Merge git://git.kernel.org/pub/scm/linux/kernel/git/lethal/sh-2.6.24
[linux-2.6] / fs / lockd / svclock.c
1 /*
2  * linux/fs/lockd/svclock.c
3  *
4  * Handling of server-side locks, mostly of the blocked variety.
5  * This is the ugliest part of lockd because we tread on very thin ice.
6  * GRANT and CANCEL calls may get stuck, meet in mid-flight, etc.
7  * IMNSHO introducing the grant callback into the NLM protocol was one
8  * of the worst ideas Sun ever had. Except maybe for the idea of doing
9  * NFS file locking at all.
10  *
11  * I'm trying hard to avoid race conditions by protecting most accesses
12  * to a file's list of blocked locks through a semaphore. The global
13  * list of blocked locks is not protected in this fashion however.
14  * Therefore, some functions (such as the RPC callback for the async grant
15  * call) move blocked locks towards the head of the list *while some other
16  * process might be traversing it*. This should not be a problem in
17  * practice, because this will only cause functions traversing the list
18  * to visit some blocks twice.
19  *
20  * Copyright (C) 1996, Olaf Kirch <okir@monad.swb.de>
21  */
22
23 #include <linux/types.h>
24 #include <linux/errno.h>
25 #include <linux/kernel.h>
26 #include <linux/sched.h>
27 #include <linux/smp_lock.h>
28 #include <linux/sunrpc/clnt.h>
29 #include <linux/sunrpc/svc.h>
30 #include <linux/lockd/nlm.h>
31 #include <linux/lockd/lockd.h>
32
33 #define NLMDBG_FACILITY         NLMDBG_SVCLOCK
34
35 #ifdef CONFIG_LOCKD_V4
36 #define nlm_deadlock    nlm4_deadlock
37 #else
38 #define nlm_deadlock    nlm_lck_denied
39 #endif
40
41 static void nlmsvc_release_block(struct nlm_block *block);
42 static void     nlmsvc_insert_block(struct nlm_block *block, unsigned long);
43 static void     nlmsvc_remove_block(struct nlm_block *block);
44
45 static int nlmsvc_setgrantargs(struct nlm_rqst *call, struct nlm_lock *lock);
46 static void nlmsvc_freegrantargs(struct nlm_rqst *call);
47 static const struct rpc_call_ops nlmsvc_grant_ops;
48
49 /*
50  * The list of blocked locks to retry
51  */
52 static LIST_HEAD(nlm_blocked);
53
54 /*
55  * Insert a blocked lock into the global list
56  */
57 static void
58 nlmsvc_insert_block(struct nlm_block *block, unsigned long when)
59 {
60         struct nlm_block *b;
61         struct list_head *pos;
62
63         dprintk("lockd: nlmsvc_insert_block(%p, %ld)\n", block, when);
64         if (list_empty(&block->b_list)) {
65                 kref_get(&block->b_count);
66         } else {
67                 list_del_init(&block->b_list);
68         }
69
70         pos = &nlm_blocked;
71         if (when != NLM_NEVER) {
72                 if ((when += jiffies) == NLM_NEVER)
73                         when ++;
74                 list_for_each(pos, &nlm_blocked) {
75                         b = list_entry(pos, struct nlm_block, b_list);
76                         if (time_after(b->b_when,when) || b->b_when == NLM_NEVER)
77                                 break;
78                 }
79                 /* On normal exit from the loop, pos == &nlm_blocked,
80                  * so we will be adding to the end of the list - good
81                  */
82         }
83
84         list_add_tail(&block->b_list, pos);
85         block->b_when = when;
86 }
87
88 /*
89  * Remove a block from the global list
90  */
91 static inline void
92 nlmsvc_remove_block(struct nlm_block *block)
93 {
94         if (!list_empty(&block->b_list)) {
95                 list_del_init(&block->b_list);
96                 nlmsvc_release_block(block);
97         }
98 }
99
100 /*
101  * Find a block for a given lock
102  */
103 static struct nlm_block *
104 nlmsvc_lookup_block(struct nlm_file *file, struct nlm_lock *lock)
105 {
106         struct nlm_block        *block;
107         struct file_lock        *fl;
108
109         dprintk("lockd: nlmsvc_lookup_block f=%p pd=%d %Ld-%Ld ty=%d\n",
110                                 file, lock->fl.fl_pid,
111                                 (long long)lock->fl.fl_start,
112                                 (long long)lock->fl.fl_end, lock->fl.fl_type);
113         list_for_each_entry(block, &nlm_blocked, b_list) {
114                 fl = &block->b_call->a_args.lock.fl;
115                 dprintk("lockd: check f=%p pd=%d %Ld-%Ld ty=%d cookie=%s\n",
116                                 block->b_file, fl->fl_pid,
117                                 (long long)fl->fl_start,
118                                 (long long)fl->fl_end, fl->fl_type,
119                                 nlmdbg_cookie2a(&block->b_call->a_args.cookie));
120                 if (block->b_file == file && nlm_compare_locks(fl, &lock->fl)) {
121                         kref_get(&block->b_count);
122                         return block;
123                 }
124         }
125
126         return NULL;
127 }
128
129 static inline int nlm_cookie_match(struct nlm_cookie *a, struct nlm_cookie *b)
130 {
131         if(a->len != b->len)
132                 return 0;
133         if(memcmp(a->data,b->data,a->len))
134                 return 0;
135         return 1;
136 }
137
138 /*
139  * Find a block with a given NLM cookie.
140  */
141 static inline struct nlm_block *
142 nlmsvc_find_block(struct nlm_cookie *cookie)
143 {
144         struct nlm_block *block;
145
146         list_for_each_entry(block, &nlm_blocked, b_list) {
147                 if (nlm_cookie_match(&block->b_call->a_args.cookie,cookie))
148                         goto found;
149         }
150
151         return NULL;
152
153 found:
154         dprintk("nlmsvc_find_block(%s): block=%p\n", nlmdbg_cookie2a(cookie), block);
155         kref_get(&block->b_count);
156         return block;
157 }
158
159 /*
160  * Create a block and initialize it.
161  *
162  * Note: we explicitly set the cookie of the grant reply to that of
163  * the blocked lock request. The spec explicitly mentions that the client
164  * should _not_ rely on the callback containing the same cookie as the
165  * request, but (as I found out later) that's because some implementations
166  * do just this. Never mind the standards comittees, they support our
167  * logging industries.
168  *
169  * 10 years later: I hope we can safely ignore these old and broken
170  * clients by now. Let's fix this so we can uniquely identify an incoming
171  * GRANTED_RES message by cookie, without having to rely on the client's IP
172  * address. --okir
173  */
174 static struct nlm_block *
175 nlmsvc_create_block(struct svc_rqst *rqstp, struct nlm_host *host,
176                     struct nlm_file *file, struct nlm_lock *lock,
177                     struct nlm_cookie *cookie)
178 {
179         struct nlm_block        *block;
180         struct nlm_rqst         *call = NULL;
181
182         call = nlm_alloc_call(host);
183         if (call == NULL)
184                 return NULL;
185
186         /* Allocate memory for block, and initialize arguments */
187         block = kzalloc(sizeof(*block), GFP_KERNEL);
188         if (block == NULL)
189                 goto failed;
190         kref_init(&block->b_count);
191         INIT_LIST_HEAD(&block->b_list);
192         INIT_LIST_HEAD(&block->b_flist);
193
194         if (!nlmsvc_setgrantargs(call, lock))
195                 goto failed_free;
196
197         /* Set notifier function for VFS, and init args */
198         call->a_args.lock.fl.fl_flags |= FL_SLEEP;
199         call->a_args.lock.fl.fl_lmops = &nlmsvc_lock_operations;
200         nlmclnt_next_cookie(&call->a_args.cookie);
201
202         dprintk("lockd: created block %p...\n", block);
203
204         /* Create and initialize the block */
205         block->b_daemon = rqstp->rq_server;
206         block->b_host   = host;
207         block->b_file   = file;
208         block->b_fl = NULL;
209         file->f_count++;
210
211         /* Add to file's list of blocks */
212         list_add(&block->b_flist, &file->f_blocks);
213
214         /* Set up RPC arguments for callback */
215         block->b_call = call;
216         call->a_flags   = RPC_TASK_ASYNC;
217         call->a_block = block;
218
219         return block;
220
221 failed_free:
222         kfree(block);
223 failed:
224         nlm_release_call(call);
225         return NULL;
226 }
227
228 /*
229  * Delete a block. If the lock was cancelled or the grant callback
230  * failed, unlock is set to 1.
231  * It is the caller's responsibility to check whether the file
232  * can be closed hereafter.
233  */
234 static int nlmsvc_unlink_block(struct nlm_block *block)
235 {
236         int status;
237         dprintk("lockd: unlinking block %p...\n", block);
238
239         /* Remove block from list */
240         status = posix_unblock_lock(block->b_file->f_file, &block->b_call->a_args.lock.fl);
241         nlmsvc_remove_block(block);
242         return status;
243 }
244
245 static void nlmsvc_free_block(struct kref *kref)
246 {
247         struct nlm_block *block = container_of(kref, struct nlm_block, b_count);
248         struct nlm_file         *file = block->b_file;
249
250         dprintk("lockd: freeing block %p...\n", block);
251
252         /* Remove block from file's list of blocks */
253         mutex_lock(&file->f_mutex);
254         list_del_init(&block->b_flist);
255         mutex_unlock(&file->f_mutex);
256
257         nlmsvc_freegrantargs(block->b_call);
258         nlm_release_call(block->b_call);
259         nlm_release_file(block->b_file);
260         kfree(block->b_fl);
261         kfree(block);
262 }
263
264 static void nlmsvc_release_block(struct nlm_block *block)
265 {
266         if (block != NULL)
267                 kref_put(&block->b_count, nlmsvc_free_block);
268 }
269
270 /*
271  * Loop over all blocks and delete blocks held by
272  * a matching host.
273  */
274 void nlmsvc_traverse_blocks(struct nlm_host *host,
275                         struct nlm_file *file,
276                         nlm_host_match_fn_t match)
277 {
278         struct nlm_block *block, *next;
279
280 restart:
281         mutex_lock(&file->f_mutex);
282         list_for_each_entry_safe(block, next, &file->f_blocks, b_flist) {
283                 if (!match(block->b_host, host))
284                         continue;
285                 /* Do not destroy blocks that are not on
286                  * the global retry list - why? */
287                 if (list_empty(&block->b_list))
288                         continue;
289                 kref_get(&block->b_count);
290                 mutex_unlock(&file->f_mutex);
291                 nlmsvc_unlink_block(block);
292                 nlmsvc_release_block(block);
293                 goto restart;
294         }
295         mutex_unlock(&file->f_mutex);
296 }
297
298 /*
299  * Initialize arguments for GRANTED call. The nlm_rqst structure
300  * has been cleared already.
301  */
302 static int nlmsvc_setgrantargs(struct nlm_rqst *call, struct nlm_lock *lock)
303 {
304         locks_copy_lock(&call->a_args.lock.fl, &lock->fl);
305         memcpy(&call->a_args.lock.fh, &lock->fh, sizeof(call->a_args.lock.fh));
306         call->a_args.lock.caller = utsname()->nodename;
307         call->a_args.lock.oh.len = lock->oh.len;
308
309         /* set default data area */
310         call->a_args.lock.oh.data = call->a_owner;
311         call->a_args.lock.svid = lock->fl.fl_pid;
312
313         if (lock->oh.len > NLMCLNT_OHSIZE) {
314                 void *data = kmalloc(lock->oh.len, GFP_KERNEL);
315                 if (!data)
316                         return 0;
317                 call->a_args.lock.oh.data = (u8 *) data;
318         }
319
320         memcpy(call->a_args.lock.oh.data, lock->oh.data, lock->oh.len);
321         return 1;
322 }
323
324 static void nlmsvc_freegrantargs(struct nlm_rqst *call)
325 {
326         if (call->a_args.lock.oh.data != call->a_owner)
327                 kfree(call->a_args.lock.oh.data);
328 }
329
330 /*
331  * Deferred lock request handling for non-blocking lock
332  */
333 static __be32
334 nlmsvc_defer_lock_rqst(struct svc_rqst *rqstp, struct nlm_block *block)
335 {
336         __be32 status = nlm_lck_denied_nolocks;
337
338         block->b_flags |= B_QUEUED;
339
340         nlmsvc_insert_block(block, NLM_TIMEOUT);
341
342         block->b_cache_req = &rqstp->rq_chandle;
343         if (rqstp->rq_chandle.defer) {
344                 block->b_deferred_req =
345                         rqstp->rq_chandle.defer(block->b_cache_req);
346                 if (block->b_deferred_req != NULL)
347                         status = nlm_drop_reply;
348         }
349         dprintk("lockd: nlmsvc_defer_lock_rqst block %p flags %d status %d\n",
350                 block, block->b_flags, ntohl(status));
351
352         return status;
353 }
354
355 /*
356  * Attempt to establish a lock, and if it can't be granted, block it
357  * if required.
358  */
359 __be32
360 nlmsvc_lock(struct svc_rqst *rqstp, struct nlm_file *file,
361                         struct nlm_lock *lock, int wait, struct nlm_cookie *cookie)
362 {
363         struct nlm_block        *block = NULL;
364         struct nlm_host         *host;
365         int                     error;
366         __be32                  ret;
367
368         dprintk("lockd: nlmsvc_lock(%s/%ld, ty=%d, pi=%d, %Ld-%Ld, bl=%d)\n",
369                                 file->f_file->f_path.dentry->d_inode->i_sb->s_id,
370                                 file->f_file->f_path.dentry->d_inode->i_ino,
371                                 lock->fl.fl_type, lock->fl.fl_pid,
372                                 (long long)lock->fl.fl_start,
373                                 (long long)lock->fl.fl_end,
374                                 wait);
375
376         /* Create host handle for callback */
377         host = nlmsvc_lookup_host(rqstp, lock->caller, lock->len);
378         if (host == NULL)
379                 return nlm_lck_denied_nolocks;
380
381         /* Lock file against concurrent access */
382         mutex_lock(&file->f_mutex);
383         /* Get existing block (in case client is busy-waiting)
384          * or create new block
385          */
386         block = nlmsvc_lookup_block(file, lock);
387         if (block == NULL) {
388                 block = nlmsvc_create_block(rqstp, nlm_get_host(host), file,
389                                 lock, cookie);
390                 ret = nlm_lck_denied_nolocks;
391                 if (block == NULL)
392                         goto out;
393                 lock = &block->b_call->a_args.lock;
394         } else
395                 lock->fl.fl_flags &= ~FL_SLEEP;
396
397         if (block->b_flags & B_QUEUED) {
398                 dprintk("lockd: nlmsvc_lock deferred block %p flags %d\n",
399                                                         block, block->b_flags);
400                 if (block->b_granted) {
401                         nlmsvc_unlink_block(block);
402                         ret = nlm_granted;
403                         goto out;
404                 }
405                 if (block->b_flags & B_TIMED_OUT) {
406                         nlmsvc_unlink_block(block);
407                         ret = nlm_lck_denied;
408                         goto out;
409                 }
410                 ret = nlm_drop_reply;
411                 goto out;
412         }
413
414         if (!wait)
415                 lock->fl.fl_flags &= ~FL_SLEEP;
416         error = vfs_lock_file(file->f_file, F_SETLK, &lock->fl, NULL);
417         lock->fl.fl_flags &= ~FL_SLEEP;
418
419         dprintk("lockd: vfs_lock_file returned %d\n", error);
420         switch(error) {
421                 case 0:
422                         ret = nlm_granted;
423                         goto out;
424                 case -EAGAIN:
425                         ret = nlm_lck_denied;
426                         break;
427                 case -EINPROGRESS:
428                         if (wait)
429                                 break;
430                         /* Filesystem lock operation is in progress
431                            Add it to the queue waiting for callback */
432                         ret = nlmsvc_defer_lock_rqst(rqstp, block);
433                         goto out;
434                 case -EDEADLK:
435                         ret = nlm_deadlock;
436                         goto out;
437                 default:                        /* includes ENOLCK */
438                         ret = nlm_lck_denied_nolocks;
439                         goto out;
440         }
441
442         ret = nlm_lck_denied;
443         if (!wait)
444                 goto out;
445
446         ret = nlm_lck_blocked;
447
448         /* Append to list of blocked */
449         nlmsvc_insert_block(block, NLM_NEVER);
450 out:
451         mutex_unlock(&file->f_mutex);
452         nlmsvc_release_block(block);
453         nlm_release_host(host);
454         dprintk("lockd: nlmsvc_lock returned %u\n", ret);
455         return ret;
456 }
457
458 /*
459  * Test for presence of a conflicting lock.
460  */
461 __be32
462 nlmsvc_testlock(struct svc_rqst *rqstp, struct nlm_file *file,
463                 struct nlm_lock *lock, struct nlm_lock *conflock,
464                 struct nlm_cookie *cookie)
465 {
466         struct nlm_block        *block = NULL;
467         int                     error;
468         __be32                  ret;
469
470         dprintk("lockd: nlmsvc_testlock(%s/%ld, ty=%d, %Ld-%Ld)\n",
471                                 file->f_file->f_path.dentry->d_inode->i_sb->s_id,
472                                 file->f_file->f_path.dentry->d_inode->i_ino,
473                                 lock->fl.fl_type,
474                                 (long long)lock->fl.fl_start,
475                                 (long long)lock->fl.fl_end);
476
477         /* Get existing block (in case client is busy-waiting) */
478         block = nlmsvc_lookup_block(file, lock);
479
480         if (block == NULL) {
481                 struct file_lock *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
482                 struct nlm_host *host;
483
484                 if (conf == NULL)
485                         return nlm_granted;
486                 /* Create host handle for callback */
487                 host = nlmsvc_lookup_host(rqstp, lock->caller, lock->len);
488                 if (host == NULL) {
489                         kfree(conf);
490                         return nlm_lck_denied_nolocks;
491                 }
492                 block = nlmsvc_create_block(rqstp, host, file, lock, cookie);
493                 if (block == NULL) {
494                         kfree(conf);
495                         return nlm_granted;
496                 }
497                 block->b_fl = conf;
498         }
499         if (block->b_flags & B_QUEUED) {
500                 dprintk("lockd: nlmsvc_testlock deferred block %p flags %d fl %p\n",
501                         block, block->b_flags, block->b_fl);
502                 if (block->b_flags & B_TIMED_OUT) {
503                         nlmsvc_unlink_block(block);
504                         return nlm_lck_denied;
505                 }
506                 if (block->b_flags & B_GOT_CALLBACK) {
507                         if (block->b_fl != NULL
508                                         && block->b_fl->fl_type != F_UNLCK) {
509                                 lock->fl = *block->b_fl;
510                                 goto conf_lock;
511                         }
512                         else {
513                                 nlmsvc_unlink_block(block);
514                                 return nlm_granted;
515                         }
516                 }
517                 return nlm_drop_reply;
518         }
519
520         error = vfs_test_lock(file->f_file, &lock->fl);
521         if (error == -EINPROGRESS)
522                 return nlmsvc_defer_lock_rqst(rqstp, block);
523         if (error) {
524                 ret = nlm_lck_denied_nolocks;
525                 goto out;
526         }
527         if (lock->fl.fl_type == F_UNLCK) {
528                 ret = nlm_granted;
529                 goto out;
530         }
531
532 conf_lock:
533         dprintk("lockd: conflicting lock(ty=%d, %Ld-%Ld)\n",
534                 lock->fl.fl_type, (long long)lock->fl.fl_start,
535                 (long long)lock->fl.fl_end);
536         conflock->caller = "somehost";  /* FIXME */
537         conflock->len = strlen(conflock->caller);
538         conflock->oh.len = 0;           /* don't return OH info */
539         conflock->svid = lock->fl.fl_pid;
540         conflock->fl.fl_type = lock->fl.fl_type;
541         conflock->fl.fl_start = lock->fl.fl_start;
542         conflock->fl.fl_end = lock->fl.fl_end;
543         ret = nlm_lck_denied;
544 out:
545         if (block)
546                 nlmsvc_release_block(block);
547         return ret;
548 }
549
550 /*
551  * Remove a lock.
552  * This implies a CANCEL call: We send a GRANT_MSG, the client replies
553  * with a GRANT_RES call which gets lost, and calls UNLOCK immediately
554  * afterwards. In this case the block will still be there, and hence
555  * must be removed.
556  */
557 __be32
558 nlmsvc_unlock(struct nlm_file *file, struct nlm_lock *lock)
559 {
560         int     error;
561
562         dprintk("lockd: nlmsvc_unlock(%s/%ld, pi=%d, %Ld-%Ld)\n",
563                                 file->f_file->f_path.dentry->d_inode->i_sb->s_id,
564                                 file->f_file->f_path.dentry->d_inode->i_ino,
565                                 lock->fl.fl_pid,
566                                 (long long)lock->fl.fl_start,
567                                 (long long)lock->fl.fl_end);
568
569         /* First, cancel any lock that might be there */
570         nlmsvc_cancel_blocked(file, lock);
571
572         lock->fl.fl_type = F_UNLCK;
573         error = vfs_lock_file(file->f_file, F_SETLK, &lock->fl, NULL);
574
575         return (error < 0)? nlm_lck_denied_nolocks : nlm_granted;
576 }
577
578 /*
579  * Cancel a previously blocked request.
580  *
581  * A cancel request always overrides any grant that may currently
582  * be in progress.
583  * The calling procedure must check whether the file can be closed.
584  */
585 __be32
586 nlmsvc_cancel_blocked(struct nlm_file *file, struct nlm_lock *lock)
587 {
588         struct nlm_block        *block;
589         int status = 0;
590
591         dprintk("lockd: nlmsvc_cancel(%s/%ld, pi=%d, %Ld-%Ld)\n",
592                                 file->f_file->f_path.dentry->d_inode->i_sb->s_id,
593                                 file->f_file->f_path.dentry->d_inode->i_ino,
594                                 lock->fl.fl_pid,
595                                 (long long)lock->fl.fl_start,
596                                 (long long)lock->fl.fl_end);
597
598         mutex_lock(&file->f_mutex);
599         block = nlmsvc_lookup_block(file, lock);
600         mutex_unlock(&file->f_mutex);
601         if (block != NULL) {
602                 vfs_cancel_lock(block->b_file->f_file,
603                                 &block->b_call->a_args.lock.fl);
604                 status = nlmsvc_unlink_block(block);
605                 nlmsvc_release_block(block);
606         }
607         return status ? nlm_lck_denied : nlm_granted;
608 }
609
610 /*
611  * This is a callback from the filesystem for VFS file lock requests.
612  * It will be used if fl_grant is defined and the filesystem can not
613  * respond to the request immediately.
614  * For GETLK request it will copy the reply to the nlm_block.
615  * For SETLK or SETLKW request it will get the local posix lock.
616  * In all cases it will move the block to the head of nlm_blocked q where
617  * nlmsvc_retry_blocked() can send back a reply for SETLKW or revisit the
618  * deferred rpc for GETLK and SETLK.
619  */
620 static void
621 nlmsvc_update_deferred_block(struct nlm_block *block, struct file_lock *conf,
622                              int result)
623 {
624         block->b_flags |= B_GOT_CALLBACK;
625         if (result == 0)
626                 block->b_granted = 1;
627         else
628                 block->b_flags |= B_TIMED_OUT;
629         if (conf) {
630                 if (block->b_fl)
631                         locks_copy_lock(block->b_fl, conf);
632         }
633 }
634
635 static int nlmsvc_grant_deferred(struct file_lock *fl, struct file_lock *conf,
636                                         int result)
637 {
638         struct nlm_block *block;
639         int rc = -ENOENT;
640
641         lock_kernel();
642         list_for_each_entry(block, &nlm_blocked, b_list) {
643                 if (nlm_compare_locks(&block->b_call->a_args.lock.fl, fl)) {
644                         dprintk("lockd: nlmsvc_notify_blocked block %p flags %d\n",
645                                                         block, block->b_flags);
646                         if (block->b_flags & B_QUEUED) {
647                                 if (block->b_flags & B_TIMED_OUT) {
648                                         rc = -ENOLCK;
649                                         break;
650                                 }
651                                 nlmsvc_update_deferred_block(block, conf, result);
652                         } else if (result == 0)
653                                 block->b_granted = 1;
654
655                         nlmsvc_insert_block(block, 0);
656                         svc_wake_up(block->b_daemon);
657                         rc = 0;
658                         break;
659                 }
660         }
661         unlock_kernel();
662         if (rc == -ENOENT)
663                 printk(KERN_WARNING "lockd: grant for unknown block\n");
664         return rc;
665 }
666
667 /*
668  * Unblock a blocked lock request. This is a callback invoked from the
669  * VFS layer when a lock on which we blocked is removed.
670  *
671  * This function doesn't grant the blocked lock instantly, but rather moves
672  * the block to the head of nlm_blocked where it can be picked up by lockd.
673  */
674 static void
675 nlmsvc_notify_blocked(struct file_lock *fl)
676 {
677         struct nlm_block        *block;
678
679         dprintk("lockd: VFS unblock notification for block %p\n", fl);
680         list_for_each_entry(block, &nlm_blocked, b_list) {
681                 if (nlm_compare_locks(&block->b_call->a_args.lock.fl, fl)) {
682                         nlmsvc_insert_block(block, 0);
683                         svc_wake_up(block->b_daemon);
684                         return;
685                 }
686         }
687
688         printk(KERN_WARNING "lockd: notification for unknown block!\n");
689 }
690
691 static int nlmsvc_same_owner(struct file_lock *fl1, struct file_lock *fl2)
692 {
693         return fl1->fl_owner == fl2->fl_owner && fl1->fl_pid == fl2->fl_pid;
694 }
695
696 struct lock_manager_operations nlmsvc_lock_operations = {
697         .fl_compare_owner = nlmsvc_same_owner,
698         .fl_notify = nlmsvc_notify_blocked,
699         .fl_grant = nlmsvc_grant_deferred,
700 };
701
702 /*
703  * Try to claim a lock that was previously blocked.
704  *
705  * Note that we use both the RPC_GRANTED_MSG call _and_ an async
706  * RPC thread when notifying the client. This seems like overkill...
707  * Here's why:
708  *  -   we don't want to use a synchronous RPC thread, otherwise
709  *      we might find ourselves hanging on a dead portmapper.
710  *  -   Some lockd implementations (e.g. HP) don't react to
711  *      RPC_GRANTED calls; they seem to insist on RPC_GRANTED_MSG calls.
712  */
713 static void
714 nlmsvc_grant_blocked(struct nlm_block *block)
715 {
716         struct nlm_file         *file = block->b_file;
717         struct nlm_lock         *lock = &block->b_call->a_args.lock;
718         int                     error;
719
720         dprintk("lockd: grant blocked lock %p\n", block);
721
722         kref_get(&block->b_count);
723
724         /* Unlink block request from list */
725         nlmsvc_unlink_block(block);
726
727         /* If b_granted is true this means we've been here before.
728          * Just retry the grant callback, possibly refreshing the RPC
729          * binding */
730         if (block->b_granted) {
731                 nlm_rebind_host(block->b_host);
732                 goto callback;
733         }
734
735         /* Try the lock operation again */
736         lock->fl.fl_flags |= FL_SLEEP;
737         error = vfs_lock_file(file->f_file, F_SETLK, &lock->fl, NULL);
738         lock->fl.fl_flags &= ~FL_SLEEP;
739
740         switch (error) {
741         case 0:
742                 break;
743         case -EAGAIN:
744         case -EINPROGRESS:
745                 dprintk("lockd: lock still blocked error %d\n", error);
746                 nlmsvc_insert_block(block, NLM_NEVER);
747                 nlmsvc_release_block(block);
748                 return;
749         default:
750                 printk(KERN_WARNING "lockd: unexpected error %d in %s!\n",
751                                 -error, __FUNCTION__);
752                 nlmsvc_insert_block(block, 10 * HZ);
753                 nlmsvc_release_block(block);
754                 return;
755         }
756
757 callback:
758         /* Lock was granted by VFS. */
759         dprintk("lockd: GRANTing blocked lock.\n");
760         block->b_granted = 1;
761
762         /* Schedule next grant callback in 30 seconds */
763         nlmsvc_insert_block(block, 30 * HZ);
764
765         /* Call the client */
766         nlm_async_call(block->b_call, NLMPROC_GRANTED_MSG, &nlmsvc_grant_ops);
767 }
768
769 /*
770  * This is the callback from the RPC layer when the NLM_GRANTED_MSG
771  * RPC call has succeeded or timed out.
772  * Like all RPC callbacks, it is invoked by the rpciod process, so it
773  * better not sleep. Therefore, we put the blocked lock on the nlm_blocked
774  * chain once more in order to have it removed by lockd itself (which can
775  * then sleep on the file semaphore without disrupting e.g. the nfs client).
776  */
777 static void nlmsvc_grant_callback(struct rpc_task *task, void *data)
778 {
779         struct nlm_rqst         *call = data;
780         struct nlm_block        *block = call->a_block;
781         unsigned long           timeout;
782
783         dprintk("lockd: GRANT_MSG RPC callback\n");
784
785         /* Technically, we should down the file semaphore here. Since we
786          * move the block towards the head of the queue only, no harm
787          * can be done, though. */
788         if (task->tk_status < 0) {
789                 /* RPC error: Re-insert for retransmission */
790                 timeout = 10 * HZ;
791         } else {
792                 /* Call was successful, now wait for client callback */
793                 timeout = 60 * HZ;
794         }
795         nlmsvc_insert_block(block, timeout);
796         svc_wake_up(block->b_daemon);
797 }
798
799 static void nlmsvc_grant_release(void *data)
800 {
801         struct nlm_rqst         *call = data;
802
803         nlmsvc_release_block(call->a_block);
804 }
805
806 static const struct rpc_call_ops nlmsvc_grant_ops = {
807         .rpc_call_done = nlmsvc_grant_callback,
808         .rpc_release = nlmsvc_grant_release,
809 };
810
811 /*
812  * We received a GRANT_RES callback. Try to find the corresponding
813  * block.
814  */
815 void
816 nlmsvc_grant_reply(struct nlm_cookie *cookie, __be32 status)
817 {
818         struct nlm_block        *block;
819
820         dprintk("grant_reply: looking for cookie %x, s=%d \n",
821                 *(unsigned int *)(cookie->data), status);
822         if (!(block = nlmsvc_find_block(cookie)))
823                 return;
824
825         if (block) {
826                 if (status == nlm_lck_denied_grace_period) {
827                         /* Try again in a couple of seconds */
828                         nlmsvc_insert_block(block, 10 * HZ);
829                 } else {
830                         /* Lock is now held by client, or has been rejected.
831                          * In both cases, the block should be removed. */
832                         nlmsvc_unlink_block(block);
833                 }
834         }
835         nlmsvc_release_block(block);
836 }
837
838 /* Helper function to handle retry of a deferred block.
839  * If it is a blocking lock, call grant_blocked.
840  * For a non-blocking lock or test lock, revisit the request.
841  */
842 static void
843 retry_deferred_block(struct nlm_block *block)
844 {
845         if (!(block->b_flags & B_GOT_CALLBACK))
846                 block->b_flags |= B_TIMED_OUT;
847         nlmsvc_insert_block(block, NLM_TIMEOUT);
848         dprintk("revisit block %p flags %d\n",  block, block->b_flags);
849         if (block->b_deferred_req) {
850                 block->b_deferred_req->revisit(block->b_deferred_req, 0);
851                 block->b_deferred_req = NULL;
852         }
853 }
854
855 /*
856  * Retry all blocked locks that have been notified. This is where lockd
857  * picks up locks that can be granted, or grant notifications that must
858  * be retransmitted.
859  */
860 unsigned long
861 nlmsvc_retry_blocked(void)
862 {
863         unsigned long   timeout = MAX_SCHEDULE_TIMEOUT;
864         struct nlm_block *block;
865
866         while (!list_empty(&nlm_blocked)) {
867                 block = list_entry(nlm_blocked.next, struct nlm_block, b_list);
868
869                 if (block->b_when == NLM_NEVER)
870                         break;
871                 if (time_after(block->b_when,jiffies)) {
872                         timeout = block->b_when - jiffies;
873                         break;
874                 }
875
876                 dprintk("nlmsvc_retry_blocked(%p, when=%ld)\n",
877                         block, block->b_when);
878                 if (block->b_flags & B_QUEUED) {
879                         dprintk("nlmsvc_retry_blocked delete block (%p, granted=%d, flags=%d)\n",
880                                 block, block->b_granted, block->b_flags);
881                         retry_deferred_block(block);
882                 } else
883                         nlmsvc_grant_blocked(block);
884         }
885
886         return timeout;
887 }