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/slab.h>
 25#include <linux/errno.h>
 26#include <linux/kernel.h>
 27#include <linux/sched.h>
 28#include <linux/sunrpc/clnt.h>
 29#include <linux/sunrpc/svc_xprt.h>
 30#include <linux/lockd/nlm.h>
 31#include <linux/lockd/lockd.h>
 32#include <linux/kthread.h>
 33
 34#define NLMDBG_FACILITY		NLMDBG_SVCLOCK
 35
 36#ifdef CONFIG_LOCKD_V4
 37#define nlm_deadlock	nlm4_deadlock
 38#else
 39#define nlm_deadlock	nlm_lck_denied
 40#endif
 41
 42static void nlmsvc_release_block(struct nlm_block *block);
 43static void	nlmsvc_insert_block(struct nlm_block *block, unsigned long);
 44static void	nlmsvc_remove_block(struct nlm_block *block);
 45
 46static int nlmsvc_setgrantargs(struct nlm_rqst *call, struct nlm_lock *lock);
 47static void nlmsvc_freegrantargs(struct nlm_rqst *call);
 48static const struct rpc_call_ops nlmsvc_grant_ops;
 49
 50/*
 51 * The list of blocked locks to retry
 52 */
 53static LIST_HEAD(nlm_blocked);
 54static DEFINE_SPINLOCK(nlm_blocked_lock);
 55
 56#ifdef LOCKD_DEBUG
 57static const char *nlmdbg_cookie2a(const struct nlm_cookie *cookie)
 58{
 59	/*
 60	 * We can get away with a static buffer because we're only
 61	 * called with BKL held.
 62	 */
 63	static char buf[2*NLM_MAXCOOKIELEN+1];
 64	unsigned int i, len = sizeof(buf);
 65	char *p = buf;
 66
 67	len--;	/* allow for trailing \0 */
 68	if (len < 3)
 69		return "???";
 70	for (i = 0 ; i < cookie->len ; i++) {
 71		if (len < 2) {
 72			strcpy(p-3, "...");
 73			break;
 74		}
 75		sprintf(p, "%02x", cookie->data[i]);
 76		p += 2;
 77		len -= 2;
 78	}
 79	*p = '\0';
 80
 81	return buf;
 82}
 83#endif
 84
 85/*
 86 * Insert a blocked lock into the global list
 87 */
 88static void
 89nlmsvc_insert_block_locked(struct nlm_block *block, unsigned long when)
 90{
 91	struct nlm_block *b;
 92	struct list_head *pos;
 93
 94	dprintk("lockd: nlmsvc_insert_block(%p, %ld)\n", block, when);
 95	if (list_empty(&block->b_list)) {
 96		kref_get(&block->b_count);
 97	} else {
 98		list_del_init(&block->b_list);
 99	}
100
101	pos = &nlm_blocked;
102	if (when != NLM_NEVER) {
103		if ((when += jiffies) == NLM_NEVER)
104			when ++;
105		list_for_each(pos, &nlm_blocked) {
106			b = list_entry(pos, struct nlm_block, b_list);
107			if (time_after(b->b_when,when) || b->b_when == NLM_NEVER)
108				break;
109		}
110		/* On normal exit from the loop, pos == &nlm_blocked,
111		 * so we will be adding to the end of the list - good
112		 */
113	}
114
115	list_add_tail(&block->b_list, pos);
116	block->b_when = when;
117}
118
119static void nlmsvc_insert_block(struct nlm_block *block, unsigned long when)
120{
121	spin_lock(&nlm_blocked_lock);
122	nlmsvc_insert_block_locked(block, when);
123	spin_unlock(&nlm_blocked_lock);
124}
125
126/*
127 * Remove a block from the global list
128 */
129static inline void
130nlmsvc_remove_block(struct nlm_block *block)
131{
132	if (!list_empty(&block->b_list)) {
133		spin_lock(&nlm_blocked_lock);
134		list_del_init(&block->b_list);
135		spin_unlock(&nlm_blocked_lock);
136		nlmsvc_release_block(block);
137	}
138}
139
140/*
141 * Find a block for a given lock
142 */
143static struct nlm_block *
144nlmsvc_lookup_block(struct nlm_file *file, struct nlm_lock *lock)
145{
146	struct nlm_block	*block;
147	struct file_lock	*fl;
148
149	dprintk("lockd: nlmsvc_lookup_block f=%p pd=%d %Ld-%Ld ty=%d\n",
150				file, lock->fl.fl_pid,
151				(long long)lock->fl.fl_start,
152				(long long)lock->fl.fl_end, lock->fl.fl_type);
153	list_for_each_entry(block, &nlm_blocked, b_list) {
154		fl = &block->b_call->a_args.lock.fl;
155		dprintk("lockd: check f=%p pd=%d %Ld-%Ld ty=%d cookie=%s\n",
156				block->b_file, fl->fl_pid,
157				(long long)fl->fl_start,
158				(long long)fl->fl_end, fl->fl_type,
159				nlmdbg_cookie2a(&block->b_call->a_args.cookie));
160		if (block->b_file == file && nlm_compare_locks(fl, &lock->fl)) {
161			kref_get(&block->b_count);
162			return block;
163		}
164	}
165
166	return NULL;
167}
168
169static inline int nlm_cookie_match(struct nlm_cookie *a, struct nlm_cookie *b)
170{
171	if (a->len != b->len)
172		return 0;
173	if (memcmp(a->data, b->data, a->len))
174		return 0;
175	return 1;
176}
177
178/*
179 * Find a block with a given NLM cookie.
180 */
181static inline struct nlm_block *
182nlmsvc_find_block(struct nlm_cookie *cookie)
183{
184	struct nlm_block *block;
185
186	list_for_each_entry(block, &nlm_blocked, b_list) {
187		if (nlm_cookie_match(&block->b_call->a_args.cookie,cookie))
188			goto found;
189	}
190
191	return NULL;
192
193found:
194	dprintk("nlmsvc_find_block(%s): block=%p\n", nlmdbg_cookie2a(cookie), block);
195	kref_get(&block->b_count);
196	return block;
197}
198
199/*
200 * Create a block and initialize it.
201 *
202 * Note: we explicitly set the cookie of the grant reply to that of
203 * the blocked lock request. The spec explicitly mentions that the client
204 * should _not_ rely on the callback containing the same cookie as the
205 * request, but (as I found out later) that's because some implementations
206 * do just this. Never mind the standards comittees, they support our
207 * logging industries.
208 *
209 * 10 years later: I hope we can safely ignore these old and broken
210 * clients by now. Let's fix this so we can uniquely identify an incoming
211 * GRANTED_RES message by cookie, without having to rely on the client's IP
212 * address. --okir
213 */
214static struct nlm_block *
215nlmsvc_create_block(struct svc_rqst *rqstp, struct nlm_host *host,
216		    struct nlm_file *file, struct nlm_lock *lock,
217		    struct nlm_cookie *cookie)
218{
219	struct nlm_block	*block;
220	struct nlm_rqst		*call = NULL;
221
 
222	call = nlm_alloc_call(host);
223	if (call == NULL)
224		return NULL;
225
226	/* Allocate memory for block, and initialize arguments */
227	block = kzalloc(sizeof(*block), GFP_KERNEL);
228	if (block == NULL)
229		goto failed;
230	kref_init(&block->b_count);
231	INIT_LIST_HEAD(&block->b_list);
232	INIT_LIST_HEAD(&block->b_flist);
233
234	if (!nlmsvc_setgrantargs(call, lock))
235		goto failed_free;
236
237	/* Set notifier function for VFS, and init args */
238	call->a_args.lock.fl.fl_flags |= FL_SLEEP;
239	call->a_args.lock.fl.fl_lmops = &nlmsvc_lock_operations;
240	nlmclnt_next_cookie(&call->a_args.cookie);
241
242	dprintk("lockd: created block %p...\n", block);
243
244	/* Create and initialize the block */
245	block->b_daemon = rqstp->rq_server;
246	block->b_host   = host;
247	block->b_file   = file;
248	block->b_fl = NULL;
249	file->f_count++;
250
251	/* Add to file's list of blocks */
252	list_add(&block->b_flist, &file->f_blocks);
253
254	/* Set up RPC arguments for callback */
255	block->b_call = call;
256	call->a_flags   = RPC_TASK_ASYNC;
257	call->a_block = block;
258
259	return block;
260
261failed_free:
262	kfree(block);
263failed:
264	nlmsvc_release_call(call);
265	return NULL;
266}
267
268/*
269 * Delete a block.
270 * It is the caller's responsibility to check whether the file
271 * can be closed hereafter.
272 */
273static int nlmsvc_unlink_block(struct nlm_block *block)
274{
275	int status;
276	dprintk("lockd: unlinking block %p...\n", block);
277
278	/* Remove block from list */
279	status = posix_unblock_lock(&block->b_call->a_args.lock.fl);
280	nlmsvc_remove_block(block);
281	return status;
282}
283
284static void nlmsvc_free_block(struct kref *kref)
285{
286	struct nlm_block *block = container_of(kref, struct nlm_block, b_count);
287	struct nlm_file		*file = block->b_file;
288
289	dprintk("lockd: freeing block %p...\n", block);
290
291	/* Remove block from file's list of blocks */
 
292	list_del_init(&block->b_flist);
293	mutex_unlock(&file->f_mutex);
294
295	nlmsvc_freegrantargs(block->b_call);
296	nlmsvc_release_call(block->b_call);
297	nlm_release_file(block->b_file);
298	kfree(block->b_fl);
299	kfree(block);
300}
301
302static void nlmsvc_release_block(struct nlm_block *block)
303{
304	if (block != NULL)
305		kref_put_mutex(&block->b_count, nlmsvc_free_block, &block->b_file->f_mutex);
306}
307
308/*
309 * Loop over all blocks and delete blocks held by
310 * a matching host.
311 */
312void nlmsvc_traverse_blocks(struct nlm_host *host,
313			struct nlm_file *file,
314			nlm_host_match_fn_t match)
315{
316	struct nlm_block *block, *next;
317
318restart:
319	mutex_lock(&file->f_mutex);
320	list_for_each_entry_safe(block, next, &file->f_blocks, b_flist) {
321		if (!match(block->b_host, host))
322			continue;
323		/* Do not destroy blocks that are not on
324		 * the global retry list - why? */
325		if (list_empty(&block->b_list))
326			continue;
327		kref_get(&block->b_count);
328		mutex_unlock(&file->f_mutex);
329		nlmsvc_unlink_block(block);
330		nlmsvc_release_block(block);
331		goto restart;
332	}
333	mutex_unlock(&file->f_mutex);
334}
335
336/*
337 * Initialize arguments for GRANTED call. The nlm_rqst structure
338 * has been cleared already.
339 */
340static int nlmsvc_setgrantargs(struct nlm_rqst *call, struct nlm_lock *lock)
341{
342	locks_copy_lock(&call->a_args.lock.fl, &lock->fl);
343	memcpy(&call->a_args.lock.fh, &lock->fh, sizeof(call->a_args.lock.fh));
344	call->a_args.lock.caller = utsname()->nodename;
345	call->a_args.lock.oh.len = lock->oh.len;
346
347	/* set default data area */
348	call->a_args.lock.oh.data = call->a_owner;
349	call->a_args.lock.svid = lock->fl.fl_pid;
350
351	if (lock->oh.len > NLMCLNT_OHSIZE) {
352		void *data = kmalloc(lock->oh.len, GFP_KERNEL);
353		if (!data)
354			return 0;
355		call->a_args.lock.oh.data = (u8 *) data;
356	}
357
358	memcpy(call->a_args.lock.oh.data, lock->oh.data, lock->oh.len);
359	return 1;
360}
361
362static void nlmsvc_freegrantargs(struct nlm_rqst *call)
363{
364	if (call->a_args.lock.oh.data != call->a_owner)
365		kfree(call->a_args.lock.oh.data);
366
367	locks_release_private(&call->a_args.lock.fl);
368}
369
370/*
371 * Deferred lock request handling for non-blocking lock
372 */
373static __be32
374nlmsvc_defer_lock_rqst(struct svc_rqst *rqstp, struct nlm_block *block)
375{
376	__be32 status = nlm_lck_denied_nolocks;
377
378	block->b_flags |= B_QUEUED;
379
380	nlmsvc_insert_block(block, NLM_TIMEOUT);
381
382	block->b_cache_req = &rqstp->rq_chandle;
383	if (rqstp->rq_chandle.defer) {
384		block->b_deferred_req =
385			rqstp->rq_chandle.defer(block->b_cache_req);
386		if (block->b_deferred_req != NULL)
387			status = nlm_drop_reply;
388	}
389	dprintk("lockd: nlmsvc_defer_lock_rqst block %p flags %d status %d\n",
390		block, block->b_flags, ntohl(status));
391
392	return status;
393}
394
395/*
396 * Attempt to establish a lock, and if it can't be granted, block it
397 * if required.
398 */
399__be32
400nlmsvc_lock(struct svc_rqst *rqstp, struct nlm_file *file,
401	    struct nlm_host *host, struct nlm_lock *lock, int wait,
402	    struct nlm_cookie *cookie, int reclaim)
403{
404	struct nlm_block	*block = NULL;
405	int			error;
406	__be32			ret;
407
408	dprintk("lockd: nlmsvc_lock(%s/%ld, ty=%d, pi=%d, %Ld-%Ld, bl=%d)\n",
409				file_inode(file->f_file)->i_sb->s_id,
410				file_inode(file->f_file)->i_ino,
411				lock->fl.fl_type, lock->fl.fl_pid,
412				(long long)lock->fl.fl_start,
413				(long long)lock->fl.fl_end,
414				wait);
415
416	/* Lock file against concurrent access */
417	mutex_lock(&file->f_mutex);
418	/* Get existing block (in case client is busy-waiting)
419	 * or create new block
420	 */
421	block = nlmsvc_lookup_block(file, lock);
422	if (block == NULL) {
423		block = nlmsvc_create_block(rqstp, host, file, lock, cookie);
424		ret = nlm_lck_denied_nolocks;
425		if (block == NULL)
426			goto out;
427		lock = &block->b_call->a_args.lock;
428	} else
429		lock->fl.fl_flags &= ~FL_SLEEP;
430
431	if (block->b_flags & B_QUEUED) {
432		dprintk("lockd: nlmsvc_lock deferred block %p flags %d\n",
433							block, block->b_flags);
434		if (block->b_granted) {
435			nlmsvc_unlink_block(block);
436			ret = nlm_granted;
437			goto out;
438		}
439		if (block->b_flags & B_TIMED_OUT) {
440			nlmsvc_unlink_block(block);
441			ret = nlm_lck_denied;
442			goto out;
443		}
444		ret = nlm_drop_reply;
445		goto out;
446	}
447
448	if (locks_in_grace(SVC_NET(rqstp)) && !reclaim) {
449		ret = nlm_lck_denied_grace_period;
450		goto out;
451	}
452	if (reclaim && !locks_in_grace(SVC_NET(rqstp))) {
453		ret = nlm_lck_denied_grace_period;
454		goto out;
455	}
456
457	if (!wait)
458		lock->fl.fl_flags &= ~FL_SLEEP;
459	error = vfs_lock_file(file->f_file, F_SETLK, &lock->fl, NULL);
460	lock->fl.fl_flags &= ~FL_SLEEP;
461
462	dprintk("lockd: vfs_lock_file returned %d\n", error);
463	switch (error) {
464		case 0:
465			ret = nlm_granted;
466			goto out;
467		case -EAGAIN:
468			/*
469			 * If this is a blocking request for an
470			 * already pending lock request then we need
471			 * to put it back on lockd's block list
472			 */
473			if (wait)
474				break;
475			ret = nlm_lck_denied;
476			goto out;
477		case FILE_LOCK_DEFERRED:
478			if (wait)
479				break;
480			/* Filesystem lock operation is in progress
481			   Add it to the queue waiting for callback */
482			ret = nlmsvc_defer_lock_rqst(rqstp, block);
483			goto out;
484		case -EDEADLK:
485			ret = nlm_deadlock;
486			goto out;
487		default:			/* includes ENOLCK */
488			ret = nlm_lck_denied_nolocks;
489			goto out;
490	}
491
492	ret = nlm_lck_blocked;
493
494	/* Append to list of blocked */
495	nlmsvc_insert_block(block, NLM_NEVER);
496out:
497	mutex_unlock(&file->f_mutex);
498	nlmsvc_release_block(block);
499	dprintk("lockd: nlmsvc_lock returned %u\n", ret);
500	return ret;
501}
502
503/*
504 * Test for presence of a conflicting lock.
505 */
506__be32
507nlmsvc_testlock(struct svc_rqst *rqstp, struct nlm_file *file,
508		struct nlm_host *host, struct nlm_lock *lock,
509		struct nlm_lock *conflock, struct nlm_cookie *cookie)
510{
511	struct nlm_block 	*block = NULL;
512	int			error;
513	__be32			ret;
514
515	dprintk("lockd: nlmsvc_testlock(%s/%ld, ty=%d, %Ld-%Ld)\n",
516				file_inode(file->f_file)->i_sb->s_id,
517				file_inode(file->f_file)->i_ino,
518				lock->fl.fl_type,
519				(long long)lock->fl.fl_start,
520				(long long)lock->fl.fl_end);
521
522	/* Get existing block (in case client is busy-waiting) */
523	block = nlmsvc_lookup_block(file, lock);
524
525	if (block == NULL) {
526		struct file_lock *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
527
528		if (conf == NULL)
529			return nlm_granted;
530		block = nlmsvc_create_block(rqstp, host, file, lock, cookie);
531		if (block == NULL) {
532			kfree(conf);
533			return nlm_granted;
534		}
535		block->b_fl = conf;
536	}
537	if (block->b_flags & B_QUEUED) {
538		dprintk("lockd: nlmsvc_testlock deferred block %p flags %d fl %p\n",
539			block, block->b_flags, block->b_fl);
540		if (block->b_flags & B_TIMED_OUT) {
541			nlmsvc_unlink_block(block);
542			ret = nlm_lck_denied;
543			goto out;
544		}
545		if (block->b_flags & B_GOT_CALLBACK) {
546			nlmsvc_unlink_block(block);
547			if (block->b_fl != NULL
548					&& block->b_fl->fl_type != F_UNLCK) {
549				lock->fl = *block->b_fl;
550				goto conf_lock;
551			} else {
552				ret = nlm_granted;
553				goto out;
554			}
555		}
556		ret = nlm_drop_reply;
557		goto out;
558	}
559
560	if (locks_in_grace(SVC_NET(rqstp))) {
561		ret = nlm_lck_denied_grace_period;
562		goto out;
563	}
564	error = vfs_test_lock(file->f_file, &lock->fl);
565	if (error == FILE_LOCK_DEFERRED) {
566		ret = nlmsvc_defer_lock_rqst(rqstp, block);
567		goto out;
568	}
569	if (error) {
570		ret = nlm_lck_denied_nolocks;
571		goto out;
572	}
573	if (lock->fl.fl_type == F_UNLCK) {
574		ret = nlm_granted;
575		goto out;
576	}
577
578conf_lock:
579	dprintk("lockd: conflicting lock(ty=%d, %Ld-%Ld)\n",
580		lock->fl.fl_type, (long long)lock->fl.fl_start,
581		(long long)lock->fl.fl_end);
582	conflock->caller = "somehost";	/* FIXME */
583	conflock->len = strlen(conflock->caller);
584	conflock->oh.len = 0;		/* don't return OH info */
585	conflock->svid = lock->fl.fl_pid;
586	conflock->fl.fl_type = lock->fl.fl_type;
587	conflock->fl.fl_start = lock->fl.fl_start;
588	conflock->fl.fl_end = lock->fl.fl_end;
589	ret = nlm_lck_denied;
590out:
591	if (block)
592		nlmsvc_release_block(block);
593	return ret;
594}
595
596/*
597 * Remove a lock.
598 * This implies a CANCEL call: We send a GRANT_MSG, the client replies
599 * with a GRANT_RES call which gets lost, and calls UNLOCK immediately
600 * afterwards. In this case the block will still be there, and hence
601 * must be removed.
602 */
603__be32
604nlmsvc_unlock(struct net *net, struct nlm_file *file, struct nlm_lock *lock)
605{
606	int	error;
607
608	dprintk("lockd: nlmsvc_unlock(%s/%ld, pi=%d, %Ld-%Ld)\n",
609				file_inode(file->f_file)->i_sb->s_id,
610				file_inode(file->f_file)->i_ino,
611				lock->fl.fl_pid,
612				(long long)lock->fl.fl_start,
613				(long long)lock->fl.fl_end);
614
615	/* First, cancel any lock that might be there */
616	nlmsvc_cancel_blocked(net, file, lock);
617
618	lock->fl.fl_type = F_UNLCK;
619	error = vfs_lock_file(file->f_file, F_SETLK, &lock->fl, NULL);
620
621	return (error < 0)? nlm_lck_denied_nolocks : nlm_granted;
622}
623
624/*
625 * Cancel a previously blocked request.
626 *
627 * A cancel request always overrides any grant that may currently
628 * be in progress.
629 * The calling procedure must check whether the file can be closed.
630 */
631__be32
632nlmsvc_cancel_blocked(struct net *net, struct nlm_file *file, struct nlm_lock *lock)
633{
634	struct nlm_block	*block;
635	int status = 0;
636
637	dprintk("lockd: nlmsvc_cancel(%s/%ld, pi=%d, %Ld-%Ld)\n",
638				file_inode(file->f_file)->i_sb->s_id,
639				file_inode(file->f_file)->i_ino,
640				lock->fl.fl_pid,
641				(long long)lock->fl.fl_start,
642				(long long)lock->fl.fl_end);
643
644	if (locks_in_grace(net))
645		return nlm_lck_denied_grace_period;
646
647	mutex_lock(&file->f_mutex);
648	block = nlmsvc_lookup_block(file, lock);
649	mutex_unlock(&file->f_mutex);
650	if (block != NULL) {
651		vfs_cancel_lock(block->b_file->f_file,
652				&block->b_call->a_args.lock.fl);
653		status = nlmsvc_unlink_block(block);
654		nlmsvc_release_block(block);
655	}
656	return status ? nlm_lck_denied : nlm_granted;
657}
658
659/*
660 * This is a callback from the filesystem for VFS file lock requests.
661 * It will be used if lm_grant is defined and the filesystem can not
662 * respond to the request immediately.
663 * For GETLK request it will copy the reply to the nlm_block.
664 * For SETLK or SETLKW request it will get the local posix lock.
665 * In all cases it will move the block to the head of nlm_blocked q where
666 * nlmsvc_retry_blocked() can send back a reply for SETLKW or revisit the
667 * deferred rpc for GETLK and SETLK.
668 */
669static void
670nlmsvc_update_deferred_block(struct nlm_block *block, struct file_lock *conf,
671			     int result)
672{
673	block->b_flags |= B_GOT_CALLBACK;
674	if (result == 0)
675		block->b_granted = 1;
676	else
677		block->b_flags |= B_TIMED_OUT;
678	if (conf) {
679		if (block->b_fl)
680			__locks_copy_lock(block->b_fl, conf);
681	}
682}
683
684static int nlmsvc_grant_deferred(struct file_lock *fl, struct file_lock *conf,
685					int result)
686{
687	struct nlm_block *block;
688	int rc = -ENOENT;
689
690	spin_lock(&nlm_blocked_lock);
691	list_for_each_entry(block, &nlm_blocked, b_list) {
692		if (nlm_compare_locks(&block->b_call->a_args.lock.fl, fl)) {
693			dprintk("lockd: nlmsvc_notify_blocked block %p flags %d\n",
694							block, block->b_flags);
695			if (block->b_flags & B_QUEUED) {
696				if (block->b_flags & B_TIMED_OUT) {
697					rc = -ENOLCK;
698					break;
699				}
700				nlmsvc_update_deferred_block(block, conf, result);
701			} else if (result == 0)
702				block->b_granted = 1;
703
704			nlmsvc_insert_block_locked(block, 0);
705			svc_wake_up(block->b_daemon);
706			rc = 0;
707			break;
708		}
709	}
710	spin_unlock(&nlm_blocked_lock);
711	if (rc == -ENOENT)
712		printk(KERN_WARNING "lockd: grant for unknown block\n");
713	return rc;
714}
715
716/*
717 * Unblock a blocked lock request. This is a callback invoked from the
718 * VFS layer when a lock on which we blocked is removed.
719 *
720 * This function doesn't grant the blocked lock instantly, but rather moves
721 * the block to the head of nlm_blocked where it can be picked up by lockd.
722 */
723static void
724nlmsvc_notify_blocked(struct file_lock *fl)
725{
726	struct nlm_block	*block;
727
728	dprintk("lockd: VFS unblock notification for block %p\n", fl);
729	spin_lock(&nlm_blocked_lock);
730	list_for_each_entry(block, &nlm_blocked, b_list) {
731		if (nlm_compare_locks(&block->b_call->a_args.lock.fl, fl)) {
732			nlmsvc_insert_block_locked(block, 0);
733			spin_unlock(&nlm_blocked_lock);
734			svc_wake_up(block->b_daemon);
735			return;
736		}
737	}
738	spin_unlock(&nlm_blocked_lock);
739	printk(KERN_WARNING "lockd: notification for unknown block!\n");
740}
741
742static int nlmsvc_same_owner(struct file_lock *fl1, struct file_lock *fl2)
743{
744	return fl1->fl_owner == fl2->fl_owner && fl1->fl_pid == fl2->fl_pid;
745}
746
747/*
748 * Since NLM uses two "keys" for tracking locks, we need to hash them down
749 * to one for the blocked_hash. Here, we're just xor'ing the host address
750 * with the pid in order to create a key value for picking a hash bucket.
751 */
752static unsigned long
753nlmsvc_owner_key(struct file_lock *fl)
754{
755	return (unsigned long)fl->fl_owner ^ (unsigned long)fl->fl_pid;
756}
757
758const struct lock_manager_operations nlmsvc_lock_operations = {
759	.lm_compare_owner = nlmsvc_same_owner,
760	.lm_owner_key = nlmsvc_owner_key,
761	.lm_notify = nlmsvc_notify_blocked,
762	.lm_grant = nlmsvc_grant_deferred,
763};
764
765/*
766 * Try to claim a lock that was previously blocked.
767 *
768 * Note that we use both the RPC_GRANTED_MSG call _and_ an async
769 * RPC thread when notifying the client. This seems like overkill...
770 * Here's why:
771 *  -	we don't want to use a synchronous RPC thread, otherwise
772 *	we might find ourselves hanging on a dead portmapper.
773 *  -	Some lockd implementations (e.g. HP) don't react to
774 *	RPC_GRANTED calls; they seem to insist on RPC_GRANTED_MSG calls.
775 */
776static void
777nlmsvc_grant_blocked(struct nlm_block *block)
778{
779	struct nlm_file		*file = block->b_file;
780	struct nlm_lock		*lock = &block->b_call->a_args.lock;
781	int			error;
782	loff_t			fl_start, fl_end;
783
784	dprintk("lockd: grant blocked lock %p\n", block);
785
786	kref_get(&block->b_count);
787
788	/* Unlink block request from list */
789	nlmsvc_unlink_block(block);
790
791	/* If b_granted is true this means we've been here before.
792	 * Just retry the grant callback, possibly refreshing the RPC
793	 * binding */
794	if (block->b_granted) {
795		nlm_rebind_host(block->b_host);
796		goto callback;
797	}
798
799	/* Try the lock operation again */
800	/* vfs_lock_file() can mangle fl_start and fl_end, but we need
801	 * them unchanged for the GRANT_MSG
802	 */
803	lock->fl.fl_flags |= FL_SLEEP;
804	fl_start = lock->fl.fl_start;
805	fl_end = lock->fl.fl_end;
806	error = vfs_lock_file(file->f_file, F_SETLK, &lock->fl, NULL);
807	lock->fl.fl_flags &= ~FL_SLEEP;
808	lock->fl.fl_start = fl_start;
809	lock->fl.fl_end = fl_end;
810
811	switch (error) {
812	case 0:
813		break;
814	case FILE_LOCK_DEFERRED:
815		dprintk("lockd: lock still blocked error %d\n", error);
816		nlmsvc_insert_block(block, NLM_NEVER);
817		nlmsvc_release_block(block);
818		return;
819	default:
820		printk(KERN_WARNING "lockd: unexpected error %d in %s!\n",
821				-error, __func__);
822		nlmsvc_insert_block(block, 10 * HZ);
823		nlmsvc_release_block(block);
824		return;
825	}
826
827callback:
828	/* Lock was granted by VFS. */
829	dprintk("lockd: GRANTing blocked lock.\n");
830	block->b_granted = 1;
831
832	/* keep block on the list, but don't reattempt until the RPC
833	 * completes or the submission fails
834	 */
835	nlmsvc_insert_block(block, NLM_NEVER);
836
837	/* Call the client -- use a soft RPC task since nlmsvc_retry_blocked
838	 * will queue up a new one if this one times out
839	 */
840	error = nlm_async_call(block->b_call, NLMPROC_GRANTED_MSG,
841				&nlmsvc_grant_ops);
842
843	/* RPC submission failed, wait a bit and retry */
844	if (error < 0)
845		nlmsvc_insert_block(block, 10 * HZ);
846}
847
848/*
849 * This is the callback from the RPC layer when the NLM_GRANTED_MSG
850 * RPC call has succeeded or timed out.
851 * Like all RPC callbacks, it is invoked by the rpciod process, so it
852 * better not sleep. Therefore, we put the blocked lock on the nlm_blocked
853 * chain once more in order to have it removed by lockd itself (which can
854 * then sleep on the file semaphore without disrupting e.g. the nfs client).
855 */
856static void nlmsvc_grant_callback(struct rpc_task *task, void *data)
857{
858	struct nlm_rqst		*call = data;
859	struct nlm_block	*block = call->a_block;
860	unsigned long		timeout;
861
862	dprintk("lockd: GRANT_MSG RPC callback\n");
863
864	spin_lock(&nlm_blocked_lock);
865	/* if the block is not on a list at this point then it has
866	 * been invalidated. Don't try to requeue it.
867	 *
868	 * FIXME: it's possible that the block is removed from the list
869	 * after this check but before the nlmsvc_insert_block. In that
870	 * case it will be added back. Perhaps we need better locking
871	 * for nlm_blocked?
872	 */
873	if (list_empty(&block->b_list))
874		goto out;
875
876	/* Technically, we should down the file semaphore here. Since we
877	 * move the block towards the head of the queue only, no harm
878	 * can be done, though. */
879	if (task->tk_status < 0) {
880		/* RPC error: Re-insert for retransmission */
881		timeout = 10 * HZ;
882	} else {
883		/* Call was successful, now wait for client callback */
884		timeout = 60 * HZ;
885	}
886	nlmsvc_insert_block_locked(block, timeout);
887	svc_wake_up(block->b_daemon);
888out:
889	spin_unlock(&nlm_blocked_lock);
890}
891
892/*
893 * FIXME: nlmsvc_release_block() grabs a mutex.  This is not allowed for an
894 * .rpc_release rpc_call_op
895 */
896static void nlmsvc_grant_release(void *data)
897{
898	struct nlm_rqst		*call = data;
899	nlmsvc_release_block(call->a_block);
900}
901
902static const struct rpc_call_ops nlmsvc_grant_ops = {
903	.rpc_call_done = nlmsvc_grant_callback,
904	.rpc_release = nlmsvc_grant_release,
905};
906
907/*
908 * We received a GRANT_RES callback. Try to find the corresponding
909 * block.
910 */
911void
912nlmsvc_grant_reply(struct nlm_cookie *cookie, __be32 status)
913{
914	struct nlm_block	*block;
915
916	dprintk("grant_reply: looking for cookie %x, s=%d \n",
917		*(unsigned int *)(cookie->data), status);
918	if (!(block = nlmsvc_find_block(cookie)))
919		return;
920
921	if (block) {
922		if (status == nlm_lck_denied_grace_period) {
923			/* Try again in a couple of seconds */
924			nlmsvc_insert_block(block, 10 * HZ);
925		} else {
926			/* Lock is now held by client, or has been rejected.
927			 * In both cases, the block should be removed. */
928			nlmsvc_unlink_block(block);
929		}
930	}
931	nlmsvc_release_block(block);
932}
933
934/* Helper function to handle retry of a deferred block.
935 * If it is a blocking lock, call grant_blocked.
936 * For a non-blocking lock or test lock, revisit the request.
937 */
938static void
939retry_deferred_block(struct nlm_block *block)
940{
941	if (!(block->b_flags & B_GOT_CALLBACK))
942		block->b_flags |= B_TIMED_OUT;
943	nlmsvc_insert_block(block, NLM_TIMEOUT);
944	dprintk("revisit block %p flags %d\n",	block, block->b_flags);
945	if (block->b_deferred_req) {
946		block->b_deferred_req->revisit(block->b_deferred_req, 0);
947		block->b_deferred_req = NULL;
948	}
949}
950
951/*
952 * Retry all blocked locks that have been notified. This is where lockd
953 * picks up locks that can be granted, or grant notifications that must
954 * be retransmitted.
955 */
956unsigned long
957nlmsvc_retry_blocked(void)
958{
959	unsigned long	timeout = MAX_SCHEDULE_TIMEOUT;
960	struct nlm_block *block;
961
962	spin_lock(&nlm_blocked_lock);
963	while (!list_empty(&nlm_blocked) && !kthread_should_stop()) {
964		block = list_entry(nlm_blocked.next, struct nlm_block, b_list);
965
966		if (block->b_when == NLM_NEVER)
967			break;
968		if (time_after(block->b_when, jiffies)) {
969			timeout = block->b_when - jiffies;
970			break;
971		}
972		spin_unlock(&nlm_blocked_lock);
973
974		dprintk("nlmsvc_retry_blocked(%p, when=%ld)\n",
975			block, block->b_when);
976		if (block->b_flags & B_QUEUED) {
977			dprintk("nlmsvc_retry_blocked delete block (%p, granted=%d, flags=%d)\n",
978				block, block->b_granted, block->b_flags);
979			retry_deferred_block(block);
980		} else
981			nlmsvc_grant_blocked(block);
982		spin_lock(&nlm_blocked_lock);
983	}
984	spin_unlock(&nlm_blocked_lock);
985
986	return timeout;
987}