svclock.c 23.6 KB
Newer Older
L
Linus Torvalds 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40
/*
 * linux/fs/lockd/svclock.c
 *
 * Handling of server-side locks, mostly of the blocked variety.
 * This is the ugliest part of lockd because we tread on very thin ice.
 * GRANT and CANCEL calls may get stuck, meet in mid-flight, etc.
 * IMNSHO introducing the grant callback into the NLM protocol was one
 * of the worst ideas Sun ever had. Except maybe for the idea of doing
 * NFS file locking at all.
 *
 * I'm trying hard to avoid race conditions by protecting most accesses
 * to a file's list of blocked locks through a semaphore. The global
 * list of blocked locks is not protected in this fashion however.
 * Therefore, some functions (such as the RPC callback for the async grant
 * call) move blocked locks towards the head of the list *while some other
 * process might be traversing it*. This should not be a problem in
 * practice, because this will only cause functions traversing the list
 * to visit some blocks twice.
 *
 * Copyright (C) 1996, Olaf Kirch <okir@monad.swb.de>
 */

#include <linux/types.h>
#include <linux/errno.h>
#include <linux/kernel.h>
#include <linux/sched.h>
#include <linux/smp_lock.h>
#include <linux/sunrpc/clnt.h>
#include <linux/sunrpc/svc.h>
#include <linux/lockd/nlm.h>
#include <linux/lockd/lockd.h>

#define NLMDBG_FACILITY		NLMDBG_SVCLOCK

#ifdef CONFIG_LOCKD_V4
#define nlm_deadlock	nlm4_deadlock
#else
#define nlm_deadlock	nlm_lck_denied
#endif

41
static void nlmsvc_release_block(struct nlm_block *block);
L
Linus Torvalds 已提交
42
static void	nlmsvc_insert_block(struct nlm_block *block, unsigned long);
43
static void	nlmsvc_remove_block(struct nlm_block *block);
44

45 46
static int nlmsvc_setgrantargs(struct nlm_rqst *call, struct nlm_lock *lock);
static void nlmsvc_freegrantargs(struct nlm_rqst *call);
47
static const struct rpc_call_ops nlmsvc_grant_ops;
L
Linus Torvalds 已提交
48 49 50 51

/*
 * The list of blocked locks to retry
 */
52
static LIST_HEAD(nlm_blocked);
L
Linus Torvalds 已提交
53 54 55 56 57 58 59

/*
 * Insert a blocked lock into the global list
 */
static void
nlmsvc_insert_block(struct nlm_block *block, unsigned long when)
{
60 61
	struct nlm_block *b;
	struct list_head *pos;
L
Linus Torvalds 已提交
62 63

	dprintk("lockd: nlmsvc_insert_block(%p, %ld)\n", block, when);
64 65 66 67 68 69 70
	if (list_empty(&block->b_list)) {
		kref_get(&block->b_count);
	} else {
		list_del_init(&block->b_list);
	}

	pos = &nlm_blocked;
L
Linus Torvalds 已提交
71 72 73
	if (when != NLM_NEVER) {
		if ((when += jiffies) == NLM_NEVER)
			when ++;
74 75 76 77 78 79 80 81 82
		list_for_each(pos, &nlm_blocked) {
			b = list_entry(pos, struct nlm_block, b_list);
			if (time_after(b->b_when,when) || b->b_when == NLM_NEVER)
				break;
		}
		/* On normal exit from the loop, pos == &nlm_blocked,
		 * so we will be adding to the end of the list - good
		 */
	}
L
Linus Torvalds 已提交
83

84
	list_add_tail(&block->b_list, pos);
L
Linus Torvalds 已提交
85 86 87 88 89 90
	block->b_when = when;
}

/*
 * Remove a block from the global list
 */
91
static inline void
L
Linus Torvalds 已提交
92 93
nlmsvc_remove_block(struct nlm_block *block)
{
94 95 96
	if (!list_empty(&block->b_list)) {
		list_del_init(&block->b_list);
		nlmsvc_release_block(block);
L
Linus Torvalds 已提交
97 98 99 100
	}
}

/*
101
 * Find a block for a given lock
L
Linus Torvalds 已提交
102 103
 */
static struct nlm_block *
104
nlmsvc_lookup_block(struct nlm_file *file, struct nlm_lock *lock)
L
Linus Torvalds 已提交
105
{
106
	struct nlm_block	*block;
L
Linus Torvalds 已提交
107 108 109 110 111 112
	struct file_lock	*fl;

	dprintk("lockd: nlmsvc_lookup_block f=%p pd=%d %Ld-%Ld ty=%d\n",
				file, lock->fl.fl_pid,
				(long long)lock->fl.fl_start,
				(long long)lock->fl.fl_end, lock->fl.fl_type);
113
	list_for_each_entry(block, &nlm_blocked, b_list) {
T
Trond Myklebust 已提交
114
		fl = &block->b_call->a_args.lock.fl;
L
Linus Torvalds 已提交
115 116 117 118
		dprintk("lockd: check f=%p pd=%d %Ld-%Ld ty=%d cookie=%s\n",
				block->b_file, fl->fl_pid,
				(long long)fl->fl_start,
				(long long)fl->fl_end, fl->fl_type,
T
Trond Myklebust 已提交
119
				nlmdbg_cookie2a(&block->b_call->a_args.cookie));
L
Linus Torvalds 已提交
120
		if (block->b_file == file && nlm_compare_locks(fl, &lock->fl)) {
121
			kref_get(&block->b_count);
L
Linus Torvalds 已提交
122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141
			return block;
		}
	}

	return NULL;
}

static inline int nlm_cookie_match(struct nlm_cookie *a, struct nlm_cookie *b)
{
	if(a->len != b->len)
		return 0;
	if(memcmp(a->data,b->data,a->len))
		return 0;
	return 1;
}

/*
 * Find a block with a given NLM cookie.
 */
static inline struct nlm_block *
142
nlmsvc_find_block(struct nlm_cookie *cookie)
L
Linus Torvalds 已提交
143 144 145
{
	struct nlm_block *block;

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;
L
Linus Torvalds 已提交
149 150
	}

151 152 153
	return NULL;

found:
154
	dprintk("nlmsvc_find_block(%s): block=%p\n", nlmdbg_cookie2a(cookie), block);
155
	kref_get(&block->b_count);
L
Linus Torvalds 已提交
156 157 158 159 160 161 162 163 164 165 166 167
	return block;
}

/*
 * Create a block and initialize it.
 *
 * Note: we explicitly set the cookie of the grant reply to that of
 * the blocked lock request. The spec explicitly mentions that the client
 * should _not_ rely on the callback containing the same cookie as the
 * request, but (as I found out later) that's because some implementations
 * do just this. Never mind the standards comittees, they support our
 * logging industries.
168 169 170 171 172
 *
 * 10 years later: I hope we can safely ignore these old and broken
 * clients by now. Let's fix this so we can uniquely identify an incoming
 * GRANTED_RES message by cookie, without having to rely on the client's IP
 * address. --okir
L
Linus Torvalds 已提交
173
 */
174 175 176 177
static struct nlm_block *
nlmsvc_create_block(struct svc_rqst *rqstp, struct nlm_host *host,
		    struct nlm_file *file, struct nlm_lock *lock,
		    struct nlm_cookie *cookie)
L
Linus Torvalds 已提交
178 179
{
	struct nlm_block	*block;
T
Trond Myklebust 已提交
180
	struct nlm_rqst		*call = NULL;
L
Linus Torvalds 已提交
181

T
Trond Myklebust 已提交
182 183 184 185
	call = nlm_alloc_call(host);
	if (call == NULL)
		return NULL;

L
Linus Torvalds 已提交
186
	/* Allocate memory for block, and initialize arguments */
T
Trond Myklebust 已提交
187 188
	block = kzalloc(sizeof(*block), GFP_KERNEL);
	if (block == NULL)
L
Linus Torvalds 已提交
189
		goto failed;
190
	kref_init(&block->b_count);
191 192
	INIT_LIST_HEAD(&block->b_list);
	INIT_LIST_HEAD(&block->b_flist);
L
Linus Torvalds 已提交
193

T
Trond Myklebust 已提交
194
	if (!nlmsvc_setgrantargs(call, lock))
L
Linus Torvalds 已提交
195 196 197
		goto failed_free;

	/* Set notifier function for VFS, and init args */
T
Trond Myklebust 已提交
198 199
	call->a_args.lock.fl.fl_flags |= FL_SLEEP;
	call->a_args.lock.fl.fl_lmops = &nlmsvc_lock_operations;
200
	nlmclnt_next_cookie(&call->a_args.cookie);
L
Linus Torvalds 已提交
201 202 203 204 205 206 207

	dprintk("lockd: created block %p...\n", block);

	/* Create and initialize the block */
	block->b_daemon = rqstp->rq_server;
	block->b_host   = host;
	block->b_file   = file;
M
Marc Eshel 已提交
208
	block->b_fl = NULL;
209
	file->f_count++;
L
Linus Torvalds 已提交
210 211

	/* Add to file's list of blocks */
212
	list_add(&block->b_flist, &file->f_blocks);
L
Linus Torvalds 已提交
213 214

	/* Set up RPC arguments for callback */
T
Trond Myklebust 已提交
215
	block->b_call = call;
L
Linus Torvalds 已提交
216
	call->a_flags   = RPC_TASK_ASYNC;
T
Trond Myklebust 已提交
217
	call->a_block = block;
L
Linus Torvalds 已提交
218 219 220 221 222 223

	return block;

failed_free:
	kfree(block);
failed:
T
Trond Myklebust 已提交
224
	nlm_release_call(call);
L
Linus Torvalds 已提交
225 226 227 228 229 230 231 232 233
	return NULL;
}

/*
 * Delete a block. If the lock was cancelled or the grant callback
 * failed, unlock is set to 1.
 * It is the caller's responsibility to check whether the file
 * can be closed hereafter.
 */
234
static int nlmsvc_unlink_block(struct nlm_block *block)
L
Linus Torvalds 已提交
235
{
236
	int status;
237
	dprintk("lockd: unlinking block %p...\n", block);
L
Linus Torvalds 已提交
238 239

	/* Remove block from list */
T
Trond Myklebust 已提交
240
	status = posix_unblock_lock(block->b_file->f_file, &block->b_call->a_args.lock.fl);
L
Linus Torvalds 已提交
241
	nlmsvc_remove_block(block);
242 243
	return status;
}
L
Linus Torvalds 已提交
244

245 246 247 248 249 250
static void nlmsvc_free_block(struct kref *kref)
{
	struct nlm_block *block = container_of(kref, struct nlm_block, b_count);
	struct nlm_file		*file = block->b_file;

	dprintk("lockd: freeing block %p...\n", block);
L
Linus Torvalds 已提交
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);
L
Linus Torvalds 已提交
256

T
Trond Myklebust 已提交
257 258
	nlmsvc_freegrantargs(block->b_call);
	nlm_release_call(block->b_call);
259
	nlm_release_file(block->b_file);
M
Marc Eshel 已提交
260
	kfree(block->b_fl);
L
Linus Torvalds 已提交
261
	kfree(block);
262 263 264 265 266 267
}

static void nlmsvc_release_block(struct nlm_block *block)
{
	if (block != NULL)
		kref_put(&block->b_count, nlmsvc_free_block);
L
Linus Torvalds 已提交
268 269
}

270 271 272 273 274 275 276
/*
 * Loop over all blocks and delete blocks held by
 * a matching host.
 */
void nlmsvc_traverse_blocks(struct nlm_host *host,
			struct nlm_file *file,
			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;
285 286 287
		/* Do not destroy blocks that are not on
		 * the global retry list - why? */
		if (list_empty(&block->b_list))
288 289
			continue;
		kref_get(&block->b_count);
290
		mutex_unlock(&file->f_mutex);
291 292 293 294
		nlmsvc_unlink_block(block);
		nlmsvc_release_block(block);
		goto restart;
	}
295
	mutex_unlock(&file->f_mutex);
296 297
}

298 299 300 301 302 303 304 305
/*
 * Initialize arguments for GRANTED call. The nlm_rqst structure
 * has been cleared already.
 */
static int nlmsvc_setgrantargs(struct nlm_rqst *call, struct nlm_lock *lock)
{
	locks_copy_lock(&call->a_args.lock.fl, &lock->fl);
	memcpy(&call->a_args.lock.fh, &lock->fh, sizeof(call->a_args.lock.fh));
306
	call->a_args.lock.caller = utsname()->nodename;
307 308 309 310 311 312 313 314
	call->a_args.lock.oh.len = lock->oh.len;

	/* set default data area */
	call->a_args.lock.oh.data = call->a_owner;
	call->a_args.lock.svid = lock->fl.fl_pid;

	if (lock->oh.len > NLMCLNT_OHSIZE) {
		void *data = kmalloc(lock->oh.len, GFP_KERNEL);
T
Trond Myklebust 已提交
315
		if (!data)
316 317 318 319 320 321 322 323 324 325
			return 0;
		call->a_args.lock.oh.data = (u8 *) data;
	}

	memcpy(call->a_args.lock.oh.data, lock->oh.data, lock->oh.len);
	return 1;
}

static void nlmsvc_freegrantargs(struct nlm_rqst *call)
{
T
Trond Myklebust 已提交
326
	if (call->a_args.lock.oh.data != call->a_owner)
327 328 329
		kfree(call->a_args.lock.oh.data);
}

M
Marc Eshel 已提交
330 331 332
/*
 * Deferred lock request handling for non-blocking lock
 */
333
static __be32
M
Marc Eshel 已提交
334 335
nlmsvc_defer_lock_rqst(struct svc_rqst *rqstp, struct nlm_block *block)
{
336
	__be32 status = nlm_lck_denied_nolocks;
M
Marc Eshel 已提交
337 338 339 340 341 342 343 344 345 346 347 348 349

	block->b_flags |= B_QUEUED;

	nlmsvc_insert_block(block, NLM_TIMEOUT);

	block->b_cache_req = &rqstp->rq_chandle;
	if (rqstp->rq_chandle.defer) {
		block->b_deferred_req =
			rqstp->rq_chandle.defer(block->b_cache_req);
		if (block->b_deferred_req != NULL)
			status = nlm_drop_reply;
	}
	dprintk("lockd: nlmsvc_defer_lock_rqst block %p flags %d status %d\n",
350
		block, block->b_flags, ntohl(status));
M
Marc Eshel 已提交
351 352 353 354

	return status;
}

L
Linus Torvalds 已提交
355 356 357 358
/*
 * Attempt to establish a lock, and if it can't be granted, block it
 * if required.
 */
A
Al Viro 已提交
359
__be32
L
Linus Torvalds 已提交
360 361 362
nlmsvc_lock(struct svc_rqst *rqstp, struct nlm_file *file,
			struct nlm_lock *lock, int wait, struct nlm_cookie *cookie)
{
363
	struct nlm_block	*block = NULL;
364
	struct nlm_host		*host;
L
Linus Torvalds 已提交
365
	int			error;
A
Al Viro 已提交
366
	__be32			ret;
L
Linus Torvalds 已提交
367 368

	dprintk("lockd: nlmsvc_lock(%s/%ld, ty=%d, pi=%d, %Ld-%Ld, bl=%d)\n",
J
Josef Sipek 已提交
369 370
				file->f_file->f_path.dentry->d_inode->i_sb->s_id,
				file->f_file->f_path.dentry->d_inode->i_ino,
L
Linus Torvalds 已提交
371 372 373 374 375
				lock->fl.fl_type, lock->fl.fl_pid,
				(long long)lock->fl.fl_start,
				(long long)lock->fl.fl_end,
				wait);

376 377 378 379
	/* Create host handle for callback */
	host = nlmsvc_lookup_host(rqstp, lock->caller, lock->len);
	if (host == NULL)
		return nlm_lck_denied_nolocks;
L
Linus Torvalds 已提交
380 381

	/* Lock file against concurrent access */
382
	mutex_lock(&file->f_mutex);
383 384 385
	/* Get existing block (in case client is busy-waiting)
	 * or create new block
	 */
386
	block = nlmsvc_lookup_block(file, lock);
387
	if (block == NULL) {
388 389
		block = nlmsvc_create_block(rqstp, nlm_get_host(host), file,
				lock, cookie);
390 391 392
		ret = nlm_lck_denied_nolocks;
		if (block == NULL)
			goto out;
T
Trond Myklebust 已提交
393
		lock = &block->b_call->a_args.lock;
394 395
	} else
		lock->fl.fl_flags &= ~FL_SLEEP;
L
Linus Torvalds 已提交
396

397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412
	if (block->b_flags & B_QUEUED) {
		dprintk("lockd: nlmsvc_lock deferred block %p flags %d\n",
							block, block->b_flags);
		if (block->b_granted) {
			nlmsvc_unlink_block(block);
			ret = nlm_granted;
			goto out;
		}
		if (block->b_flags & B_TIMED_OUT) {
			nlmsvc_unlink_block(block);
			ret = nlm_lck_denied;
			goto out;
		}
		ret = nlm_drop_reply;
		goto out;
	}
L
Linus Torvalds 已提交
413

414 415 416 417
	if (!wait)
		lock->fl.fl_flags &= ~FL_SLEEP;
	error = vfs_lock_file(file->f_file, F_SETLK, &lock->fl, NULL);
	lock->fl.fl_flags &= ~FL_SLEEP;
418

419
	dprintk("lockd: vfs_lock_file returned %d\n", error);
420
	switch(error) {
L
Linus Torvalds 已提交
421
		case 0:
A
Andy Adamson 已提交
422 423
			ret = nlm_granted;
			goto out;
424
		case -EAGAIN:
425
			ret = nlm_lck_denied;
426
			break;
427 428 429 430 431 432 433
		case -EINPROGRESS:
			if (wait)
				break;
			/* Filesystem lock operation is in progress
			   Add it to the queue waiting for callback */
			ret = nlmsvc_defer_lock_rqst(rqstp, block);
			goto out;
434
		case -EDEADLK:
A
Andy Adamson 已提交
435 436
			ret = nlm_deadlock;
			goto out;
L
Linus Torvalds 已提交
437
		default:			/* includes ENOLCK */
A
Andy Adamson 已提交
438 439
			ret = nlm_lck_denied_nolocks;
			goto out;
L
Linus Torvalds 已提交
440 441
	}

442 443 444 445 446
	ret = nlm_lck_denied;
	if (!wait)
		goto out;

	ret = nlm_lck_blocked;
L
Linus Torvalds 已提交
447 448

	/* Append to list of blocked */
449
	nlmsvc_insert_block(block, NLM_NEVER);
A
Andy Adamson 已提交
450
out:
451
	mutex_unlock(&file->f_mutex);
452
	nlmsvc_release_block(block);
453
	nlm_release_host(host);
A
Andy Adamson 已提交
454 455
	dprintk("lockd: nlmsvc_lock returned %u\n", ret);
	return ret;
L
Linus Torvalds 已提交
456 457 458 459 460
}

/*
 * Test for presence of a conflicting lock.
 */
A
Al Viro 已提交
461
__be32
462 463 464
nlmsvc_testlock(struct svc_rqst *rqstp, struct nlm_file *file,
		struct nlm_lock *lock, struct nlm_lock *conflock,
		struct nlm_cookie *cookie)
L
Linus Torvalds 已提交
465
{
M
Marc Eshel 已提交
466 467 468 469
	struct nlm_block 	*block = NULL;
	int			error;
	__be32			ret;

L
Linus Torvalds 已提交
470
	dprintk("lockd: nlmsvc_testlock(%s/%ld, ty=%d, %Ld-%Ld)\n",
J
Josef Sipek 已提交
471 472
				file->f_file->f_path.dentry->d_inode->i_sb->s_id,
				file->f_file->f_path.dentry->d_inode->i_ino,
L
Linus Torvalds 已提交
473 474 475 476
				lock->fl.fl_type,
				(long long)lock->fl.fl_start,
				(long long)lock->fl.fl_end);

M
Marc Eshel 已提交
477 478 479 480 481
	/* Get existing block (in case client is busy-waiting) */
	block = nlmsvc_lookup_block(file, lock);

	if (block == NULL) {
		struct file_lock *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
482
		struct nlm_host	*host;
M
Marc Eshel 已提交
483 484 485

		if (conf == NULL)
			return nlm_granted;
486 487
		/* Create host handle for callback */
		host = nlmsvc_lookup_host(rqstp, lock->caller, lock->len);
488 489
		if (host == NULL) {
			kfree(conf);
490
			return nlm_lck_denied_nolocks;
491
		}
492
		block = nlmsvc_create_block(rqstp, host, file, lock, cookie);
M
Marc Eshel 已提交
493 494 495 496 497 498 499 500 501 502 503
		if (block == NULL) {
			kfree(conf);
			return nlm_granted;
		}
		block->b_fl = conf;
	}
	if (block->b_flags & B_QUEUED) {
		dprintk("lockd: nlmsvc_testlock deferred block %p flags %d fl %p\n",
			block, block->b_flags, block->b_fl);
		if (block->b_flags & B_TIMED_OUT) {
			nlmsvc_unlink_block(block);
504 505
			ret = nlm_lck_denied;
			goto out;
M
Marc Eshel 已提交
506 507
		}
		if (block->b_flags & B_GOT_CALLBACK) {
508
			nlmsvc_unlink_block(block);
M
Marc Eshel 已提交
509 510 511 512
			if (block->b_fl != NULL
					&& block->b_fl->fl_type != F_UNLCK) {
				lock->fl = *block->b_fl;
				goto conf_lock;
513 514 515
			} else {
				ret = nlm_granted;
				goto out;
M
Marc Eshel 已提交
516 517
			}
		}
518 519
		ret = nlm_drop_reply;
		goto out;
L
Linus Torvalds 已提交
520 521
	}

M
Marc Eshel 已提交
522
	error = vfs_test_lock(file->f_file, &lock->fl);
523 524 525 526
	if (error == -EINPROGRESS) {
		ret = nlmsvc_defer_lock_rqst(rqstp, block);
		goto out;
	}
M
Marc Eshel 已提交
527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551
	if (error) {
		ret = nlm_lck_denied_nolocks;
		goto out;
	}
	if (lock->fl.fl_type == F_UNLCK) {
		ret = nlm_granted;
		goto out;
	}

conf_lock:
	dprintk("lockd: conflicting lock(ty=%d, %Ld-%Ld)\n",
		lock->fl.fl_type, (long long)lock->fl.fl_start,
		(long long)lock->fl.fl_end);
	conflock->caller = "somehost";	/* FIXME */
	conflock->len = strlen(conflock->caller);
	conflock->oh.len = 0;		/* don't return OH info */
	conflock->svid = lock->fl.fl_pid;
	conflock->fl.fl_type = lock->fl.fl_type;
	conflock->fl.fl_start = lock->fl.fl_start;
	conflock->fl.fl_end = lock->fl.fl_end;
	ret = nlm_lck_denied;
out:
	if (block)
		nlmsvc_release_block(block);
	return ret;
L
Linus Torvalds 已提交
552 553 554 555 556 557 558 559 560
}

/*
 * Remove a lock.
 * This implies a CANCEL call: We send a GRANT_MSG, the client replies
 * with a GRANT_RES call which gets lost, and calls UNLOCK immediately
 * afterwards. In this case the block will still be there, and hence
 * must be removed.
 */
A
Al Viro 已提交
561
__be32
L
Linus Torvalds 已提交
562 563 564 565 566
nlmsvc_unlock(struct nlm_file *file, struct nlm_lock *lock)
{
	int	error;

	dprintk("lockd: nlmsvc_unlock(%s/%ld, pi=%d, %Ld-%Ld)\n",
J
Josef Sipek 已提交
567 568
				file->f_file->f_path.dentry->d_inode->i_sb->s_id,
				file->f_file->f_path.dentry->d_inode->i_ino,
L
Linus Torvalds 已提交
569 570 571 572 573 574 575 576
				lock->fl.fl_pid,
				(long long)lock->fl.fl_start,
				(long long)lock->fl.fl_end);

	/* First, cancel any lock that might be there */
	nlmsvc_cancel_blocked(file, lock);

	lock->fl.fl_type = F_UNLCK;
577
	error = vfs_lock_file(file->f_file, F_SETLK, &lock->fl, NULL);
L
Linus Torvalds 已提交
578 579 580 581 582 583 584 585 586 587 588

	return (error < 0)? nlm_lck_denied_nolocks : nlm_granted;
}

/*
 * Cancel a previously blocked request.
 *
 * A cancel request always overrides any grant that may currently
 * be in progress.
 * The calling procedure must check whether the file can be closed.
 */
A
Al Viro 已提交
589
__be32
L
Linus Torvalds 已提交
590 591 592
nlmsvc_cancel_blocked(struct nlm_file *file, struct nlm_lock *lock)
{
	struct nlm_block	*block;
J
J. Bruce Fields 已提交
593
	int status = 0;
L
Linus Torvalds 已提交
594 595

	dprintk("lockd: nlmsvc_cancel(%s/%ld, pi=%d, %Ld-%Ld)\n",
J
Josef Sipek 已提交
596 597
				file->f_file->f_path.dentry->d_inode->i_sb->s_id,
				file->f_file->f_path.dentry->d_inode->i_ino,
L
Linus Torvalds 已提交
598 599 600 601
				lock->fl.fl_pid,
				(long long)lock->fl.fl_start,
				(long long)lock->fl.fl_end);

602
	mutex_lock(&file->f_mutex);
603
	block = nlmsvc_lookup_block(file, lock);
604
	mutex_unlock(&file->f_mutex);
605
	if (block != NULL) {
606 607
		vfs_cancel_lock(block->b_file->f_file,
				&block->b_call->a_args.lock.fl);
608 609 610
		status = nlmsvc_unlink_block(block);
		nlmsvc_release_block(block);
	}
J
J. Bruce Fields 已提交
611
	return status ? nlm_lck_denied : nlm_granted;
L
Linus Torvalds 已提交
612 613
}

M
Marc Eshel 已提交
614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670
/*
 * This is a callback from the filesystem for VFS file lock requests.
 * It will be used if fl_grant is defined and the filesystem can not
 * respond to the request immediately.
 * For GETLK request it will copy the reply to the nlm_block.
 * For SETLK or SETLKW request it will get the local posix lock.
 * In all cases it will move the block to the head of nlm_blocked q where
 * nlmsvc_retry_blocked() can send back a reply for SETLKW or revisit the
 * deferred rpc for GETLK and SETLK.
 */
static void
nlmsvc_update_deferred_block(struct nlm_block *block, struct file_lock *conf,
			     int result)
{
	block->b_flags |= B_GOT_CALLBACK;
	if (result == 0)
		block->b_granted = 1;
	else
		block->b_flags |= B_TIMED_OUT;
	if (conf) {
		if (block->b_fl)
			locks_copy_lock(block->b_fl, conf);
	}
}

static int nlmsvc_grant_deferred(struct file_lock *fl, struct file_lock *conf,
					int result)
{
	struct nlm_block *block;
	int rc = -ENOENT;

	lock_kernel();
	list_for_each_entry(block, &nlm_blocked, b_list) {
		if (nlm_compare_locks(&block->b_call->a_args.lock.fl, fl)) {
			dprintk("lockd: nlmsvc_notify_blocked block %p flags %d\n",
							block, block->b_flags);
			if (block->b_flags & B_QUEUED) {
				if (block->b_flags & B_TIMED_OUT) {
					rc = -ENOLCK;
					break;
				}
				nlmsvc_update_deferred_block(block, conf, result);
			} else if (result == 0)
				block->b_granted = 1;

			nlmsvc_insert_block(block, 0);
			svc_wake_up(block->b_daemon);
			rc = 0;
			break;
		}
	}
	unlock_kernel();
	if (rc == -ENOENT)
		printk(KERN_WARNING "lockd: grant for unknown block\n");
	return rc;
}

L
Linus Torvalds 已提交
671 672 673 674 675 676 677 678 679 680
/*
 * Unblock a blocked lock request. This is a callback invoked from the
 * VFS layer when a lock on which we blocked is removed.
 *
 * This function doesn't grant the blocked lock instantly, but rather moves
 * the block to the head of nlm_blocked where it can be picked up by lockd.
 */
static void
nlmsvc_notify_blocked(struct file_lock *fl)
{
681
	struct nlm_block	*block;
L
Linus Torvalds 已提交
682 683

	dprintk("lockd: VFS unblock notification for block %p\n", fl);
684
	list_for_each_entry(block, &nlm_blocked, b_list) {
T
Trond Myklebust 已提交
685
		if (nlm_compare_locks(&block->b_call->a_args.lock.fl, fl)) {
L
Linus Torvalds 已提交
686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702
			nlmsvc_insert_block(block, 0);
			svc_wake_up(block->b_daemon);
			return;
		}
	}

	printk(KERN_WARNING "lockd: notification for unknown block!\n");
}

static int nlmsvc_same_owner(struct file_lock *fl1, struct file_lock *fl2)
{
	return fl1->fl_owner == fl2->fl_owner && fl1->fl_pid == fl2->fl_pid;
}

struct lock_manager_operations nlmsvc_lock_operations = {
	.fl_compare_owner = nlmsvc_same_owner,
	.fl_notify = nlmsvc_notify_blocked,
M
Marc Eshel 已提交
703
	.fl_grant = nlmsvc_grant_deferred,
L
Linus Torvalds 已提交
704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720
};

/*
 * Try to claim a lock that was previously blocked.
 *
 * Note that we use both the RPC_GRANTED_MSG call _and_ an async
 * RPC thread when notifying the client. This seems like overkill...
 * Here's why:
 *  -	we don't want to use a synchronous RPC thread, otherwise
 *	we might find ourselves hanging on a dead portmapper.
 *  -	Some lockd implementations (e.g. HP) don't react to
 *	RPC_GRANTED calls; they seem to insist on RPC_GRANTED_MSG calls.
 */
static void
nlmsvc_grant_blocked(struct nlm_block *block)
{
	struct nlm_file		*file = block->b_file;
T
Trond Myklebust 已提交
721
	struct nlm_lock		*lock = &block->b_call->a_args.lock;
L
Linus Torvalds 已提交
722 723 724 725
	int			error;

	dprintk("lockd: grant blocked lock %p\n", block);

M
Marc Eshel 已提交
726 727
	kref_get(&block->b_count);

L
Linus Torvalds 已提交
728
	/* Unlink block request from list */
729
	nlmsvc_unlink_block(block);
L
Linus Torvalds 已提交
730 731 732 733 734 735 736 737 738 739

	/* If b_granted is true this means we've been here before.
	 * Just retry the grant callback, possibly refreshing the RPC
	 * binding */
	if (block->b_granted) {
		nlm_rebind_host(block->b_host);
		goto callback;
	}

	/* Try the lock operation again */
740
	lock->fl.fl_flags |= FL_SLEEP;
741
	error = vfs_lock_file(file->f_file, F_SETLK, &lock->fl, NULL);
742 743
	lock->fl.fl_flags &= ~FL_SLEEP;

744 745 746 747
	switch (error) {
	case 0:
		break;
	case -EAGAIN:
748 749
	case -EINPROGRESS:
		dprintk("lockd: lock still blocked error %d\n", error);
L
Linus Torvalds 已提交
750
		nlmsvc_insert_block(block, NLM_NEVER);
M
Marc Eshel 已提交
751
		nlmsvc_release_block(block);
752
		return;
753
	default:
L
Linus Torvalds 已提交
754 755 756
		printk(KERN_WARNING "lockd: unexpected error %d in %s!\n",
				-error, __FUNCTION__);
		nlmsvc_insert_block(block, 10 * HZ);
M
Marc Eshel 已提交
757
		nlmsvc_release_block(block);
758
		return;
L
Linus Torvalds 已提交
759 760 761 762 763 764 765 766 767 768 769
	}

callback:
	/* Lock was granted by VFS. */
	dprintk("lockd: GRANTing blocked lock.\n");
	block->b_granted = 1;

	/* Schedule next grant callback in 30 seconds */
	nlmsvc_insert_block(block, 30 * HZ);

	/* Call the client */
770
	nlm_async_call(block->b_call, NLMPROC_GRANTED_MSG, &nlmsvc_grant_ops);
L
Linus Torvalds 已提交
771 772 773 774 775 776 777 778 779 780
}

/*
 * This is the callback from the RPC layer when the NLM_GRANTED_MSG
 * RPC call has succeeded or timed out.
 * Like all RPC callbacks, it is invoked by the rpciod process, so it
 * better not sleep. Therefore, we put the blocked lock on the nlm_blocked
 * chain once more in order to have it removed by lockd itself (which can
 * then sleep on the file semaphore without disrupting e.g. the nfs client).
 */
781
static void nlmsvc_grant_callback(struct rpc_task *task, void *data)
L
Linus Torvalds 已提交
782
{
783
	struct nlm_rqst		*call = data;
T
Trond Myklebust 已提交
784
	struct nlm_block	*block = call->a_block;
L
Linus Torvalds 已提交
785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800
	unsigned long		timeout;

	dprintk("lockd: GRANT_MSG RPC callback\n");

	/* Technically, we should down the file semaphore here. Since we
	 * move the block towards the head of the queue only, no harm
	 * can be done, though. */
	if (task->tk_status < 0) {
		/* RPC error: Re-insert for retransmission */
		timeout = 10 * HZ;
	} else {
		/* Call was successful, now wait for client callback */
		timeout = 60 * HZ;
	}
	nlmsvc_insert_block(block, timeout);
	svc_wake_up(block->b_daemon);
801 802
}

A
Adrian Bunk 已提交
803
static void nlmsvc_grant_release(void *data)
804
{
805 806 807
	struct nlm_rqst		*call = data;

	nlmsvc_release_block(call->a_block);
L
Linus Torvalds 已提交
808 809
}

810 811
static const struct rpc_call_ops nlmsvc_grant_ops = {
	.rpc_call_done = nlmsvc_grant_callback,
812
	.rpc_release = nlmsvc_grant_release,
813 814
};

L
Linus Torvalds 已提交
815 816 817 818 819
/*
 * We received a GRANT_RES callback. Try to find the corresponding
 * block.
 */
void
A
Al Viro 已提交
820
nlmsvc_grant_reply(struct nlm_cookie *cookie, __be32 status)
L
Linus Torvalds 已提交
821 822 823
{
	struct nlm_block	*block;

824 825 826
	dprintk("grant_reply: looking for cookie %x, s=%d \n",
		*(unsigned int *)(cookie->data), status);
	if (!(block = nlmsvc_find_block(cookie)))
L
Linus Torvalds 已提交
827 828
		return;

829
	if (block) {
A
Al Viro 已提交
830
		if (status == nlm_lck_denied_grace_period) {
L
Linus Torvalds 已提交
831 832 833 834 835
			/* Try again in a couple of seconds */
			nlmsvc_insert_block(block, 10 * HZ);
		} else {
			/* Lock is now held by client, or has been rejected.
			 * In both cases, the block should be removed. */
836
			nlmsvc_unlink_block(block);
L
Linus Torvalds 已提交
837 838
		}
	}
839
	nlmsvc_release_block(block);
L
Linus Torvalds 已提交
840 841
}

M
Marc Eshel 已提交
842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858
/* Helper function to handle retry of a deferred block.
 * If it is a blocking lock, call grant_blocked.
 * For a non-blocking lock or test lock, revisit the request.
 */
static void
retry_deferred_block(struct nlm_block *block)
{
	if (!(block->b_flags & B_GOT_CALLBACK))
		block->b_flags |= B_TIMED_OUT;
	nlmsvc_insert_block(block, NLM_TIMEOUT);
	dprintk("revisit block %p flags %d\n",	block, block->b_flags);
	if (block->b_deferred_req) {
		block->b_deferred_req->revisit(block->b_deferred_req, 0);
		block->b_deferred_req = NULL;
	}
}

L
Linus Torvalds 已提交
859 860 861 862 863 864 865 866
/*
 * Retry all blocked locks that have been notified. This is where lockd
 * picks up locks that can be granted, or grant notifications that must
 * be retransmitted.
 */
unsigned long
nlmsvc_retry_blocked(void)
{
867 868 869 870 871
	unsigned long	timeout = MAX_SCHEDULE_TIMEOUT;
	struct nlm_block *block;

	while (!list_empty(&nlm_blocked)) {
		block = list_entry(nlm_blocked.next, struct nlm_block, b_list);
L
Linus Torvalds 已提交
872 873 874

		if (block->b_when == NLM_NEVER)
			break;
875 876
	        if (time_after(block->b_when,jiffies)) {
			timeout = block->b_when - jiffies;
L
Linus Torvalds 已提交
877
			break;
878 879
		}

J
J. Bruce Fields 已提交
880 881
		dprintk("nlmsvc_retry_blocked(%p, when=%ld)\n",
			block, block->b_when);
M
Marc Eshel 已提交
882 883 884 885 886 887
		if (block->b_flags & B_QUEUED) {
			dprintk("nlmsvc_retry_blocked delete block (%p, granted=%d, flags=%d)\n",
				block, block->b_granted, block->b_flags);
			retry_deferred_block(block);
		} else
			nlmsvc_grant_blocked(block);
L
Linus Torvalds 已提交
888 889
	}

890
	return timeout;
L
Linus Torvalds 已提交
891
}