xref: /linux/fs/afs/flock.c (revision 79790b6818e96c58fe2bffee1b418c16e64e7b80)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* AFS file locking support
3  *
4  * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
5  * Written by David Howells (dhowells@redhat.com)
6  */
7 
8 #include "internal.h"
9 
10 #define AFS_LOCK_GRANTED	0
11 #define AFS_LOCK_PENDING	1
12 #define AFS_LOCK_YOUR_TRY	2
13 
14 struct workqueue_struct *afs_lock_manager;
15 
16 static void afs_next_locker(struct afs_vnode *vnode, int error);
17 static void afs_fl_copy_lock(struct file_lock *new, struct file_lock *fl);
18 static void afs_fl_release_private(struct file_lock *fl);
19 
20 static const struct file_lock_operations afs_lock_ops = {
21 	.fl_copy_lock		= afs_fl_copy_lock,
22 	.fl_release_private	= afs_fl_release_private,
23 };
24 
afs_set_lock_state(struct afs_vnode * vnode,enum afs_lock_state state)25 static inline void afs_set_lock_state(struct afs_vnode *vnode, enum afs_lock_state state)
26 {
27 	_debug("STATE %u -> %u", vnode->lock_state, state);
28 	vnode->lock_state = state;
29 }
30 
31 static atomic_t afs_file_lock_debug_id;
32 
33 /*
34  * if the callback is broken on this vnode, then the lock may now be available
35  */
afs_lock_may_be_available(struct afs_vnode * vnode)36 void afs_lock_may_be_available(struct afs_vnode *vnode)
37 {
38 	_enter("{%llx:%llu}", vnode->fid.vid, vnode->fid.vnode);
39 
40 	spin_lock(&vnode->lock);
41 	if (vnode->lock_state == AFS_VNODE_LOCK_WAITING_FOR_CB)
42 		afs_next_locker(vnode, 0);
43 	trace_afs_flock_ev(vnode, NULL, afs_flock_callback_break, 0);
44 	spin_unlock(&vnode->lock);
45 }
46 
47 /*
48  * the lock will time out in 5 minutes unless we extend it, so schedule
49  * extension in a bit less than that time
50  */
afs_schedule_lock_extension(struct afs_vnode * vnode)51 static void afs_schedule_lock_extension(struct afs_vnode *vnode)
52 {
53 	ktime_t expires_at, now, duration;
54 	u64 duration_j;
55 
56 	expires_at = ktime_add_ms(vnode->locked_at, AFS_LOCKWAIT * 1000 / 2);
57 	now = ktime_get_real();
58 	duration = ktime_sub(expires_at, now);
59 	if (duration <= 0)
60 		duration_j = 0;
61 	else
62 		duration_j = nsecs_to_jiffies(ktime_to_ns(duration));
63 
64 	queue_delayed_work(afs_lock_manager, &vnode->lock_work, duration_j);
65 }
66 
67 /*
68  * In the case of successful completion of a lock operation, record the time
69  * the reply appeared and start the lock extension timer.
70  */
afs_lock_op_done(struct afs_call * call)71 void afs_lock_op_done(struct afs_call *call)
72 {
73 	struct afs_operation *op = call->op;
74 	struct afs_vnode *vnode = op->file[0].vnode;
75 
76 	if (call->error == 0) {
77 		spin_lock(&vnode->lock);
78 		trace_afs_flock_ev(vnode, NULL, afs_flock_timestamp, 0);
79 		vnode->locked_at = call->issue_time;
80 		afs_schedule_lock_extension(vnode);
81 		spin_unlock(&vnode->lock);
82 	}
83 }
84 
85 /*
86  * grant one or more locks (readlocks are allowed to jump the queue if the
87  * first lock in the queue is itself a readlock)
88  * - the caller must hold the vnode lock
89  */
afs_grant_locks(struct afs_vnode * vnode)90 static void afs_grant_locks(struct afs_vnode *vnode)
91 {
92 	struct file_lock *p, *_p;
93 	bool exclusive = (vnode->lock_type == AFS_LOCK_WRITE);
94 
95 	list_for_each_entry_safe(p, _p, &vnode->pending_locks, fl_u.afs.link) {
96 		if (!exclusive && lock_is_write(p))
97 			continue;
98 
99 		list_move_tail(&p->fl_u.afs.link, &vnode->granted_locks);
100 		p->fl_u.afs.state = AFS_LOCK_GRANTED;
101 		trace_afs_flock_op(vnode, p, afs_flock_op_grant);
102 		locks_wake_up(p);
103 	}
104 }
105 
106 /*
107  * If an error is specified, reject every pending lock that matches the
108  * authentication and type of the lock we failed to get.  If there are any
109  * remaining lockers, try to wake up one of them to have a go.
110  */
afs_next_locker(struct afs_vnode * vnode,int error)111 static void afs_next_locker(struct afs_vnode *vnode, int error)
112 {
113 	struct file_lock *p, *_p, *next = NULL;
114 	struct key *key = vnode->lock_key;
115 	unsigned int type = F_RDLCK;
116 
117 	_enter("");
118 
119 	if (vnode->lock_type == AFS_LOCK_WRITE)
120 		type = F_WRLCK;
121 
122 	list_for_each_entry_safe(p, _p, &vnode->pending_locks, fl_u.afs.link) {
123 		if (error &&
124 		    p->c.flc_type == type &&
125 		    afs_file_key(p->c.flc_file) == key) {
126 			list_del_init(&p->fl_u.afs.link);
127 			p->fl_u.afs.state = error;
128 			locks_wake_up(p);
129 		}
130 
131 		/* Select the next locker to hand off to. */
132 		if (next && (lock_is_write(next) || lock_is_read(p)))
133 			continue;
134 		next = p;
135 	}
136 
137 	vnode->lock_key = NULL;
138 	key_put(key);
139 
140 	if (next) {
141 		afs_set_lock_state(vnode, AFS_VNODE_LOCK_SETTING);
142 		next->fl_u.afs.state = AFS_LOCK_YOUR_TRY;
143 		trace_afs_flock_op(vnode, next, afs_flock_op_wake);
144 		locks_wake_up(next);
145 	} else {
146 		afs_set_lock_state(vnode, AFS_VNODE_LOCK_NONE);
147 		trace_afs_flock_ev(vnode, NULL, afs_flock_no_lockers, 0);
148 	}
149 
150 	_leave("");
151 }
152 
153 /*
154  * Kill off all waiters in the the pending lock queue due to the vnode being
155  * deleted.
156  */
afs_kill_lockers_enoent(struct afs_vnode * vnode)157 static void afs_kill_lockers_enoent(struct afs_vnode *vnode)
158 {
159 	struct file_lock *p;
160 
161 	afs_set_lock_state(vnode, AFS_VNODE_LOCK_DELETED);
162 
163 	while (!list_empty(&vnode->pending_locks)) {
164 		p = list_entry(vnode->pending_locks.next,
165 			       struct file_lock, fl_u.afs.link);
166 		list_del_init(&p->fl_u.afs.link);
167 		p->fl_u.afs.state = -ENOENT;
168 		locks_wake_up(p);
169 	}
170 
171 	key_put(vnode->lock_key);
172 	vnode->lock_key = NULL;
173 }
174 
afs_lock_success(struct afs_operation * op)175 static void afs_lock_success(struct afs_operation *op)
176 {
177 	_enter("op=%08x", op->debug_id);
178 	afs_vnode_commit_status(op, &op->file[0]);
179 }
180 
181 static const struct afs_operation_ops afs_set_lock_operation = {
182 	.issue_afs_rpc	= afs_fs_set_lock,
183 	.issue_yfs_rpc	= yfs_fs_set_lock,
184 	.success	= afs_lock_success,
185 	.aborted	= afs_check_for_remote_deletion,
186 };
187 
188 /*
189  * Get a lock on a file
190  */
afs_set_lock(struct afs_vnode * vnode,struct key * key,afs_lock_type_t type)191 static int afs_set_lock(struct afs_vnode *vnode, struct key *key,
192 			afs_lock_type_t type)
193 {
194 	struct afs_operation *op;
195 
196 	_enter("%s{%llx:%llu.%u},%x,%u",
197 	       vnode->volume->name,
198 	       vnode->fid.vid,
199 	       vnode->fid.vnode,
200 	       vnode->fid.unique,
201 	       key_serial(key), type);
202 
203 	op = afs_alloc_operation(key, vnode->volume);
204 	if (IS_ERR(op))
205 		return PTR_ERR(op);
206 
207 	afs_op_set_vnode(op, 0, vnode);
208 
209 	op->lock.type	= type;
210 	op->ops		= &afs_set_lock_operation;
211 	return afs_do_sync_operation(op);
212 }
213 
214 static const struct afs_operation_ops afs_extend_lock_operation = {
215 	.issue_afs_rpc	= afs_fs_extend_lock,
216 	.issue_yfs_rpc	= yfs_fs_extend_lock,
217 	.success	= afs_lock_success,
218 };
219 
220 /*
221  * Extend a lock on a file
222  */
afs_extend_lock(struct afs_vnode * vnode,struct key * key)223 static int afs_extend_lock(struct afs_vnode *vnode, struct key *key)
224 {
225 	struct afs_operation *op;
226 
227 	_enter("%s{%llx:%llu.%u},%x",
228 	       vnode->volume->name,
229 	       vnode->fid.vid,
230 	       vnode->fid.vnode,
231 	       vnode->fid.unique,
232 	       key_serial(key));
233 
234 	op = afs_alloc_operation(key, vnode->volume);
235 	if (IS_ERR(op))
236 		return PTR_ERR(op);
237 
238 	afs_op_set_vnode(op, 0, vnode);
239 
240 	op->flags	|= AFS_OPERATION_UNINTR;
241 	op->ops		= &afs_extend_lock_operation;
242 	return afs_do_sync_operation(op);
243 }
244 
245 static const struct afs_operation_ops afs_release_lock_operation = {
246 	.issue_afs_rpc	= afs_fs_release_lock,
247 	.issue_yfs_rpc	= yfs_fs_release_lock,
248 	.success	= afs_lock_success,
249 };
250 
251 /*
252  * Release a lock on a file
253  */
afs_release_lock(struct afs_vnode * vnode,struct key * key)254 static int afs_release_lock(struct afs_vnode *vnode, struct key *key)
255 {
256 	struct afs_operation *op;
257 
258 	_enter("%s{%llx:%llu.%u},%x",
259 	       vnode->volume->name,
260 	       vnode->fid.vid,
261 	       vnode->fid.vnode,
262 	       vnode->fid.unique,
263 	       key_serial(key));
264 
265 	op = afs_alloc_operation(key, vnode->volume);
266 	if (IS_ERR(op))
267 		return PTR_ERR(op);
268 
269 	afs_op_set_vnode(op, 0, vnode);
270 
271 	op->flags	|= AFS_OPERATION_UNINTR;
272 	op->ops		= &afs_release_lock_operation;
273 	return afs_do_sync_operation(op);
274 }
275 
276 /*
277  * do work for a lock, including:
278  * - probing for a lock we're waiting on but didn't get immediately
279  * - extending a lock that's close to timing out
280  */
afs_lock_work(struct work_struct * work)281 void afs_lock_work(struct work_struct *work)
282 {
283 	struct afs_vnode *vnode =
284 		container_of(work, struct afs_vnode, lock_work.work);
285 	struct key *key;
286 	int ret;
287 
288 	_enter("{%llx:%llu}", vnode->fid.vid, vnode->fid.vnode);
289 
290 	spin_lock(&vnode->lock);
291 
292 again:
293 	_debug("wstate %u for %p", vnode->lock_state, vnode);
294 	switch (vnode->lock_state) {
295 	case AFS_VNODE_LOCK_NEED_UNLOCK:
296 		afs_set_lock_state(vnode, AFS_VNODE_LOCK_UNLOCKING);
297 		trace_afs_flock_ev(vnode, NULL, afs_flock_work_unlocking, 0);
298 		spin_unlock(&vnode->lock);
299 
300 		/* attempt to release the server lock; if it fails, we just
301 		 * wait 5 minutes and it'll expire anyway */
302 		ret = afs_release_lock(vnode, vnode->lock_key);
303 		if (ret < 0 && vnode->lock_state != AFS_VNODE_LOCK_DELETED) {
304 			trace_afs_flock_ev(vnode, NULL, afs_flock_release_fail,
305 					   ret);
306 			printk(KERN_WARNING "AFS:"
307 			       " Failed to release lock on {%llx:%llx} error %d\n",
308 			       vnode->fid.vid, vnode->fid.vnode, ret);
309 		}
310 
311 		spin_lock(&vnode->lock);
312 		if (ret == -ENOENT)
313 			afs_kill_lockers_enoent(vnode);
314 		else
315 			afs_next_locker(vnode, 0);
316 		spin_unlock(&vnode->lock);
317 		return;
318 
319 	/* If we've already got a lock, then it must be time to extend that
320 	 * lock as AFS locks time out after 5 minutes.
321 	 */
322 	case AFS_VNODE_LOCK_GRANTED:
323 		_debug("extend");
324 
325 		ASSERT(!list_empty(&vnode->granted_locks));
326 
327 		key = key_get(vnode->lock_key);
328 		afs_set_lock_state(vnode, AFS_VNODE_LOCK_EXTENDING);
329 		trace_afs_flock_ev(vnode, NULL, afs_flock_work_extending, 0);
330 		spin_unlock(&vnode->lock);
331 
332 		ret = afs_extend_lock(vnode, key); /* RPC */
333 		key_put(key);
334 
335 		if (ret < 0) {
336 			trace_afs_flock_ev(vnode, NULL, afs_flock_extend_fail,
337 					   ret);
338 			pr_warn("AFS: Failed to extend lock on {%llx:%llx} error %d\n",
339 				vnode->fid.vid, vnode->fid.vnode, ret);
340 		}
341 
342 		spin_lock(&vnode->lock);
343 
344 		if (ret == -ENOENT) {
345 			afs_kill_lockers_enoent(vnode);
346 			spin_unlock(&vnode->lock);
347 			return;
348 		}
349 
350 		if (vnode->lock_state != AFS_VNODE_LOCK_EXTENDING)
351 			goto again;
352 		afs_set_lock_state(vnode, AFS_VNODE_LOCK_GRANTED);
353 
354 		if (ret != 0)
355 			queue_delayed_work(afs_lock_manager, &vnode->lock_work,
356 					   HZ * 10);
357 		spin_unlock(&vnode->lock);
358 		_leave(" [ext]");
359 		return;
360 
361 	/* If we're waiting for a callback to indicate lock release, we can't
362 	 * actually rely on this, so need to recheck at regular intervals.  The
363 	 * problem is that the server might not notify us if the lock just
364 	 * expires (say because a client died) rather than being explicitly
365 	 * released.
366 	 */
367 	case AFS_VNODE_LOCK_WAITING_FOR_CB:
368 		_debug("retry");
369 		afs_next_locker(vnode, 0);
370 		spin_unlock(&vnode->lock);
371 		return;
372 
373 	case AFS_VNODE_LOCK_DELETED:
374 		afs_kill_lockers_enoent(vnode);
375 		spin_unlock(&vnode->lock);
376 		return;
377 
378 	default:
379 		/* Looks like a lock request was withdrawn. */
380 		spin_unlock(&vnode->lock);
381 		_leave(" [no]");
382 		return;
383 	}
384 }
385 
386 /*
387  * pass responsibility for the unlocking of a vnode on the server to the
388  * manager thread, lest a pending signal in the calling thread interrupt
389  * AF_RXRPC
390  * - the caller must hold the vnode lock
391  */
afs_defer_unlock(struct afs_vnode * vnode)392 static void afs_defer_unlock(struct afs_vnode *vnode)
393 {
394 	_enter("%u", vnode->lock_state);
395 
396 	if (list_empty(&vnode->granted_locks) &&
397 	    (vnode->lock_state == AFS_VNODE_LOCK_GRANTED ||
398 	     vnode->lock_state == AFS_VNODE_LOCK_EXTENDING)) {
399 		cancel_delayed_work(&vnode->lock_work);
400 
401 		afs_set_lock_state(vnode, AFS_VNODE_LOCK_NEED_UNLOCK);
402 		trace_afs_flock_ev(vnode, NULL, afs_flock_defer_unlock, 0);
403 		queue_delayed_work(afs_lock_manager, &vnode->lock_work, 0);
404 	}
405 }
406 
407 /*
408  * Check that our view of the file metadata is up to date and check to see
409  * whether we think that we have a locking permit.
410  */
afs_do_setlk_check(struct afs_vnode * vnode,struct key * key,enum afs_flock_mode mode,afs_lock_type_t type)411 static int afs_do_setlk_check(struct afs_vnode *vnode, struct key *key,
412 			      enum afs_flock_mode mode, afs_lock_type_t type)
413 {
414 	afs_access_t access;
415 	int ret;
416 
417 	/* Make sure we've got a callback on this file and that our view of the
418 	 * data version is up to date.
419 	 */
420 	ret = afs_validate(vnode, key);
421 	if (ret < 0)
422 		return ret;
423 
424 	/* Check the permission set to see if we're actually going to be
425 	 * allowed to get a lock on this file.
426 	 */
427 	ret = afs_check_permit(vnode, key, &access);
428 	if (ret < 0)
429 		return ret;
430 
431 	/* At a rough estimation, you need LOCK, WRITE or INSERT perm to
432 	 * read-lock a file and WRITE or INSERT perm to write-lock a file.
433 	 *
434 	 * We can't rely on the server to do this for us since if we want to
435 	 * share a read lock that we already have, we won't go the server.
436 	 */
437 	if (type == AFS_LOCK_READ) {
438 		if (!(access & (AFS_ACE_INSERT | AFS_ACE_WRITE | AFS_ACE_LOCK)))
439 			return -EACCES;
440 	} else {
441 		if (!(access & (AFS_ACE_INSERT | AFS_ACE_WRITE)))
442 			return -EACCES;
443 	}
444 
445 	return 0;
446 }
447 
448 /*
449  * request a lock on a file on the server
450  */
afs_do_setlk(struct file * file,struct file_lock * fl)451 static int afs_do_setlk(struct file *file, struct file_lock *fl)
452 {
453 	struct inode *inode = file_inode(file);
454 	struct afs_vnode *vnode = AFS_FS_I(inode);
455 	enum afs_flock_mode mode = AFS_FS_S(inode->i_sb)->flock_mode;
456 	afs_lock_type_t type;
457 	struct key *key = afs_file_key(file);
458 	bool partial, no_server_lock = false;
459 	int ret;
460 
461 	if (mode == afs_flock_mode_unset)
462 		mode = afs_flock_mode_openafs;
463 
464 	_enter("{%llx:%llu},%llu-%llu,%u,%u",
465 	       vnode->fid.vid, vnode->fid.vnode,
466 	       fl->fl_start, fl->fl_end, fl->c.flc_type, mode);
467 
468 	fl->fl_ops = &afs_lock_ops;
469 	INIT_LIST_HEAD(&fl->fl_u.afs.link);
470 	fl->fl_u.afs.state = AFS_LOCK_PENDING;
471 
472 	partial = (fl->fl_start != 0 || fl->fl_end != OFFSET_MAX);
473 	type = lock_is_read(fl) ? AFS_LOCK_READ : AFS_LOCK_WRITE;
474 	if (mode == afs_flock_mode_write && partial)
475 		type = AFS_LOCK_WRITE;
476 
477 	ret = afs_do_setlk_check(vnode, key, mode, type);
478 	if (ret < 0)
479 		return ret;
480 
481 	trace_afs_flock_op(vnode, fl, afs_flock_op_set_lock);
482 
483 	/* AFS3 protocol only supports full-file locks and doesn't provide any
484 	 * method of upgrade/downgrade, so we need to emulate for partial-file
485 	 * locks.
486 	 *
487 	 * The OpenAFS client only gets a server lock for a full-file lock and
488 	 * keeps partial-file locks local.  Allow this behaviour to be emulated
489 	 * (as the default).
490 	 */
491 	if (mode == afs_flock_mode_local ||
492 	    (partial && mode == afs_flock_mode_openafs)) {
493 		no_server_lock = true;
494 		goto skip_server_lock;
495 	}
496 
497 	spin_lock(&vnode->lock);
498 	list_add_tail(&fl->fl_u.afs.link, &vnode->pending_locks);
499 
500 	ret = -ENOENT;
501 	if (vnode->lock_state == AFS_VNODE_LOCK_DELETED)
502 		goto error_unlock;
503 
504 	/* If we've already got a lock on the server then try to move to having
505 	 * the VFS grant the requested lock.  Note that this means that other
506 	 * clients may get starved out.
507 	 */
508 	_debug("try %u", vnode->lock_state);
509 	if (vnode->lock_state == AFS_VNODE_LOCK_GRANTED) {
510 		if (type == AFS_LOCK_READ) {
511 			_debug("instant readlock");
512 			list_move_tail(&fl->fl_u.afs.link, &vnode->granted_locks);
513 			fl->fl_u.afs.state = AFS_LOCK_GRANTED;
514 			goto vnode_is_locked_u;
515 		}
516 
517 		if (vnode->lock_type == AFS_LOCK_WRITE) {
518 			_debug("instant writelock");
519 			list_move_tail(&fl->fl_u.afs.link, &vnode->granted_locks);
520 			fl->fl_u.afs.state = AFS_LOCK_GRANTED;
521 			goto vnode_is_locked_u;
522 		}
523 	}
524 
525 	if (vnode->lock_state == AFS_VNODE_LOCK_NONE &&
526 	    !(fl->c.flc_flags & FL_SLEEP)) {
527 		ret = -EAGAIN;
528 		if (type == AFS_LOCK_READ) {
529 			if (vnode->status.lock_count == -1)
530 				goto lock_is_contended; /* Write locked */
531 		} else {
532 			if (vnode->status.lock_count != 0)
533 				goto lock_is_contended; /* Locked */
534 		}
535 	}
536 
537 	if (vnode->lock_state != AFS_VNODE_LOCK_NONE)
538 		goto need_to_wait;
539 
540 try_to_lock:
541 	/* We don't have a lock on this vnode and we aren't currently waiting
542 	 * for one either, so ask the server for a lock.
543 	 *
544 	 * Note that we need to be careful if we get interrupted by a signal
545 	 * after dispatching the request as we may still get the lock, even
546 	 * though we don't wait for the reply (it's not too bad a problem - the
547 	 * lock will expire in 5 mins anyway).
548 	 */
549 	trace_afs_flock_ev(vnode, fl, afs_flock_try_to_lock, 0);
550 	vnode->lock_key = key_get(key);
551 	vnode->lock_type = type;
552 	afs_set_lock_state(vnode, AFS_VNODE_LOCK_SETTING);
553 	spin_unlock(&vnode->lock);
554 
555 	ret = afs_set_lock(vnode, key, type); /* RPC */
556 
557 	spin_lock(&vnode->lock);
558 	switch (ret) {
559 	case -EKEYREJECTED:
560 	case -EKEYEXPIRED:
561 	case -EKEYREVOKED:
562 	case -EPERM:
563 	case -EACCES:
564 		fl->fl_u.afs.state = ret;
565 		trace_afs_flock_ev(vnode, fl, afs_flock_fail_perm, ret);
566 		list_del_init(&fl->fl_u.afs.link);
567 		afs_next_locker(vnode, ret);
568 		goto error_unlock;
569 
570 	case -ENOENT:
571 		fl->fl_u.afs.state = ret;
572 		trace_afs_flock_ev(vnode, fl, afs_flock_fail_other, ret);
573 		list_del_init(&fl->fl_u.afs.link);
574 		afs_kill_lockers_enoent(vnode);
575 		goto error_unlock;
576 
577 	default:
578 		fl->fl_u.afs.state = ret;
579 		trace_afs_flock_ev(vnode, fl, afs_flock_fail_other, ret);
580 		list_del_init(&fl->fl_u.afs.link);
581 		afs_next_locker(vnode, 0);
582 		goto error_unlock;
583 
584 	case -EWOULDBLOCK:
585 		/* The server doesn't have a lock-waiting queue, so the client
586 		 * will have to retry.  The server will break the outstanding
587 		 * callbacks on a file when a lock is released.
588 		 */
589 		ASSERT(list_empty(&vnode->granted_locks));
590 		ASSERTCMP(vnode->pending_locks.next, ==, &fl->fl_u.afs.link);
591 		goto lock_is_contended;
592 
593 	case 0:
594 		afs_set_lock_state(vnode, AFS_VNODE_LOCK_GRANTED);
595 		trace_afs_flock_ev(vnode, fl, afs_flock_acquired, type);
596 		afs_grant_locks(vnode);
597 		goto vnode_is_locked_u;
598 	}
599 
600 vnode_is_locked_u:
601 	spin_unlock(&vnode->lock);
602 vnode_is_locked:
603 	/* the lock has been granted by the server... */
604 	ASSERTCMP(fl->fl_u.afs.state, ==, AFS_LOCK_GRANTED);
605 
606 skip_server_lock:
607 	/* ... but the VFS still needs to distribute access on this client. */
608 	trace_afs_flock_ev(vnode, fl, afs_flock_vfs_locking, 0);
609 	ret = locks_lock_file_wait(file, fl);
610 	trace_afs_flock_ev(vnode, fl, afs_flock_vfs_lock, ret);
611 	if (ret < 0)
612 		goto vfs_rejected_lock;
613 
614 	/* Again, make sure we've got a callback on this file and, again, make
615 	 * sure that our view of the data version is up to date (we ignore
616 	 * errors incurred here and deal with the consequences elsewhere).
617 	 */
618 	afs_validate(vnode, key);
619 	_leave(" = 0");
620 	return 0;
621 
622 lock_is_contended:
623 	if (!(fl->c.flc_flags & FL_SLEEP)) {
624 		list_del_init(&fl->fl_u.afs.link);
625 		afs_next_locker(vnode, 0);
626 		ret = -EAGAIN;
627 		goto error_unlock;
628 	}
629 
630 	afs_set_lock_state(vnode, AFS_VNODE_LOCK_WAITING_FOR_CB);
631 	trace_afs_flock_ev(vnode, fl, afs_flock_would_block, ret);
632 	queue_delayed_work(afs_lock_manager, &vnode->lock_work, HZ * 5);
633 
634 need_to_wait:
635 	/* We're going to have to wait.  Either this client doesn't have a lock
636 	 * on the server yet and we need to wait for a callback to occur, or
637 	 * the client does have a lock on the server, but it's shared and we
638 	 * need an exclusive lock.
639 	 */
640 	spin_unlock(&vnode->lock);
641 
642 	trace_afs_flock_ev(vnode, fl, afs_flock_waiting, 0);
643 	ret = wait_event_interruptible(fl->c.flc_wait,
644 				       fl->fl_u.afs.state != AFS_LOCK_PENDING);
645 	trace_afs_flock_ev(vnode, fl, afs_flock_waited, ret);
646 
647 	if (fl->fl_u.afs.state >= 0 && fl->fl_u.afs.state != AFS_LOCK_GRANTED) {
648 		spin_lock(&vnode->lock);
649 
650 		switch (fl->fl_u.afs.state) {
651 		case AFS_LOCK_YOUR_TRY:
652 			fl->fl_u.afs.state = AFS_LOCK_PENDING;
653 			goto try_to_lock;
654 		case AFS_LOCK_PENDING:
655 			if (ret > 0) {
656 				/* We need to retry the lock.  We may not be
657 				 * notified by the server if it just expired
658 				 * rather than being released.
659 				 */
660 				ASSERTCMP(vnode->lock_state, ==, AFS_VNODE_LOCK_WAITING_FOR_CB);
661 				afs_set_lock_state(vnode, AFS_VNODE_LOCK_SETTING);
662 				fl->fl_u.afs.state = AFS_LOCK_PENDING;
663 				goto try_to_lock;
664 			}
665 			goto error_unlock;
666 		case AFS_LOCK_GRANTED:
667 		default:
668 			break;
669 		}
670 
671 		spin_unlock(&vnode->lock);
672 	}
673 
674 	if (fl->fl_u.afs.state == AFS_LOCK_GRANTED)
675 		goto vnode_is_locked;
676 	ret = fl->fl_u.afs.state;
677 	goto error;
678 
679 vfs_rejected_lock:
680 	/* The VFS rejected the lock we just obtained, so we have to discard
681 	 * what we just got.  We defer this to the lock manager work item to
682 	 * deal with.
683 	 */
684 	_debug("vfs refused %d", ret);
685 	if (no_server_lock)
686 		goto error;
687 	spin_lock(&vnode->lock);
688 	list_del_init(&fl->fl_u.afs.link);
689 	afs_defer_unlock(vnode);
690 
691 error_unlock:
692 	spin_unlock(&vnode->lock);
693 error:
694 	_leave(" = %d", ret);
695 	return ret;
696 }
697 
698 /*
699  * unlock on a file on the server
700  */
afs_do_unlk(struct file * file,struct file_lock * fl)701 static int afs_do_unlk(struct file *file, struct file_lock *fl)
702 {
703 	struct afs_vnode *vnode = AFS_FS_I(file_inode(file));
704 	int ret;
705 
706 	_enter("{%llx:%llu},%u", vnode->fid.vid, vnode->fid.vnode,
707 	       fl->c.flc_type);
708 
709 	trace_afs_flock_op(vnode, fl, afs_flock_op_unlock);
710 
711 	/* Flush all pending writes before doing anything with locks. */
712 	vfs_fsync(file, 0);
713 
714 	ret = locks_lock_file_wait(file, fl);
715 	_leave(" = %d [%u]", ret, vnode->lock_state);
716 	return ret;
717 }
718 
719 /*
720  * return information about a lock we currently hold, if indeed we hold one
721  */
afs_do_getlk(struct file * file,struct file_lock * fl)722 static int afs_do_getlk(struct file *file, struct file_lock *fl)
723 {
724 	struct afs_vnode *vnode = AFS_FS_I(file_inode(file));
725 	struct key *key = afs_file_key(file);
726 	int ret, lock_count;
727 
728 	_enter("");
729 
730 	if (vnode->lock_state == AFS_VNODE_LOCK_DELETED)
731 		return -ENOENT;
732 
733 	fl->c.flc_type = F_UNLCK;
734 
735 	/* check local lock records first */
736 	posix_test_lock(file, fl);
737 	if (lock_is_unlock(fl)) {
738 		/* no local locks; consult the server */
739 		ret = afs_fetch_status(vnode, key, false, NULL);
740 		if (ret < 0)
741 			goto error;
742 
743 		lock_count = READ_ONCE(vnode->status.lock_count);
744 		if (lock_count != 0) {
745 			if (lock_count > 0)
746 				fl->c.flc_type = F_RDLCK;
747 			else
748 				fl->c.flc_type = F_WRLCK;
749 			fl->fl_start = 0;
750 			fl->fl_end = OFFSET_MAX;
751 			fl->c.flc_pid = 0;
752 		}
753 	}
754 
755 	ret = 0;
756 error:
757 	_leave(" = %d [%hd]", ret, fl->c.flc_type);
758 	return ret;
759 }
760 
761 /*
762  * manage POSIX locks on a file
763  */
afs_lock(struct file * file,int cmd,struct file_lock * fl)764 int afs_lock(struct file *file, int cmd, struct file_lock *fl)
765 {
766 	struct afs_vnode *vnode = AFS_FS_I(file_inode(file));
767 	enum afs_flock_operation op;
768 	int ret;
769 
770 	_enter("{%llx:%llu},%d,{t=%x,fl=%x,r=%Ld:%Ld}",
771 	       vnode->fid.vid, vnode->fid.vnode, cmd,
772 	       fl->c.flc_type, fl->c.flc_flags,
773 	       (long long) fl->fl_start, (long long) fl->fl_end);
774 
775 	if (IS_GETLK(cmd))
776 		return afs_do_getlk(file, fl);
777 
778 	fl->fl_u.afs.debug_id = atomic_inc_return(&afs_file_lock_debug_id);
779 	trace_afs_flock_op(vnode, fl, afs_flock_op_lock);
780 
781 	if (lock_is_unlock(fl))
782 		ret = afs_do_unlk(file, fl);
783 	else
784 		ret = afs_do_setlk(file, fl);
785 
786 	switch (ret) {
787 	case 0:		op = afs_flock_op_return_ok; break;
788 	case -EAGAIN:	op = afs_flock_op_return_eagain; break;
789 	case -EDEADLK:	op = afs_flock_op_return_edeadlk; break;
790 	default:	op = afs_flock_op_return_error; break;
791 	}
792 	trace_afs_flock_op(vnode, fl, op);
793 	return ret;
794 }
795 
796 /*
797  * manage FLOCK locks on a file
798  */
afs_flock(struct file * file,int cmd,struct file_lock * fl)799 int afs_flock(struct file *file, int cmd, struct file_lock *fl)
800 {
801 	struct afs_vnode *vnode = AFS_FS_I(file_inode(file));
802 	enum afs_flock_operation op;
803 	int ret;
804 
805 	_enter("{%llx:%llu},%d,{t=%x,fl=%x}",
806 	       vnode->fid.vid, vnode->fid.vnode, cmd,
807 	       fl->c.flc_type, fl->c.flc_flags);
808 
809 	/*
810 	 * No BSD flocks over NFS allowed.
811 	 * Note: we could try to fake a POSIX lock request here by
812 	 * using ((u32) filp | 0x80000000) or some such as the pid.
813 	 * Not sure whether that would be unique, though, or whether
814 	 * that would break in other places.
815 	 */
816 	if (!(fl->c.flc_flags & FL_FLOCK))
817 		return -ENOLCK;
818 
819 	fl->fl_u.afs.debug_id = atomic_inc_return(&afs_file_lock_debug_id);
820 	trace_afs_flock_op(vnode, fl, afs_flock_op_flock);
821 
822 	/* we're simulating flock() locks using posix locks on the server */
823 	if (lock_is_unlock(fl))
824 		ret = afs_do_unlk(file, fl);
825 	else
826 		ret = afs_do_setlk(file, fl);
827 
828 	switch (ret) {
829 	case 0:		op = afs_flock_op_return_ok; break;
830 	case -EAGAIN:	op = afs_flock_op_return_eagain; break;
831 	case -EDEADLK:	op = afs_flock_op_return_edeadlk; break;
832 	default:	op = afs_flock_op_return_error; break;
833 	}
834 	trace_afs_flock_op(vnode, fl, op);
835 	return ret;
836 }
837 
838 /*
839  * the POSIX lock management core VFS code copies the lock record and adds the
840  * copy into its own list, so we need to add that copy to the vnode's lock
841  * queue in the same place as the original (which will be deleted shortly
842  * after)
843  */
afs_fl_copy_lock(struct file_lock * new,struct file_lock * fl)844 static void afs_fl_copy_lock(struct file_lock *new, struct file_lock *fl)
845 {
846 	struct afs_vnode *vnode = AFS_FS_I(file_inode(fl->c.flc_file));
847 
848 	_enter("");
849 
850 	new->fl_u.afs.debug_id = atomic_inc_return(&afs_file_lock_debug_id);
851 
852 	spin_lock(&vnode->lock);
853 	trace_afs_flock_op(vnode, new, afs_flock_op_copy_lock);
854 	list_add(&new->fl_u.afs.link, &fl->fl_u.afs.link);
855 	spin_unlock(&vnode->lock);
856 }
857 
858 /*
859  * need to remove this lock from the vnode queue when it's removed from the
860  * VFS's list
861  */
afs_fl_release_private(struct file_lock * fl)862 static void afs_fl_release_private(struct file_lock *fl)
863 {
864 	struct afs_vnode *vnode = AFS_FS_I(file_inode(fl->c.flc_file));
865 
866 	_enter("");
867 
868 	spin_lock(&vnode->lock);
869 
870 	trace_afs_flock_op(vnode, fl, afs_flock_op_release_lock);
871 	list_del_init(&fl->fl_u.afs.link);
872 	if (list_empty(&vnode->granted_locks))
873 		afs_defer_unlock(vnode);
874 
875 	_debug("state %u for %p", vnode->lock_state, vnode);
876 	spin_unlock(&vnode->lock);
877 }
878