1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/lockd/svclock.c
5 * Handling of server-side locks, mostly of the blocked variety.
6 * This is the ugliest part of lockd because we tread on very thin ice.
7 * GRANT and CANCEL calls may get stuck, meet in mid-flight, etc.
8 * IMNSHO introducing the grant callback into the NLM protocol was one
9 * of the worst ideas Sun ever had. Except maybe for the idea of doing
10 * NFS file locking at all.
12 * I'm trying hard to avoid race conditions by protecting most accesses
13 * to a file's list of blocked locks through a semaphore. The global
14 * list of blocked locks is not protected in this fashion however.
15 * Therefore, some functions (such as the RPC callback for the async grant
16 * call) move blocked locks towards the head of the list *while some other
17 * process might be traversing it*. This should not be a problem in
18 * practice, because this will only cause functions traversing the list
19 * to visit some blocks twice.
21 * Copyright (C) 1996, Olaf Kirch <okir@monad.swb.de>
24 #include <linux/types.h>
25 #include <linux/slab.h>
26 #include <linux/errno.h>
27 #include <linux/kernel.h>
28 #include <linux/sched.h>
29 #include <linux/sunrpc/clnt.h>
30 #include <linux/sunrpc/svc_xprt.h>
31 #include <linux/lockd/nlm.h>
32 #include <linux/lockd/lockd.h>
33 #include <linux/exportfs.h>
35 #define NLMDBG_FACILITY NLMDBG_SVCLOCK
37 #ifdef CONFIG_LOCKD_V4
38 #define nlm_deadlock nlm4_deadlock
40 #define nlm_deadlock nlm_lck_denied
43 static void nlmsvc_release_block(struct nlm_block *block);
44 static void nlmsvc_insert_block(struct nlm_block *block, unsigned long);
45 static void nlmsvc_remove_block(struct nlm_block *block);
47 static int nlmsvc_setgrantargs(struct nlm_rqst *call, struct nlm_lock *lock);
48 static void nlmsvc_freegrantargs(struct nlm_rqst *call);
49 static const struct rpc_call_ops nlmsvc_grant_ops;
52 * The list of blocked locks to retry
54 static LIST_HEAD(nlm_blocked);
55 static DEFINE_SPINLOCK(nlm_blocked_lock);
57 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
58 static const char *nlmdbg_cookie2a(const struct nlm_cookie *cookie)
61 * We can get away with a static buffer because this is only called
62 * from lockd, which is single-threaded.
64 static char buf[2*NLM_MAXCOOKIELEN+1];
65 unsigned int i, len = sizeof(buf);
68 len--; /* allow for trailing \0 */
71 for (i = 0 ; i < cookie->len ; i++) {
76 sprintf(p, "%02x", cookie->data[i]);
87 * Insert a blocked lock into the global list
90 nlmsvc_insert_block_locked(struct nlm_block *block, unsigned long when)
93 struct list_head *pos;
95 dprintk("lockd: nlmsvc_insert_block(%p, %ld)\n", block, when);
96 if (list_empty(&block->b_list)) {
97 kref_get(&block->b_count);
99 list_del_init(&block->b_list);
103 if (when != NLM_NEVER) {
104 if ((when += jiffies) == NLM_NEVER)
106 list_for_each(pos, &nlm_blocked) {
107 b = list_entry(pos, struct nlm_block, b_list);
108 if (time_after(b->b_when,when) || b->b_when == NLM_NEVER)
111 /* On normal exit from the loop, pos == &nlm_blocked,
112 * so we will be adding to the end of the list - good
116 list_add_tail(&block->b_list, pos);
117 block->b_when = when;
120 static void nlmsvc_insert_block(struct nlm_block *block, unsigned long when)
122 spin_lock(&nlm_blocked_lock);
123 nlmsvc_insert_block_locked(block, when);
124 spin_unlock(&nlm_blocked_lock);
128 * Remove a block from the global list
131 nlmsvc_remove_block(struct nlm_block *block)
133 spin_lock(&nlm_blocked_lock);
134 if (!list_empty(&block->b_list)) {
135 list_del_init(&block->b_list);
136 spin_unlock(&nlm_blocked_lock);
137 nlmsvc_release_block(block);
140 spin_unlock(&nlm_blocked_lock);
144 * Find a block for a given lock
146 static struct nlm_block *
147 nlmsvc_lookup_block(struct nlm_file *file, struct nlm_lock *lock)
149 struct nlm_block *block;
150 struct file_lock *fl;
152 dprintk("lockd: nlmsvc_lookup_block f=%p pd=%d %Ld-%Ld ty=%d\n",
153 file, lock->fl.fl_pid,
154 (long long)lock->fl.fl_start,
155 (long long)lock->fl.fl_end, lock->fl.fl_type);
156 spin_lock(&nlm_blocked_lock);
157 list_for_each_entry(block, &nlm_blocked, b_list) {
158 fl = &block->b_call->a_args.lock.fl;
159 dprintk("lockd: check f=%p pd=%d %Ld-%Ld ty=%d cookie=%s\n",
160 block->b_file, fl->fl_pid,
161 (long long)fl->fl_start,
162 (long long)fl->fl_end, fl->fl_type,
163 nlmdbg_cookie2a(&block->b_call->a_args.cookie));
164 if (block->b_file == file && nlm_compare_locks(fl, &lock->fl)) {
165 kref_get(&block->b_count);
166 spin_unlock(&nlm_blocked_lock);
170 spin_unlock(&nlm_blocked_lock);
175 static inline int nlm_cookie_match(struct nlm_cookie *a, struct nlm_cookie *b)
177 if (a->len != b->len)
179 if (memcmp(a->data, b->data, a->len))
185 * Find a block with a given NLM cookie.
187 static inline struct nlm_block *
188 nlmsvc_find_block(struct nlm_cookie *cookie)
190 struct nlm_block *block;
192 spin_lock(&nlm_blocked_lock);
193 list_for_each_entry(block, &nlm_blocked, b_list) {
194 if (nlm_cookie_match(&block->b_call->a_args.cookie,cookie))
197 spin_unlock(&nlm_blocked_lock);
202 dprintk("nlmsvc_find_block(%s): block=%p\n", nlmdbg_cookie2a(cookie), block);
203 kref_get(&block->b_count);
204 spin_unlock(&nlm_blocked_lock);
209 * Create a block and initialize it.
211 * Note: we explicitly set the cookie of the grant reply to that of
212 * the blocked lock request. The spec explicitly mentions that the client
213 * should _not_ rely on the callback containing the same cookie as the
214 * request, but (as I found out later) that's because some implementations
215 * do just this. Never mind the standards comittees, they support our
216 * logging industries.
218 * 10 years later: I hope we can safely ignore these old and broken
219 * clients by now. Let's fix this so we can uniquely identify an incoming
220 * GRANTED_RES message by cookie, without having to rely on the client's IP
223 static struct nlm_block *
224 nlmsvc_create_block(struct svc_rqst *rqstp, struct nlm_host *host,
225 struct nlm_file *file, struct nlm_lock *lock,
226 struct nlm_cookie *cookie)
228 struct nlm_block *block;
229 struct nlm_rqst *call = NULL;
231 call = nlm_alloc_call(host);
235 /* Allocate memory for block, and initialize arguments */
236 block = kzalloc(sizeof(*block), GFP_KERNEL);
239 kref_init(&block->b_count);
240 INIT_LIST_HEAD(&block->b_list);
241 INIT_LIST_HEAD(&block->b_flist);
243 if (!nlmsvc_setgrantargs(call, lock))
246 /* Set notifier function for VFS, and init args */
247 call->a_args.lock.fl.fl_flags |= FL_SLEEP;
248 call->a_args.lock.fl.fl_lmops = &nlmsvc_lock_operations;
249 nlmclnt_next_cookie(&call->a_args.cookie);
251 dprintk("lockd: created block %p...\n", block);
253 /* Create and initialize the block */
254 block->b_daemon = rqstp->rq_server;
255 block->b_host = host;
256 block->b_file = file;
259 /* Add to file's list of blocks */
260 list_add(&block->b_flist, &file->f_blocks);
262 /* Set up RPC arguments for callback */
263 block->b_call = call;
264 call->a_flags = RPC_TASK_ASYNC;
265 call->a_block = block;
272 nlmsvc_release_call(call);
278 * It is the caller's responsibility to check whether the file
279 * can be closed hereafter.
281 static int nlmsvc_unlink_block(struct nlm_block *block)
284 dprintk("lockd: unlinking block %p...\n", block);
286 /* Remove block from list */
287 status = locks_delete_block(&block->b_call->a_args.lock.fl);
288 nlmsvc_remove_block(block);
292 static void nlmsvc_free_block(struct kref *kref)
294 struct nlm_block *block = container_of(kref, struct nlm_block, b_count);
295 struct nlm_file *file = block->b_file;
297 dprintk("lockd: freeing block %p...\n", block);
299 /* Remove block from file's list of blocks */
300 list_del_init(&block->b_flist);
301 mutex_unlock(&file->f_mutex);
303 nlmsvc_freegrantargs(block->b_call);
304 nlmsvc_release_call(block->b_call);
305 nlm_release_file(block->b_file);
309 static void nlmsvc_release_block(struct nlm_block *block)
312 kref_put_mutex(&block->b_count, nlmsvc_free_block, &block->b_file->f_mutex);
316 * Loop over all blocks and delete blocks held by
319 void nlmsvc_traverse_blocks(struct nlm_host *host,
320 struct nlm_file *file,
321 nlm_host_match_fn_t match)
323 struct nlm_block *block, *next;
326 mutex_lock(&file->f_mutex);
327 spin_lock(&nlm_blocked_lock);
328 list_for_each_entry_safe(block, next, &file->f_blocks, b_flist) {
329 if (!match(block->b_host, host))
331 /* Do not destroy blocks that are not on
332 * the global retry list - why? */
333 if (list_empty(&block->b_list))
335 kref_get(&block->b_count);
336 spin_unlock(&nlm_blocked_lock);
337 mutex_unlock(&file->f_mutex);
338 nlmsvc_unlink_block(block);
339 nlmsvc_release_block(block);
342 spin_unlock(&nlm_blocked_lock);
343 mutex_unlock(&file->f_mutex);
346 static struct nlm_lockowner *
347 nlmsvc_get_lockowner(struct nlm_lockowner *lockowner)
349 refcount_inc(&lockowner->count);
353 void nlmsvc_put_lockowner(struct nlm_lockowner *lockowner)
355 if (!refcount_dec_and_lock(&lockowner->count, &lockowner->host->h_lock))
357 list_del(&lockowner->list);
358 spin_unlock(&lockowner->host->h_lock);
359 nlmsvc_release_host(lockowner->host);
363 static struct nlm_lockowner *__nlmsvc_find_lockowner(struct nlm_host *host, pid_t pid)
365 struct nlm_lockowner *lockowner;
366 list_for_each_entry(lockowner, &host->h_lockowners, list) {
367 if (lockowner->pid != pid)
369 return nlmsvc_get_lockowner(lockowner);
374 static struct nlm_lockowner *nlmsvc_find_lockowner(struct nlm_host *host, pid_t pid)
376 struct nlm_lockowner *res, *new = NULL;
378 spin_lock(&host->h_lock);
379 res = __nlmsvc_find_lockowner(host, pid);
382 spin_unlock(&host->h_lock);
383 new = kmalloc(sizeof(*res), GFP_KERNEL);
384 spin_lock(&host->h_lock);
385 res = __nlmsvc_find_lockowner(host, pid);
386 if (res == NULL && new != NULL) {
388 /* fs/locks.c will manage the refcount through lock_ops */
389 refcount_set(&new->count, 1);
391 new->host = nlm_get_host(host);
392 list_add(&new->list, &host->h_lockowners);
397 spin_unlock(&host->h_lock);
403 nlmsvc_release_lockowner(struct nlm_lock *lock)
405 if (lock->fl.fl_owner)
406 nlmsvc_put_lockowner(lock->fl.fl_owner);
409 void nlmsvc_locks_init_private(struct file_lock *fl, struct nlm_host *host,
412 fl->fl_owner = nlmsvc_find_lockowner(host, pid);
416 * Initialize arguments for GRANTED call. The nlm_rqst structure
417 * has been cleared already.
419 static int nlmsvc_setgrantargs(struct nlm_rqst *call, struct nlm_lock *lock)
421 locks_copy_lock(&call->a_args.lock.fl, &lock->fl);
422 memcpy(&call->a_args.lock.fh, &lock->fh, sizeof(call->a_args.lock.fh));
423 call->a_args.lock.caller = utsname()->nodename;
424 call->a_args.lock.oh.len = lock->oh.len;
426 /* set default data area */
427 call->a_args.lock.oh.data = call->a_owner;
428 call->a_args.lock.svid = ((struct nlm_lockowner *)lock->fl.fl_owner)->pid;
430 if (lock->oh.len > NLMCLNT_OHSIZE) {
431 void *data = kmalloc(lock->oh.len, GFP_KERNEL);
434 call->a_args.lock.oh.data = (u8 *) data;
437 memcpy(call->a_args.lock.oh.data, lock->oh.data, lock->oh.len);
441 static void nlmsvc_freegrantargs(struct nlm_rqst *call)
443 if (call->a_args.lock.oh.data != call->a_owner)
444 kfree(call->a_args.lock.oh.data);
446 locks_release_private(&call->a_args.lock.fl);
450 * Deferred lock request handling for non-blocking lock
453 nlmsvc_defer_lock_rqst(struct svc_rqst *rqstp, struct nlm_block *block)
455 __be32 status = nlm_lck_denied_nolocks;
457 block->b_flags |= B_QUEUED;
459 nlmsvc_insert_block(block, NLM_TIMEOUT);
461 block->b_cache_req = &rqstp->rq_chandle;
462 if (rqstp->rq_chandle.defer) {
463 block->b_deferred_req =
464 rqstp->rq_chandle.defer(block->b_cache_req);
465 if (block->b_deferred_req != NULL)
466 status = nlm_drop_reply;
468 dprintk("lockd: nlmsvc_defer_lock_rqst block %p flags %d status %d\n",
469 block, block->b_flags, ntohl(status));
475 * Attempt to establish a lock, and if it can't be granted, block it
479 nlmsvc_lock(struct svc_rqst *rqstp, struct nlm_file *file,
480 struct nlm_host *host, struct nlm_lock *lock, int wait,
481 struct nlm_cookie *cookie, int reclaim)
483 struct inode *inode = nlmsvc_file_inode(file);
484 struct nlm_block *block = NULL;
490 dprintk("lockd: nlmsvc_lock(%s/%ld, ty=%d, pi=%d, %Ld-%Ld, bl=%d)\n",
491 inode->i_sb->s_id, inode->i_ino,
492 lock->fl.fl_type, lock->fl.fl_pid,
493 (long long)lock->fl.fl_start,
494 (long long)lock->fl.fl_end,
497 if (!exportfs_lock_op_is_async(inode->i_sb->s_export_op)) {
502 /* Lock file against concurrent access */
503 mutex_lock(&file->f_mutex);
504 /* Get existing block (in case client is busy-waiting)
505 * or create new block
507 block = nlmsvc_lookup_block(file, lock);
509 block = nlmsvc_create_block(rqstp, host, file, lock, cookie);
510 ret = nlm_lck_denied_nolocks;
513 lock = &block->b_call->a_args.lock;
515 lock->fl.fl_flags &= ~FL_SLEEP;
517 if (block->b_flags & B_QUEUED) {
518 dprintk("lockd: nlmsvc_lock deferred block %p flags %d\n",
519 block, block->b_flags);
520 if (block->b_granted) {
521 nlmsvc_unlink_block(block);
525 if (block->b_flags & B_TIMED_OUT) {
526 nlmsvc_unlink_block(block);
527 ret = nlm_lck_denied;
530 ret = nlm_drop_reply;
534 if (locks_in_grace(SVC_NET(rqstp)) && !reclaim) {
535 ret = nlm_lck_denied_grace_period;
538 if (reclaim && !locks_in_grace(SVC_NET(rqstp))) {
539 ret = nlm_lck_denied_grace_period;
543 spin_lock(&nlm_blocked_lock);
545 * If this is a lock request for an already pending
546 * lock request we return nlm_lck_blocked without calling
547 * vfs_lock_file() again. Otherwise we have two pending
548 * requests on the underlaying ->lock() implementation but
549 * only one nlm_block to being granted by lm_grant().
551 if (exportfs_lock_op_is_async(inode->i_sb->s_export_op) &&
552 !list_empty(&block->b_list)) {
553 spin_unlock(&nlm_blocked_lock);
554 ret = nlm_lck_blocked;
558 /* Append to list of blocked */
559 nlmsvc_insert_block_locked(block, NLM_NEVER);
560 spin_unlock(&nlm_blocked_lock);
563 lock->fl.fl_flags &= ~FL_SLEEP;
564 mode = lock_to_openmode(&lock->fl);
565 error = vfs_lock_file(file->f_file[mode], F_SETLK, &lock->fl, NULL);
566 lock->fl.fl_flags &= ~FL_SLEEP;
568 dprintk("lockd: vfs_lock_file returned %d\n", error);
571 nlmsvc_remove_block(block);
576 nlmsvc_remove_block(block);
577 ret = async_block ? nlm_lck_blocked : nlm_lck_denied;
579 case FILE_LOCK_DEFERRED:
582 /* Filesystem lock operation is in progress
583 Add it to the queue waiting for callback */
584 ret = nlmsvc_defer_lock_rqst(rqstp, block);
587 nlmsvc_remove_block(block);
590 default: /* includes ENOLCK */
591 nlmsvc_remove_block(block);
592 ret = nlm_lck_denied_nolocks;
596 ret = nlm_lck_blocked;
598 mutex_unlock(&file->f_mutex);
599 nlmsvc_release_block(block);
600 dprintk("lockd: nlmsvc_lock returned %u\n", ret);
605 * Test for presence of a conflicting lock.
608 nlmsvc_testlock(struct svc_rqst *rqstp, struct nlm_file *file,
609 struct nlm_host *host, struct nlm_lock *lock,
610 struct nlm_lock *conflock, struct nlm_cookie *cookie)
616 dprintk("lockd: nlmsvc_testlock(%s/%ld, ty=%d, %Ld-%Ld)\n",
617 nlmsvc_file_inode(file)->i_sb->s_id,
618 nlmsvc_file_inode(file)->i_ino,
620 (long long)lock->fl.fl_start,
621 (long long)lock->fl.fl_end);
623 if (locks_in_grace(SVC_NET(rqstp))) {
624 ret = nlm_lck_denied_grace_period;
628 mode = lock_to_openmode(&lock->fl);
629 error = vfs_test_lock(file->f_file[mode], &lock->fl);
631 /* We can't currently deal with deferred test requests */
632 if (error == FILE_LOCK_DEFERRED)
635 ret = nlm_lck_denied_nolocks;
639 if (lock->fl.fl_type == F_UNLCK) {
644 dprintk("lockd: conflicting lock(ty=%d, %Ld-%Ld)\n",
645 lock->fl.fl_type, (long long)lock->fl.fl_start,
646 (long long)lock->fl.fl_end);
647 conflock->caller = "somehost"; /* FIXME */
648 conflock->len = strlen(conflock->caller);
649 conflock->oh.len = 0; /* don't return OH info */
650 conflock->svid = lock->fl.fl_pid;
651 conflock->fl.fl_type = lock->fl.fl_type;
652 conflock->fl.fl_start = lock->fl.fl_start;
653 conflock->fl.fl_end = lock->fl.fl_end;
654 locks_release_private(&lock->fl);
656 ret = nlm_lck_denied;
663 * This implies a CANCEL call: We send a GRANT_MSG, the client replies
664 * with a GRANT_RES call which gets lost, and calls UNLOCK immediately
665 * afterwards. In this case the block will still be there, and hence
669 nlmsvc_unlock(struct net *net, struct nlm_file *file, struct nlm_lock *lock)
673 dprintk("lockd: nlmsvc_unlock(%s/%ld, pi=%d, %Ld-%Ld)\n",
674 nlmsvc_file_inode(file)->i_sb->s_id,
675 nlmsvc_file_inode(file)->i_ino,
677 (long long)lock->fl.fl_start,
678 (long long)lock->fl.fl_end);
680 /* First, cancel any lock that might be there */
681 nlmsvc_cancel_blocked(net, file, lock);
683 lock->fl.fl_type = F_UNLCK;
684 lock->fl.fl_file = file->f_file[O_RDONLY];
685 if (lock->fl.fl_file)
686 error = vfs_lock_file(lock->fl.fl_file, F_SETLK,
688 lock->fl.fl_file = file->f_file[O_WRONLY];
689 if (lock->fl.fl_file)
690 error |= vfs_lock_file(lock->fl.fl_file, F_SETLK,
693 return (error < 0)? nlm_lck_denied_nolocks : nlm_granted;
697 * Cancel a previously blocked request.
699 * A cancel request always overrides any grant that may currently
701 * The calling procedure must check whether the file can be closed.
704 nlmsvc_cancel_blocked(struct net *net, struct nlm_file *file, struct nlm_lock *lock)
706 struct nlm_block *block;
710 dprintk("lockd: nlmsvc_cancel(%s/%ld, pi=%d, %Ld-%Ld)\n",
711 nlmsvc_file_inode(file)->i_sb->s_id,
712 nlmsvc_file_inode(file)->i_ino,
714 (long long)lock->fl.fl_start,
715 (long long)lock->fl.fl_end);
717 if (locks_in_grace(net))
718 return nlm_lck_denied_grace_period;
720 mutex_lock(&file->f_mutex);
721 block = nlmsvc_lookup_block(file, lock);
722 mutex_unlock(&file->f_mutex);
724 struct file_lock *fl = &block->b_call->a_args.lock.fl;
726 mode = lock_to_openmode(fl);
727 vfs_cancel_lock(block->b_file->f_file[mode], fl);
728 status = nlmsvc_unlink_block(block);
729 nlmsvc_release_block(block);
731 return status ? nlm_lck_denied : nlm_granted;
735 * This is a callback from the filesystem for VFS file lock requests.
736 * It will be used if lm_grant is defined and the filesystem can not
737 * respond to the request immediately.
738 * For SETLK or SETLKW request it will get the local posix lock.
739 * In all cases it will move the block to the head of nlm_blocked q where
740 * nlmsvc_retry_blocked() can send back a reply for SETLKW or revisit the
741 * deferred rpc for GETLK and SETLK.
744 nlmsvc_update_deferred_block(struct nlm_block *block, int result)
746 block->b_flags |= B_GOT_CALLBACK;
748 block->b_granted = 1;
750 block->b_flags |= B_TIMED_OUT;
753 static int nlmsvc_grant_deferred(struct file_lock *fl, int result)
755 struct nlm_block *block;
758 spin_lock(&nlm_blocked_lock);
759 list_for_each_entry(block, &nlm_blocked, b_list) {
760 if (nlm_compare_locks(&block->b_call->a_args.lock.fl, fl)) {
761 dprintk("lockd: nlmsvc_notify_blocked block %p flags %d\n",
762 block, block->b_flags);
763 if (block->b_flags & B_QUEUED) {
764 if (block->b_flags & B_TIMED_OUT) {
768 nlmsvc_update_deferred_block(block, result);
769 } else if (result == 0)
770 block->b_granted = 1;
772 nlmsvc_insert_block_locked(block, 0);
773 svc_wake_up(block->b_daemon);
778 spin_unlock(&nlm_blocked_lock);
780 printk(KERN_WARNING "lockd: grant for unknown block\n");
785 * Unblock a blocked lock request. This is a callback invoked from the
786 * VFS layer when a lock on which we blocked is removed.
788 * This function doesn't grant the blocked lock instantly, but rather moves
789 * the block to the head of nlm_blocked where it can be picked up by lockd.
792 nlmsvc_notify_blocked(struct file_lock *fl)
794 struct nlm_block *block;
796 dprintk("lockd: VFS unblock notification for block %p\n", fl);
797 spin_lock(&nlm_blocked_lock);
798 list_for_each_entry(block, &nlm_blocked, b_list) {
799 if (nlm_compare_locks(&block->b_call->a_args.lock.fl, fl)) {
800 nlmsvc_insert_block_locked(block, 0);
801 spin_unlock(&nlm_blocked_lock);
802 svc_wake_up(block->b_daemon);
806 spin_unlock(&nlm_blocked_lock);
807 printk(KERN_WARNING "lockd: notification for unknown block!\n");
810 static fl_owner_t nlmsvc_get_owner(fl_owner_t owner)
812 return nlmsvc_get_lockowner(owner);
815 static void nlmsvc_put_owner(fl_owner_t owner)
817 nlmsvc_put_lockowner(owner);
820 const struct lock_manager_operations nlmsvc_lock_operations = {
821 .lm_notify = nlmsvc_notify_blocked,
822 .lm_grant = nlmsvc_grant_deferred,
823 .lm_get_owner = nlmsvc_get_owner,
824 .lm_put_owner = nlmsvc_put_owner,
828 * Try to claim a lock that was previously blocked.
830 * Note that we use both the RPC_GRANTED_MSG call _and_ an async
831 * RPC thread when notifying the client. This seems like overkill...
833 * - we don't want to use a synchronous RPC thread, otherwise
834 * we might find ourselves hanging on a dead portmapper.
835 * - Some lockd implementations (e.g. HP) don't react to
836 * RPC_GRANTED calls; they seem to insist on RPC_GRANTED_MSG calls.
839 nlmsvc_grant_blocked(struct nlm_block *block)
841 struct nlm_file *file = block->b_file;
842 struct nlm_lock *lock = &block->b_call->a_args.lock;
845 loff_t fl_start, fl_end;
847 dprintk("lockd: grant blocked lock %p\n", block);
849 kref_get(&block->b_count);
851 /* Unlink block request from list */
852 nlmsvc_unlink_block(block);
854 /* If b_granted is true this means we've been here before.
855 * Just retry the grant callback, possibly refreshing the RPC
857 if (block->b_granted) {
858 nlm_rebind_host(block->b_host);
862 /* Try the lock operation again */
863 /* vfs_lock_file() can mangle fl_start and fl_end, but we need
864 * them unchanged for the GRANT_MSG
866 lock->fl.fl_flags |= FL_SLEEP;
867 fl_start = lock->fl.fl_start;
868 fl_end = lock->fl.fl_end;
869 mode = lock_to_openmode(&lock->fl);
870 error = vfs_lock_file(file->f_file[mode], F_SETLK, &lock->fl, NULL);
871 lock->fl.fl_flags &= ~FL_SLEEP;
872 lock->fl.fl_start = fl_start;
873 lock->fl.fl_end = fl_end;
878 case FILE_LOCK_DEFERRED:
879 dprintk("lockd: lock still blocked error %d\n", error);
880 nlmsvc_insert_block(block, NLM_NEVER);
881 nlmsvc_release_block(block);
884 printk(KERN_WARNING "lockd: unexpected error %d in %s!\n",
886 nlmsvc_insert_block(block, 10 * HZ);
887 nlmsvc_release_block(block);
892 /* Lock was granted by VFS. */
893 dprintk("lockd: GRANTing blocked lock.\n");
894 block->b_granted = 1;
896 /* keep block on the list, but don't reattempt until the RPC
897 * completes or the submission fails
899 nlmsvc_insert_block(block, NLM_NEVER);
901 /* Call the client -- use a soft RPC task since nlmsvc_retry_blocked
902 * will queue up a new one if this one times out
904 error = nlm_async_call(block->b_call, NLMPROC_GRANTED_MSG,
907 /* RPC submission failed, wait a bit and retry */
909 nlmsvc_insert_block(block, 10 * HZ);
913 * This is the callback from the RPC layer when the NLM_GRANTED_MSG
914 * RPC call has succeeded or timed out.
915 * Like all RPC callbacks, it is invoked by the rpciod process, so it
916 * better not sleep. Therefore, we put the blocked lock on the nlm_blocked
917 * chain once more in order to have it removed by lockd itself (which can
918 * then sleep on the file semaphore without disrupting e.g. the nfs client).
920 static void nlmsvc_grant_callback(struct rpc_task *task, void *data)
922 struct nlm_rqst *call = data;
923 struct nlm_block *block = call->a_block;
924 unsigned long timeout;
926 dprintk("lockd: GRANT_MSG RPC callback\n");
928 spin_lock(&nlm_blocked_lock);
929 /* if the block is not on a list at this point then it has
930 * been invalidated. Don't try to requeue it.
932 * FIXME: it's possible that the block is removed from the list
933 * after this check but before the nlmsvc_insert_block. In that
934 * case it will be added back. Perhaps we need better locking
937 if (list_empty(&block->b_list))
940 /* Technically, we should down the file semaphore here. Since we
941 * move the block towards the head of the queue only, no harm
942 * can be done, though. */
943 if (task->tk_status < 0) {
944 /* RPC error: Re-insert for retransmission */
947 /* Call was successful, now wait for client callback */
950 nlmsvc_insert_block_locked(block, timeout);
951 svc_wake_up(block->b_daemon);
953 spin_unlock(&nlm_blocked_lock);
957 * FIXME: nlmsvc_release_block() grabs a mutex. This is not allowed for an
958 * .rpc_release rpc_call_op
960 static void nlmsvc_grant_release(void *data)
962 struct nlm_rqst *call = data;
963 nlmsvc_release_block(call->a_block);
966 static const struct rpc_call_ops nlmsvc_grant_ops = {
967 .rpc_call_done = nlmsvc_grant_callback,
968 .rpc_release = nlmsvc_grant_release,
972 * We received a GRANT_RES callback. Try to find the corresponding
976 nlmsvc_grant_reply(struct nlm_cookie *cookie, __be32 status)
978 struct nlm_block *block;
979 struct file_lock *fl;
982 dprintk("grant_reply: looking for cookie %x, s=%d \n",
983 *(unsigned int *)(cookie->data), status);
984 if (!(block = nlmsvc_find_block(cookie)))
988 case nlm_lck_denied_grace_period:
989 /* Try again in a couple of seconds */
990 nlmsvc_insert_block(block, 10 * HZ);
993 /* Client doesn't want it, just unlock it */
994 nlmsvc_unlink_block(block);
995 fl = &block->b_call->a_args.lock.fl;
996 fl->fl_type = F_UNLCK;
997 error = vfs_lock_file(fl->fl_file, F_SETLK, fl, NULL);
999 pr_warn("lockd: unable to unlock lock rejected by client!\n");
1003 * Either it was accepted or the status makes no sense
1004 * just unlink it either way.
1006 nlmsvc_unlink_block(block);
1008 nlmsvc_release_block(block);
1011 /* Helper function to handle retry of a deferred block.
1012 * If it is a blocking lock, call grant_blocked.
1013 * For a non-blocking lock or test lock, revisit the request.
1016 retry_deferred_block(struct nlm_block *block)
1018 if (!(block->b_flags & B_GOT_CALLBACK))
1019 block->b_flags |= B_TIMED_OUT;
1020 nlmsvc_insert_block(block, NLM_TIMEOUT);
1021 dprintk("revisit block %p flags %d\n", block, block->b_flags);
1022 if (block->b_deferred_req) {
1023 block->b_deferred_req->revisit(block->b_deferred_req, 0);
1024 block->b_deferred_req = NULL;
1029 * Retry all blocked locks that have been notified. This is where lockd
1030 * picks up locks that can be granted, or grant notifications that must
1034 nlmsvc_retry_blocked(struct svc_rqst *rqstp)
1036 unsigned long timeout = MAX_SCHEDULE_TIMEOUT;
1037 struct nlm_block *block;
1039 spin_lock(&nlm_blocked_lock);
1040 while (!list_empty(&nlm_blocked) && !svc_thread_should_stop(rqstp)) {
1041 block = list_entry(nlm_blocked.next, struct nlm_block, b_list);
1043 if (block->b_when == NLM_NEVER)
1045 if (time_after(block->b_when, jiffies)) {
1046 timeout = block->b_when - jiffies;
1049 spin_unlock(&nlm_blocked_lock);
1051 dprintk("nlmsvc_retry_blocked(%p, when=%ld)\n",
1052 block, block->b_when);
1053 if (block->b_flags & B_QUEUED) {
1054 dprintk("nlmsvc_retry_blocked delete block (%p, granted=%d, flags=%d)\n",
1055 block, block->b_granted, block->b_flags);
1056 retry_deferred_block(block);
1058 nlmsvc_grant_blocked(block);
1059 spin_lock(&nlm_blocked_lock);
1061 spin_unlock(&nlm_blocked_lock);
1063 if (timeout < MAX_SCHEDULE_TIMEOUT)
1064 mod_timer(&nlmsvc_retry, jiffies + timeout);