xref: /linux/fs/afs/flock.c (revision 8b4a40809e5330c9da5d20107d693d92d73b31dc)
1 /* AFS file locking support
2  *
3  * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
4  * Written by David Howells (dhowells@redhat.com)
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public License
8  * as published by the Free Software Foundation; either version
9  * 2 of the License, or (at your option) any later version.
10  */
11 
12 #include <linux/smp_lock.h>
13 #include "internal.h"
14 
15 #define AFS_LOCK_GRANTED	0
16 #define AFS_LOCK_PENDING	1
17 
18 static void afs_fl_copy_lock(struct file_lock *new, struct file_lock *fl);
19 static void afs_fl_release_private(struct file_lock *fl);
20 
21 static struct workqueue_struct *afs_lock_manager;
22 
23 static struct file_lock_operations afs_lock_ops = {
24 	.fl_copy_lock		= afs_fl_copy_lock,
25 	.fl_release_private	= afs_fl_release_private,
26 };
27 
28 /*
29  * initialise the lock manager thread if it isn't already running
30  */
31 static int afs_init_lock_manager(void)
32 {
33 	if (!afs_lock_manager) {
34 		afs_lock_manager = create_singlethread_workqueue("kafs_lockd");
35 		if (!afs_lock_manager)
36 			return -ENOMEM;
37 	}
38 	return 0;
39 }
40 
41 /*
42  * destroy the lock manager thread if it's running
43  */
44 void __exit afs_kill_lock_manager(void)
45 {
46 	if (afs_lock_manager)
47 		destroy_workqueue(afs_lock_manager);
48 }
49 
50 /*
51  * if the callback is broken on this vnode, then the lock may now be available
52  */
53 void afs_lock_may_be_available(struct afs_vnode *vnode)
54 {
55 	_enter("{%x:%u}", vnode->fid.vid, vnode->fid.vnode);
56 
57 	queue_delayed_work(afs_lock_manager, &vnode->lock_work, 0);
58 }
59 
60 /*
61  * the lock will time out in 5 minutes unless we extend it, so schedule
62  * extension in a bit less than that time
63  */
64 static void afs_schedule_lock_extension(struct afs_vnode *vnode)
65 {
66 	queue_delayed_work(afs_lock_manager, &vnode->lock_work,
67 			   AFS_LOCKWAIT * HZ / 2);
68 }
69 
70 /*
71  * do work for a lock, including:
72  * - probing for a lock we're waiting on but didn't get immediately
73  * - extending a lock that's close to timing out
74  */
75 void afs_lock_work(struct work_struct *work)
76 {
77 	struct afs_vnode *vnode =
78 		container_of(work, struct afs_vnode, lock_work.work);
79 	struct file_lock *fl;
80 	afs_lock_type_t type;
81 	struct key *key;
82 	int ret;
83 
84 	_enter("{%x:%u}", vnode->fid.vid, vnode->fid.vnode);
85 
86 	spin_lock(&vnode->lock);
87 
88 	if (test_bit(AFS_VNODE_UNLOCKING, &vnode->flags)) {
89 		_debug("unlock");
90 		spin_unlock(&vnode->lock);
91 
92 		/* attempt to release the server lock; if it fails, we just
93 		 * wait 5 minutes and it'll time out anyway */
94 		ret = afs_vnode_release_lock(vnode, vnode->unlock_key);
95 		if (ret < 0)
96 			printk(KERN_WARNING "AFS:"
97 			       " Failed to release lock on {%x:%x} error %d\n",
98 			       vnode->fid.vid, vnode->fid.vnode, ret);
99 
100 		spin_lock(&vnode->lock);
101 		key_put(vnode->unlock_key);
102 		vnode->unlock_key = NULL;
103 		clear_bit(AFS_VNODE_UNLOCKING, &vnode->flags);
104 	}
105 
106 	/* if we've got a lock, then it must be time to extend that lock as AFS
107 	 * locks time out after 5 minutes */
108 	if (!list_empty(&vnode->granted_locks)) {
109 		_debug("extend");
110 
111 		if (test_and_set_bit(AFS_VNODE_LOCKING, &vnode->flags))
112 			BUG();
113 		fl = list_entry(vnode->granted_locks.next,
114 				struct file_lock, fl_u.afs.link);
115 		key = key_get(fl->fl_file->private_data);
116 		spin_unlock(&vnode->lock);
117 
118 		ret = afs_vnode_extend_lock(vnode, key);
119 		clear_bit(AFS_VNODE_LOCKING, &vnode->flags);
120 		key_put(key);
121 		switch (ret) {
122 		case 0:
123 			afs_schedule_lock_extension(vnode);
124 			break;
125 		default:
126 			/* ummm... we failed to extend the lock - retry
127 			 * extension shortly */
128 			printk(KERN_WARNING "AFS:"
129 			       " Failed to extend lock on {%x:%x} error %d\n",
130 			       vnode->fid.vid, vnode->fid.vnode, ret);
131 			queue_delayed_work(afs_lock_manager, &vnode->lock_work,
132 					   HZ * 10);
133 			break;
134 		}
135 		_leave(" [extend]");
136 		return;
137 	}
138 
139 	/* if we don't have a granted lock, then we must've been called back by
140 	 * the server, and so if might be possible to get a lock we're
141 	 * currently waiting for */
142 	if (!list_empty(&vnode->pending_locks)) {
143 		_debug("get");
144 
145 		if (test_and_set_bit(AFS_VNODE_LOCKING, &vnode->flags))
146 			BUG();
147 		fl = list_entry(vnode->pending_locks.next,
148 				struct file_lock, fl_u.afs.link);
149 		key = key_get(fl->fl_file->private_data);
150 		type = (fl->fl_type == F_RDLCK) ?
151 			AFS_LOCK_READ : AFS_LOCK_WRITE;
152 		spin_unlock(&vnode->lock);
153 
154 		ret = afs_vnode_set_lock(vnode, key, type);
155 		clear_bit(AFS_VNODE_LOCKING, &vnode->flags);
156 		switch (ret) {
157 		case -EWOULDBLOCK:
158 			_debug("blocked");
159 			break;
160 		case 0:
161 			_debug("acquired");
162 			if (type == AFS_LOCK_READ)
163 				set_bit(AFS_VNODE_READLOCKED, &vnode->flags);
164 			else
165 				set_bit(AFS_VNODE_WRITELOCKED, &vnode->flags);
166 			ret = AFS_LOCK_GRANTED;
167 		default:
168 			spin_lock(&vnode->lock);
169 			/* the pending lock may have been withdrawn due to a
170 			 * signal */
171 			if (list_entry(vnode->pending_locks.next,
172 				       struct file_lock, fl_u.afs.link) == fl) {
173 				fl->fl_u.afs.state = ret;
174 				if (ret == AFS_LOCK_GRANTED)
175 					list_move_tail(&fl->fl_u.afs.link,
176 						       &vnode->granted_locks);
177 				else
178 					list_del_init(&fl->fl_u.afs.link);
179 				wake_up(&fl->fl_wait);
180 				spin_unlock(&vnode->lock);
181 			} else {
182 				_debug("withdrawn");
183 				clear_bit(AFS_VNODE_READLOCKED, &vnode->flags);
184 				clear_bit(AFS_VNODE_WRITELOCKED, &vnode->flags);
185 				spin_unlock(&vnode->lock);
186 				afs_vnode_release_lock(vnode, key);
187 				if (!list_empty(&vnode->pending_locks))
188 					afs_lock_may_be_available(vnode);
189 			}
190 			break;
191 		}
192 		key_put(key);
193 		_leave(" [pend]");
194 		return;
195 	}
196 
197 	/* looks like the lock request was withdrawn on a signal */
198 	spin_unlock(&vnode->lock);
199 	_leave(" [no locks]");
200 }
201 
202 /*
203  * pass responsibility for the unlocking of a vnode on the server to the
204  * manager thread, lest a pending signal in the calling thread interrupt
205  * AF_RXRPC
206  * - the caller must hold the vnode lock
207  */
208 static void afs_defer_unlock(struct afs_vnode *vnode, struct key *key)
209 {
210 	cancel_delayed_work(&vnode->lock_work);
211 	if (!test_and_clear_bit(AFS_VNODE_READLOCKED, &vnode->flags) &&
212 	    !test_and_clear_bit(AFS_VNODE_WRITELOCKED, &vnode->flags))
213 		BUG();
214 	if (test_and_set_bit(AFS_VNODE_UNLOCKING, &vnode->flags))
215 		BUG();
216 	vnode->unlock_key = key_get(key);
217 	afs_lock_may_be_available(vnode);
218 }
219 
220 /*
221  * request a lock on a file on the server
222  */
223 static int afs_do_setlk(struct file *file, struct file_lock *fl)
224 {
225 	struct afs_vnode *vnode = AFS_FS_I(file->f_mapping->host);
226 	afs_lock_type_t type;
227 	struct key *key = file->private_data;
228 	int ret;
229 
230 	_enter("{%x:%u},%u", vnode->fid.vid, vnode->fid.vnode, fl->fl_type);
231 
232 	/* only whole-file locks are supported */
233 	if (fl->fl_start != 0 || fl->fl_end != OFFSET_MAX)
234 		return -EINVAL;
235 
236 	ret = afs_init_lock_manager();
237 	if (ret < 0)
238 		return ret;
239 
240 	fl->fl_ops = &afs_lock_ops;
241 	INIT_LIST_HEAD(&fl->fl_u.afs.link);
242 	fl->fl_u.afs.state = AFS_LOCK_PENDING;
243 
244 	type = (fl->fl_type == F_RDLCK) ? AFS_LOCK_READ : AFS_LOCK_WRITE;
245 
246 	lock_kernel();
247 
248 	/* make sure we've got a callback on this file and that our view of the
249 	 * data version is up to date */
250 	ret = afs_vnode_fetch_status(vnode, NULL, key);
251 	if (ret < 0)
252 		goto error;
253 
254 	if (vnode->status.lock_count != 0 && !(fl->fl_flags & FL_SLEEP)) {
255 		ret = -EAGAIN;
256 		goto error;
257 	}
258 
259 	spin_lock(&vnode->lock);
260 
261 	if (list_empty(&vnode->pending_locks)) {
262 		/* if there's no-one else with a lock on this vnode, then we
263 		 * need to ask the server for a lock */
264 		if (list_empty(&vnode->granted_locks)) {
265 			_debug("not locked");
266 			ASSERTCMP(vnode->flags &
267 				  ((1 << AFS_VNODE_LOCKING) |
268 				   (1 << AFS_VNODE_READLOCKED) |
269 				   (1 << AFS_VNODE_WRITELOCKED)), ==, 0);
270 			list_add_tail(&fl->fl_u.afs.link, &vnode->pending_locks);
271 			set_bit(AFS_VNODE_LOCKING, &vnode->flags);
272 			spin_unlock(&vnode->lock);
273 
274 			ret = afs_vnode_set_lock(vnode, key, type);
275 			clear_bit(AFS_VNODE_LOCKING, &vnode->flags);
276 			switch (ret) {
277 			case 0:
278 				goto acquired_server_lock;
279 			case -EWOULDBLOCK:
280 				spin_lock(&vnode->lock);
281 				ASSERT(list_empty(&vnode->granted_locks));
282 				ASSERTCMP(vnode->pending_locks.next, ==,
283 					  &fl->fl_u.afs.link);
284 				goto wait;
285 			default:
286 				spin_lock(&vnode->lock);
287 				list_del_init(&fl->fl_u.afs.link);
288 				spin_unlock(&vnode->lock);
289 				goto error;
290 			}
291 		}
292 
293 		/* if we've already got a readlock on the server and no waiting
294 		 * writelocks, then we might be able to instantly grant another
295 		 * readlock */
296 		if (type == AFS_LOCK_READ &&
297 		    vnode->flags & (1 << AFS_VNODE_READLOCKED)) {
298 			_debug("instant readlock");
299 			ASSERTCMP(vnode->flags &
300 				  ((1 << AFS_VNODE_LOCKING) |
301 				   (1 << AFS_VNODE_WRITELOCKED)), ==, 0);
302 			ASSERT(!list_empty(&vnode->granted_locks));
303 			goto sharing_existing_lock;
304 		}
305 	}
306 
307 	/* otherwise, we need to wait for a local lock to become available */
308 	_debug("wait local");
309 	list_add_tail(&fl->fl_u.afs.link, &vnode->pending_locks);
310 wait:
311 	if (!(fl->fl_flags & FL_SLEEP)) {
312 		_debug("noblock");
313 		ret = -EAGAIN;
314 		goto abort_attempt;
315 	}
316 	spin_unlock(&vnode->lock);
317 
318 	/* now we need to sleep and wait for the lock manager thread to get the
319 	 * lock from the server */
320 	_debug("sleep");
321 	ret = wait_event_interruptible(fl->fl_wait,
322 				       fl->fl_u.afs.state <= AFS_LOCK_GRANTED);
323 	if (fl->fl_u.afs.state <= AFS_LOCK_GRANTED) {
324 		ret = fl->fl_u.afs.state;
325 		if (ret < 0)
326 			goto error;
327 		spin_lock(&vnode->lock);
328 		goto given_lock;
329 	}
330 
331 	/* we were interrupted, but someone may still be in the throes of
332 	 * giving us the lock */
333 	_debug("intr");
334 	ASSERTCMP(ret, ==, -ERESTARTSYS);
335 
336 	spin_lock(&vnode->lock);
337 	if (fl->fl_u.afs.state <= AFS_LOCK_GRANTED) {
338 		ret = fl->fl_u.afs.state;
339 		if (ret < 0) {
340 			spin_unlock(&vnode->lock);
341 			goto error;
342 		}
343 		goto given_lock;
344 	}
345 
346 abort_attempt:
347 	/* we aren't going to get the lock, either because we're unwilling to
348 	 * wait, or because some signal happened */
349 	_debug("abort");
350 	if (list_empty(&vnode->granted_locks) &&
351 	    vnode->pending_locks.next == &fl->fl_u.afs.link) {
352 		if (vnode->pending_locks.prev != &fl->fl_u.afs.link) {
353 			/* kick the next pending lock into having a go */
354 			list_del_init(&fl->fl_u.afs.link);
355 			afs_lock_may_be_available(vnode);
356 		}
357 	} else {
358 		list_del_init(&fl->fl_u.afs.link);
359 	}
360 	spin_unlock(&vnode->lock);
361 	goto error;
362 
363 acquired_server_lock:
364 	/* we've acquired a server lock, but it needs to be renewed after 5
365 	 * mins */
366 	spin_lock(&vnode->lock);
367 	afs_schedule_lock_extension(vnode);
368 	if (type == AFS_LOCK_READ)
369 		set_bit(AFS_VNODE_READLOCKED, &vnode->flags);
370 	else
371 		set_bit(AFS_VNODE_WRITELOCKED, &vnode->flags);
372 sharing_existing_lock:
373 	/* the lock has been granted as far as we're concerned... */
374 	fl->fl_u.afs.state = AFS_LOCK_GRANTED;
375 	list_move_tail(&fl->fl_u.afs.link, &vnode->granted_locks);
376 given_lock:
377 	/* ... but we do still need to get the VFS's blessing */
378 	ASSERT(!(vnode->flags & (1 << AFS_VNODE_LOCKING)));
379 	ASSERT((vnode->flags & ((1 << AFS_VNODE_READLOCKED) |
380 				(1 << AFS_VNODE_WRITELOCKED))) != 0);
381 	ret = posix_lock_file(file, fl, NULL);
382 	if (ret < 0)
383 		goto vfs_rejected_lock;
384 	spin_unlock(&vnode->lock);
385 
386 	/* again, make sure we've got a callback on this file and, again, make
387 	 * sure that our view of the data version is up to date (we ignore
388 	 * errors incurred here and deal with the consequences elsewhere) */
389 	afs_vnode_fetch_status(vnode, NULL, key);
390 
391 error:
392 	unlock_kernel();
393 	_leave(" = %d", ret);
394 	return ret;
395 
396 vfs_rejected_lock:
397 	/* the VFS rejected the lock we just obtained, so we have to discard
398 	 * what we just got */
399 	_debug("vfs refused %d", ret);
400 	list_del_init(&fl->fl_u.afs.link);
401 	if (list_empty(&vnode->granted_locks))
402 		afs_defer_unlock(vnode, key);
403 	spin_unlock(&vnode->lock);
404 	goto abort_attempt;
405 }
406 
407 /*
408  * unlock on a file on the server
409  */
410 static int afs_do_unlk(struct file *file, struct file_lock *fl)
411 {
412 	struct afs_vnode *vnode = AFS_FS_I(file->f_mapping->host);
413 	struct key *key = file->private_data;
414 	int ret;
415 
416 	_enter("{%x:%u},%u", vnode->fid.vid, vnode->fid.vnode, fl->fl_type);
417 
418 	/* only whole-file unlocks are supported */
419 	if (fl->fl_start != 0 || fl->fl_end != OFFSET_MAX)
420 		return -EINVAL;
421 
422 	fl->fl_ops = &afs_lock_ops;
423 	INIT_LIST_HEAD(&fl->fl_u.afs.link);
424 	fl->fl_u.afs.state = AFS_LOCK_PENDING;
425 
426 	spin_lock(&vnode->lock);
427 	ret = posix_lock_file(file, fl, NULL);
428 	if (ret < 0) {
429 		spin_unlock(&vnode->lock);
430 		_leave(" = %d [vfs]", ret);
431 		return ret;
432 	}
433 
434 	/* discard the server lock only if all granted locks are gone */
435 	if (list_empty(&vnode->granted_locks))
436 		afs_defer_unlock(vnode, key);
437 	spin_unlock(&vnode->lock);
438 	_leave(" = 0");
439 	return 0;
440 }
441 
442 /*
443  * return information about a lock we currently hold, if indeed we hold one
444  */
445 static int afs_do_getlk(struct file *file, struct file_lock *fl)
446 {
447 	struct afs_vnode *vnode = AFS_FS_I(file->f_mapping->host);
448 	struct key *key = file->private_data;
449 	int ret, lock_count;
450 
451 	_enter("");
452 
453 	fl->fl_type = F_UNLCK;
454 
455 	mutex_lock(&vnode->vfs_inode.i_mutex);
456 
457 	/* check local lock records first */
458 	ret = 0;
459 	if (posix_test_lock(file, fl) == 0) {
460 		/* no local locks; consult the server */
461 		ret = afs_vnode_fetch_status(vnode, NULL, key);
462 		if (ret < 0)
463 			goto error;
464 		lock_count = vnode->status.lock_count;
465 		if (lock_count) {
466 			if (lock_count > 0)
467 				fl->fl_type = F_RDLCK;
468 			else
469 				fl->fl_type = F_WRLCK;
470 			fl->fl_start = 0;
471 			fl->fl_end = OFFSET_MAX;
472 		}
473 	}
474 
475 error:
476 	mutex_unlock(&vnode->vfs_inode.i_mutex);
477 	_leave(" = %d [%hd]", ret, fl->fl_type);
478 	return ret;
479 }
480 
481 /*
482  * manage POSIX locks on a file
483  */
484 int afs_lock(struct file *file, int cmd, struct file_lock *fl)
485 {
486 	struct afs_vnode *vnode = AFS_FS_I(file->f_dentry->d_inode);
487 
488 	_enter("{%x:%u},%d,{t=%x,fl=%x,r=%Ld:%Ld}",
489 	       vnode->fid.vid, vnode->fid.vnode, cmd,
490 	       fl->fl_type, fl->fl_flags,
491 	       (long long) fl->fl_start, (long long) fl->fl_end);
492 
493 	/* AFS doesn't support mandatory locks */
494 	if ((vnode->vfs_inode.i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
495 	    fl->fl_type != F_UNLCK)
496 		return -ENOLCK;
497 
498 	if (IS_GETLK(cmd))
499 		return afs_do_getlk(file, fl);
500 	if (fl->fl_type == F_UNLCK)
501 		return afs_do_unlk(file, fl);
502 	return afs_do_setlk(file, fl);
503 }
504 
505 /*
506  * manage FLOCK locks on a file
507  */
508 int afs_flock(struct file *file, int cmd, struct file_lock *fl)
509 {
510 	struct afs_vnode *vnode = AFS_FS_I(file->f_dentry->d_inode);
511 
512 	_enter("{%x:%u},%d,{t=%x,fl=%x}",
513 	       vnode->fid.vid, vnode->fid.vnode, cmd,
514 	       fl->fl_type, fl->fl_flags);
515 
516 	/*
517 	 * No BSD flocks over NFS allowed.
518 	 * Note: we could try to fake a POSIX lock request here by
519 	 * using ((u32) filp | 0x80000000) or some such as the pid.
520 	 * Not sure whether that would be unique, though, or whether
521 	 * that would break in other places.
522 	 */
523 	if (!(fl->fl_flags & FL_FLOCK))
524 		return -ENOLCK;
525 
526 	/* we're simulating flock() locks using posix locks on the server */
527 	fl->fl_owner = (fl_owner_t) file;
528 	fl->fl_start = 0;
529 	fl->fl_end = OFFSET_MAX;
530 
531 	if (fl->fl_type == F_UNLCK)
532 		return afs_do_unlk(file, fl);
533 	return afs_do_setlk(file, fl);
534 }
535 
536 /*
537  * the POSIX lock management core VFS code copies the lock record and adds the
538  * copy into its own list, so we need to add that copy to the vnode's lock
539  * queue in the same place as the original (which will be deleted shortly
540  * after)
541  */
542 static void afs_fl_copy_lock(struct file_lock *new, struct file_lock *fl)
543 {
544 	_enter("");
545 
546 	list_add(&new->fl_u.afs.link, &fl->fl_u.afs.link);
547 }
548 
549 /*
550  * need to remove this lock from the vnode queue when it's removed from the
551  * VFS's list
552  */
553 static void afs_fl_release_private(struct file_lock *fl)
554 {
555 	_enter("");
556 
557 	list_del_init(&fl->fl_u.afs.link);
558 }
559