_states);
|
|
|
|
else
|
|
|
|
list_move_tail(&state->open_states, &state->owner->so_states);
|
|
|
|
}
|
|
|
|
state->state = mode;
|
|
|
|
}
|
|
|
|
|
2005-04-17 00:20:36 +02:00
|
|
|
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) {
|
2007-07-03 20:41:19 +02:00
|
|
|
if (state->owner != owner)
|
2005-04-17 00:20:36 +02:00
|
|
|
continue;
|
2007-07-03 20:41:19 +02:00
|
|
|
if (atomic_inc_not_zero(&state->count))
|
2005-04-17 00:20:36 +02:00
|
|
|
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();
|
2005-10-20 23:22:47 +02:00
|
|
|
spin_lock(&owner->so_lock);
|
2005-04-17 00:20:36 +02:00
|
|
|
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);
|
|
|
|
state->inode = igrab(inode);
|
|
|
|
spin_unlock(&inode->i_lock);
|
2005-10-20 23:22:47 +02:00
|
|
|
/* 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 00:20:36 +02:00
|
|
|
} else {
|
|
|
|
spin_unlock(&inode->i_lock);
|
2005-10-20 23:22:47 +02:00
|
|
|
spin_unlock(&owner->so_lock);
|
2005-04-17 00:20:36 +02:00
|
|
|
if (new)
|
|
|
|
nfs4_free_open_state(new);
|
|
|
|
}
|
|
|
|
out:
|
|
|
|
return state;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Beware! Caller must be holding exactly one
|
2005-10-18 23:20:13 +02:00
|
|
|
* reference to clp->cl_sem!
|
2005-04-17 00:20:36 +02:00
|
|
|
*/
|
|
|
|
void nfs4_put_open_state(struct nfs4_state *state)
|
|
|
|
{
|
|
|
|
struct inode *inode = state->inode;
|
|
|
|
struct nfs4_state_owner *owner = state->owner;
|
|
|
|
|
2005-10-20 23:22:47 +02:00
|
|
|
if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
|
2005-04-17 00:20:36 +02:00
|
|
|
return;
|
2005-10-20 23:22:47 +02:00
|
|
|
spin_lock(&inode->i_lock);
|
2007-07-27 16:23:05 +02:00
|
|
|
list_del(&state->inode_states);
|
2005-04-17 00:20:36 +02:00
|
|
|
list_del(&state->open_states);
|
2005-10-20 23:22:47 +02:00
|
|
|
spin_unlock(&inode->i_lock);
|
|
|
|
spin_unlock(&owner->so_lock);
|
2005-04-17 00:20:36 +02:00
|
|
|
iput(inode);
|
|
|
|
nfs4_free_open_state(state);
|
|
|
|
nfs4_put_state_owner(owner);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
2005-10-18 23:20:13 +02:00
|
|
|
* Close the current file.
|
2005-04-17 00:20:36 +02:00
|
|
|
*/
|
2007-10-19 00:03:27 +02:00
|
|
|
static void __nfs4_close(struct path *path, struct nfs4_state *state, mode_t mode, int wait)
|
2005-04-17 00:20:36 +02:00
|
|
|
{
|
|
|
|
struct nfs4_state_owner *owner = state->owner;
|
2007-07-06 00:07:55 +02:00
|
|
|
int call_close = 0;
|
|
|
|
int newstate;
|
2005-04-17 00:20:36 +02:00
|
|
|
|
|
|
|
atomic_inc(&owner->so_count);
|
|
|
|
/* Protect against nfs4_find_state() */
|
2005-10-20 23:22:47 +02:00
|
|
|
spin_lock(&owner->so_lock);
|
2006-01-03 09:55:13 +01:00
|
|
|
switch (mode & (FMODE_READ | FMODE_WRITE)) {
|
|
|
|
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 00:07:55 +02:00
|
|
|
newstate = FMODE_READ|FMODE_WRITE;
|
2006-01-03 09:55:13 +01:00
|
|
|
if (state->n_rdwr == 0) {
|
2007-07-06 00:07:55 +02:00
|
|
|
if (state->n_rdonly == 0) {
|
2006-01-03 09:55:13 +01:00
|
|
|
newstate &= ~FMODE_READ;
|
2007-07-06 00:07:55 +02:00
|
|
|
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 09:55:13 +01:00
|
|
|
newstate &= ~FMODE_WRITE;
|
2007-07-06 00:07:55 +02:00
|
|
|
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 09:55:13 +01:00
|
|
|
}
|
2007-07-06 00:07:55 +02:00
|
|
|
nfs4_state_set_mode_locked(state, newstate);
|
2005-10-20 23:22:47 +02:00
|
|
|
spin_unlock(&owner->so_lock);
|
2005-11-04 21:32:58 +01:00
|
|
|
|
2007-07-06 00:07:55 +02:00
|
|
|
if (!call_close) {
|
2007-06-12 05:05:07 +02:00
|
|
|
nfs4_put_open_state(state);
|
|
|
|
nfs4_put_state_owner(owner);
|
|
|
|
} else
|
2007-10-19 00:03:27 +02:00
|
|
|
nfs4_do_close(path, state, wait);
|
|
|
|
}
|
|
|
|
|
|
|
|
void nfs4_close_state(struct path *path, struct nfs4_state *state, mode_t mode)
|
|
|
|
{
|
|
|
|
__nfs4_close(path, state, mode, 0);
|
|
|
|
}
|
|
|
|
|
|
|
|
void nfs4_close_sync(struct path *path, struct nfs4_state *state, mode_t mode)
|
|
|
|
{
|
|
|
|
__nfs4_close(path, state, mode, 1);
|
2005-04-17 00:20:36 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Search the state->lock_states for an existing lock_owner
|
|
|
|
* that is compatible with current->files
|
|
|
|
*/
|
|
|
|
static struct nfs4_lock_state *
|
|
|
|
__nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
|
|
|
|
{
|
|
|
|
struct nfs4_lock_state *pos;
|
|
|
|
list_for_each_entry(pos, &state->lock_states, ls_locks) {
|
|
|
|
if (pos->ls_owner != fl_owner)
|
|
|
|
continue;
|
|
|
|
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.
|
|
|
|
*
|
|
|
|
*/
|
|
|
|
static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
|
|
|
|
{
|
|
|
|
struct nfs4_lock_state *lsp;
|
2006-08-23 02:06:08 +02:00
|
|
|
struct nfs_client *clp = state->owner->so_client;
|
2005-04-17 00:20:36 +02:00
|
|
|
|
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>
2005-10-18 23:20:12 +02:00
|
|
|
lsp = kzalloc(sizeof(*lsp), GFP_KERNEL);
|
2005-04-17 00:20:36 +02:00
|
|
|
if (lsp == NULL)
|
|
|
|
return NULL;
|
2008-01-10 22:07:54 +01:00
|
|
|
rpc_init_wait_queue(&lsp->ls_sequence.wait, "lock_seqid_waitqueue");
|
|
|
|
spin_lock_init(&lsp->ls_sequence.lock);
|
|
|
|
INIT_LIST_HEAD(&lsp->ls_sequence.list);
|
|
|
|
lsp->ls_seqid.sequence = &lsp->ls_sequence;
|
2005-04-17 00:20:36 +02:00
|
|
|
atomic_set(&lsp->ls_count, 1);
|
|
|
|
lsp->ls_owner = fl_owner;
|
|
|
|
spin_lock(&clp->cl_lock);
|
2007-07-02 19:58:33 +02:00
|
|
|
nfs_alloc_unique_id(&clp->cl_lockowner_id, &lsp->ls_id, 1, 64);
|
2005-04-17 00:20:36 +02:00
|
|
|
spin_unlock(&clp->cl_lock);
|
2005-06-22 19:16:32 +02:00
|
|
|
INIT_LIST_HEAD(&lsp->ls_locks);
|
2005-04-17 00:20:36 +02:00
|
|
|
return lsp;
|
|
|
|
}
|
|
|
|
|
2007-07-02 19:58:33 +02:00
|
|
|
static void nfs4_free_lock_state(struct nfs4_lock_state *lsp)
|
|
|
|
{
|
|
|
|
struct nfs_client *clp = lsp->ls_state->owner->so_client;
|
|
|
|
|
|
|
|
spin_lock(&clp->cl_lock);
|
|
|
|
nfs_free_unique_id(&clp->cl_lockowner_id, &lsp->ls_id);
|
|
|
|
spin_unlock(&clp->cl_lock);
|
|
|
|
kfree(lsp);
|
|
|
|
}
|
|
|
|
|
2005-04-17 00:20:36 +02:00
|
|
|
/*
|
|
|
|
* Return a compatible lock_state. If no initialized lock_state structure
|
|
|
|
* exists, return an uninitialized one.
|
|
|
|
*
|
2005-10-18 23:20:13 +02:00
|
|
|
* The caller must be holding clp->cl_sem
|
2005-04-17 00:20:36 +02:00
|
|
|
*/
|
2005-06-22 19:16:32 +02:00
|
|
|
static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
|
2005-04-17 00:20:36 +02:00
|
|
|
{
|
2005-06-22 19:16:32 +02:00
|
|
|
struct nfs4_lock_state *lsp, *new = NULL;
|
2005-04-17 00:20:36 +02:00
|
|
|
|
2005-06-22 19:16:32 +02:00
|
|
|
for(;;) {
|
|
|
|
spin_lock(&state->state_lock);
|
|
|
|
lsp = __nfs4_find_lock_state(state, owner);
|
|
|
|
if (lsp != NULL)
|
|
|
|
break;
|
|
|
|
if (new != NULL) {
|
|
|
|
new->ls_state = state;
|
|
|
|
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);
|
|
|
|
new = nfs4_alloc_lock_state(state, owner);
|
|
|
|
if (new == NULL)
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
spin_unlock(&state->state_lock);
|
2007-07-02 19:58:33 +02:00
|
|
|
if (new != NULL)
|
|
|
|
nfs4_free_lock_state(new);
|
2005-04-17 00:20:36 +02:00
|
|
|
return lsp;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
2005-06-22 19:16:32 +02:00
|
|
|
* Release reference to lock_state, and free it if we see that
|
|
|
|
* it is no longer in use
|
2005-04-17 00:20:36 +02:00
|
|
|
*/
|
2005-10-18 23:20:15 +02:00
|
|
|
void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
|
2005-04-17 00:20:36 +02:00
|
|
|
{
|
2005-06-22 19:16:32 +02:00
|
|
|
struct nfs4_state *state;
|
2005-04-17 00:20:36 +02:00
|
|
|
|
2005-06-22 19:16:32 +02:00
|
|
|
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);
|
2007-07-02 19:58:33 +02:00
|
|
|
nfs4_free_lock_state(lsp);
|
2005-04-17 00:20:36 +02:00
|
|
|
}
|
|
|
|
|
2005-06-22 19:16:32 +02:00
|
|
|
static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
|
2005-04-17 00:20:36 +02:00
|
|
|
{
|
2005-06-22 19:16:32 +02:00
|
|
|
struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
|
2005-04-17 00:20:36 +02:00
|
|
|
|
2005-06-22 19:16:32 +02:00
|
|
|
dst->fl_u.nfs4_fl.owner = lsp;
|
|
|
|
atomic_inc(&lsp->ls_count);
|
|
|
|
}
|
2005-04-17 00:20:36 +02:00
|
|
|
|
2005-06-22 19:16:32 +02:00
|
|
|
static void nfs4_fl_release_lock(struct file_lock *fl)
|
2005-04-17 00:20:36 +02:00
|
|
|
{
|
2005-06-22 19:16:32 +02:00
|
|
|
nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
|
2005-04-17 00:20:36 +02:00
|
|
|
}
|
|
|
|
|
2005-06-22 19:16:32 +02:00
|
|
|
static struct file_lock_operations nfs4_fl_lock_ops = {
|
|
|
|
.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 00:20:36 +02:00
|
|
|
{
|
2005-06-22 19:16:32 +02:00
|
|
|
struct nfs4_lock_state *lsp;
|
|
|
|
|
|
|
|
if (fl->fl_ops != NULL)
|
|
|
|
return 0;
|
|
|
|
lsp = nfs4_get_lock_state(state, fl->fl_owner);
|
|
|
|
if (lsp == NULL)
|
|
|
|
return -ENOMEM;
|
|
|
|
fl->fl_u.nfs4_fl.owner = lsp;
|
|
|
|
fl->fl_ops = &nfs4_fl_lock_ops;
|
|
|
|
return 0;
|
2005-04-17 00:20:36 +02:00
|
|
|
}
|
|
|
|
|
2005-06-22 19:16:32 +02:00
|
|
|
/*
|
|
|
|
* Byte-range lock aware utility to initialize the stateid of read/write
|
|
|
|
* requests.
|
2005-04-17 00:20:36 +02:00
|
|
|
*/
|
2005-06-22 19:16:32 +02:00
|
|
|
void nfs4_copy_stateid(nfs4_stateid *dst, struct nfs4_state *state, fl_owner_t fl_owner)
|
2005-04-17 00:20:36 +02:00
|
|
|
{
|
2005-06-22 19:16:32 +02:00
|
|
|
struct nfs4_lock_state *lsp;
|
2007-07-09 16:45:42 +02:00
|
|
|
int seq;
|
2005-04-17 00:20:36 +02:00
|
|
|
|
2007-07-09 16:45:42 +02:00
|
|
|
do {
|
|
|
|
seq = read_seqbegin(&state->seqlock);
|
|
|
|
memcpy(dst, &state->stateid, sizeof(*dst));
|
|
|
|
} while (read_seqretry(&state->seqlock, seq));
|
2005-06-22 19:16:32 +02:00
|
|
|
if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
|
|
|
|
return;
|
2005-04-17 00:20:36 +02:00
|
|
|
|
2005-06-22 19:16:32 +02:00
|
|
|
spin_lock(&state->state_lock);
|
|
|
|
lsp = __nfs4_find_lock_state(state, fl_owner);
|
|
|
|
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 00:20:36 +02:00
|
|
|
nfs4_put_lock_state(lsp);
|
|
|
|
}
|
|
|
|
|
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>
2005-10-18 23:20:12 +02:00
|
|
|
struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter)
|
|
|
|
{
|
|
|
|
struct nfs_seqid *new;
|
|
|
|
|
|
|
|
new = kmalloc(sizeof(*new), GFP_KERNEL);
|
|
|
|
if (new != NULL) {
|
|
|
|
new->sequence = counter;
|
2008-01-08 23:56:07 +01:00
|
|
|
INIT_LIST_HEAD(&new->list);
|
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>
2005-10-18 23:20:12 +02:00
|
|
|
}
|
|
|
|
return new;
|
|
|
|
}
|
|
|
|
|
|
|
|
void nfs_free_seqid(struct nfs_seqid *seqid)
|
2005-04-17 00:20:36 +02:00
|
|
|
{
|
2008-01-08 23:56:07 +01:00
|
|
|
if (!list_empty(&seqid->list)) {
|
|
|
|
struct rpc_sequence *sequence = seqid->sequence->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>
2005-10-18 23:20:12 +02:00
|
|
|
|
2008-01-08 23:56:07 +01:00
|
|
|
spin_lock(&sequence->lock);
|
|
|
|
list_del(&seqid->list);
|
|
|
|
spin_unlock(&sequence->lock);
|
|
|
|
rpc_wake_up(&sequence->wait);
|
|
|
|
}
|
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>
2005-10-18 23:20:12 +02:00
|
|
|
kfree(seqid);
|
2005-04-17 00:20:36 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
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>
2005-10-18 23:20:12 +02:00
|
|
|
* 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()
|
|
|
|
*/
|
2007-07-02 20:03:03 +02:00
|
|
|
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>
2005-10-18 23:20:12 +02:00
|
|
|
{
|
2008-01-08 23:56:07 +01:00
|
|
|
BUG_ON(list_first_entry(&seqid->sequence->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>
2005-10-18 23:20:12 +02:00
|
|
|
switch (status) {
|
|
|
|
case 0:
|
|
|
|
break;
|
|
|
|
case -NFS4ERR_BAD_SEQID:
|
2007-07-08 22:49:11 +02:00
|
|
|
if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
|
|
|
|
return;
|
|
|
|
printk(KERN_WARNING "NFS: v4 server returned a bad"
|
2008-02-13 22:09:35 +01:00
|
|
|
" sequence-id error on an"
|
|
|
|
" unconfirmed sequence %p!\n",
|
2007-07-08 22:49:11 +02:00
|
|
|
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>
2005-10-18 23:20:12 +02:00
|
|
|
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)
|
|
|
|
{
|
|
|
|
if (status == -NFS4ERR_BAD_SEQID) {
|
|
|
|
struct nfs4_state_owner *sp = container_of(seqid->sequence,
|
|
|
|
struct nfs4_state_owner, so_seqid);
|
2005-04-17 00:20:36 +02:00
|
|
|
nfs4_drop_state_owner(sp);
|
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>
2005-10-18 23:20:12 +02:00
|
|
|
}
|
2007-07-02 20:03:03 +02:00
|
|
|
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>
2005-10-18 23:20:12 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* 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)
|
|
|
|
{
|
2007-07-02 20:03:03 +02:00
|
|
|
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>
2005-10-18 23:20:12 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
|
|
|
|
{
|
|
|
|
struct rpc_sequence *sequence = seqid->sequence->sequence;
|
|
|
|
int status = 0;
|
|
|
|
|
|
|
|
spin_lock(&sequence->lock);
|
2008-01-08 23:56:07 +01:00
|
|
|
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;
|
|
|
|
rpc_sleep_on(&sequence->wait, task, NULL, NULL);
|
|
|
|
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>
2005-10-18 23:20:12 +02:00
|
|
|
spin_unlock(&sequence->lock);
|
|
|
|
return status;
|
2005-04-17 00:20:36 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
static int reclaimer(void *);
|
|
|
|
|
2006-08-23 02:06:08 +02:00
|
|
|
static inline void nfs4_clear_recover_bit(struct nfs_client *clp)
|
2006-01-03 09:55:22 +01:00
|
|
|
{
|
|
|
|
smp_mb__before_clear_bit();
|
|
|
|
clear_bit(NFS4CLNT_STATE_RECOVER, &clp->cl_state);
|
|
|
|
smp_mb__after_clear_bit();
|
|
|
|
wake_up_bit(&clp->cl_state, NFS4CLNT_STATE_RECOVER);
|
|
|
|
rpc_wake_up(&clp->cl_rpcwaitq);
|
|
|
|
}
|
|
|
|
|
2005-04-17 00:20:36 +02:00
|
|
|
/*
|
|
|
|
* State recovery routine
|
|
|
|
*/
|
2006-08-23 02:06:08 +02:00
|
|
|
static void nfs4_recover_state(struct nfs_client *clp)
|
2005-04-17 00:20:36 +02:00
|
|
|
{
|
2006-01-03 09:55:23 +01:00
|
|
|
struct task_struct *task;
|
2005-04-17 00:20:36 +02:00
|
|
|
|
2006-01-03 09:55:23 +01:00
|
|
|
__module_get(THIS_MODULE);
|
|
|
|
atomic_inc(&clp->cl_count);
|
2007-12-10 20:57:16 +01:00
|
|
|
task = kthread_run(reclaimer, clp, "%s-reclaim",
|
|
|
|
rpc_peeraddr2str(clp->cl_rpcclient,
|
|
|
|
RPC_DISPLAY_ADDR));
|
2006-01-03 09:55:23 +01:00
|
|
|
if (!IS_ERR(task))
|
|
|
|
return;
|
2006-01-03 09:55:22 +01:00
|
|
|
nfs4_clear_recover_bit(clp);
|
2006-08-23 02:06:10 +02:00
|
|
|
nfs_put_client(clp);
|
2006-01-03 09:55:23 +01:00
|
|
|
module_put(THIS_MODULE);
|
2005-04-17 00:20:36 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Schedule a state recovery attempt
|
|
|
|
*/
|
2006-08-23 02:06:08 +02:00
|
|
|
void nfs4_schedule_state_recovery(struct nfs_client *clp)
|
2005-04-17 00:20:36 +02:00
|
|
|
{
|
|
|
|
if (!clp)
|
|
|
|
return;
|
2006-01-03 09:55:22 +01:00
|
|
|
if (test_and_set_bit(NFS4CLNT_STATE_RECOVER, &clp->cl_state) == 0)
|
2006-01-03 09:55:23 +01:00
|
|
|
nfs4_recover_state(clp);
|
2005-04-17 00:20:36 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
static int nfs4_reclaim_locks(struct nfs4_state_recovery_ops *ops, struct nfs4_state *state)
|
|
|
|
{
|
|
|
|
struct inode *inode = state->inode;
|
|
|
|
struct file_lock *fl;
|
|
|
|
int status = 0;
|
|
|
|
|
|
|
|
for (fl = inode->i_flock; fl != 0; fl = fl->fl_next) {
|
2005-11-04 21:35:30 +01:00
|
|
|
if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
|
2005-04-17 00:20:36 +02:00
|
|
|
continue;
|
2007-08-10 23:44:32 +02:00
|
|
|
if (nfs_file_open_context(fl->fl_file)->state != state)
|
2005-04-17 00:20:36 +02:00
|
|
|
continue;
|
|
|
|
status = ops->recover_lock(state, fl);
|
|
|
|
if (status >= 0)
|
|
|
|
continue;
|
|
|
|
switch (status) {
|
|
|
|
default:
|
|
|
|
printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
|
|
|
|
__FUNCTION__, status);
|
|
|
|
case -NFS4ERR_EXPIRED:
|
|
|
|
case -NFS4ERR_NO_GRACE:
|
|
|
|
case -NFS4ERR_RECLAIM_BAD:
|
|
|
|
case -NFS4ERR_RECLAIM_CONFLICT:
|
2005-11-04 21:35:30 +01:00
|
|
|
/* kill_proc(fl->fl_pid, SIGLOST, 1); */
|
2005-04-17 00:20:36 +02:00
|
|
|
break;
|
|
|
|
case -NFS4ERR_STALE_CLIENTID:
|
|
|
|
goto out_err;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return 0;
|
|
|
|
out_err:
|
|
|
|
return status;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int nfs4_reclaim_open_state(struct nfs4_state_recovery_ops *ops, struct nfs4_state_owner *sp)
|
|
|
|
{
|
|
|
|
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.
|
|
|
|
*/
|
|
|
|
list_for_each_entry(state, &sp->so_states, open_states) {
|
|
|
|
if (state->state == 0)
|
|
|
|
continue;
|
|
|
|
status = ops->recover_open(sp, state);
|
|
|
|
if (status >= 0) {
|
|
|
|
status = nfs4_reclaim_locks(ops, state);
|
|
|
|
if (status < 0)
|
|
|
|
goto out_err;
|
|
|
|
list_for_each_entry(lock, &state->lock_states, ls_locks) {
|
|
|
|
if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
|
|
|
|
printk("%s: Lock reclaim failed!\n",
|
|
|
|
__FUNCTION__);
|
|
|
|
}
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
switch (status) {
|
|
|
|
default:
|
|
|
|
printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
|
|
|
|
__FUNCTION__, status);
|
|
|
|
case -ENOENT:
|
|
|
|
case -NFS4ERR_RECLAIM_BAD:
|
|
|
|
case -NFS4ERR_RECLAIM_CONFLICT:
|
|
|
|
/*
|
|
|
|
* 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;
|
|
|
|
case -NFS4ERR_EXPIRED:
|
|
|
|
case -NFS4ERR_NO_GRACE:
|
|
|
|
case -NFS4ERR_STALE_CLIENTID:
|
|
|
|
goto out_err;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return 0;
|
|
|
|
out_err:
|
|
|
|
return status;
|
|
|
|
}
|
|
|
|
|
2006-08-23 02:06:08 +02:00
|
|
|
static void nfs4_state_mark_reclaim(struct nfs_client *clp)
|
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>
2005-10-18 23:20:12 +02:00
|
|
|
{
|
|
|
|
struct nfs4_state_owner *sp;
|
2007-07-02 19:58:33 +02:00
|
|
|
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>
2005-10-18 23:20:12 +02:00
|
|
|
struct nfs4_state *state;
|
|
|
|
struct nfs4_lock_state *lock;
|
|
|
|
|
|
|
|
/* Reset all sequence ids to zero */
|
2007-07-02 19:58:33 +02:00
|
|
|
for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
|
|
|
|
sp = rb_entry(pos, struct nfs4_state_owner, so_client_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>
2005-10-18 23:20:12 +02:00
|
|
|
sp->so_seqid.counter = 0;
|
|
|
|
sp->so_seqid.flags = 0;
|
2005-10-20 23:22:47 +02:00
|
|
|
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>
2005-10-18 23:20:12 +02:00
|
|
|
list_for_each_entry(state, &sp->so_states, open_states) {
|
2007-07-06 00:07:55 +02:00
|
|
|
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);
|
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>
2005-10-18 23:20:12 +02:00
|
|
|
list_for_each_entry(lock, &state->lock_states, ls_locks) {
|
|
|
|
lock->ls_seqid.counter = 0;
|
|
|
|
lock->ls_seqid.flags = 0;
|
|
|
|
lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
|
|
|
|
}
|
|
|
|
}
|
2005-10-20 23:22:47 +02:00
|
|
|
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>
2005-10-18 23:20:12 +02:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2005-04-17 00:20:36 +02:00
|
|
|
static int reclaimer(void *ptr)
|
|
|
|
{
|
2006-08-23 02:06:08 +02:00
|
|
|
struct nfs_client *clp = ptr;
|
2005-04-17 00:20:36 +02:00
|
|
|
struct nfs4_state_owner *sp;
|
2007-07-02 19:58:33 +02:00
|
|
|
struct rb_node *pos;
|
2005-04-17 00:20:36 +02:00
|
|
|
struct nfs4_state_recovery_ops *ops;
|
2006-01-03 09:55:26 +01:00
|
|
|
struct rpc_cred *cred;
|
2005-04-17 00:20:36 +02:00
|
|
|
int status = 0;
|
|
|
|
|
|
|
|
allow_signal(SIGKILL);
|
|
|
|
|
|
|
|
/* Ensure exclusive access to NFSv4 state */
|
|
|
|
lock_kernel();
|
|
|
|
down_write(&clp->cl_sem);
|
|
|
|
/* Are there any NFS mounts out there? */
|
|
|
|
if (list_empty(&clp->cl_superblocks))
|
|
|
|
goto out;
|
|
|
|
restart_loop:
|
2006-01-03 09:55:26 +01:00
|
|
|
ops = &nfs4_network_partition_recovery_ops;
|
|
|
|
/* Are there any open files on this volume? */
|
|
|
|
cred = nfs4_get_renew_cred(clp);
|
|
|
|
if (cred != NULL) {
|
|
|
|
/* Yes there are: try to renew the old lease */
|
|
|
|
status = nfs4_proc_renew(clp, cred);
|
|
|
|
switch (status) {
|
|
|
|
case 0:
|
|
|
|
case -NFS4ERR_CB_PATH_DOWN:
|
|
|
|
put_rpccred(cred);
|
|
|
|
goto out;
|
|
|
|
case -NFS4ERR_STALE_CLIENTID:
|
|
|
|
case -NFS4ERR_LEASE_MOVED:
|
|
|
|
ops = &nfs4_reboot_recovery_ops;
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
/* "reboot" to ensure we clear all state on the server */
|
|
|
|
clp->cl_boot_time = CURRENT_TIME;
|
|
|
|
cred = nfs4_get_setclientid_cred(clp);
|
|
|
|
}
|
|
|
|
/* We're going to have to re-establish a clientid */
|
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>
2005-10-18 23:20:12 +02:00
|
|
|
nfs4_state_mark_reclaim(clp);
|
2006-01-03 09:55:26 +01:00
|
|
|
status = -ENOENT;
|
|
|
|
if (cred != NULL) {
|
|
|
|
status = nfs4_init_client(clp, cred);
|
|
|
|
put_rpccred(cred);
|
|
|
|
}
|
2005-04-17 00:20:36 +02:00
|
|
|
if (status)
|
|
|
|
goto out_error;
|
|
|
|
/* Mark all delegations for reclaim */
|
|
|
|
nfs_delegation_mark_reclaim(clp);
|
|
|
|
/* Note: list is protected by exclusive lock on cl->cl_sem */
|
2007-07-02 19:58:33 +02:00
|
|
|
for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
|
|
|
|
sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
|
2005-04-17 00:20:36 +02:00
|
|
|
status = nfs4_reclaim_open_state(ops, sp);
|
|
|
|
if (status < 0) {
|
|
|
|
if (status == -NFS4ERR_NO_GRACE) {
|
|
|
|
ops = &nfs4_network_partition_recovery_ops;
|
|
|
|
status = nfs4_reclaim_open_state(ops, sp);
|
|
|
|
}
|
|
|
|
if (status == -NFS4ERR_STALE_CLIENTID)
|
|
|
|
goto restart_loop;
|
|
|
|
if (status == -NFS4ERR_EXPIRED)
|
|
|
|
goto restart_loop;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
nfs_delegation_reap_unclaimed(clp);
|
|
|
|
out:
|
|
|
|
up_write(&clp->cl_sem);
|
|
|
|
unlock_kernel();
|
|
|
|
if (status == -NFS4ERR_CB_PATH_DOWN)
|
|
|
|
nfs_handle_cb_pathdown(clp);
|
2006-01-03 09:55:22 +01:00
|
|
|
nfs4_clear_recover_bit(clp);
|
2006-08-23 02:06:10 +02:00
|
|
|
nfs_put_client(clp);
|
2006-01-03 09:55:23 +01:00
|
|
|
module_put_and_exit(0);
|
2005-04-17 00:20:36 +02:00
|
|
|
return 0;
|
|
|
|
out_error:
|
2007-12-10 20:57:16 +01:00
|
|
|
printk(KERN_WARNING "Error: state recovery failed on NFSv4 server %s"
|
|
|
|
" with error %d\n", clp->cl_hostname, -status);
|
2006-03-20 19:44:47 +01:00
|
|
|
set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
|
2005-04-17 00:20:36 +02:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Local variables:
|
|
|
|
* c-basic-offset: 8
|
|
|
|
* End:
|
|
|
|
*/
|