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->reply_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 && p->fl_type == F_WRLCK)
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 		wake_up(&p->fl_wait);
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 fl_type = F_RDLCK;
116 
117 	_enter("");
118 
119 	if (vnode->lock_type == AFS_LOCK_WRITE)
120 		fl_type = F_WRLCK;
121 
122 	list_for_each_entry_safe(p, _p, &vnode->pending_locks, fl_u.afs.link) {
123 		if (error &&
124 		    p->fl_type == fl_type &&
125 		    afs_file_key(p->fl_file) == key) {
126 			list_del_init(&p->fl_u.afs.link);
127 			p->fl_u.afs.state = error;
128 			wake_up(&p->fl_wait);
129 		}
130 
131 		/* Select the next locker to hand off to. */
132 		if (next &&
133 		    (next->fl_type == F_WRLCK || p->fl_type == F_RDLCK))
134 			continue;
135 		next = p;
136 	}
137 
138 	vnode->lock_key = NULL;
139 	key_put(key);
140 
141 	if (next) {
142 		afs_set_lock_state(vnode, AFS_VNODE_LOCK_SETTING);
143 		next->fl_u.afs.state = AFS_LOCK_YOUR_TRY;
144 		trace_afs_flock_op(vnode, next, afs_flock_op_wake);
145 		wake_up(&next->fl_wait);
146 	} else {
147 		afs_set_lock_state(vnode, AFS_VNODE_LOCK_NONE);
148 		trace_afs_flock_ev(vnode, NULL, afs_flock_no_lockers, 0);
149 	}
150 
151 	_leave("");
152 }
153 
154 /*
155  * Kill off all waiters in the the pending lock queue due to the vnode being
156  * deleted.
157  */
afs_kill_lockers_enoent(struct afs_vnode * vnode)158 static void afs_kill_lockers_enoent(struct afs_vnode *vnode)
159 {
160 	struct file_lock *p;
161 
162 	afs_set_lock_state(vnode, AFS_VNODE_LOCK_DELETED);
163 
164 	while (!list_empty(&vnode->pending_locks)) {
165 		p = list_entry(vnode->pending_locks.next,
166 			       struct file_lock, fl_u.afs.link);
167 		list_del_init(&p->fl_u.afs.link);
168 		p->fl_u.afs.state = -ENOENT;
169 		wake_up(&p->fl_wait);
170 	}
171 
172 	key_put(vnode->lock_key);
173 	vnode->lock_key = NULL;
174 }
175 
afs_lock_success(struct afs_operation * op)176 static void afs_lock_success(struct afs_operation *op)
177 {
178 	_enter("op=%08x", op->debug_id);
179 	afs_vnode_commit_status(op, &op->file[0]);
180 }
181 
182 static const struct afs_operation_ops afs_set_lock_operation = {
183 	.issue_afs_rpc	= afs_fs_set_lock,
184 	.issue_yfs_rpc	= yfs_fs_set_lock,
185 	.success	= afs_lock_success,
186 	.aborted	= afs_check_for_remote_deletion,
187 };
188 
189 /*
190  * Get a lock on a file
191  */
afs_set_lock(struct afs_vnode * vnode,struct key * key,afs_lock_type_t type)192 static int afs_set_lock(struct afs_vnode *vnode, struct key *key,
193 			afs_lock_type_t type)
194 {
195 	struct afs_operation *op;
196 
197 	_enter("%s{%llx:%llu.%u},%x,%u",
198 	       vnode->volume->name,
199 	       vnode->fid.vid,
200 	       vnode->fid.vnode,
201 	       vnode->fid.unique,
202 	       key_serial(key), type);
203 
204 	op = afs_alloc_operation(key, vnode->volume);
205 	if (IS_ERR(op))
206 		return PTR_ERR(op);
207 
208 	afs_op_set_vnode(op, 0, vnode);
209 
210 	op->lock.type	= type;
211 	op->ops		= &afs_set_lock_operation;
212 	return afs_do_sync_operation(op);
213 }
214 
215 static const struct afs_operation_ops afs_extend_lock_operation = {
216 	.issue_afs_rpc	= afs_fs_extend_lock,
217 	.issue_yfs_rpc	= yfs_fs_extend_lock,
218 	.success	= afs_lock_success,
219 };
220 
221 /*
222  * Extend a lock on a file
223  */
afs_extend_lock(struct afs_vnode * vnode,struct key * key)224 static int afs_extend_lock(struct afs_vnode *vnode, struct key *key)
225 {
226 	struct afs_operation *op;
227 
228 	_enter("%s{%llx:%llu.%u},%x",
229 	       vnode->volume->name,
230 	       vnode->fid.vid,
231 	       vnode->fid.vnode,
232 	       vnode->fid.unique,
233 	       key_serial(key));
234 
235 	op = afs_alloc_operation(key, vnode->volume);
236 	if (IS_ERR(op))
237 		return PTR_ERR(op);
238 
239 	afs_op_set_vnode(op, 0, vnode);
240 
241 	op->flags	|= AFS_OPERATION_UNINTR;
242 	op->ops		= &afs_extend_lock_operation;
243 	return afs_do_sync_operation(op);
244 }
245 
246 static const struct afs_operation_ops afs_release_lock_operation = {
247 	.issue_afs_rpc	= afs_fs_release_lock,
248 	.issue_yfs_rpc	= yfs_fs_release_lock,
249 	.success	= afs_lock_success,
250 };
251 
252 /*
253  * Release a lock on a file
254  */
afs_release_lock(struct afs_vnode * vnode,struct key * key)255 static int afs_release_lock(struct afs_vnode *vnode, struct key *key)
256 {
257 	struct afs_operation *op;
258 
259 	_enter("%s{%llx:%llu.%u},%x",
260 	       vnode->volume->name,
261 	       vnode->fid.vid,
262 	       vnode->fid.vnode,
263 	       vnode->fid.unique,
264 	       key_serial(key));
265 
266 	op = afs_alloc_operation(key, vnode->volume);
267 	if (IS_ERR(op))
268 		return PTR_ERR(op);
269 
270 	afs_op_set_vnode(op, 0, vnode);
271 
272 	op->flags	|= AFS_OPERATION_UNINTR;
273 	op->ops		= &afs_release_lock_operation;
274 	return afs_do_sync_operation(op);
275 }
276 
277 /*
278  * do work for a lock, including:
279  * - probing for a lock we're waiting on but didn't get immediately
280  * - extending a lock that's close to timing out
281  */
afs_lock_work(struct work_struct * work)282 void afs_lock_work(struct work_struct *work)
283 {
284 	struct afs_vnode *vnode =
285 		container_of(work, struct afs_vnode, lock_work.work);
286 	struct key *key;
287 	int ret;
288 
289 	_enter("{%llx:%llu}", vnode->fid.vid, vnode->fid.vnode);
290 
291 	spin_lock(&vnode->lock);
292 
293 again:
294 	_debug("wstate %u for %p", vnode->lock_state, vnode);
295 	switch (vnode->lock_state) {
296 	case AFS_VNODE_LOCK_NEED_UNLOCK:
297 		afs_set_lock_state(vnode, AFS_VNODE_LOCK_UNLOCKING);
298 		trace_afs_flock_ev(vnode, NULL, afs_flock_work_unlocking, 0);
299 		spin_unlock(&vnode->lock);
300 
301 		/* attempt to release the server lock; if it fails, we just
302 		 * wait 5 minutes and it'll expire anyway */
303 		ret = afs_release_lock(vnode, vnode->lock_key);
304 		if (ret < 0 && vnode->lock_state != AFS_VNODE_LOCK_DELETED) {
305 			trace_afs_flock_ev(vnode, NULL, afs_flock_release_fail,
306 					   ret);
307 			printk(KERN_WARNING "AFS:"
308 			       " Failed to release lock on {%llx:%llx} error %d\n",
309 			       vnode->fid.vid, vnode->fid.vnode, ret);
310 		}
311 
312 		spin_lock(&vnode->lock);
313 		if (ret == -ENOENT)
314 			afs_kill_lockers_enoent(vnode);
315 		else
316 			afs_next_locker(vnode, 0);
317 		spin_unlock(&vnode->lock);
318 		return;
319 
320 	/* If we've already got a lock, then it must be time to extend that
321 	 * lock as AFS locks time out after 5 minutes.
322 	 */
323 	case AFS_VNODE_LOCK_GRANTED:
324 		_debug("extend");
325 
326 		ASSERT(!list_empty(&vnode->granted_locks));
327 
328 		key = key_get(vnode->lock_key);
329 		afs_set_lock_state(vnode, AFS_VNODE_LOCK_EXTENDING);
330 		trace_afs_flock_ev(vnode, NULL, afs_flock_work_extending, 0);
331 		spin_unlock(&vnode->lock);
332 
333 		ret = afs_extend_lock(vnode, key); /* RPC */
334 		key_put(key);
335 
336 		if (ret < 0) {
337 			trace_afs_flock_ev(vnode, NULL, afs_flock_extend_fail,
338 					   ret);
339 			pr_warn("AFS: Failed to extend lock on {%llx:%llx} error %d\n",
340 				vnode->fid.vid, vnode->fid.vnode, ret);
341 		}
342 
343 		spin_lock(&vnode->lock);
344 
345 		if (ret == -ENOENT) {
346 			afs_kill_lockers_enoent(vnode);
347 			spin_unlock(&vnode->lock);
348 			return;
349 		}
350 
351 		if (vnode->lock_state != AFS_VNODE_LOCK_EXTENDING)
352 			goto again;
353 		afs_set_lock_state(vnode, AFS_VNODE_LOCK_GRANTED);
354 
355 		if (ret != 0)
356 			queue_delayed_work(afs_lock_manager, &vnode->lock_work,
357 					   HZ * 10);
358 		spin_unlock(&vnode->lock);
359 		_leave(" [ext]");
360 		return;
361 
362 	/* If we're waiting for a callback to indicate lock release, we can't
363 	 * actually rely on this, so need to recheck at regular intervals.  The
364 	 * problem is that the server might not notify us if the lock just
365 	 * expires (say because a client died) rather than being explicitly
366 	 * released.
367 	 */
368 	case AFS_VNODE_LOCK_WAITING_FOR_CB:
369 		_debug("retry");
370 		afs_next_locker(vnode, 0);
371 		spin_unlock(&vnode->lock);
372 		return;
373 
374 	case AFS_VNODE_LOCK_DELETED:
375 		afs_kill_lockers_enoent(vnode);
376 		spin_unlock(&vnode->lock);
377 		return;
378 
379 	default:
380 		/* Looks like a lock request was withdrawn. */
381 		spin_unlock(&vnode->lock);
382 		_leave(" [no]");
383 		return;
384 	}
385 }
386 
387 /*
388  * pass responsibility for the unlocking of a vnode on the server to the
389  * manager thread, lest a pending signal in the calling thread interrupt
390  * AF_RXRPC
391  * - the caller must hold the vnode lock
392  */
afs_defer_unlock(struct afs_vnode * vnode)393 static void afs_defer_unlock(struct afs_vnode *vnode)
394 {
395 	_enter("%u", vnode->lock_state);
396 
397 	if (list_empty(&vnode->granted_locks) &&
398 	    (vnode->lock_state == AFS_VNODE_LOCK_GRANTED ||
399 	     vnode->lock_state == AFS_VNODE_LOCK_EXTENDING)) {
400 		cancel_delayed_work(&vnode->lock_work);
401 
402 		afs_set_lock_state(vnode, AFS_VNODE_LOCK_NEED_UNLOCK);
403 		trace_afs_flock_ev(vnode, NULL, afs_flock_defer_unlock, 0);
404 		queue_delayed_work(afs_lock_manager, &vnode->lock_work, 0);
405 	}
406 }
407 
408 /*
409  * Check that our view of the file metadata is up to date and check to see
410  * whether we think that we have a locking permit.
411  */
afs_do_setlk_check(struct afs_vnode * vnode,struct key * key,enum afs_flock_mode mode,afs_lock_type_t type)412 static int afs_do_setlk_check(struct afs_vnode *vnode, struct key *key,
413 			      enum afs_flock_mode mode, afs_lock_type_t type)
414 {
415 	afs_access_t access;
416 	int ret;
417 
418 	/* Make sure we've got a callback on this file and that our view of the
419 	 * data version is up to date.
420 	 */
421 	ret = afs_validate(vnode, key);
422 	if (ret < 0)
423 		return ret;
424 
425 	/* Check the permission set to see if we're actually going to be
426 	 * allowed to get a lock on this file.
427 	 */
428 	ret = afs_check_permit(vnode, key, &access);
429 	if (ret < 0)
430 		return ret;
431 
432 	/* At a rough estimation, you need LOCK, WRITE or INSERT perm to
433 	 * read-lock a file and WRITE or INSERT perm to write-lock a file.
434 	 *
435 	 * We can't rely on the server to do this for us since if we want to
436 	 * share a read lock that we already have, we won't go the server.
437 	 */
438 	if (type == AFS_LOCK_READ) {
439 		if (!(access & (AFS_ACE_INSERT | AFS_ACE_WRITE | AFS_ACE_LOCK)))
440 			return -EACCES;
441 	} else {
442 		if (!(access & (AFS_ACE_INSERT | AFS_ACE_WRITE)))
443 			return -EACCES;
444 	}
445 
446 	return 0;
447 }
448 
449 /*
450  * request a lock on a file on the server
451  */
afs_do_setlk(struct file * file,struct file_lock * fl)452 static int afs_do_setlk(struct file *file, struct file_lock *fl)
453 {
454 	struct inode *inode = locks_inode(file);
455 	struct afs_vnode *vnode = AFS_FS_I(inode);
456 	enum afs_flock_mode mode = AFS_FS_S(inode->i_sb)->flock_mode;
457 	afs_lock_type_t type;
458 	struct key *key = afs_file_key(file);
459 	bool partial, no_server_lock = false;
460 	int ret;
461 
462 	if (mode == afs_flock_mode_unset)
463 		mode = afs_flock_mode_openafs;
464 
465 	_enter("{%llx:%llu},%llu-%llu,%u,%u",
466 	       vnode->fid.vid, vnode->fid.vnode,
467 	       fl->fl_start, fl->fl_end, fl->fl_type, mode);
468 
469 	fl->fl_ops = &afs_lock_ops;
470 	INIT_LIST_HEAD(&fl->fl_u.afs.link);
471 	fl->fl_u.afs.state = AFS_LOCK_PENDING;
472 
473 	partial = (fl->fl_start != 0 || fl->fl_end != OFFSET_MAX);
474 	type = (fl->fl_type == F_RDLCK) ? AFS_LOCK_READ : AFS_LOCK_WRITE;
475 	if (mode == afs_flock_mode_write && partial)
476 		type = AFS_LOCK_WRITE;
477 
478 	ret = afs_do_setlk_check(vnode, key, mode, type);
479 	if (ret < 0)
480 		return ret;
481 
482 	trace_afs_flock_op(vnode, fl, afs_flock_op_set_lock);
483 
484 	/* AFS3 protocol only supports full-file locks and doesn't provide any
485 	 * method of upgrade/downgrade, so we need to emulate for partial-file
486 	 * locks.
487 	 *
488 	 * The OpenAFS client only gets a server lock for a full-file lock and
489 	 * keeps partial-file locks local.  Allow this behaviour to be emulated
490 	 * (as the default).
491 	 */
492 	if (mode == afs_flock_mode_local ||
493 	    (partial && mode == afs_flock_mode_openafs)) {
494 		no_server_lock = true;
495 		goto skip_server_lock;
496 	}
497 
498 	spin_lock(&vnode->lock);
499 	list_add_tail(&fl->fl_u.afs.link, &vnode->pending_locks);
500 
501 	ret = -ENOENT;
502 	if (vnode->lock_state == AFS_VNODE_LOCK_DELETED)
503 		goto error_unlock;
504 
505 	/* If we've already got a lock on the server then try to move to having
506 	 * the VFS grant the requested lock.  Note that this means that other
507 	 * clients may get starved out.
508 	 */
509 	_debug("try %u", vnode->lock_state);
510 	if (vnode->lock_state == AFS_VNODE_LOCK_GRANTED) {
511 		if (type == AFS_LOCK_READ) {
512 			_debug("instant readlock");
513 			list_move_tail(&fl->fl_u.afs.link, &vnode->granted_locks);
514 			fl->fl_u.afs.state = AFS_LOCK_GRANTED;
515 			goto vnode_is_locked_u;
516 		}
517 
518 		if (vnode->lock_type == AFS_LOCK_WRITE) {
519 			_debug("instant writelock");
520 			list_move_tail(&fl->fl_u.afs.link, &vnode->granted_locks);
521 			fl->fl_u.afs.state = AFS_LOCK_GRANTED;
522 			goto vnode_is_locked_u;
523 		}
524 	}
525 
526 	if (vnode->lock_state == AFS_VNODE_LOCK_NONE &&
527 	    !(fl->fl_flags & FL_SLEEP)) {
528 		ret = -EAGAIN;
529 		if (type == AFS_LOCK_READ) {
530 			if (vnode->status.lock_count == -1)
531 				goto lock_is_contended; /* Write locked */
532 		} else {
533 			if (vnode->status.lock_count != 0)
534 				goto lock_is_contended; /* Locked */
535 		}
536 	}
537 
538 	if (vnode->lock_state != AFS_VNODE_LOCK_NONE)
539 		goto need_to_wait;
540 
541 try_to_lock:
542 	/* We don't have a lock on this vnode and we aren't currently waiting
543 	 * for one either, so ask the server for a lock.
544 	 *
545 	 * Note that we need to be careful if we get interrupted by a signal
546 	 * after dispatching the request as we may still get the lock, even
547 	 * though we don't wait for the reply (it's not too bad a problem - the
548 	 * lock will expire in 5 mins anyway).
549 	 */
550 	trace_afs_flock_ev(vnode, fl, afs_flock_try_to_lock, 0);
551 	vnode->lock_key = key_get(key);
552 	vnode->lock_type = type;
553 	afs_set_lock_state(vnode, AFS_VNODE_LOCK_SETTING);
554 	spin_unlock(&vnode->lock);
555 
556 	ret = afs_set_lock(vnode, key, type); /* RPC */
557 
558 	spin_lock(&vnode->lock);
559 	switch (ret) {
560 	case -EKEYREJECTED:
561 	case -EKEYEXPIRED:
562 	case -EKEYREVOKED:
563 	case -EPERM:
564 	case -EACCES:
565 		fl->fl_u.afs.state = ret;
566 		trace_afs_flock_ev(vnode, fl, afs_flock_fail_perm, ret);
567 		list_del_init(&fl->fl_u.afs.link);
568 		afs_next_locker(vnode, ret);
569 		goto error_unlock;
570 
571 	case -ENOENT:
572 		fl->fl_u.afs.state = ret;
573 		trace_afs_flock_ev(vnode, fl, afs_flock_fail_other, ret);
574 		list_del_init(&fl->fl_u.afs.link);
575 		afs_kill_lockers_enoent(vnode);
576 		goto error_unlock;
577 
578 	default:
579 		fl->fl_u.afs.state = ret;
580 		trace_afs_flock_ev(vnode, fl, afs_flock_fail_other, ret);
581 		list_del_init(&fl->fl_u.afs.link);
582 		afs_next_locker(vnode, 0);
583 		goto error_unlock;
584 
585 	case -EWOULDBLOCK:
586 		/* The server doesn't have a lock-waiting queue, so the client
587 		 * will have to retry.  The server will break the outstanding
588 		 * callbacks on a file when a lock is released.
589 		 */
590 		ASSERT(list_empty(&vnode->granted_locks));
591 		ASSERTCMP(vnode->pending_locks.next, ==, &fl->fl_u.afs.link);
592 		goto lock_is_contended;
593 
594 	case 0:
595 		afs_set_lock_state(vnode, AFS_VNODE_LOCK_GRANTED);
596 		trace_afs_flock_ev(vnode, fl, afs_flock_acquired, type);
597 		afs_grant_locks(vnode);
598 		goto vnode_is_locked_u;
599 	}
600 
601 vnode_is_locked_u:
602 	spin_unlock(&vnode->lock);
603 vnode_is_locked:
604 	/* the lock has been granted by the server... */
605 	ASSERTCMP(fl->fl_u.afs.state, ==, AFS_LOCK_GRANTED);
606 
607 skip_server_lock:
608 	/* ... but the VFS still needs to distribute access on this client. */
609 	trace_afs_flock_ev(vnode, fl, afs_flock_vfs_locking, 0);
610 	ret = locks_lock_file_wait(file, fl);
611 	trace_afs_flock_ev(vnode, fl, afs_flock_vfs_lock, ret);
612 	if (ret < 0)
613 		goto vfs_rejected_lock;
614 
615 	/* Again, make sure we've got a callback on this file and, again, make
616 	 * sure that our view of the data version is up to date (we ignore
617 	 * errors incurred here and deal with the consequences elsewhere).
618 	 */
619 	afs_validate(vnode, key);
620 	_leave(" = 0");
621 	return 0;
622 
623 lock_is_contended:
624 	if (!(fl->fl_flags & FL_SLEEP)) {
625 		list_del_init(&fl->fl_u.afs.link);
626 		afs_next_locker(vnode, 0);
627 		ret = -EAGAIN;
628 		goto error_unlock;
629 	}
630 
631 	afs_set_lock_state(vnode, AFS_VNODE_LOCK_WAITING_FOR_CB);
632 	trace_afs_flock_ev(vnode, fl, afs_flock_would_block, ret);
633 	queue_delayed_work(afs_lock_manager, &vnode->lock_work, HZ * 5);
634 
635 need_to_wait:
636 	/* We're going to have to wait.  Either this client doesn't have a lock
637 	 * on the server yet and we need to wait for a callback to occur, or
638 	 * the client does have a lock on the server, but it's shared and we
639 	 * need an exclusive lock.
640 	 */
641 	spin_unlock(&vnode->lock);
642 
643 	trace_afs_flock_ev(vnode, fl, afs_flock_waiting, 0);
644 	ret = wait_event_interruptible(fl->fl_wait,
645 				       fl->fl_u.afs.state != AFS_LOCK_PENDING);
646 	trace_afs_flock_ev(vnode, fl, afs_flock_waited, ret);
647 
648 	if (fl->fl_u.afs.state >= 0 && fl->fl_u.afs.state != AFS_LOCK_GRANTED) {
649 		spin_lock(&vnode->lock);
650 
651 		switch (fl->fl_u.afs.state) {
652 		case AFS_LOCK_YOUR_TRY:
653 			fl->fl_u.afs.state = AFS_LOCK_PENDING;
654 			goto try_to_lock;
655 		case AFS_LOCK_PENDING:
656 			if (ret > 0) {
657 				/* We need to retry the lock.  We may not be
658 				 * notified by the server if it just expired
659 				 * rather than being released.
660 				 */
661 				ASSERTCMP(vnode->lock_state, ==, AFS_VNODE_LOCK_WAITING_FOR_CB);
662 				afs_set_lock_state(vnode, AFS_VNODE_LOCK_SETTING);
663 				fl->fl_u.afs.state = AFS_LOCK_PENDING;
664 				goto try_to_lock;
665 			}
666 			goto error_unlock;
667 		case AFS_LOCK_GRANTED:
668 		default:
669 			break;
670 		}
671 
672 		spin_unlock(&vnode->lock);
673 	}
674 
675 	if (fl->fl_u.afs.state == AFS_LOCK_GRANTED)
676 		goto vnode_is_locked;
677 	ret = fl->fl_u.afs.state;
678 	goto error;
679 
680 vfs_rejected_lock:
681 	/* The VFS rejected the lock we just obtained, so we have to discard
682 	 * what we just got.  We defer this to the lock manager work item to
683 	 * deal with.
684 	 */
685 	_debug("vfs refused %d", ret);
686 	if (no_server_lock)
687 		goto error;
688 	spin_lock(&vnode->lock);
689 	list_del_init(&fl->fl_u.afs.link);
690 	afs_defer_unlock(vnode);
691 
692 error_unlock:
693 	spin_unlock(&vnode->lock);
694 error:
695 	_leave(" = %d", ret);
696 	return ret;
697 }
698 
699 /*
700  * unlock on a file on the server
701  */
afs_do_unlk(struct file * file,struct file_lock * fl)702 static int afs_do_unlk(struct file *file, struct file_lock *fl)
703 {
704 	struct afs_vnode *vnode = AFS_FS_I(locks_inode(file));
705 	int ret;
706 
707 	_enter("{%llx:%llu},%u", vnode->fid.vid, vnode->fid.vnode, fl->fl_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(locks_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->fl_type = F_UNLCK;
734 
735 	/* check local lock records first */
736 	posix_test_lock(file, fl);
737 	if (fl->fl_type == F_UNLCK) {
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->fl_type = F_RDLCK;
747 			else
748 				fl->fl_type = F_WRLCK;
749 			fl->fl_start = 0;
750 			fl->fl_end = OFFSET_MAX;
751 			fl->fl_pid = 0;
752 		}
753 	}
754 
755 	ret = 0;
756 error:
757 	_leave(" = %d [%hd]", ret, fl->fl_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(locks_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->fl_type, fl->fl_flags,
773 	       (long long) fl->fl_start, (long long) fl->fl_end);
774 
775 	/* AFS doesn't support mandatory locks */
776 	if (__mandatory_lock(&vnode->vfs_inode) && fl->fl_type != F_UNLCK)
777 		return -ENOLCK;
778 
779 	if (IS_GETLK(cmd))
780 		return afs_do_getlk(file, fl);
781 
782 	fl->fl_u.afs.debug_id = atomic_inc_return(&afs_file_lock_debug_id);
783 	trace_afs_flock_op(vnode, fl, afs_flock_op_lock);
784 
785 	if (fl->fl_type == F_UNLCK)
786 		ret = afs_do_unlk(file, fl);
787 	else
788 		ret = afs_do_setlk(file, fl);
789 
790 	switch (ret) {
791 	case 0:		op = afs_flock_op_return_ok; break;
792 	case -EAGAIN:	op = afs_flock_op_return_eagain; break;
793 	case -EDEADLK:	op = afs_flock_op_return_edeadlk; break;
794 	default:	op = afs_flock_op_return_error; break;
795 	}
796 	trace_afs_flock_op(vnode, fl, op);
797 	return ret;
798 }
799 
800 /*
801  * manage FLOCK locks on a file
802  */
afs_flock(struct file * file,int cmd,struct file_lock * fl)803 int afs_flock(struct file *file, int cmd, struct file_lock *fl)
804 {
805 	struct afs_vnode *vnode = AFS_FS_I(locks_inode(file));
806 	enum afs_flock_operation op;
807 	int ret;
808 
809 	_enter("{%llx:%llu},%d,{t=%x,fl=%x}",
810 	       vnode->fid.vid, vnode->fid.vnode, cmd,
811 	       fl->fl_type, fl->fl_flags);
812 
813 	/*
814 	 * No BSD flocks over NFS allowed.
815 	 * Note: we could try to fake a POSIX lock request here by
816 	 * using ((u32) filp | 0x80000000) or some such as the pid.
817 	 * Not sure whether that would be unique, though, or whether
818 	 * that would break in other places.
819 	 */
820 	if (!(fl->fl_flags & FL_FLOCK))
821 		return -ENOLCK;
822 
823 	fl->fl_u.afs.debug_id = atomic_inc_return(&afs_file_lock_debug_id);
824 	trace_afs_flock_op(vnode, fl, afs_flock_op_flock);
825 
826 	/* we're simulating flock() locks using posix locks on the server */
827 	if (fl->fl_type == F_UNLCK)
828 		ret = afs_do_unlk(file, fl);
829 	else
830 		ret = afs_do_setlk(file, fl);
831 
832 	switch (ret) {
833 	case 0:		op = afs_flock_op_return_ok; break;
834 	case -EAGAIN:	op = afs_flock_op_return_eagain; break;
835 	case -EDEADLK:	op = afs_flock_op_return_edeadlk; break;
836 	default:	op = afs_flock_op_return_error; break;
837 	}
838 	trace_afs_flock_op(vnode, fl, op);
839 	return ret;
840 }
841 
842 /*
843  * the POSIX lock management core VFS code copies the lock record and adds the
844  * copy into its own list, so we need to add that copy to the vnode's lock
845  * queue in the same place as the original (which will be deleted shortly
846  * after)
847  */
afs_fl_copy_lock(struct file_lock * new,struct file_lock * fl)848 static void afs_fl_copy_lock(struct file_lock *new, struct file_lock *fl)
849 {
850 	struct afs_vnode *vnode = AFS_FS_I(locks_inode(fl->fl_file));
851 
852 	_enter("");
853 
854 	new->fl_u.afs.debug_id = atomic_inc_return(&afs_file_lock_debug_id);
855 
856 	spin_lock(&vnode->lock);
857 	trace_afs_flock_op(vnode, new, afs_flock_op_copy_lock);
858 	list_add(&new->fl_u.afs.link, &fl->fl_u.afs.link);
859 	spin_unlock(&vnode->lock);
860 }
861 
862 /*
863  * need to remove this lock from the vnode queue when it's removed from the
864  * VFS's list
865  */
afs_fl_release_private(struct file_lock * fl)866 static void afs_fl_release_private(struct file_lock *fl)
867 {
868 	struct afs_vnode *vnode = AFS_FS_I(locks_inode(fl->fl_file));
869 
870 	_enter("");
871 
872 	spin_lock(&vnode->lock);
873 
874 	trace_afs_flock_op(vnode, fl, afs_flock_op_release_lock);
875 	list_del_init(&fl->fl_u.afs.link);
876 	if (list_empty(&vnode->granted_locks))
877 		afs_defer_unlock(vnode);
878 
879 	_debug("state %u for %p", vnode->lock_state, vnode);
880 	spin_unlock(&vnode->lock);
881 }
882