]>
Commit | Line | Data |
---|---|---|
1 | // SPDX-License-Identifier: GPL-2.0 | |
2 | /* | |
3 | * linux/fs/lockd/svclock.c | |
4 | * | |
5 | * Handling of server-side locks, mostly of the blocked variety. | |
6 | * This is the ugliest part of lockd because we tread on very thin ice. | |
7 | * GRANT and CANCEL calls may get stuck, meet in mid-flight, etc. | |
8 | * IMNSHO introducing the grant callback into the NLM protocol was one | |
9 | * of the worst ideas Sun ever had. Except maybe for the idea of doing | |
10 | * NFS file locking at all. | |
11 | * | |
12 | * I'm trying hard to avoid race conditions by protecting most accesses | |
13 | * to a file's list of blocked locks through a semaphore. The global | |
14 | * list of blocked locks is not protected in this fashion however. | |
15 | * Therefore, some functions (such as the RPC callback for the async grant | |
16 | * call) move blocked locks towards the head of the list *while some other | |
17 | * process might be traversing it*. This should not be a problem in | |
18 | * practice, because this will only cause functions traversing the list | |
19 | * to visit some blocks twice. | |
20 | * | |
21 | * Copyright (C) 1996, Olaf Kirch <okir@monad.swb.de> | |
22 | */ | |
23 | ||
24 | #include <linux/types.h> | |
25 | #include <linux/slab.h> | |
26 | #include <linux/errno.h> | |
27 | #include <linux/kernel.h> | |
28 | #include <linux/sched.h> | |
29 | #include <linux/sunrpc/clnt.h> | |
30 | #include <linux/sunrpc/svc_xprt.h> | |
31 | #include <linux/lockd/nlm.h> | |
32 | #include <linux/lockd/lockd.h> | |
33 | #include <linux/kthread.h> | |
34 | ||
35 | #define NLMDBG_FACILITY NLMDBG_SVCLOCK | |
36 | ||
37 | #ifdef CONFIG_LOCKD_V4 | |
38 | #define nlm_deadlock nlm4_deadlock | |
39 | #else | |
40 | #define nlm_deadlock nlm_lck_denied | |
41 | #endif | |
42 | ||
43 | static void nlmsvc_release_block(struct nlm_block *block); | |
44 | static void nlmsvc_insert_block(struct nlm_block *block, unsigned long); | |
45 | static void nlmsvc_remove_block(struct nlm_block *block); | |
46 | ||
47 | static int nlmsvc_setgrantargs(struct nlm_rqst *call, struct nlm_lock *lock); | |
48 | static void nlmsvc_freegrantargs(struct nlm_rqst *call); | |
49 | static const struct rpc_call_ops nlmsvc_grant_ops; | |
50 | ||
51 | /* | |
52 | * The list of blocked locks to retry | |
53 | */ | |
54 | static LIST_HEAD(nlm_blocked); | |
55 | static DEFINE_SPINLOCK(nlm_blocked_lock); | |
56 | ||
57 | #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) | |
58 | static const char *nlmdbg_cookie2a(const struct nlm_cookie *cookie) | |
59 | { | |
60 | /* | |
61 | * We can get away with a static buffer because this is only called | |
62 | * from lockd, which is single-threaded. | |
63 | */ | |
64 | static char buf[2*NLM_MAXCOOKIELEN+1]; | |
65 | unsigned int i, len = sizeof(buf); | |
66 | char *p = buf; | |
67 | ||
68 | len--; /* allow for trailing \0 */ | |
69 | if (len < 3) | |
70 | return "???"; | |
71 | for (i = 0 ; i < cookie->len ; i++) { | |
72 | if (len < 2) { | |
73 | strcpy(p-3, "..."); | |
74 | break; | |
75 | } | |
76 | sprintf(p, "%02x", cookie->data[i]); | |
77 | p += 2; | |
78 | len -= 2; | |
79 | } | |
80 | *p = '\0'; | |
81 | ||
82 | return buf; | |
83 | } | |
84 | #endif | |
85 | ||
86 | /* | |
87 | * Insert a blocked lock into the global list | |
88 | */ | |
89 | static void | |
90 | nlmsvc_insert_block_locked(struct nlm_block *block, unsigned long when) | |
91 | { | |
92 | struct nlm_block *b; | |
93 | struct list_head *pos; | |
94 | ||
95 | dprintk("lockd: nlmsvc_insert_block(%p, %ld)\n", block, when); | |
96 | if (list_empty(&block->b_list)) { | |
97 | kref_get(&block->b_count); | |
98 | } else { | |
99 | list_del_init(&block->b_list); | |
100 | } | |
101 | ||
102 | pos = &nlm_blocked; | |
103 | if (when != NLM_NEVER) { | |
104 | if ((when += jiffies) == NLM_NEVER) | |
105 | when ++; | |
106 | list_for_each(pos, &nlm_blocked) { | |
107 | b = list_entry(pos, struct nlm_block, b_list); | |
108 | if (time_after(b->b_when,when) || b->b_when == NLM_NEVER) | |
109 | break; | |
110 | } | |
111 | /* On normal exit from the loop, pos == &nlm_blocked, | |
112 | * so we will be adding to the end of the list - good | |
113 | */ | |
114 | } | |
115 | ||
116 | list_add_tail(&block->b_list, pos); | |
117 | block->b_when = when; | |
118 | } | |
119 | ||
120 | static void nlmsvc_insert_block(struct nlm_block *block, unsigned long when) | |
121 | { | |
122 | spin_lock(&nlm_blocked_lock); | |
123 | nlmsvc_insert_block_locked(block, when); | |
124 | spin_unlock(&nlm_blocked_lock); | |
125 | } | |
126 | ||
127 | /* | |
128 | * Remove a block from the global list | |
129 | */ | |
130 | static inline void | |
131 | nlmsvc_remove_block(struct nlm_block *block) | |
132 | { | |
133 | if (!list_empty(&block->b_list)) { | |
134 | spin_lock(&nlm_blocked_lock); | |
135 | list_del_init(&block->b_list); | |
136 | spin_unlock(&nlm_blocked_lock); | |
137 | nlmsvc_release_block(block); | |
138 | } | |
139 | } | |
140 | ||
141 | /* | |
142 | * Find a block for a given lock | |
143 | */ | |
144 | static struct nlm_block * | |
145 | nlmsvc_lookup_block(struct nlm_file *file, struct nlm_lock *lock) | |
146 | { | |
147 | struct nlm_block *block; | |
148 | struct file_lock *fl; | |
149 | ||
150 | dprintk("lockd: nlmsvc_lookup_block f=%p pd=%d %Ld-%Ld ty=%d\n", | |
151 | file, lock->fl.fl_pid, | |
152 | (long long)lock->fl.fl_start, | |
153 | (long long)lock->fl.fl_end, lock->fl.fl_type); | |
154 | list_for_each_entry(block, &nlm_blocked, b_list) { | |
155 | fl = &block->b_call->a_args.lock.fl; | |
156 | dprintk("lockd: check f=%p pd=%d %Ld-%Ld ty=%d cookie=%s\n", | |
157 | block->b_file, fl->fl_pid, | |
158 | (long long)fl->fl_start, | |
159 | (long long)fl->fl_end, fl->fl_type, | |
160 | nlmdbg_cookie2a(&block->b_call->a_args.cookie)); | |
161 | if (block->b_file == file && nlm_compare_locks(fl, &lock->fl)) { | |
162 | kref_get(&block->b_count); | |
163 | return block; | |
164 | } | |
165 | } | |
166 | ||
167 | return NULL; | |
168 | } | |
169 | ||
170 | static inline int nlm_cookie_match(struct nlm_cookie *a, struct nlm_cookie *b) | |
171 | { | |
172 | if (a->len != b->len) | |
173 | return 0; | |
174 | if (memcmp(a->data, b->data, a->len)) | |
175 | return 0; | |
176 | return 1; | |
177 | } | |
178 | ||
179 | /* | |
180 | * Find a block with a given NLM cookie. | |
181 | */ | |
182 | static inline struct nlm_block * | |
183 | nlmsvc_find_block(struct nlm_cookie *cookie) | |
184 | { | |
185 | struct nlm_block *block; | |
186 | ||
187 | list_for_each_entry(block, &nlm_blocked, b_list) { | |
188 | if (nlm_cookie_match(&block->b_call->a_args.cookie,cookie)) | |
189 | goto found; | |
190 | } | |
191 | ||
192 | return NULL; | |
193 | ||
194 | found: | |
195 | dprintk("nlmsvc_find_block(%s): block=%p\n", nlmdbg_cookie2a(cookie), block); | |
196 | kref_get(&block->b_count); | |
197 | return block; | |
198 | } | |
199 | ||
200 | /* | |
201 | * Create a block and initialize it. | |
202 | * | |
203 | * Note: we explicitly set the cookie of the grant reply to that of | |
204 | * the blocked lock request. The spec explicitly mentions that the client | |
205 | * should _not_ rely on the callback containing the same cookie as the | |
206 | * request, but (as I found out later) that's because some implementations | |
207 | * do just this. Never mind the standards comittees, they support our | |
208 | * logging industries. | |
209 | * | |
210 | * 10 years later: I hope we can safely ignore these old and broken | |
211 | * clients by now. Let's fix this so we can uniquely identify an incoming | |
212 | * GRANTED_RES message by cookie, without having to rely on the client's IP | |
213 | * address. --okir | |
214 | */ | |
215 | static struct nlm_block * | |
216 | nlmsvc_create_block(struct svc_rqst *rqstp, struct nlm_host *host, | |
217 | struct nlm_file *file, struct nlm_lock *lock, | |
218 | struct nlm_cookie *cookie) | |
219 | { | |
220 | struct nlm_block *block; | |
221 | struct nlm_rqst *call = NULL; | |
222 | ||
223 | call = nlm_alloc_call(host); | |
224 | if (call == NULL) | |
225 | return NULL; | |
226 | ||
227 | /* Allocate memory for block, and initialize arguments */ | |
228 | block = kzalloc(sizeof(*block), GFP_KERNEL); | |
229 | if (block == NULL) | |
230 | goto failed; | |
231 | kref_init(&block->b_count); | |
232 | INIT_LIST_HEAD(&block->b_list); | |
233 | INIT_LIST_HEAD(&block->b_flist); | |
234 | ||
235 | if (!nlmsvc_setgrantargs(call, lock)) | |
236 | goto failed_free; | |
237 | ||
238 | /* Set notifier function for VFS, and init args */ | |
239 | call->a_args.lock.fl.fl_flags |= FL_SLEEP; | |
240 | call->a_args.lock.fl.fl_lmops = &nlmsvc_lock_operations; | |
241 | nlmclnt_next_cookie(&call->a_args.cookie); | |
242 | ||
243 | dprintk("lockd: created block %p...\n", block); | |
244 | ||
245 | /* Create and initialize the block */ | |
246 | block->b_daemon = rqstp->rq_server; | |
247 | block->b_host = host; | |
248 | block->b_file = file; | |
249 | file->f_count++; | |
250 | ||
251 | /* Add to file's list of blocks */ | |
252 | list_add(&block->b_flist, &file->f_blocks); | |
253 | ||
254 | /* Set up RPC arguments for callback */ | |
255 | block->b_call = call; | |
256 | call->a_flags = RPC_TASK_ASYNC; | |
257 | call->a_block = block; | |
258 | ||
259 | return block; | |
260 | ||
261 | failed_free: | |
262 | kfree(block); | |
263 | failed: | |
264 | nlmsvc_release_call(call); | |
265 | return NULL; | |
266 | } | |
267 | ||
268 | /* | |
269 | * Delete a block. | |
270 | * It is the caller's responsibility to check whether the file | |
271 | * can be closed hereafter. | |
272 | */ | |
273 | static int nlmsvc_unlink_block(struct nlm_block *block) | |
274 | { | |
275 | int status; | |
276 | dprintk("lockd: unlinking block %p...\n", block); | |
277 | ||
278 | /* Remove block from list */ | |
279 | status = locks_delete_block(&block->b_call->a_args.lock.fl); | |
280 | nlmsvc_remove_block(block); | |
281 | return status; | |
282 | } | |
283 | ||
284 | static void nlmsvc_free_block(struct kref *kref) | |
285 | { | |
286 | struct nlm_block *block = container_of(kref, struct nlm_block, b_count); | |
287 | struct nlm_file *file = block->b_file; | |
288 | ||
289 | dprintk("lockd: freeing block %p...\n", block); | |
290 | ||
291 | /* Remove block from file's list of blocks */ | |
292 | list_del_init(&block->b_flist); | |
293 | mutex_unlock(&file->f_mutex); | |
294 | ||
295 | nlmsvc_freegrantargs(block->b_call); | |
296 | nlmsvc_release_call(block->b_call); | |
297 | nlm_release_file(block->b_file); | |
298 | kfree(block); | |
299 | } | |
300 | ||
301 | static void nlmsvc_release_block(struct nlm_block *block) | |
302 | { | |
303 | if (block != NULL) | |
304 | kref_put_mutex(&block->b_count, nlmsvc_free_block, &block->b_file->f_mutex); | |
305 | } | |
306 | ||
307 | /* | |
308 | * Loop over all blocks and delete blocks held by | |
309 | * a matching host. | |
310 | */ | |
311 | void nlmsvc_traverse_blocks(struct nlm_host *host, | |
312 | struct nlm_file *file, | |
313 | nlm_host_match_fn_t match) | |
314 | { | |
315 | struct nlm_block *block, *next; | |
316 | ||
317 | restart: | |
318 | mutex_lock(&file->f_mutex); | |
319 | list_for_each_entry_safe(block, next, &file->f_blocks, b_flist) { | |
320 | if (!match(block->b_host, host)) | |
321 | continue; | |
322 | /* Do not destroy blocks that are not on | |
323 | * the global retry list - why? */ | |
324 | if (list_empty(&block->b_list)) | |
325 | continue; | |
326 | kref_get(&block->b_count); | |
327 | mutex_unlock(&file->f_mutex); | |
328 | nlmsvc_unlink_block(block); | |
329 | nlmsvc_release_block(block); | |
330 | goto restart; | |
331 | } | |
332 | mutex_unlock(&file->f_mutex); | |
333 | } | |
334 | ||
335 | static struct nlm_lockowner * | |
336 | nlmsvc_get_lockowner(struct nlm_lockowner *lockowner) | |
337 | { | |
338 | refcount_inc(&lockowner->count); | |
339 | return lockowner; | |
340 | } | |
341 | ||
342 | static void nlmsvc_put_lockowner(struct nlm_lockowner *lockowner) | |
343 | { | |
344 | if (!refcount_dec_and_lock(&lockowner->count, &lockowner->host->h_lock)) | |
345 | return; | |
346 | list_del(&lockowner->list); | |
347 | spin_unlock(&lockowner->host->h_lock); | |
348 | nlmsvc_release_host(lockowner->host); | |
349 | kfree(lockowner); | |
350 | } | |
351 | ||
352 | static struct nlm_lockowner *__nlmsvc_find_lockowner(struct nlm_host *host, pid_t pid) | |
353 | { | |
354 | struct nlm_lockowner *lockowner; | |
355 | list_for_each_entry(lockowner, &host->h_lockowners, list) { | |
356 | if (lockowner->pid != pid) | |
357 | continue; | |
358 | return nlmsvc_get_lockowner(lockowner); | |
359 | } | |
360 | return NULL; | |
361 | } | |
362 | ||
363 | static struct nlm_lockowner *nlmsvc_find_lockowner(struct nlm_host *host, pid_t pid) | |
364 | { | |
365 | struct nlm_lockowner *res, *new = NULL; | |
366 | ||
367 | spin_lock(&host->h_lock); | |
368 | res = __nlmsvc_find_lockowner(host, pid); | |
369 | ||
370 | if (res == NULL) { | |
371 | spin_unlock(&host->h_lock); | |
372 | new = kmalloc(sizeof(*res), GFP_KERNEL); | |
373 | spin_lock(&host->h_lock); | |
374 | res = __nlmsvc_find_lockowner(host, pid); | |
375 | if (res == NULL && new != NULL) { | |
376 | res = new; | |
377 | /* fs/locks.c will manage the refcount through lock_ops */ | |
378 | refcount_set(&new->count, 1); | |
379 | new->pid = pid; | |
380 | new->host = nlm_get_host(host); | |
381 | list_add(&new->list, &host->h_lockowners); | |
382 | new = NULL; | |
383 | } | |
384 | } | |
385 | ||
386 | spin_unlock(&host->h_lock); | |
387 | kfree(new); | |
388 | return res; | |
389 | } | |
390 | ||
391 | void | |
392 | nlmsvc_release_lockowner(struct nlm_lock *lock) | |
393 | { | |
394 | if (lock->fl.fl_owner) | |
395 | nlmsvc_put_lockowner(lock->fl.fl_owner); | |
396 | } | |
397 | ||
398 | static void nlmsvc_locks_copy_lock(struct file_lock *new, struct file_lock *fl) | |
399 | { | |
400 | struct nlm_lockowner *nlm_lo = (struct nlm_lockowner *)fl->fl_owner; | |
401 | new->fl_owner = nlmsvc_get_lockowner(nlm_lo); | |
402 | } | |
403 | ||
404 | static void nlmsvc_locks_release_private(struct file_lock *fl) | |
405 | { | |
406 | nlmsvc_put_lockowner((struct nlm_lockowner *)fl->fl_owner); | |
407 | } | |
408 | ||
409 | static const struct file_lock_operations nlmsvc_lock_ops = { | |
410 | .fl_copy_lock = nlmsvc_locks_copy_lock, | |
411 | .fl_release_private = nlmsvc_locks_release_private, | |
412 | }; | |
413 | ||
414 | void nlmsvc_locks_init_private(struct file_lock *fl, struct nlm_host *host, | |
415 | pid_t pid) | |
416 | { | |
417 | fl->fl_owner = nlmsvc_find_lockowner(host, pid); | |
418 | if (fl->fl_owner != NULL) | |
419 | fl->fl_ops = &nlmsvc_lock_ops; | |
420 | } | |
421 | ||
422 | /* | |
423 | * Initialize arguments for GRANTED call. The nlm_rqst structure | |
424 | * has been cleared already. | |
425 | */ | |
426 | static int nlmsvc_setgrantargs(struct nlm_rqst *call, struct nlm_lock *lock) | |
427 | { | |
428 | locks_copy_lock(&call->a_args.lock.fl, &lock->fl); | |
429 | memcpy(&call->a_args.lock.fh, &lock->fh, sizeof(call->a_args.lock.fh)); | |
430 | call->a_args.lock.caller = utsname()->nodename; | |
431 | call->a_args.lock.oh.len = lock->oh.len; | |
432 | ||
433 | /* set default data area */ | |
434 | call->a_args.lock.oh.data = call->a_owner; | |
435 | call->a_args.lock.svid = ((struct nlm_lockowner *)lock->fl.fl_owner)->pid; | |
436 | ||
437 | if (lock->oh.len > NLMCLNT_OHSIZE) { | |
438 | void *data = kmalloc(lock->oh.len, GFP_KERNEL); | |
439 | if (!data) | |
440 | return 0; | |
441 | call->a_args.lock.oh.data = (u8 *) data; | |
442 | } | |
443 | ||
444 | memcpy(call->a_args.lock.oh.data, lock->oh.data, lock->oh.len); | |
445 | return 1; | |
446 | } | |
447 | ||
448 | static void nlmsvc_freegrantargs(struct nlm_rqst *call) | |
449 | { | |
450 | if (call->a_args.lock.oh.data != call->a_owner) | |
451 | kfree(call->a_args.lock.oh.data); | |
452 | ||
453 | locks_release_private(&call->a_args.lock.fl); | |
454 | } | |
455 | ||
456 | /* | |
457 | * Deferred lock request handling for non-blocking lock | |
458 | */ | |
459 | static __be32 | |
460 | nlmsvc_defer_lock_rqst(struct svc_rqst *rqstp, struct nlm_block *block) | |
461 | { | |
462 | __be32 status = nlm_lck_denied_nolocks; | |
463 | ||
464 | block->b_flags |= B_QUEUED; | |
465 | ||
466 | nlmsvc_insert_block(block, NLM_TIMEOUT); | |
467 | ||
468 | block->b_cache_req = &rqstp->rq_chandle; | |
469 | if (rqstp->rq_chandle.defer) { | |
470 | block->b_deferred_req = | |
471 | rqstp->rq_chandle.defer(block->b_cache_req); | |
472 | if (block->b_deferred_req != NULL) | |
473 | status = nlm_drop_reply; | |
474 | } | |
475 | dprintk("lockd: nlmsvc_defer_lock_rqst block %p flags %d status %d\n", | |
476 | block, block->b_flags, ntohl(status)); | |
477 | ||
478 | return status; | |
479 | } | |
480 | ||
481 | /* | |
482 | * Attempt to establish a lock, and if it can't be granted, block it | |
483 | * if required. | |
484 | */ | |
485 | __be32 | |
486 | nlmsvc_lock(struct svc_rqst *rqstp, struct nlm_file *file, | |
487 | struct nlm_host *host, struct nlm_lock *lock, int wait, | |
488 | struct nlm_cookie *cookie, int reclaim) | |
489 | { | |
490 | struct nlm_block *block = NULL; | |
491 | int error; | |
492 | __be32 ret; | |
493 | ||
494 | dprintk("lockd: nlmsvc_lock(%s/%ld, ty=%d, pi=%d, %Ld-%Ld, bl=%d)\n", | |
495 | locks_inode(file->f_file)->i_sb->s_id, | |
496 | locks_inode(file->f_file)->i_ino, | |
497 | lock->fl.fl_type, lock->fl.fl_pid, | |
498 | (long long)lock->fl.fl_start, | |
499 | (long long)lock->fl.fl_end, | |
500 | wait); | |
501 | ||
502 | /* Lock file against concurrent access */ | |
503 | mutex_lock(&file->f_mutex); | |
504 | /* Get existing block (in case client is busy-waiting) | |
505 | * or create new block | |
506 | */ | |
507 | block = nlmsvc_lookup_block(file, lock); | |
508 | if (block == NULL) { | |
509 | block = nlmsvc_create_block(rqstp, host, file, lock, cookie); | |
510 | ret = nlm_lck_denied_nolocks; | |
511 | if (block == NULL) | |
512 | goto out; | |
513 | lock = &block->b_call->a_args.lock; | |
514 | } else | |
515 | lock->fl.fl_flags &= ~FL_SLEEP; | |
516 | ||
517 | if (block->b_flags & B_QUEUED) { | |
518 | dprintk("lockd: nlmsvc_lock deferred block %p flags %d\n", | |
519 | block, block->b_flags); | |
520 | if (block->b_granted) { | |
521 | nlmsvc_unlink_block(block); | |
522 | ret = nlm_granted; | |
523 | goto out; | |
524 | } | |
525 | if (block->b_flags & B_TIMED_OUT) { | |
526 | nlmsvc_unlink_block(block); | |
527 | ret = nlm_lck_denied; | |
528 | goto out; | |
529 | } | |
530 | ret = nlm_drop_reply; | |
531 | goto out; | |
532 | } | |
533 | ||
534 | if (locks_in_grace(SVC_NET(rqstp)) && !reclaim) { | |
535 | ret = nlm_lck_denied_grace_period; | |
536 | goto out; | |
537 | } | |
538 | if (reclaim && !locks_in_grace(SVC_NET(rqstp))) { | |
539 | ret = nlm_lck_denied_grace_period; | |
540 | goto out; | |
541 | } | |
542 | ||
543 | if (!wait) | |
544 | lock->fl.fl_flags &= ~FL_SLEEP; | |
545 | error = vfs_lock_file(file->f_file, F_SETLK, &lock->fl, NULL); | |
546 | lock->fl.fl_flags &= ~FL_SLEEP; | |
547 | ||
548 | dprintk("lockd: vfs_lock_file returned %d\n", error); | |
549 | switch (error) { | |
550 | case 0: | |
551 | ret = nlm_granted; | |
552 | goto out; | |
553 | case -EAGAIN: | |
554 | /* | |
555 | * If this is a blocking request for an | |
556 | * already pending lock request then we need | |
557 | * to put it back on lockd's block list | |
558 | */ | |
559 | if (wait) | |
560 | break; | |
561 | ret = nlm_lck_denied; | |
562 | goto out; | |
563 | case FILE_LOCK_DEFERRED: | |
564 | if (wait) | |
565 | break; | |
566 | /* Filesystem lock operation is in progress | |
567 | Add it to the queue waiting for callback */ | |
568 | ret = nlmsvc_defer_lock_rqst(rqstp, block); | |
569 | goto out; | |
570 | case -EDEADLK: | |
571 | ret = nlm_deadlock; | |
572 | goto out; | |
573 | default: /* includes ENOLCK */ | |
574 | ret = nlm_lck_denied_nolocks; | |
575 | goto out; | |
576 | } | |
577 | ||
578 | ret = nlm_lck_blocked; | |
579 | ||
580 | /* Append to list of blocked */ | |
581 | nlmsvc_insert_block(block, NLM_NEVER); | |
582 | out: | |
583 | mutex_unlock(&file->f_mutex); | |
584 | nlmsvc_release_block(block); | |
585 | dprintk("lockd: nlmsvc_lock returned %u\n", ret); | |
586 | return ret; | |
587 | } | |
588 | ||
589 | /* | |
590 | * Test for presence of a conflicting lock. | |
591 | */ | |
592 | __be32 | |
593 | nlmsvc_testlock(struct svc_rqst *rqstp, struct nlm_file *file, | |
594 | struct nlm_host *host, struct nlm_lock *lock, | |
595 | struct nlm_lock *conflock, struct nlm_cookie *cookie) | |
596 | { | |
597 | int error; | |
598 | __be32 ret; | |
599 | struct nlm_lockowner *test_owner; | |
600 | ||
601 | dprintk("lockd: nlmsvc_testlock(%s/%ld, ty=%d, %Ld-%Ld)\n", | |
602 | locks_inode(file->f_file)->i_sb->s_id, | |
603 | locks_inode(file->f_file)->i_ino, | |
604 | lock->fl.fl_type, | |
605 | (long long)lock->fl.fl_start, | |
606 | (long long)lock->fl.fl_end); | |
607 | ||
608 | if (locks_in_grace(SVC_NET(rqstp))) { | |
609 | ret = nlm_lck_denied_grace_period; | |
610 | goto out; | |
611 | } | |
612 | ||
613 | /* If there's a conflicting lock, remember to clean up the test lock */ | |
614 | test_owner = (struct nlm_lockowner *)lock->fl.fl_owner; | |
615 | ||
616 | error = vfs_test_lock(file->f_file, &lock->fl); | |
617 | if (error) { | |
618 | /* We can't currently deal with deferred test requests */ | |
619 | if (error == FILE_LOCK_DEFERRED) | |
620 | WARN_ON_ONCE(1); | |
621 | ||
622 | ret = nlm_lck_denied_nolocks; | |
623 | goto out; | |
624 | } | |
625 | ||
626 | if (lock->fl.fl_type == F_UNLCK) { | |
627 | ret = nlm_granted; | |
628 | goto out; | |
629 | } | |
630 | ||
631 | dprintk("lockd: conflicting lock(ty=%d, %Ld-%Ld)\n", | |
632 | lock->fl.fl_type, (long long)lock->fl.fl_start, | |
633 | (long long)lock->fl.fl_end); | |
634 | conflock->caller = "somehost"; /* FIXME */ | |
635 | conflock->len = strlen(conflock->caller); | |
636 | conflock->oh.len = 0; /* don't return OH info */ | |
637 | conflock->svid = ((struct nlm_lockowner *)lock->fl.fl_owner)->pid; | |
638 | conflock->fl.fl_type = lock->fl.fl_type; | |
639 | conflock->fl.fl_start = lock->fl.fl_start; | |
640 | conflock->fl.fl_end = lock->fl.fl_end; | |
641 | locks_release_private(&lock->fl); | |
642 | ||
643 | /* Clean up the test lock */ | |
644 | lock->fl.fl_owner = NULL; | |
645 | nlmsvc_put_lockowner(test_owner); | |
646 | ||
647 | ret = nlm_lck_denied; | |
648 | out: | |
649 | return ret; | |
650 | } | |
651 | ||
652 | /* | |
653 | * Remove a lock. | |
654 | * This implies a CANCEL call: We send a GRANT_MSG, the client replies | |
655 | * with a GRANT_RES call which gets lost, and calls UNLOCK immediately | |
656 | * afterwards. In this case the block will still be there, and hence | |
657 | * must be removed. | |
658 | */ | |
659 | __be32 | |
660 | nlmsvc_unlock(struct net *net, struct nlm_file *file, struct nlm_lock *lock) | |
661 | { | |
662 | int error; | |
663 | ||
664 | dprintk("lockd: nlmsvc_unlock(%s/%ld, pi=%d, %Ld-%Ld)\n", | |
665 | locks_inode(file->f_file)->i_sb->s_id, | |
666 | locks_inode(file->f_file)->i_ino, | |
667 | lock->fl.fl_pid, | |
668 | (long long)lock->fl.fl_start, | |
669 | (long long)lock->fl.fl_end); | |
670 | ||
671 | /* First, cancel any lock that might be there */ | |
672 | nlmsvc_cancel_blocked(net, file, lock); | |
673 | ||
674 | lock->fl.fl_type = F_UNLCK; | |
675 | error = vfs_lock_file(file->f_file, F_SETLK, &lock->fl, NULL); | |
676 | ||
677 | return (error < 0)? nlm_lck_denied_nolocks : nlm_granted; | |
678 | } | |
679 | ||
680 | /* | |
681 | * Cancel a previously blocked request. | |
682 | * | |
683 | * A cancel request always overrides any grant that may currently | |
684 | * be in progress. | |
685 | * The calling procedure must check whether the file can be closed. | |
686 | */ | |
687 | __be32 | |
688 | nlmsvc_cancel_blocked(struct net *net, struct nlm_file *file, struct nlm_lock *lock) | |
689 | { | |
690 | struct nlm_block *block; | |
691 | int status = 0; | |
692 | ||
693 | dprintk("lockd: nlmsvc_cancel(%s/%ld, pi=%d, %Ld-%Ld)\n", | |
694 | locks_inode(file->f_file)->i_sb->s_id, | |
695 | locks_inode(file->f_file)->i_ino, | |
696 | lock->fl.fl_pid, | |
697 | (long long)lock->fl.fl_start, | |
698 | (long long)lock->fl.fl_end); | |
699 | ||
700 | if (locks_in_grace(net)) | |
701 | return nlm_lck_denied_grace_period; | |
702 | ||
703 | mutex_lock(&file->f_mutex); | |
704 | block = nlmsvc_lookup_block(file, lock); | |
705 | mutex_unlock(&file->f_mutex); | |
706 | if (block != NULL) { | |
707 | vfs_cancel_lock(block->b_file->f_file, | |
708 | &block->b_call->a_args.lock.fl); | |
709 | status = nlmsvc_unlink_block(block); | |
710 | nlmsvc_release_block(block); | |
711 | } | |
712 | return status ? nlm_lck_denied : nlm_granted; | |
713 | } | |
714 | ||
715 | /* | |
716 | * This is a callback from the filesystem for VFS file lock requests. | |
717 | * It will be used if lm_grant is defined and the filesystem can not | |
718 | * respond to the request immediately. | |
719 | * For SETLK or SETLKW request it will get the local posix lock. | |
720 | * In all cases it will move the block to the head of nlm_blocked q where | |
721 | * nlmsvc_retry_blocked() can send back a reply for SETLKW or revisit the | |
722 | * deferred rpc for GETLK and SETLK. | |
723 | */ | |
724 | static void | |
725 | nlmsvc_update_deferred_block(struct nlm_block *block, int result) | |
726 | { | |
727 | block->b_flags |= B_GOT_CALLBACK; | |
728 | if (result == 0) | |
729 | block->b_granted = 1; | |
730 | else | |
731 | block->b_flags |= B_TIMED_OUT; | |
732 | } | |
733 | ||
734 | static int nlmsvc_grant_deferred(struct file_lock *fl, int result) | |
735 | { | |
736 | struct nlm_block *block; | |
737 | int rc = -ENOENT; | |
738 | ||
739 | spin_lock(&nlm_blocked_lock); | |
740 | list_for_each_entry(block, &nlm_blocked, b_list) { | |
741 | if (nlm_compare_locks(&block->b_call->a_args.lock.fl, fl)) { | |
742 | dprintk("lockd: nlmsvc_notify_blocked block %p flags %d\n", | |
743 | block, block->b_flags); | |
744 | if (block->b_flags & B_QUEUED) { | |
745 | if (block->b_flags & B_TIMED_OUT) { | |
746 | rc = -ENOLCK; | |
747 | break; | |
748 | } | |
749 | nlmsvc_update_deferred_block(block, result); | |
750 | } else if (result == 0) | |
751 | block->b_granted = 1; | |
752 | ||
753 | nlmsvc_insert_block_locked(block, 0); | |
754 | svc_wake_up(block->b_daemon); | |
755 | rc = 0; | |
756 | break; | |
757 | } | |
758 | } | |
759 | spin_unlock(&nlm_blocked_lock); | |
760 | if (rc == -ENOENT) | |
761 | printk(KERN_WARNING "lockd: grant for unknown block\n"); | |
762 | return rc; | |
763 | } | |
764 | ||
765 | /* | |
766 | * Unblock a blocked lock request. This is a callback invoked from the | |
767 | * VFS layer when a lock on which we blocked is removed. | |
768 | * | |
769 | * This function doesn't grant the blocked lock instantly, but rather moves | |
770 | * the block to the head of nlm_blocked where it can be picked up by lockd. | |
771 | */ | |
772 | static void | |
773 | nlmsvc_notify_blocked(struct file_lock *fl) | |
774 | { | |
775 | struct nlm_block *block; | |
776 | ||
777 | dprintk("lockd: VFS unblock notification for block %p\n", fl); | |
778 | spin_lock(&nlm_blocked_lock); | |
779 | list_for_each_entry(block, &nlm_blocked, b_list) { | |
780 | if (nlm_compare_locks(&block->b_call->a_args.lock.fl, fl)) { | |
781 | nlmsvc_insert_block_locked(block, 0); | |
782 | spin_unlock(&nlm_blocked_lock); | |
783 | svc_wake_up(block->b_daemon); | |
784 | return; | |
785 | } | |
786 | } | |
787 | spin_unlock(&nlm_blocked_lock); | |
788 | printk(KERN_WARNING "lockd: notification for unknown block!\n"); | |
789 | } | |
790 | ||
791 | const struct lock_manager_operations nlmsvc_lock_operations = { | |
792 | .lm_notify = nlmsvc_notify_blocked, | |
793 | .lm_grant = nlmsvc_grant_deferred, | |
794 | }; | |
795 | ||
796 | /* | |
797 | * Try to claim a lock that was previously blocked. | |
798 | * | |
799 | * Note that we use both the RPC_GRANTED_MSG call _and_ an async | |
800 | * RPC thread when notifying the client. This seems like overkill... | |
801 | * Here's why: | |
802 | * - we don't want to use a synchronous RPC thread, otherwise | |
803 | * we might find ourselves hanging on a dead portmapper. | |
804 | * - Some lockd implementations (e.g. HP) don't react to | |
805 | * RPC_GRANTED calls; they seem to insist on RPC_GRANTED_MSG calls. | |
806 | */ | |
807 | static void | |
808 | nlmsvc_grant_blocked(struct nlm_block *block) | |
809 | { | |
810 | struct nlm_file *file = block->b_file; | |
811 | struct nlm_lock *lock = &block->b_call->a_args.lock; | |
812 | int error; | |
813 | loff_t fl_start, fl_end; | |
814 | ||
815 | dprintk("lockd: grant blocked lock %p\n", block); | |
816 | ||
817 | kref_get(&block->b_count); | |
818 | ||
819 | /* Unlink block request from list */ | |
820 | nlmsvc_unlink_block(block); | |
821 | ||
822 | /* If b_granted is true this means we've been here before. | |
823 | * Just retry the grant callback, possibly refreshing the RPC | |
824 | * binding */ | |
825 | if (block->b_granted) { | |
826 | nlm_rebind_host(block->b_host); | |
827 | goto callback; | |
828 | } | |
829 | ||
830 | /* Try the lock operation again */ | |
831 | /* vfs_lock_file() can mangle fl_start and fl_end, but we need | |
832 | * them unchanged for the GRANT_MSG | |
833 | */ | |
834 | lock->fl.fl_flags |= FL_SLEEP; | |
835 | fl_start = lock->fl.fl_start; | |
836 | fl_end = lock->fl.fl_end; | |
837 | error = vfs_lock_file(file->f_file, F_SETLK, &lock->fl, NULL); | |
838 | lock->fl.fl_flags &= ~FL_SLEEP; | |
839 | lock->fl.fl_start = fl_start; | |
840 | lock->fl.fl_end = fl_end; | |
841 | ||
842 | switch (error) { | |
843 | case 0: | |
844 | break; | |
845 | case FILE_LOCK_DEFERRED: | |
846 | dprintk("lockd: lock still blocked error %d\n", error); | |
847 | nlmsvc_insert_block(block, NLM_NEVER); | |
848 | nlmsvc_release_block(block); | |
849 | return; | |
850 | default: | |
851 | printk(KERN_WARNING "lockd: unexpected error %d in %s!\n", | |
852 | -error, __func__); | |
853 | nlmsvc_insert_block(block, 10 * HZ); | |
854 | nlmsvc_release_block(block); | |
855 | return; | |
856 | } | |
857 | ||
858 | callback: | |
859 | /* Lock was granted by VFS. */ | |
860 | dprintk("lockd: GRANTing blocked lock.\n"); | |
861 | block->b_granted = 1; | |
862 | ||
863 | /* keep block on the list, but don't reattempt until the RPC | |
864 | * completes or the submission fails | |
865 | */ | |
866 | nlmsvc_insert_block(block, NLM_NEVER); | |
867 | ||
868 | /* Call the client -- use a soft RPC task since nlmsvc_retry_blocked | |
869 | * will queue up a new one if this one times out | |
870 | */ | |
871 | error = nlm_async_call(block->b_call, NLMPROC_GRANTED_MSG, | |
872 | &nlmsvc_grant_ops); | |
873 | ||
874 | /* RPC submission failed, wait a bit and retry */ | |
875 | if (error < 0) | |
876 | nlmsvc_insert_block(block, 10 * HZ); | |
877 | } | |
878 | ||
879 | /* | |
880 | * This is the callback from the RPC layer when the NLM_GRANTED_MSG | |
881 | * RPC call has succeeded or timed out. | |
882 | * Like all RPC callbacks, it is invoked by the rpciod process, so it | |
883 | * better not sleep. Therefore, we put the blocked lock on the nlm_blocked | |
884 | * chain once more in order to have it removed by lockd itself (which can | |
885 | * then sleep on the file semaphore without disrupting e.g. the nfs client). | |
886 | */ | |
887 | static void nlmsvc_grant_callback(struct rpc_task *task, void *data) | |
888 | { | |
889 | struct nlm_rqst *call = data; | |
890 | struct nlm_block *block = call->a_block; | |
891 | unsigned long timeout; | |
892 | ||
893 | dprintk("lockd: GRANT_MSG RPC callback\n"); | |
894 | ||
895 | spin_lock(&nlm_blocked_lock); | |
896 | /* if the block is not on a list at this point then it has | |
897 | * been invalidated. Don't try to requeue it. | |
898 | * | |
899 | * FIXME: it's possible that the block is removed from the list | |
900 | * after this check but before the nlmsvc_insert_block. In that | |
901 | * case it will be added back. Perhaps we need better locking | |
902 | * for nlm_blocked? | |
903 | */ | |
904 | if (list_empty(&block->b_list)) | |
905 | goto out; | |
906 | ||
907 | /* Technically, we should down the file semaphore here. Since we | |
908 | * move the block towards the head of the queue only, no harm | |
909 | * can be done, though. */ | |
910 | if (task->tk_status < 0) { | |
911 | /* RPC error: Re-insert for retransmission */ | |
912 | timeout = 10 * HZ; | |
913 | } else { | |
914 | /* Call was successful, now wait for client callback */ | |
915 | timeout = 60 * HZ; | |
916 | } | |
917 | nlmsvc_insert_block_locked(block, timeout); | |
918 | svc_wake_up(block->b_daemon); | |
919 | out: | |
920 | spin_unlock(&nlm_blocked_lock); | |
921 | } | |
922 | ||
923 | /* | |
924 | * FIXME: nlmsvc_release_block() grabs a mutex. This is not allowed for an | |
925 | * .rpc_release rpc_call_op | |
926 | */ | |
927 | static void nlmsvc_grant_release(void *data) | |
928 | { | |
929 | struct nlm_rqst *call = data; | |
930 | nlmsvc_release_block(call->a_block); | |
931 | } | |
932 | ||
933 | static const struct rpc_call_ops nlmsvc_grant_ops = { | |
934 | .rpc_call_done = nlmsvc_grant_callback, | |
935 | .rpc_release = nlmsvc_grant_release, | |
936 | }; | |
937 | ||
938 | /* | |
939 | * We received a GRANT_RES callback. Try to find the corresponding | |
940 | * block. | |
941 | */ | |
942 | void | |
943 | nlmsvc_grant_reply(struct nlm_cookie *cookie, __be32 status) | |
944 | { | |
945 | struct nlm_block *block; | |
946 | ||
947 | dprintk("grant_reply: looking for cookie %x, s=%d \n", | |
948 | *(unsigned int *)(cookie->data), status); | |
949 | if (!(block = nlmsvc_find_block(cookie))) | |
950 | return; | |
951 | ||
952 | if (status == nlm_lck_denied_grace_period) { | |
953 | /* Try again in a couple of seconds */ | |
954 | nlmsvc_insert_block(block, 10 * HZ); | |
955 | } else { | |
956 | /* | |
957 | * Lock is now held by client, or has been rejected. | |
958 | * In both cases, the block should be removed. | |
959 | */ | |
960 | nlmsvc_unlink_block(block); | |
961 | } | |
962 | nlmsvc_release_block(block); | |
963 | } | |
964 | ||
965 | /* Helper function to handle retry of a deferred block. | |
966 | * If it is a blocking lock, call grant_blocked. | |
967 | * For a non-blocking lock or test lock, revisit the request. | |
968 | */ | |
969 | static void | |
970 | retry_deferred_block(struct nlm_block *block) | |
971 | { | |
972 | if (!(block->b_flags & B_GOT_CALLBACK)) | |
973 | block->b_flags |= B_TIMED_OUT; | |
974 | nlmsvc_insert_block(block, NLM_TIMEOUT); | |
975 | dprintk("revisit block %p flags %d\n", block, block->b_flags); | |
976 | if (block->b_deferred_req) { | |
977 | block->b_deferred_req->revisit(block->b_deferred_req, 0); | |
978 | block->b_deferred_req = NULL; | |
979 | } | |
980 | } | |
981 | ||
982 | /* | |
983 | * Retry all blocked locks that have been notified. This is where lockd | |
984 | * picks up locks that can be granted, or grant notifications that must | |
985 | * be retransmitted. | |
986 | */ | |
987 | unsigned long | |
988 | nlmsvc_retry_blocked(void) | |
989 | { | |
990 | unsigned long timeout = MAX_SCHEDULE_TIMEOUT; | |
991 | struct nlm_block *block; | |
992 | ||
993 | spin_lock(&nlm_blocked_lock); | |
994 | while (!list_empty(&nlm_blocked) && !kthread_should_stop()) { | |
995 | block = list_entry(nlm_blocked.next, struct nlm_block, b_list); | |
996 | ||
997 | if (block->b_when == NLM_NEVER) | |
998 | break; | |
999 | if (time_after(block->b_when, jiffies)) { | |
1000 | timeout = block->b_when - jiffies; | |
1001 | break; | |
1002 | } | |
1003 | spin_unlock(&nlm_blocked_lock); | |
1004 | ||
1005 | dprintk("nlmsvc_retry_blocked(%p, when=%ld)\n", | |
1006 | block, block->b_when); | |
1007 | if (block->b_flags & B_QUEUED) { | |
1008 | dprintk("nlmsvc_retry_blocked delete block (%p, granted=%d, flags=%d)\n", | |
1009 | block, block->b_granted, block->b_flags); | |
1010 | retry_deferred_block(block); | |
1011 | } else | |
1012 | nlmsvc_grant_blocked(block); | |
1013 | spin_lock(&nlm_blocked_lock); | |
1014 | } | |
1015 | spin_unlock(&nlm_blocked_lock); | |
1016 | ||
1017 | return timeout; | |
1018 | } |