]>
Commit | Line | Data |
---|---|---|
2874c5fd | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
e8d6c554 DH |
2 | /* AFS file locking support |
3 | * | |
4 | * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved. | |
5 | * Written by David Howells (dhowells@redhat.com) | |
e8d6c554 DH |
6 | */ |
7 | ||
e8d6c554 DH |
8 | #include "internal.h" |
9 | ||
10 | #define AFS_LOCK_GRANTED 0 | |
11 | #define AFS_LOCK_PENDING 1 | |
4be5975a | 12 | #define AFS_LOCK_YOUR_TRY 2 |
e8d6c554 | 13 | |
f044c884 DH |
14 | struct workqueue_struct *afs_lock_manager; |
15 | ||
4be5975a | 16 | static void afs_next_locker(struct afs_vnode *vnode, int error); |
e8d6c554 DH |
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 | ||
6aed6285 | 20 | static const struct file_lock_operations afs_lock_ops = { |
e8d6c554 DH |
21 | .fl_copy_lock = afs_fl_copy_lock, |
22 | .fl_release_private = afs_fl_release_private, | |
23 | }; | |
24 | ||
4be5975a DH |
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 | ||
d4696601 DH |
31 | static atomic_t afs_file_lock_debug_id; |
32 | ||
e8d6c554 DH |
33 | /* |
34 | * if the callback is broken on this vnode, then the lock may now be available | |
35 | */ | |
36 | void afs_lock_may_be_available(struct afs_vnode *vnode) | |
37 | { | |
3b6492df | 38 | _enter("{%llx:%llu}", vnode->fid.vid, vnode->fid.vnode); |
e8d6c554 | 39 | |
4be5975a DH |
40 | spin_lock(&vnode->lock); |
41 | if (vnode->lock_state == AFS_VNODE_LOCK_WAITING_FOR_CB) | |
42 | afs_next_locker(vnode, 0); | |
d4696601 | 43 | trace_afs_flock_ev(vnode, NULL, afs_flock_callback_break, 0); |
4be5975a | 44 | spin_unlock(&vnode->lock); |
e8d6c554 DH |
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 | */ | |
51 | static void afs_schedule_lock_extension(struct afs_vnode *vnode) | |
52 | { | |
a690f60a DH |
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 | */ | |
71 | void afs_lock_op_done(struct afs_call *call) | |
72 | { | |
a58823ac | 73 | struct afs_vnode *vnode = call->lvnode; |
a690f60a DH |
74 | |
75 | if (call->error == 0) { | |
76 | spin_lock(&vnode->lock); | |
d4696601 | 77 | trace_afs_flock_ev(vnode, NULL, afs_flock_timestamp, 0); |
a690f60a DH |
78 | vnode->locked_at = call->reply_time; |
79 | afs_schedule_lock_extension(vnode); | |
80 | spin_unlock(&vnode->lock); | |
81 | } | |
e8d6c554 DH |
82 | } |
83 | ||
ff8e210a DH |
84 | /* |
85 | * grant one or more locks (readlocks are allowed to jump the queue if the | |
86 | * first lock in the queue is itself a readlock) | |
87 | * - the caller must hold the vnode lock | |
88 | */ | |
4be5975a | 89 | static void afs_grant_locks(struct afs_vnode *vnode) |
ff8e210a DH |
90 | { |
91 | struct file_lock *p, *_p; | |
4be5975a | 92 | bool exclusive = (vnode->lock_type == AFS_LOCK_WRITE); |
ff8e210a | 93 | |
4be5975a DH |
94 | list_for_each_entry_safe(p, _p, &vnode->pending_locks, fl_u.afs.link) { |
95 | if (!exclusive && p->fl_type == F_WRLCK) | |
96 | continue; | |
97 | ||
98 | list_move_tail(&p->fl_u.afs.link, &vnode->granted_locks); | |
99 | p->fl_u.afs.state = AFS_LOCK_GRANTED; | |
d4696601 | 100 | trace_afs_flock_op(vnode, p, afs_flock_op_grant); |
4be5975a DH |
101 | wake_up(&p->fl_wait); |
102 | } | |
103 | } | |
104 | ||
105 | /* | |
106 | * If an error is specified, reject every pending lock that matches the | |
107 | * authentication and type of the lock we failed to get. If there are any | |
108 | * remaining lockers, try to wake up one of them to have a go. | |
109 | */ | |
110 | static void afs_next_locker(struct afs_vnode *vnode, int error) | |
111 | { | |
112 | struct file_lock *p, *_p, *next = NULL; | |
113 | struct key *key = vnode->lock_key; | |
114 | unsigned int fl_type = F_RDLCK; | |
115 | ||
116 | _enter(""); | |
117 | ||
118 | if (vnode->lock_type == AFS_LOCK_WRITE) | |
119 | fl_type = F_WRLCK; | |
120 | ||
121 | list_for_each_entry_safe(p, _p, &vnode->pending_locks, fl_u.afs.link) { | |
122 | if (error && | |
123 | p->fl_type == fl_type && | |
124 | afs_file_key(p->fl_file) == key) { | |
125 | list_del_init(&p->fl_u.afs.link); | |
126 | p->fl_u.afs.state = error; | |
127 | wake_up(&p->fl_wait); | |
ff8e210a | 128 | } |
4be5975a DH |
129 | |
130 | /* Select the next locker to hand off to. */ | |
131 | if (next && | |
132 | (next->fl_type == F_WRLCK || p->fl_type == F_RDLCK)) | |
133 | continue; | |
134 | next = p; | |
ff8e210a | 135 | } |
4be5975a DH |
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; | |
d4696601 | 143 | trace_afs_flock_op(vnode, next, afs_flock_op_wake); |
4be5975a DH |
144 | wake_up(&next->fl_wait); |
145 | } else { | |
146 | afs_set_lock_state(vnode, AFS_VNODE_LOCK_NONE); | |
d4696601 | 147 | trace_afs_flock_ev(vnode, NULL, afs_flock_no_lockers, 0); |
4be5975a DH |
148 | } |
149 | ||
150 | _leave(""); | |
ff8e210a DH |
151 | } |
152 | ||
cdfb26b4 DH |
153 | /* |
154 | * Kill off all waiters in the the pending lock queue due to the vnode being | |
155 | * deleted. | |
156 | */ | |
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 | wake_up(&p->fl_wait); | |
169 | } | |
170 | ||
171 | key_put(vnode->lock_key); | |
172 | vnode->lock_key = NULL; | |
ff8e210a DH |
173 | } |
174 | ||
d2ddc776 DH |
175 | /* |
176 | * Get a lock on a file | |
177 | */ | |
178 | static int afs_set_lock(struct afs_vnode *vnode, struct key *key, | |
179 | afs_lock_type_t type) | |
180 | { | |
a58823ac | 181 | struct afs_status_cb *scb; |
d2ddc776 DH |
182 | struct afs_fs_cursor fc; |
183 | int ret; | |
184 | ||
3b6492df | 185 | _enter("%s{%llx:%llu.%u},%x,%u", |
d2ddc776 DH |
186 | vnode->volume->name, |
187 | vnode->fid.vid, | |
188 | vnode->fid.vnode, | |
189 | vnode->fid.unique, | |
190 | key_serial(key), type); | |
191 | ||
a58823ac DH |
192 | scb = kzalloc(sizeof(struct afs_status_cb), GFP_KERNEL); |
193 | if (!scb) | |
194 | return -ENOMEM; | |
195 | ||
d2ddc776 | 196 | ret = -ERESTARTSYS; |
20b8391f | 197 | if (afs_begin_vnode_operation(&fc, vnode, key, true)) { |
d2ddc776 | 198 | while (afs_select_fileserver(&fc)) { |
68251f0a | 199 | fc.cb_break = afs_calc_vnode_cb_break(vnode); |
a58823ac | 200 | afs_fs_set_lock(&fc, type, scb); |
d2ddc776 DH |
201 | } |
202 | ||
a58823ac DH |
203 | afs_check_for_remote_deletion(&fc, vnode); |
204 | afs_vnode_commit_status(&fc, vnode, fc.cb_break, NULL, scb); | |
d2ddc776 DH |
205 | ret = afs_end_vnode_operation(&fc); |
206 | } | |
207 | ||
a58823ac | 208 | kfree(scb); |
d2ddc776 DH |
209 | _leave(" = %d", ret); |
210 | return ret; | |
211 | } | |
212 | ||
213 | /* | |
214 | * Extend a lock on a file | |
215 | */ | |
216 | static int afs_extend_lock(struct afs_vnode *vnode, struct key *key) | |
217 | { | |
a58823ac | 218 | struct afs_status_cb *scb; |
d2ddc776 DH |
219 | struct afs_fs_cursor fc; |
220 | int ret; | |
221 | ||
3b6492df | 222 | _enter("%s{%llx:%llu.%u},%x", |
d2ddc776 DH |
223 | vnode->volume->name, |
224 | vnode->fid.vid, | |
225 | vnode->fid.vnode, | |
226 | vnode->fid.unique, | |
227 | key_serial(key)); | |
228 | ||
a58823ac DH |
229 | scb = kzalloc(sizeof(struct afs_status_cb), GFP_KERNEL); |
230 | if (!scb) | |
231 | return -ENOMEM; | |
232 | ||
d2ddc776 | 233 | ret = -ERESTARTSYS; |
20b8391f | 234 | if (afs_begin_vnode_operation(&fc, vnode, key, false)) { |
d2ddc776 | 235 | while (afs_select_current_fileserver(&fc)) { |
68251f0a | 236 | fc.cb_break = afs_calc_vnode_cb_break(vnode); |
a58823ac | 237 | afs_fs_extend_lock(&fc, scb); |
d2ddc776 DH |
238 | } |
239 | ||
a58823ac DH |
240 | afs_check_for_remote_deletion(&fc, vnode); |
241 | afs_vnode_commit_status(&fc, vnode, fc.cb_break, NULL, scb); | |
d2ddc776 DH |
242 | ret = afs_end_vnode_operation(&fc); |
243 | } | |
244 | ||
a58823ac | 245 | kfree(scb); |
d2ddc776 DH |
246 | _leave(" = %d", ret); |
247 | return ret; | |
248 | } | |
249 | ||
250 | /* | |
251 | * Release a lock on a file | |
252 | */ | |
253 | static int afs_release_lock(struct afs_vnode *vnode, struct key *key) | |
254 | { | |
a58823ac | 255 | struct afs_status_cb *scb; |
d2ddc776 DH |
256 | struct afs_fs_cursor fc; |
257 | int ret; | |
258 | ||
3b6492df | 259 | _enter("%s{%llx:%llu.%u},%x", |
d2ddc776 DH |
260 | vnode->volume->name, |
261 | vnode->fid.vid, | |
262 | vnode->fid.vnode, | |
263 | vnode->fid.unique, | |
264 | key_serial(key)); | |
265 | ||
a58823ac DH |
266 | scb = kzalloc(sizeof(struct afs_status_cb), GFP_KERNEL); |
267 | if (!scb) | |
268 | return -ENOMEM; | |
269 | ||
d2ddc776 | 270 | ret = -ERESTARTSYS; |
20b8391f | 271 | if (afs_begin_vnode_operation(&fc, vnode, key, false)) { |
d2ddc776 | 272 | while (afs_select_current_fileserver(&fc)) { |
68251f0a | 273 | fc.cb_break = afs_calc_vnode_cb_break(vnode); |
a58823ac | 274 | afs_fs_release_lock(&fc, scb); |
d2ddc776 DH |
275 | } |
276 | ||
a58823ac DH |
277 | afs_check_for_remote_deletion(&fc, vnode); |
278 | afs_vnode_commit_status(&fc, vnode, fc.cb_break, NULL, scb); | |
d2ddc776 DH |
279 | ret = afs_end_vnode_operation(&fc); |
280 | } | |
281 | ||
a58823ac | 282 | kfree(scb); |
d2ddc776 DH |
283 | _leave(" = %d", ret); |
284 | return ret; | |
285 | } | |
286 | ||
e8d6c554 DH |
287 | /* |
288 | * do work for a lock, including: | |
289 | * - probing for a lock we're waiting on but didn't get immediately | |
290 | * - extending a lock that's close to timing out | |
291 | */ | |
292 | void afs_lock_work(struct work_struct *work) | |
293 | { | |
294 | struct afs_vnode *vnode = | |
295 | container_of(work, struct afs_vnode, lock_work.work); | |
e8d6c554 DH |
296 | struct key *key; |
297 | int ret; | |
298 | ||
3b6492df | 299 | _enter("{%llx:%llu}", vnode->fid.vid, vnode->fid.vnode); |
e8d6c554 DH |
300 | |
301 | spin_lock(&vnode->lock); | |
302 | ||
0fafdc9f DH |
303 | again: |
304 | _debug("wstate %u for %p", vnode->lock_state, vnode); | |
305 | switch (vnode->lock_state) { | |
306 | case AFS_VNODE_LOCK_NEED_UNLOCK: | |
4be5975a | 307 | afs_set_lock_state(vnode, AFS_VNODE_LOCK_UNLOCKING); |
d4696601 | 308 | trace_afs_flock_ev(vnode, NULL, afs_flock_work_unlocking, 0); |
e8d6c554 DH |
309 | spin_unlock(&vnode->lock); |
310 | ||
311 | /* attempt to release the server lock; if it fails, we just | |
0fafdc9f DH |
312 | * wait 5 minutes and it'll expire anyway */ |
313 | ret = afs_release_lock(vnode, vnode->lock_key); | |
79ddbfa5 | 314 | if (ret < 0 && vnode->lock_state != AFS_VNODE_LOCK_DELETED) { |
cdfb26b4 DH |
315 | trace_afs_flock_ev(vnode, NULL, afs_flock_release_fail, |
316 | ret); | |
e8d6c554 | 317 | printk(KERN_WARNING "AFS:" |
3b6492df | 318 | " Failed to release lock on {%llx:%llx} error %d\n", |
e8d6c554 | 319 | vnode->fid.vid, vnode->fid.vnode, ret); |
0fafdc9f DH |
320 | } |
321 | ||
e8d6c554 | 322 | spin_lock(&vnode->lock); |
cdfb26b4 DH |
323 | if (ret == -ENOENT) |
324 | afs_kill_lockers_enoent(vnode); | |
325 | else | |
326 | afs_next_locker(vnode, 0); | |
4be5975a DH |
327 | spin_unlock(&vnode->lock); |
328 | return; | |
e8d6c554 | 329 | |
0fafdc9f DH |
330 | /* If we've already got a lock, then it must be time to extend that |
331 | * lock as AFS locks time out after 5 minutes. | |
332 | */ | |
333 | case AFS_VNODE_LOCK_GRANTED: | |
e8d6c554 DH |
334 | _debug("extend"); |
335 | ||
0fafdc9f DH |
336 | ASSERT(!list_empty(&vnode->granted_locks)); |
337 | ||
338 | key = key_get(vnode->lock_key); | |
4be5975a | 339 | afs_set_lock_state(vnode, AFS_VNODE_LOCK_EXTENDING); |
d4696601 | 340 | trace_afs_flock_ev(vnode, NULL, afs_flock_work_extending, 0); |
e8d6c554 DH |
341 | spin_unlock(&vnode->lock); |
342 | ||
0fafdc9f | 343 | ret = afs_extend_lock(vnode, key); /* RPC */ |
e8d6c554 | 344 | key_put(key); |
0fafdc9f | 345 | |
cdfb26b4 DH |
346 | if (ret < 0) { |
347 | trace_afs_flock_ev(vnode, NULL, afs_flock_extend_fail, | |
348 | ret); | |
3b6492df | 349 | pr_warning("AFS: Failed to extend lock on {%llx:%llx} error %d\n", |
0fafdc9f | 350 | vnode->fid.vid, vnode->fid.vnode, ret); |
cdfb26b4 | 351 | } |
0fafdc9f DH |
352 | |
353 | spin_lock(&vnode->lock); | |
354 | ||
cdfb26b4 DH |
355 | if (ret == -ENOENT) { |
356 | afs_kill_lockers_enoent(vnode); | |
357 | spin_unlock(&vnode->lock); | |
358 | return; | |
359 | } | |
360 | ||
0fafdc9f DH |
361 | if (vnode->lock_state != AFS_VNODE_LOCK_EXTENDING) |
362 | goto again; | |
4be5975a | 363 | afs_set_lock_state(vnode, AFS_VNODE_LOCK_GRANTED); |
0fafdc9f | 364 | |
4be5975a | 365 | if (ret != 0) |
e8d6c554 DH |
366 | queue_delayed_work(afs_lock_manager, &vnode->lock_work, |
367 | HZ * 10); | |
0fafdc9f DH |
368 | spin_unlock(&vnode->lock); |
369 | _leave(" [ext]"); | |
e8d6c554 | 370 | return; |
e8d6c554 | 371 | |
4be5975a DH |
372 | /* If we're waiting for a callback to indicate lock release, we can't |
373 | * actually rely on this, so need to recheck at regular intervals. The | |
374 | * problem is that the server might not notify us if the lock just | |
375 | * expires (say because a client died) rather than being explicitly | |
376 | * released. | |
377 | */ | |
0fafdc9f | 378 | case AFS_VNODE_LOCK_WAITING_FOR_CB: |
4be5975a DH |
379 | _debug("retry"); |
380 | afs_next_locker(vnode, 0); | |
e8d6c554 | 381 | spin_unlock(&vnode->lock); |
4be5975a | 382 | return; |
e8d6c554 | 383 | |
cdfb26b4 DH |
384 | case AFS_VNODE_LOCK_DELETED: |
385 | afs_kill_lockers_enoent(vnode); | |
386 | spin_unlock(&vnode->lock); | |
387 | return; | |
0fafdc9f | 388 | |
e690c9e3 | 389 | /* Fall through */ |
0fafdc9f DH |
390 | default: |
391 | /* Looks like a lock request was withdrawn. */ | |
392 | spin_unlock(&vnode->lock); | |
393 | _leave(" [no]"); | |
e8d6c554 DH |
394 | return; |
395 | } | |
e8d6c554 DH |
396 | } |
397 | ||
398 | /* | |
399 | * pass responsibility for the unlocking of a vnode on the server to the | |
400 | * manager thread, lest a pending signal in the calling thread interrupt | |
401 | * AF_RXRPC | |
402 | * - the caller must hold the vnode lock | |
403 | */ | |
0fafdc9f | 404 | static void afs_defer_unlock(struct afs_vnode *vnode) |
e8d6c554 | 405 | { |
4be5975a | 406 | _enter("%u", vnode->lock_state); |
0fafdc9f | 407 | |
4be5975a DH |
408 | if (list_empty(&vnode->granted_locks) && |
409 | (vnode->lock_state == AFS_VNODE_LOCK_GRANTED || | |
410 | vnode->lock_state == AFS_VNODE_LOCK_EXTENDING)) { | |
0fafdc9f DH |
411 | cancel_delayed_work(&vnode->lock_work); |
412 | ||
4be5975a | 413 | afs_set_lock_state(vnode, AFS_VNODE_LOCK_NEED_UNLOCK); |
d4696601 | 414 | trace_afs_flock_ev(vnode, NULL, afs_flock_defer_unlock, 0); |
4be5975a | 415 | queue_delayed_work(afs_lock_manager, &vnode->lock_work, 0); |
0fafdc9f DH |
416 | } |
417 | } | |
418 | ||
419 | /* | |
420 | * Check that our view of the file metadata is up to date and check to see | |
421 | * whether we think that we have a locking permit. | |
422 | */ | |
423 | static int afs_do_setlk_check(struct afs_vnode *vnode, struct key *key, | |
6c6c1d63 | 424 | enum afs_flock_mode mode, afs_lock_type_t type) |
0fafdc9f DH |
425 | { |
426 | afs_access_t access; | |
427 | int ret; | |
428 | ||
429 | /* Make sure we've got a callback on this file and that our view of the | |
430 | * data version is up to date. | |
431 | */ | |
432 | ret = afs_validate(vnode, key); | |
433 | if (ret < 0) | |
434 | return ret; | |
435 | ||
436 | /* Check the permission set to see if we're actually going to be | |
437 | * allowed to get a lock on this file. | |
438 | */ | |
439 | ret = afs_check_permit(vnode, key, &access); | |
440 | if (ret < 0) | |
441 | return ret; | |
442 | ||
443 | /* At a rough estimation, you need LOCK, WRITE or INSERT perm to | |
444 | * read-lock a file and WRITE or INSERT perm to write-lock a file. | |
445 | * | |
446 | * We can't rely on the server to do this for us since if we want to | |
447 | * share a read lock that we already have, we won't go the server. | |
448 | */ | |
449 | if (type == AFS_LOCK_READ) { | |
450 | if (!(access & (AFS_ACE_INSERT | AFS_ACE_WRITE | AFS_ACE_LOCK))) | |
451 | return -EACCES; | |
0fafdc9f DH |
452 | } else { |
453 | if (!(access & (AFS_ACE_INSERT | AFS_ACE_WRITE))) | |
454 | return -EACCES; | |
0fafdc9f DH |
455 | } |
456 | ||
457 | return 0; | |
458 | } | |
459 | ||
e8d6c554 DH |
460 | /* |
461 | * request a lock on a file on the server | |
462 | */ | |
463 | static int afs_do_setlk(struct file *file, struct file_lock *fl) | |
464 | { | |
0fafdc9f | 465 | struct inode *inode = locks_inode(file); |
1c8c601a | 466 | struct afs_vnode *vnode = AFS_FS_I(inode); |
6c6c1d63 | 467 | enum afs_flock_mode mode = AFS_FS_S(inode->i_sb)->flock_mode; |
e8d6c554 | 468 | afs_lock_type_t type; |
215804a9 | 469 | struct key *key = afs_file_key(file); |
6c6c1d63 | 470 | bool partial, no_server_lock = false; |
e8d6c554 DH |
471 | int ret; |
472 | ||
6c6c1d63 DH |
473 | if (mode == afs_flock_mode_unset) |
474 | mode = afs_flock_mode_openafs; | |
e8d6c554 | 475 | |
6c6c1d63 DH |
476 | _enter("{%llx:%llu},%llu-%llu,%u,%u", |
477 | vnode->fid.vid, vnode->fid.vnode, | |
478 | fl->fl_start, fl->fl_end, fl->fl_type, mode); | |
e8d6c554 | 479 | |
e8d6c554 DH |
480 | fl->fl_ops = &afs_lock_ops; |
481 | INIT_LIST_HEAD(&fl->fl_u.afs.link); | |
482 | fl->fl_u.afs.state = AFS_LOCK_PENDING; | |
483 | ||
6c6c1d63 | 484 | partial = (fl->fl_start != 0 || fl->fl_end != OFFSET_MAX); |
e8d6c554 | 485 | type = (fl->fl_type == F_RDLCK) ? AFS_LOCK_READ : AFS_LOCK_WRITE; |
6c6c1d63 DH |
486 | if (mode == afs_flock_mode_write && partial) |
487 | type = AFS_LOCK_WRITE; | |
e8d6c554 | 488 | |
6c6c1d63 | 489 | ret = afs_do_setlk_check(vnode, key, mode, type); |
e8d6c554 | 490 | if (ret < 0) |
0fafdc9f | 491 | return ret; |
e8d6c554 | 492 | |
d4696601 | 493 | trace_afs_flock_op(vnode, fl, afs_flock_op_set_lock); |
e8d6c554 | 494 | |
6c6c1d63 DH |
495 | /* AFS3 protocol only supports full-file locks and doesn't provide any |
496 | * method of upgrade/downgrade, so we need to emulate for partial-file | |
497 | * locks. | |
498 | * | |
499 | * The OpenAFS client only gets a server lock for a full-file lock and | |
500 | * keeps partial-file locks local. Allow this behaviour to be emulated | |
501 | * (as the default). | |
0fafdc9f | 502 | */ |
6c6c1d63 DH |
503 | if (mode == afs_flock_mode_local || |
504 | (partial && mode == afs_flock_mode_openafs)) { | |
505 | no_server_lock = true; | |
506 | goto skip_server_lock; | |
ff8e210a | 507 | } |
e8d6c554 | 508 | |
e8d6c554 | 509 | spin_lock(&vnode->lock); |
0fafdc9f | 510 | list_add_tail(&fl->fl_u.afs.link, &vnode->pending_locks); |
e8d6c554 | 511 | |
cdfb26b4 DH |
512 | ret = -ENOENT; |
513 | if (vnode->lock_state == AFS_VNODE_LOCK_DELETED) | |
514 | goto error_unlock; | |
515 | ||
4be5975a DH |
516 | /* If we've already got a lock on the server then try to move to having |
517 | * the VFS grant the requested lock. Note that this means that other | |
518 | * clients may get starved out. | |
0fafdc9f | 519 | */ |
4be5975a DH |
520 | _debug("try %u", vnode->lock_state); |
521 | if (vnode->lock_state == AFS_VNODE_LOCK_GRANTED) { | |
522 | if (type == AFS_LOCK_READ) { | |
523 | _debug("instant readlock"); | |
524 | list_move_tail(&fl->fl_u.afs.link, &vnode->granted_locks); | |
525 | fl->fl_u.afs.state = AFS_LOCK_GRANTED; | |
526 | goto vnode_is_locked_u; | |
527 | } | |
e8d6c554 | 528 | |
4be5975a DH |
529 | if (vnode->lock_type == AFS_LOCK_WRITE) { |
530 | _debug("instant writelock"); | |
531 | list_move_tail(&fl->fl_u.afs.link, &vnode->granted_locks); | |
532 | fl->fl_u.afs.state = AFS_LOCK_GRANTED; | |
533 | goto vnode_is_locked_u; | |
534 | } | |
535 | } | |
e8d6c554 | 536 | |
6c6c1d63 DH |
537 | if (vnode->lock_state == AFS_VNODE_LOCK_NONE && |
538 | !(fl->fl_flags & FL_SLEEP)) { | |
539 | ret = -EAGAIN; | |
540 | if (type == AFS_LOCK_READ) { | |
541 | if (vnode->status.lock_count == -1) | |
542 | goto lock_is_contended; /* Write locked */ | |
543 | } else { | |
544 | if (vnode->status.lock_count != 0) | |
545 | goto lock_is_contended; /* Locked */ | |
546 | } | |
547 | } | |
548 | ||
0fafdc9f DH |
549 | if (vnode->lock_state != AFS_VNODE_LOCK_NONE) |
550 | goto need_to_wait; | |
e8d6c554 | 551 | |
4be5975a | 552 | try_to_lock: |
0fafdc9f DH |
553 | /* We don't have a lock on this vnode and we aren't currently waiting |
554 | * for one either, so ask the server for a lock. | |
555 | * | |
556 | * Note that we need to be careful if we get interrupted by a signal | |
557 | * after dispatching the request as we may still get the lock, even | |
558 | * though we don't wait for the reply (it's not too bad a problem - the | |
4be5975a | 559 | * lock will expire in 5 mins anyway). |
0fafdc9f | 560 | */ |
d4696601 | 561 | trace_afs_flock_ev(vnode, fl, afs_flock_try_to_lock, 0); |
0fafdc9f DH |
562 | vnode->lock_key = key_get(key); |
563 | vnode->lock_type = type; | |
4be5975a | 564 | afs_set_lock_state(vnode, AFS_VNODE_LOCK_SETTING); |
e8d6c554 DH |
565 | spin_unlock(&vnode->lock); |
566 | ||
0fafdc9f | 567 | ret = afs_set_lock(vnode, key, type); /* RPC */ |
e8d6c554 DH |
568 | |
569 | spin_lock(&vnode->lock); | |
0fafdc9f | 570 | switch (ret) { |
4be5975a DH |
571 | case -EKEYREJECTED: |
572 | case -EKEYEXPIRED: | |
573 | case -EKEYREVOKED: | |
574 | case -EPERM: | |
575 | case -EACCES: | |
576 | fl->fl_u.afs.state = ret; | |
d4696601 | 577 | trace_afs_flock_ev(vnode, fl, afs_flock_fail_perm, ret); |
4be5975a DH |
578 | list_del_init(&fl->fl_u.afs.link); |
579 | afs_next_locker(vnode, ret); | |
580 | goto error_unlock; | |
581 | ||
cdfb26b4 DH |
582 | case -ENOENT: |
583 | fl->fl_u.afs.state = ret; | |
584 | trace_afs_flock_ev(vnode, fl, afs_flock_fail_other, ret); | |
585 | list_del_init(&fl->fl_u.afs.link); | |
586 | afs_kill_lockers_enoent(vnode); | |
587 | goto error_unlock; | |
588 | ||
0fafdc9f | 589 | default: |
4be5975a | 590 | fl->fl_u.afs.state = ret; |
d4696601 | 591 | trace_afs_flock_ev(vnode, fl, afs_flock_fail_other, ret); |
4be5975a DH |
592 | list_del_init(&fl->fl_u.afs.link); |
593 | afs_next_locker(vnode, 0); | |
594 | goto error_unlock; | |
e8d6c554 | 595 | |
0fafdc9f DH |
596 | case -EWOULDBLOCK: |
597 | /* The server doesn't have a lock-waiting queue, so the client | |
598 | * will have to retry. The server will break the outstanding | |
599 | * callbacks on a file when a lock is released. | |
600 | */ | |
0fafdc9f DH |
601 | ASSERT(list_empty(&vnode->granted_locks)); |
602 | ASSERTCMP(vnode->pending_locks.next, ==, &fl->fl_u.afs.link); | |
4be5975a | 603 | goto lock_is_contended; |
0fafdc9f DH |
604 | |
605 | case 0: | |
4be5975a | 606 | afs_set_lock_state(vnode, AFS_VNODE_LOCK_GRANTED); |
d4696601 | 607 | trace_afs_flock_ev(vnode, fl, afs_flock_acquired, type); |
4be5975a DH |
608 | afs_grant_locks(vnode); |
609 | goto vnode_is_locked_u; | |
e8d6c554 | 610 | } |
e8d6c554 | 611 | |
4be5975a | 612 | vnode_is_locked_u: |
0fafdc9f | 613 | spin_unlock(&vnode->lock); |
4be5975a DH |
614 | vnode_is_locked: |
615 | /* the lock has been granted by the server... */ | |
616 | ASSERTCMP(fl->fl_u.afs.state, ==, AFS_LOCK_GRANTED); | |
0fafdc9f | 617 | |
6c6c1d63 | 618 | skip_server_lock: |
4be5975a | 619 | /* ... but the VFS still needs to distribute access on this client. */ |
d4696601 | 620 | trace_afs_flock_ev(vnode, fl, afs_flock_vfs_locking, 0); |
4be5975a | 621 | ret = locks_lock_file_wait(file, fl); |
d4696601 | 622 | trace_afs_flock_ev(vnode, fl, afs_flock_vfs_lock, ret); |
e8d6c554 DH |
623 | if (ret < 0) |
624 | goto vfs_rejected_lock; | |
e8d6c554 | 625 | |
0fafdc9f | 626 | /* Again, make sure we've got a callback on this file and, again, make |
e8d6c554 | 627 | * sure that our view of the data version is up to date (we ignore |
0fafdc9f DH |
628 | * errors incurred here and deal with the consequences elsewhere). |
629 | */ | |
d2ddc776 | 630 | afs_validate(vnode, key); |
0fafdc9f DH |
631 | _leave(" = 0"); |
632 | return 0; | |
e8d6c554 | 633 | |
4be5975a DH |
634 | lock_is_contended: |
635 | if (!(fl->fl_flags & FL_SLEEP)) { | |
636 | list_del_init(&fl->fl_u.afs.link); | |
637 | afs_next_locker(vnode, 0); | |
638 | ret = -EAGAIN; | |
639 | goto error_unlock; | |
640 | } | |
641 | ||
642 | afs_set_lock_state(vnode, AFS_VNODE_LOCK_WAITING_FOR_CB); | |
d4696601 | 643 | trace_afs_flock_ev(vnode, fl, afs_flock_would_block, ret); |
4be5975a DH |
644 | queue_delayed_work(afs_lock_manager, &vnode->lock_work, HZ * 5); |
645 | ||
0fafdc9f DH |
646 | need_to_wait: |
647 | /* We're going to have to wait. Either this client doesn't have a lock | |
648 | * on the server yet and we need to wait for a callback to occur, or | |
4be5975a DH |
649 | * the client does have a lock on the server, but it's shared and we |
650 | * need an exclusive lock. | |
0fafdc9f | 651 | */ |
4be5975a | 652 | spin_unlock(&vnode->lock); |
0fafdc9f | 653 | |
d4696601 | 654 | trace_afs_flock_ev(vnode, fl, afs_flock_waiting, 0); |
4be5975a DH |
655 | ret = wait_event_interruptible(fl->fl_wait, |
656 | fl->fl_u.afs.state != AFS_LOCK_PENDING); | |
d4696601 | 657 | trace_afs_flock_ev(vnode, fl, afs_flock_waited, ret); |
0fafdc9f | 658 | |
4be5975a | 659 | if (fl->fl_u.afs.state >= 0 && fl->fl_u.afs.state != AFS_LOCK_GRANTED) { |
0fafdc9f | 660 | spin_lock(&vnode->lock); |
0fafdc9f | 661 | |
4be5975a DH |
662 | switch (fl->fl_u.afs.state) { |
663 | case AFS_LOCK_YOUR_TRY: | |
664 | fl->fl_u.afs.state = AFS_LOCK_PENDING; | |
665 | goto try_to_lock; | |
666 | case AFS_LOCK_PENDING: | |
667 | if (ret > 0) { | |
668 | /* We need to retry the lock. We may not be | |
669 | * notified by the server if it just expired | |
670 | * rather than being released. | |
671 | */ | |
672 | ASSERTCMP(vnode->lock_state, ==, AFS_VNODE_LOCK_WAITING_FOR_CB); | |
673 | afs_set_lock_state(vnode, AFS_VNODE_LOCK_SETTING); | |
674 | fl->fl_u.afs.state = AFS_LOCK_PENDING; | |
675 | goto try_to_lock; | |
676 | } | |
677 | goto error_unlock; | |
678 | case AFS_LOCK_GRANTED: | |
679 | default: | |
680 | break; | |
681 | } | |
0fafdc9f | 682 | |
4be5975a DH |
683 | spin_unlock(&vnode->lock); |
684 | } | |
0fafdc9f | 685 | |
4be5975a DH |
686 | if (fl->fl_u.afs.state == AFS_LOCK_GRANTED) |
687 | goto vnode_is_locked; | |
688 | ret = fl->fl_u.afs.state; | |
689 | goto error; | |
e8d6c554 DH |
690 | |
691 | vfs_rejected_lock: | |
0fafdc9f DH |
692 | /* The VFS rejected the lock we just obtained, so we have to discard |
693 | * what we just got. We defer this to the lock manager work item to | |
694 | * deal with. | |
695 | */ | |
e8d6c554 | 696 | _debug("vfs refused %d", ret); |
6c6c1d63 DH |
697 | if (no_server_lock) |
698 | goto error; | |
0fafdc9f | 699 | spin_lock(&vnode->lock); |
e8d6c554 | 700 | list_del_init(&fl->fl_u.afs.link); |
4be5975a DH |
701 | afs_defer_unlock(vnode); |
702 | ||
703 | error_unlock: | |
704 | spin_unlock(&vnode->lock); | |
705 | error: | |
706 | _leave(" = %d", ret); | |
707 | return ret; | |
e8d6c554 DH |
708 | } |
709 | ||
710 | /* | |
711 | * unlock on a file on the server | |
712 | */ | |
713 | static int afs_do_unlk(struct file *file, struct file_lock *fl) | |
714 | { | |
0fafdc9f | 715 | struct afs_vnode *vnode = AFS_FS_I(locks_inode(file)); |
e8d6c554 DH |
716 | int ret; |
717 | ||
3b6492df | 718 | _enter("{%llx:%llu},%u", vnode->fid.vid, vnode->fid.vnode, fl->fl_type); |
e8d6c554 | 719 | |
d4696601 DH |
720 | trace_afs_flock_op(vnode, fl, afs_flock_op_unlock); |
721 | ||
0fafdc9f DH |
722 | /* Flush all pending writes before doing anything with locks. */ |
723 | vfs_fsync(file, 0); | |
724 | ||
4be5975a | 725 | ret = locks_lock_file_wait(file, fl); |
0fafdc9f DH |
726 | _leave(" = %d [%u]", ret, vnode->lock_state); |
727 | return ret; | |
e8d6c554 DH |
728 | } |
729 | ||
730 | /* | |
731 | * return information about a lock we currently hold, if indeed we hold one | |
732 | */ | |
733 | static int afs_do_getlk(struct file *file, struct file_lock *fl) | |
734 | { | |
0fafdc9f | 735 | struct afs_vnode *vnode = AFS_FS_I(locks_inode(file)); |
215804a9 | 736 | struct key *key = afs_file_key(file); |
e8d6c554 DH |
737 | int ret, lock_count; |
738 | ||
739 | _enter(""); | |
740 | ||
cdfb26b4 DH |
741 | if (vnode->lock_state == AFS_VNODE_LOCK_DELETED) |
742 | return -ENOENT; | |
743 | ||
e8d6c554 DH |
744 | fl->fl_type = F_UNLCK; |
745 | ||
e8d6c554 | 746 | /* check local lock records first */ |
275afcac AM |
747 | posix_test_lock(file, fl); |
748 | if (fl->fl_type == F_UNLCK) { | |
e8d6c554 | 749 | /* no local locks; consult the server */ |
a58823ac | 750 | ret = afs_fetch_status(vnode, key, false, NULL); |
e8d6c554 DH |
751 | if (ret < 0) |
752 | goto error; | |
0fafdc9f DH |
753 | |
754 | lock_count = READ_ONCE(vnode->status.lock_count); | |
68ce801f DH |
755 | if (lock_count != 0) { |
756 | if (lock_count > 0) | |
757 | fl->fl_type = F_RDLCK; | |
758 | else | |
759 | fl->fl_type = F_WRLCK; | |
760 | fl->fl_start = 0; | |
761 | fl->fl_end = OFFSET_MAX; | |
762 | fl->fl_pid = 0; | |
763 | } | |
e8d6c554 DH |
764 | } |
765 | ||
0fafdc9f | 766 | ret = 0; |
e8d6c554 | 767 | error: |
e8d6c554 DH |
768 | _leave(" = %d [%hd]", ret, fl->fl_type); |
769 | return ret; | |
770 | } | |
771 | ||
772 | /* | |
773 | * manage POSIX locks on a file | |
774 | */ | |
775 | int afs_lock(struct file *file, int cmd, struct file_lock *fl) | |
776 | { | |
0fafdc9f | 777 | struct afs_vnode *vnode = AFS_FS_I(locks_inode(file)); |
d4696601 DH |
778 | enum afs_flock_operation op; |
779 | int ret; | |
e8d6c554 | 780 | |
3b6492df | 781 | _enter("{%llx:%llu},%d,{t=%x,fl=%x,r=%Ld:%Ld}", |
e8d6c554 DH |
782 | vnode->fid.vid, vnode->fid.vnode, cmd, |
783 | fl->fl_type, fl->fl_flags, | |
784 | (long long) fl->fl_start, (long long) fl->fl_end); | |
785 | ||
786 | /* AFS doesn't support mandatory locks */ | |
fc5846e5 | 787 | if (__mandatory_lock(&vnode->vfs_inode) && fl->fl_type != F_UNLCK) |
e8d6c554 DH |
788 | return -ENOLCK; |
789 | ||
790 | if (IS_GETLK(cmd)) | |
791 | return afs_do_getlk(file, fl); | |
d4696601 DH |
792 | |
793 | fl->fl_u.afs.debug_id = atomic_inc_return(&afs_file_lock_debug_id); | |
794 | trace_afs_flock_op(vnode, fl, afs_flock_op_lock); | |
795 | ||
e8d6c554 | 796 | if (fl->fl_type == F_UNLCK) |
d4696601 DH |
797 | ret = afs_do_unlk(file, fl); |
798 | else | |
799 | ret = afs_do_setlk(file, fl); | |
800 | ||
801 | switch (ret) { | |
802 | case 0: op = afs_flock_op_return_ok; break; | |
803 | case -EAGAIN: op = afs_flock_op_return_eagain; break; | |
804 | case -EDEADLK: op = afs_flock_op_return_edeadlk; break; | |
805 | default: op = afs_flock_op_return_error; break; | |
806 | } | |
807 | trace_afs_flock_op(vnode, fl, op); | |
808 | return ret; | |
e8d6c554 DH |
809 | } |
810 | ||
811 | /* | |
812 | * manage FLOCK locks on a file | |
813 | */ | |
814 | int afs_flock(struct file *file, int cmd, struct file_lock *fl) | |
815 | { | |
0fafdc9f | 816 | struct afs_vnode *vnode = AFS_FS_I(locks_inode(file)); |
d4696601 DH |
817 | enum afs_flock_operation op; |
818 | int ret; | |
e8d6c554 | 819 | |
3b6492df | 820 | _enter("{%llx:%llu},%d,{t=%x,fl=%x}", |
e8d6c554 DH |
821 | vnode->fid.vid, vnode->fid.vnode, cmd, |
822 | fl->fl_type, fl->fl_flags); | |
823 | ||
824 | /* | |
825 | * No BSD flocks over NFS allowed. | |
826 | * Note: we could try to fake a POSIX lock request here by | |
827 | * using ((u32) filp | 0x80000000) or some such as the pid. | |
828 | * Not sure whether that would be unique, though, or whether | |
829 | * that would break in other places. | |
830 | */ | |
831 | if (!(fl->fl_flags & FL_FLOCK)) | |
832 | return -ENOLCK; | |
833 | ||
d4696601 DH |
834 | fl->fl_u.afs.debug_id = atomic_inc_return(&afs_file_lock_debug_id); |
835 | trace_afs_flock_op(vnode, fl, afs_flock_op_flock); | |
836 | ||
e8d6c554 | 837 | /* we're simulating flock() locks using posix locks on the server */ |
e8d6c554 | 838 | if (fl->fl_type == F_UNLCK) |
d4696601 DH |
839 | ret = afs_do_unlk(file, fl); |
840 | else | |
841 | ret = afs_do_setlk(file, fl); | |
842 | ||
843 | switch (ret) { | |
844 | case 0: op = afs_flock_op_return_ok; break; | |
845 | case -EAGAIN: op = afs_flock_op_return_eagain; break; | |
846 | case -EDEADLK: op = afs_flock_op_return_edeadlk; break; | |
847 | default: op = afs_flock_op_return_error; break; | |
848 | } | |
849 | trace_afs_flock_op(vnode, fl, op); | |
850 | return ret; | |
e8d6c554 DH |
851 | } |
852 | ||
853 | /* | |
854 | * the POSIX lock management core VFS code copies the lock record and adds the | |
855 | * copy into its own list, so we need to add that copy to the vnode's lock | |
856 | * queue in the same place as the original (which will be deleted shortly | |
857 | * after) | |
858 | */ | |
859 | static void afs_fl_copy_lock(struct file_lock *new, struct file_lock *fl) | |
860 | { | |
0fafdc9f DH |
861 | struct afs_vnode *vnode = AFS_FS_I(locks_inode(fl->fl_file)); |
862 | ||
e8d6c554 DH |
863 | _enter(""); |
864 | ||
d4696601 DH |
865 | new->fl_u.afs.debug_id = atomic_inc_return(&afs_file_lock_debug_id); |
866 | ||
0fafdc9f | 867 | spin_lock(&vnode->lock); |
d4696601 | 868 | trace_afs_flock_op(vnode, new, afs_flock_op_copy_lock); |
e8d6c554 | 869 | list_add(&new->fl_u.afs.link, &fl->fl_u.afs.link); |
0fafdc9f | 870 | spin_unlock(&vnode->lock); |
e8d6c554 DH |
871 | } |
872 | ||
873 | /* | |
874 | * need to remove this lock from the vnode queue when it's removed from the | |
875 | * VFS's list | |
876 | */ | |
877 | static void afs_fl_release_private(struct file_lock *fl) | |
878 | { | |
0fafdc9f DH |
879 | struct afs_vnode *vnode = AFS_FS_I(locks_inode(fl->fl_file)); |
880 | ||
e8d6c554 DH |
881 | _enter(""); |
882 | ||
0fafdc9f | 883 | spin_lock(&vnode->lock); |
4be5975a | 884 | |
d4696601 | 885 | trace_afs_flock_op(vnode, fl, afs_flock_op_release_lock); |
4be5975a DH |
886 | list_del_init(&fl->fl_u.afs.link); |
887 | if (list_empty(&vnode->granted_locks)) | |
888 | afs_defer_unlock(vnode); | |
889 | ||
0fafdc9f DH |
890 | _debug("state %u for %p", vnode->lock_state, vnode); |
891 | spin_unlock(&vnode->lock); | |
e8d6c554 | 892 | } |