linux/fs/lockd/clntlock.c
Jeff Layton 2005f5b9c3 lockd: fix races in client GRANTED_MSG wait logic
After the wait for a grant is done (for whatever reason), nlmclnt_block
updates the status of the nlm_rqst with the status of the block. At the
point it does this, however, the block is still queued its status could
change at any time.

This is particularly a problem when the waiting task is signaled during
the wait. We can end up giving up on the lock just before the GRANTED_MSG
callback comes in, and accept it even though the lock request gets back
an error, leaving a dangling lock on the server.

Since the nlm_wait never lives beyond the end of nlmclnt_lock, put it on
the stack and add functions to allow us to enqueue and dequeue the
block. Enqueue it just before the lock/wait loop, and dequeue it
just after we exit the loop instead of waiting until the end of
the function. Also, scrape the status at the time that we dequeue it to
ensure that it's final.

Reported-by: Yongcheng Yang <yoyang@redhat.com>
Link: https://bugzilla.redhat.com/show_bug.cgi?id=2063818
Signed-off-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
2023-04-26 09:05:00 -04:00

288 lines
7.1 KiB
C

// SPDX-License-Identifier: GPL-2.0-only
/*
* linux/fs/lockd/clntlock.c
*
* Lock handling for the client side NLM implementation
*
* Copyright (C) 1996, Olaf Kirch <okir@monad.swb.de>
*/
#include <linux/module.h>
#include <linux/types.h>
#include <linux/slab.h>
#include <linux/time.h>
#include <linux/nfs_fs.h>
#include <linux/sunrpc/addr.h>
#include <linux/sunrpc/svc.h>
#include <linux/lockd/lockd.h>
#include <linux/kthread.h>
#define NLMDBG_FACILITY NLMDBG_CLIENT
/*
* Local function prototypes
*/
static int reclaimer(void *ptr);
/*
* The following functions handle blocking and granting from the
* client perspective.
*/
static LIST_HEAD(nlm_blocked);
static DEFINE_SPINLOCK(nlm_blocked_lock);
/**
* nlmclnt_init - Set up per-NFS mount point lockd data structures
* @nlm_init: pointer to arguments structure
*
* Returns pointer to an appropriate nlm_host struct,
* or an ERR_PTR value.
*/
struct nlm_host *nlmclnt_init(const struct nlmclnt_initdata *nlm_init)
{
struct nlm_host *host;
u32 nlm_version = (nlm_init->nfs_version == 2) ? 1 : 4;
int status;
status = lockd_up(nlm_init->net, nlm_init->cred);
if (status < 0)
return ERR_PTR(status);
host = nlmclnt_lookup_host(nlm_init->address, nlm_init->addrlen,
nlm_init->protocol, nlm_version,
nlm_init->hostname, nlm_init->noresvport,
nlm_init->net, nlm_init->cred);
if (host == NULL)
goto out_nohost;
if (host->h_rpcclnt == NULL && nlm_bind_host(host) == NULL)
goto out_nobind;
host->h_nlmclnt_ops = nlm_init->nlmclnt_ops;
return host;
out_nobind:
nlmclnt_release_host(host);
out_nohost:
lockd_down(nlm_init->net);
return ERR_PTR(-ENOLCK);
}
EXPORT_SYMBOL_GPL(nlmclnt_init);
/**
* nlmclnt_done - Release resources allocated by nlmclnt_init()
* @host: nlm_host structure reserved by nlmclnt_init()
*
*/
void nlmclnt_done(struct nlm_host *host)
{
struct net *net = host->net;
nlmclnt_release_host(host);
lockd_down(net);
}
EXPORT_SYMBOL_GPL(nlmclnt_done);
void nlmclnt_prepare_block(struct nlm_wait *block, struct nlm_host *host, struct file_lock *fl)
{
block->b_host = host;
block->b_lock = fl;
init_waitqueue_head(&block->b_wait);
block->b_status = nlm_lck_blocked;
}
/*
* Queue up a lock for blocking so that the GRANTED request can see it
*/
void nlmclnt_queue_block(struct nlm_wait *block)
{
spin_lock(&nlm_blocked_lock);
list_add(&block->b_list, &nlm_blocked);
spin_unlock(&nlm_blocked_lock);
}
/*
* Dequeue the block and return its final status
*/
__be32 nlmclnt_dequeue_block(struct nlm_wait *block)
{
__be32 status;
spin_lock(&nlm_blocked_lock);
list_del(&block->b_list);
status = block->b_status;
spin_unlock(&nlm_blocked_lock);
return status;
}
/*
* Block on a lock
*/
int nlmclnt_wait(struct nlm_wait *block, struct nlm_rqst *req, long timeout)
{
long ret;
/* A borken server might ask us to block even if we didn't
* request it. Just say no!
*/
if (block == NULL)
return -EAGAIN;
/* Go to sleep waiting for GRANT callback. Some servers seem
* to lose callbacks, however, so we're going to poll from
* time to time just to make sure.
*
* For now, the retry frequency is pretty high; normally
* a 1 minute timeout would do. See the comment before
* nlmclnt_lock for an explanation.
*/
ret = wait_event_interruptible_timeout(block->b_wait,
block->b_status != nlm_lck_blocked,
timeout);
if (ret < 0)
return -ERESTARTSYS;
/* Reset the lock status after a server reboot so we resend */
if (block->b_status == nlm_lck_denied_grace_period)
block->b_status = nlm_lck_blocked;
return 0;
}
/*
* The server lockd has called us back to tell us the lock was granted
*/
__be32 nlmclnt_grant(const struct sockaddr *addr, const struct nlm_lock *lock)
{
const struct file_lock *fl = &lock->fl;
const struct nfs_fh *fh = &lock->fh;
struct nlm_wait *block;
__be32 res = nlm_lck_denied;
/*
* Look up blocked request based on arguments.
* Warning: must not use cookie to match it!
*/
spin_lock(&nlm_blocked_lock);
list_for_each_entry(block, &nlm_blocked, b_list) {
struct file_lock *fl_blocked = block->b_lock;
if (fl_blocked->fl_start != fl->fl_start)
continue;
if (fl_blocked->fl_end != fl->fl_end)
continue;
/*
* Careful! The NLM server will return the 32-bit "pid" that
* we put on the wire: in this case the lockowner "pid".
*/
if (fl_blocked->fl_u.nfs_fl.owner->pid != lock->svid)
continue;
if (!rpc_cmp_addr(nlm_addr(block->b_host), addr))
continue;
if (nfs_compare_fh(NFS_FH(file_inode(fl_blocked->fl_file)), fh) != 0)
continue;
/* Alright, we found a lock. Set the return status
* and wake up the caller
*/
block->b_status = nlm_granted;
wake_up(&block->b_wait);
res = nlm_granted;
}
spin_unlock(&nlm_blocked_lock);
return res;
}
/*
* The following procedures deal with the recovery of locks after a
* server crash.
*/
/*
* Reclaim all locks on server host. We do this by spawning a separate
* reclaimer thread.
*/
void
nlmclnt_recovery(struct nlm_host *host)
{
struct task_struct *task;
if (!host->h_reclaiming++) {
nlm_get_host(host);
task = kthread_run(reclaimer, host, "%s-reclaim", host->h_name);
if (IS_ERR(task))
printk(KERN_ERR "lockd: unable to spawn reclaimer "
"thread. Locks for %s won't be reclaimed! "
"(%ld)\n", host->h_name, PTR_ERR(task));
}
}
static int
reclaimer(void *ptr)
{
struct nlm_host *host = (struct nlm_host *) ptr;
struct nlm_wait *block;
struct nlm_rqst *req;
struct file_lock *fl, *next;
u32 nsmstate;
struct net *net = host->net;
req = kmalloc(sizeof(*req), GFP_KERNEL);
if (!req)
return 0;
allow_signal(SIGKILL);
down_write(&host->h_rwsem);
lockd_up(net, NULL); /* note: this cannot fail as lockd is already running */
dprintk("lockd: reclaiming locks for host %s\n", host->h_name);
restart:
nsmstate = host->h_nsmstate;
/* Force a portmap getport - the peer's lockd will
* most likely end up on a different port.
*/
host->h_nextrebind = jiffies;
nlm_rebind_host(host);
/* First, reclaim all locks that have been granted. */
list_splice_init(&host->h_granted, &host->h_reclaim);
list_for_each_entry_safe(fl, next, &host->h_reclaim, fl_u.nfs_fl.list) {
list_del_init(&fl->fl_u.nfs_fl.list);
/*
* sending this thread a SIGKILL will result in any unreclaimed
* locks being removed from the h_granted list. This means that
* the kernel will not attempt to reclaim them again if a new
* reclaimer thread is spawned for this host.
*/
if (signalled())
continue;
if (nlmclnt_reclaim(host, fl, req) != 0)
continue;
list_add_tail(&fl->fl_u.nfs_fl.list, &host->h_granted);
if (host->h_nsmstate != nsmstate) {
/* Argh! The server rebooted again! */
goto restart;
}
}
host->h_reclaiming = 0;
up_write(&host->h_rwsem);
dprintk("NLM: done reclaiming locks for host %s\n", host->h_name);
/* Now, wake up all processes that sleep on a blocked lock */
spin_lock(&nlm_blocked_lock);
list_for_each_entry(block, &nlm_blocked, b_list) {
if (block->b_host == host) {
block->b_status = nlm_lck_denied_grace_period;
wake_up(&block->b_wait);
}
}
spin_unlock(&nlm_blocked_lock);
/* Release host handle after use */
nlmclnt_release_host(host);
lockd_down(net);
kfree(req);
return 0;
}