Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/lockd/svclock.c | |
3 | * | |
4 | * Handling of server-side locks, mostly of the blocked variety. | |
5 | * This is the ugliest part of lockd because we tread on very thin ice. | |
6 | * GRANT and CANCEL calls may get stuck, meet in mid-flight, etc. | |
7 | * IMNSHO introducing the grant callback into the NLM protocol was one | |
8 | * of the worst ideas Sun ever had. Except maybe for the idea of doing | |
9 | * NFS file locking at all. | |
10 | * | |
11 | * I'm trying hard to avoid race conditions by protecting most accesses | |
12 | * to a file's list of blocked locks through a semaphore. The global | |
13 | * list of blocked locks is not protected in this fashion however. | |
14 | * Therefore, some functions (such as the RPC callback for the async grant | |
15 | * call) move blocked locks towards the head of the list *while some other | |
16 | * process might be traversing it*. This should not be a problem in | |
17 | * practice, because this will only cause functions traversing the list | |
18 | * to visit some blocks twice. | |
19 | * | |
20 | * Copyright (C) 1996, Olaf Kirch <okir@monad.swb.de> | |
21 | */ | |
22 | ||
1da177e4 LT |
23 | #include <linux/types.h> |
24 | #include <linux/errno.h> | |
25 | #include <linux/kernel.h> | |
26 | #include <linux/sched.h> | |
27 | #include <linux/smp_lock.h> | |
28 | #include <linux/sunrpc/clnt.h> | |
29 | #include <linux/sunrpc/svc.h> | |
30 | #include <linux/lockd/nlm.h> | |
31 | #include <linux/lockd/lockd.h> | |
d751a7cd | 32 | #include <linux/kthread.h> |
1da177e4 LT |
33 | |
34 | #define NLMDBG_FACILITY NLMDBG_SVCLOCK | |
35 | ||
36 | #ifdef CONFIG_LOCKD_V4 | |
37 | #define nlm_deadlock nlm4_deadlock | |
38 | #else | |
39 | #define nlm_deadlock nlm_lck_denied | |
40 | #endif | |
41 | ||
6849c0ca | 42 | static void nlmsvc_release_block(struct nlm_block *block); |
1da177e4 | 43 | static void nlmsvc_insert_block(struct nlm_block *block, unsigned long); |
68a2d76c | 44 | static void nlmsvc_remove_block(struct nlm_block *block); |
963d8fe5 | 45 | |
5e1abf8c TM |
46 | static int nlmsvc_setgrantargs(struct nlm_rqst *call, struct nlm_lock *lock); |
47 | static void nlmsvc_freegrantargs(struct nlm_rqst *call); | |
963d8fe5 | 48 | static const struct rpc_call_ops nlmsvc_grant_ops; |
1da177e4 LT |
49 | |
50 | /* | |
51 | * The list of blocked locks to retry | |
52 | */ | |
68a2d76c | 53 | static LIST_HEAD(nlm_blocked); |
1da177e4 LT |
54 | |
55 | /* | |
56 | * Insert a blocked lock into the global list | |
57 | */ | |
58 | static void | |
59 | nlmsvc_insert_block(struct nlm_block *block, unsigned long when) | |
60 | { | |
68a2d76c OK |
61 | struct nlm_block *b; |
62 | struct list_head *pos; | |
1da177e4 LT |
63 | |
64 | dprintk("lockd: nlmsvc_insert_block(%p, %ld)\n", block, when); | |
68a2d76c OK |
65 | if (list_empty(&block->b_list)) { |
66 | kref_get(&block->b_count); | |
67 | } else { | |
68 | list_del_init(&block->b_list); | |
69 | } | |
70 | ||
71 | pos = &nlm_blocked; | |
1da177e4 LT |
72 | if (when != NLM_NEVER) { |
73 | if ((when += jiffies) == NLM_NEVER) | |
74 | when ++; | |
68a2d76c OK |
75 | list_for_each(pos, &nlm_blocked) { |
76 | b = list_entry(pos, struct nlm_block, b_list); | |
77 | if (time_after(b->b_when,when) || b->b_when == NLM_NEVER) | |
78 | break; | |
79 | } | |
80 | /* On normal exit from the loop, pos == &nlm_blocked, | |
81 | * so we will be adding to the end of the list - good | |
82 | */ | |
83 | } | |
1da177e4 | 84 | |
68a2d76c | 85 | list_add_tail(&block->b_list, pos); |
1da177e4 | 86 | block->b_when = when; |
1da177e4 LT |
87 | } |
88 | ||
89 | /* | |
90 | * Remove a block from the global list | |
91 | */ | |
68a2d76c | 92 | static inline void |
1da177e4 LT |
93 | nlmsvc_remove_block(struct nlm_block *block) |
94 | { | |
68a2d76c OK |
95 | if (!list_empty(&block->b_list)) { |
96 | list_del_init(&block->b_list); | |
97 | nlmsvc_release_block(block); | |
1da177e4 | 98 | } |
1da177e4 LT |
99 | } |
100 | ||
101 | /* | |
d9f6eb75 | 102 | * Find a block for a given lock |
1da177e4 LT |
103 | */ |
104 | static struct nlm_block * | |
d9f6eb75 | 105 | nlmsvc_lookup_block(struct nlm_file *file, struct nlm_lock *lock) |
1da177e4 | 106 | { |
68a2d76c | 107 | struct nlm_block *block; |
1da177e4 LT |
108 | struct file_lock *fl; |
109 | ||
110 | dprintk("lockd: nlmsvc_lookup_block f=%p pd=%d %Ld-%Ld ty=%d\n", | |
111 | file, lock->fl.fl_pid, | |
112 | (long long)lock->fl.fl_start, | |
113 | (long long)lock->fl.fl_end, lock->fl.fl_type); | |
68a2d76c | 114 | list_for_each_entry(block, &nlm_blocked, b_list) { |
92737230 | 115 | fl = &block->b_call->a_args.lock.fl; |
1da177e4 LT |
116 | dprintk("lockd: check f=%p pd=%d %Ld-%Ld ty=%d cookie=%s\n", |
117 | block->b_file, fl->fl_pid, | |
118 | (long long)fl->fl_start, | |
119 | (long long)fl->fl_end, fl->fl_type, | |
92737230 | 120 | nlmdbg_cookie2a(&block->b_call->a_args.cookie)); |
1da177e4 | 121 | if (block->b_file == file && nlm_compare_locks(fl, &lock->fl)) { |
6849c0ca | 122 | kref_get(&block->b_count); |
1da177e4 LT |
123 | return block; |
124 | } | |
125 | } | |
126 | ||
127 | return NULL; | |
128 | } | |
129 | ||
130 | static inline int nlm_cookie_match(struct nlm_cookie *a, struct nlm_cookie *b) | |
131 | { | |
6d7bbbba | 132 | if (a->len != b->len) |
1da177e4 | 133 | return 0; |
6d7bbbba | 134 | if (memcmp(a->data, b->data, a->len)) |
1da177e4 LT |
135 | return 0; |
136 | return 1; | |
137 | } | |
138 | ||
139 | /* | |
140 | * Find a block with a given NLM cookie. | |
141 | */ | |
142 | static inline struct nlm_block * | |
39be4502 | 143 | nlmsvc_find_block(struct nlm_cookie *cookie) |
1da177e4 LT |
144 | { |
145 | struct nlm_block *block; | |
146 | ||
68a2d76c | 147 | list_for_each_entry(block, &nlm_blocked, b_list) { |
39be4502 | 148 | if (nlm_cookie_match(&block->b_call->a_args.cookie,cookie)) |
68a2d76c | 149 | goto found; |
1da177e4 LT |
150 | } |
151 | ||
68a2d76c OK |
152 | return NULL; |
153 | ||
154 | found: | |
39be4502 | 155 | dprintk("nlmsvc_find_block(%s): block=%p\n", nlmdbg_cookie2a(cookie), block); |
68a2d76c | 156 | kref_get(&block->b_count); |
1da177e4 LT |
157 | return block; |
158 | } | |
159 | ||
160 | /* | |
161 | * Create a block and initialize it. | |
162 | * | |
163 | * Note: we explicitly set the cookie of the grant reply to that of | |
164 | * the blocked lock request. The spec explicitly mentions that the client | |
165 | * should _not_ rely on the callback containing the same cookie as the | |
166 | * request, but (as I found out later) that's because some implementations | |
167 | * do just this. Never mind the standards comittees, they support our | |
168 | * logging industries. | |
39be4502 OK |
169 | * |
170 | * 10 years later: I hope we can safely ignore these old and broken | |
171 | * clients by now. Let's fix this so we can uniquely identify an incoming | |
172 | * GRANTED_RES message by cookie, without having to rely on the client's IP | |
173 | * address. --okir | |
1da177e4 | 174 | */ |
255129d1 TM |
175 | static struct nlm_block * |
176 | nlmsvc_create_block(struct svc_rqst *rqstp, struct nlm_host *host, | |
177 | struct nlm_file *file, struct nlm_lock *lock, | |
178 | struct nlm_cookie *cookie) | |
1da177e4 LT |
179 | { |
180 | struct nlm_block *block; | |
92737230 | 181 | struct nlm_rqst *call = NULL; |
1da177e4 | 182 | |
560de0e6 | 183 | nlm_get_host(host); |
92737230 TM |
184 | call = nlm_alloc_call(host); |
185 | if (call == NULL) | |
186 | return NULL; | |
187 | ||
1da177e4 | 188 | /* Allocate memory for block, and initialize arguments */ |
92737230 TM |
189 | block = kzalloc(sizeof(*block), GFP_KERNEL); |
190 | if (block == NULL) | |
1da177e4 | 191 | goto failed; |
6849c0ca | 192 | kref_init(&block->b_count); |
68a2d76c OK |
193 | INIT_LIST_HEAD(&block->b_list); |
194 | INIT_LIST_HEAD(&block->b_flist); | |
1da177e4 | 195 | |
92737230 | 196 | if (!nlmsvc_setgrantargs(call, lock)) |
1da177e4 LT |
197 | goto failed_free; |
198 | ||
199 | /* Set notifier function for VFS, and init args */ | |
92737230 TM |
200 | call->a_args.lock.fl.fl_flags |= FL_SLEEP; |
201 | call->a_args.lock.fl.fl_lmops = &nlmsvc_lock_operations; | |
39be4502 | 202 | nlmclnt_next_cookie(&call->a_args.cookie); |
1da177e4 LT |
203 | |
204 | dprintk("lockd: created block %p...\n", block); | |
205 | ||
206 | /* Create and initialize the block */ | |
207 | block->b_daemon = rqstp->rq_server; | |
208 | block->b_host = host; | |
209 | block->b_file = file; | |
5ea0d750 | 210 | block->b_fl = NULL; |
d9f6eb75 | 211 | file->f_count++; |
1da177e4 LT |
212 | |
213 | /* Add to file's list of blocks */ | |
68a2d76c | 214 | list_add(&block->b_flist, &file->f_blocks); |
1da177e4 LT |
215 | |
216 | /* Set up RPC arguments for callback */ | |
92737230 | 217 | block->b_call = call; |
1da177e4 | 218 | call->a_flags = RPC_TASK_ASYNC; |
92737230 | 219 | call->a_block = block; |
1da177e4 LT |
220 | |
221 | return block; | |
222 | ||
223 | failed_free: | |
224 | kfree(block); | |
225 | failed: | |
92737230 | 226 | nlm_release_call(call); |
1da177e4 LT |
227 | return NULL; |
228 | } | |
229 | ||
230 | /* | |
3c61eecb | 231 | * Delete a block. |
1da177e4 LT |
232 | * It is the caller's responsibility to check whether the file |
233 | * can be closed hereafter. | |
234 | */ | |
6849c0ca | 235 | static int nlmsvc_unlink_block(struct nlm_block *block) |
1da177e4 | 236 | { |
09c7938c | 237 | int status; |
6849c0ca | 238 | dprintk("lockd: unlinking block %p...\n", block); |
1da177e4 LT |
239 | |
240 | /* Remove block from list */ | |
92737230 | 241 | status = posix_unblock_lock(block->b_file->f_file, &block->b_call->a_args.lock.fl); |
1da177e4 | 242 | nlmsvc_remove_block(block); |
6849c0ca TM |
243 | return status; |
244 | } | |
1da177e4 | 245 | |
6849c0ca TM |
246 | static void nlmsvc_free_block(struct kref *kref) |
247 | { | |
248 | struct nlm_block *block = container_of(kref, struct nlm_block, b_count); | |
249 | struct nlm_file *file = block->b_file; | |
6849c0ca TM |
250 | |
251 | dprintk("lockd: freeing block %p...\n", block); | |
1da177e4 LT |
252 | |
253 | /* Remove block from file's list of blocks */ | |
89e63ef6 | 254 | mutex_lock(&file->f_mutex); |
68a2d76c | 255 | list_del_init(&block->b_flist); |
89e63ef6 | 256 | mutex_unlock(&file->f_mutex); |
1da177e4 | 257 | |
92737230 TM |
258 | nlmsvc_freegrantargs(block->b_call); |
259 | nlm_release_call(block->b_call); | |
d9f6eb75 | 260 | nlm_release_file(block->b_file); |
0e4ac9d9 | 261 | kfree(block->b_fl); |
1da177e4 | 262 | kfree(block); |
6849c0ca TM |
263 | } |
264 | ||
265 | static void nlmsvc_release_block(struct nlm_block *block) | |
266 | { | |
267 | if (block != NULL) | |
268 | kref_put(&block->b_count, nlmsvc_free_block); | |
1da177e4 LT |
269 | } |
270 | ||
f2af793d OK |
271 | /* |
272 | * Loop over all blocks and delete blocks held by | |
273 | * a matching host. | |
274 | */ | |
275 | void nlmsvc_traverse_blocks(struct nlm_host *host, | |
276 | struct nlm_file *file, | |
277 | nlm_host_match_fn_t match) | |
d9f6eb75 | 278 | { |
68a2d76c | 279 | struct nlm_block *block, *next; |
d9f6eb75 TM |
280 | |
281 | restart: | |
89e63ef6 | 282 | mutex_lock(&file->f_mutex); |
68a2d76c | 283 | list_for_each_entry_safe(block, next, &file->f_blocks, b_flist) { |
f2af793d | 284 | if (!match(block->b_host, host)) |
d9f6eb75 | 285 | continue; |
68a2d76c OK |
286 | /* Do not destroy blocks that are not on |
287 | * the global retry list - why? */ | |
288 | if (list_empty(&block->b_list)) | |
d9f6eb75 TM |
289 | continue; |
290 | kref_get(&block->b_count); | |
89e63ef6 | 291 | mutex_unlock(&file->f_mutex); |
d9f6eb75 TM |
292 | nlmsvc_unlink_block(block); |
293 | nlmsvc_release_block(block); | |
294 | goto restart; | |
295 | } | |
89e63ef6 | 296 | mutex_unlock(&file->f_mutex); |
d9f6eb75 TM |
297 | } |
298 | ||
5e1abf8c TM |
299 | /* |
300 | * Initialize arguments for GRANTED call. The nlm_rqst structure | |
301 | * has been cleared already. | |
302 | */ | |
303 | static int nlmsvc_setgrantargs(struct nlm_rqst *call, struct nlm_lock *lock) | |
304 | { | |
305 | locks_copy_lock(&call->a_args.lock.fl, &lock->fl); | |
306 | memcpy(&call->a_args.lock.fh, &lock->fh, sizeof(call->a_args.lock.fh)); | |
e9ff3990 | 307 | call->a_args.lock.caller = utsname()->nodename; |
5e1abf8c TM |
308 | call->a_args.lock.oh.len = lock->oh.len; |
309 | ||
310 | /* set default data area */ | |
311 | call->a_args.lock.oh.data = call->a_owner; | |
312 | call->a_args.lock.svid = lock->fl.fl_pid; | |
313 | ||
314 | if (lock->oh.len > NLMCLNT_OHSIZE) { | |
315 | void *data = kmalloc(lock->oh.len, GFP_KERNEL); | |
92737230 | 316 | if (!data) |
5e1abf8c | 317 | return 0; |
5e1abf8c TM |
318 | call->a_args.lock.oh.data = (u8 *) data; |
319 | } | |
320 | ||
321 | memcpy(call->a_args.lock.oh.data, lock->oh.data, lock->oh.len); | |
322 | return 1; | |
323 | } | |
324 | ||
325 | static void nlmsvc_freegrantargs(struct nlm_rqst *call) | |
326 | { | |
92737230 | 327 | if (call->a_args.lock.oh.data != call->a_owner) |
5e1abf8c | 328 | kfree(call->a_args.lock.oh.data); |
a9e61e25 FB |
329 | |
330 | locks_release_private(&call->a_args.lock.fl); | |
5e1abf8c TM |
331 | } |
332 | ||
2b36f412 ME |
333 | /* |
334 | * Deferred lock request handling for non-blocking lock | |
335 | */ | |
ca5c8cde | 336 | static __be32 |
2b36f412 ME |
337 | nlmsvc_defer_lock_rqst(struct svc_rqst *rqstp, struct nlm_block *block) |
338 | { | |
ca5c8cde | 339 | __be32 status = nlm_lck_denied_nolocks; |
2b36f412 ME |
340 | |
341 | block->b_flags |= B_QUEUED; | |
342 | ||
343 | nlmsvc_insert_block(block, NLM_TIMEOUT); | |
344 | ||
345 | block->b_cache_req = &rqstp->rq_chandle; | |
346 | if (rqstp->rq_chandle.defer) { | |
347 | block->b_deferred_req = | |
348 | rqstp->rq_chandle.defer(block->b_cache_req); | |
349 | if (block->b_deferred_req != NULL) | |
350 | status = nlm_drop_reply; | |
351 | } | |
352 | dprintk("lockd: nlmsvc_defer_lock_rqst block %p flags %d status %d\n", | |
ca5c8cde | 353 | block, block->b_flags, ntohl(status)); |
2b36f412 ME |
354 | |
355 | return status; | |
356 | } | |
357 | ||
1da177e4 LT |
358 | /* |
359 | * Attempt to establish a lock, and if it can't be granted, block it | |
360 | * if required. | |
361 | */ | |
52921e02 | 362 | __be32 |
1da177e4 | 363 | nlmsvc_lock(struct svc_rqst *rqstp, struct nlm_file *file, |
6cde4de8 | 364 | struct nlm_host *host, struct nlm_lock *lock, int wait, |
b2b50289 | 365 | struct nlm_cookie *cookie, int reclaim) |
1da177e4 | 366 | { |
f8120480 | 367 | struct nlm_block *block = NULL; |
1da177e4 | 368 | int error; |
52921e02 | 369 | __be32 ret; |
1da177e4 LT |
370 | |
371 | dprintk("lockd: nlmsvc_lock(%s/%ld, ty=%d, pi=%d, %Ld-%Ld, bl=%d)\n", | |
225a719f JS |
372 | file->f_file->f_path.dentry->d_inode->i_sb->s_id, |
373 | file->f_file->f_path.dentry->d_inode->i_ino, | |
1da177e4 LT |
374 | lock->fl.fl_type, lock->fl.fl_pid, |
375 | (long long)lock->fl.fl_start, | |
376 | (long long)lock->fl.fl_end, | |
377 | wait); | |
378 | ||
1da177e4 | 379 | /* Lock file against concurrent access */ |
89e63ef6 | 380 | mutex_lock(&file->f_mutex); |
f8120480 ME |
381 | /* Get existing block (in case client is busy-waiting) |
382 | * or create new block | |
383 | */ | |
d9f6eb75 | 384 | block = nlmsvc_lookup_block(file, lock); |
09c7938c | 385 | if (block == NULL) { |
560de0e6 | 386 | block = nlmsvc_create_block(rqstp, host, file, lock, cookie); |
f8120480 ME |
387 | ret = nlm_lck_denied_nolocks; |
388 | if (block == NULL) | |
389 | goto out; | |
92737230 | 390 | lock = &block->b_call->a_args.lock; |
f8120480 ME |
391 | } else |
392 | lock->fl.fl_flags &= ~FL_SLEEP; | |
1da177e4 | 393 | |
1a8322b2 ME |
394 | if (block->b_flags & B_QUEUED) { |
395 | dprintk("lockd: nlmsvc_lock deferred block %p flags %d\n", | |
396 | block, block->b_flags); | |
397 | if (block->b_granted) { | |
398 | nlmsvc_unlink_block(block); | |
399 | ret = nlm_granted; | |
400 | goto out; | |
401 | } | |
402 | if (block->b_flags & B_TIMED_OUT) { | |
403 | nlmsvc_unlink_block(block); | |
404 | ret = nlm_lck_denied; | |
405 | goto out; | |
406 | } | |
407 | ret = nlm_drop_reply; | |
408 | goto out; | |
409 | } | |
1da177e4 | 410 | |
b2b50289 BF |
411 | if (locks_in_grace() && !reclaim) { |
412 | ret = nlm_lck_denied_grace_period; | |
413 | goto out; | |
414 | } | |
d22b1cff BF |
415 | if (reclaim && !locks_in_grace()) { |
416 | ret = nlm_lck_denied_grace_period; | |
417 | goto out; | |
418 | } | |
b2b50289 | 419 | |
1a8322b2 ME |
420 | if (!wait) |
421 | lock->fl.fl_flags &= ~FL_SLEEP; | |
422 | error = vfs_lock_file(file->f_file, F_SETLK, &lock->fl, NULL); | |
423 | lock->fl.fl_flags &= ~FL_SLEEP; | |
a85f193e | 424 | |
1a8322b2 | 425 | dprintk("lockd: vfs_lock_file returned %d\n", error); |
6d7bbbba | 426 | switch (error) { |
1da177e4 | 427 | case 0: |
15dadef9 AA |
428 | ret = nlm_granted; |
429 | goto out; | |
09c7938c | 430 | case -EAGAIN: |
e33d1ea6 MS |
431 | /* |
432 | * If this is a blocking request for an | |
433 | * already pending lock request then we need | |
434 | * to put it back on lockd's block list | |
435 | */ | |
436 | if (wait) | |
437 | break; | |
1a8322b2 | 438 | ret = nlm_lck_denied; |
e33d1ea6 | 439 | goto out; |
bde74e4b | 440 | case FILE_LOCK_DEFERRED: |
1a8322b2 ME |
441 | if (wait) |
442 | break; | |
443 | /* Filesystem lock operation is in progress | |
444 | Add it to the queue waiting for callback */ | |
445 | ret = nlmsvc_defer_lock_rqst(rqstp, block); | |
446 | goto out; | |
09c7938c | 447 | case -EDEADLK: |
15dadef9 AA |
448 | ret = nlm_deadlock; |
449 | goto out; | |
1da177e4 | 450 | default: /* includes ENOLCK */ |
15dadef9 AA |
451 | ret = nlm_lck_denied_nolocks; |
452 | goto out; | |
1da177e4 LT |
453 | } |
454 | ||
09c7938c | 455 | ret = nlm_lck_blocked; |
1da177e4 LT |
456 | |
457 | /* Append to list of blocked */ | |
f8120480 | 458 | nlmsvc_insert_block(block, NLM_NEVER); |
15dadef9 | 459 | out: |
89e63ef6 | 460 | mutex_unlock(&file->f_mutex); |
6849c0ca | 461 | nlmsvc_release_block(block); |
15dadef9 AA |
462 | dprintk("lockd: nlmsvc_lock returned %u\n", ret); |
463 | return ret; | |
1da177e4 LT |
464 | } |
465 | ||
466 | /* | |
467 | * Test for presence of a conflicting lock. | |
468 | */ | |
52921e02 | 469 | __be32 |
85f3f1b3 | 470 | nlmsvc_testlock(struct svc_rqst *rqstp, struct nlm_file *file, |
8f920d5e JL |
471 | struct nlm_host *host, struct nlm_lock *lock, |
472 | struct nlm_lock *conflock, struct nlm_cookie *cookie) | |
1da177e4 | 473 | { |
5ea0d750 ME |
474 | struct nlm_block *block = NULL; |
475 | int error; | |
476 | __be32 ret; | |
477 | ||
1da177e4 | 478 | dprintk("lockd: nlmsvc_testlock(%s/%ld, ty=%d, %Ld-%Ld)\n", |
225a719f JS |
479 | file->f_file->f_path.dentry->d_inode->i_sb->s_id, |
480 | file->f_file->f_path.dentry->d_inode->i_ino, | |
1da177e4 LT |
481 | lock->fl.fl_type, |
482 | (long long)lock->fl.fl_start, | |
483 | (long long)lock->fl.fl_end); | |
484 | ||
5ea0d750 ME |
485 | /* Get existing block (in case client is busy-waiting) */ |
486 | block = nlmsvc_lookup_block(file, lock); | |
487 | ||
488 | if (block == NULL) { | |
489 | struct file_lock *conf = kzalloc(sizeof(*conf), GFP_KERNEL); | |
490 | ||
491 | if (conf == NULL) | |
492 | return nlm_granted; | |
255129d1 | 493 | block = nlmsvc_create_block(rqstp, host, file, lock, cookie); |
5ea0d750 ME |
494 | if (block == NULL) { |
495 | kfree(conf); | |
496 | return nlm_granted; | |
497 | } | |
498 | block->b_fl = conf; | |
499 | } | |
500 | if (block->b_flags & B_QUEUED) { | |
501 | dprintk("lockd: nlmsvc_testlock deferred block %p flags %d fl %p\n", | |
502 | block, block->b_flags, block->b_fl); | |
503 | if (block->b_flags & B_TIMED_OUT) { | |
504 | nlmsvc_unlink_block(block); | |
29dbf546 OD |
505 | ret = nlm_lck_denied; |
506 | goto out; | |
5ea0d750 ME |
507 | } |
508 | if (block->b_flags & B_GOT_CALLBACK) { | |
54ca95eb | 509 | nlmsvc_unlink_block(block); |
5ea0d750 ME |
510 | if (block->b_fl != NULL |
511 | && block->b_fl->fl_type != F_UNLCK) { | |
512 | lock->fl = *block->b_fl; | |
513 | goto conf_lock; | |
29dbf546 | 514 | } else { |
29dbf546 OD |
515 | ret = nlm_granted; |
516 | goto out; | |
5ea0d750 ME |
517 | } |
518 | } | |
29dbf546 OD |
519 | ret = nlm_drop_reply; |
520 | goto out; | |
1da177e4 LT |
521 | } |
522 | ||
b2b50289 BF |
523 | if (locks_in_grace()) { |
524 | ret = nlm_lck_denied_grace_period; | |
525 | goto out; | |
526 | } | |
5ea0d750 | 527 | error = vfs_test_lock(file->f_file, &lock->fl); |
bde74e4b | 528 | if (error == FILE_LOCK_DEFERRED) { |
29dbf546 OD |
529 | ret = nlmsvc_defer_lock_rqst(rqstp, block); |
530 | goto out; | |
531 | } | |
5ea0d750 ME |
532 | if (error) { |
533 | ret = nlm_lck_denied_nolocks; | |
534 | goto out; | |
535 | } | |
536 | if (lock->fl.fl_type == F_UNLCK) { | |
537 | ret = nlm_granted; | |
538 | goto out; | |
539 | } | |
540 | ||
541 | conf_lock: | |
542 | dprintk("lockd: conflicting lock(ty=%d, %Ld-%Ld)\n", | |
543 | lock->fl.fl_type, (long long)lock->fl.fl_start, | |
544 | (long long)lock->fl.fl_end); | |
545 | conflock->caller = "somehost"; /* FIXME */ | |
546 | conflock->len = strlen(conflock->caller); | |
547 | conflock->oh.len = 0; /* don't return OH info */ | |
548 | conflock->svid = lock->fl.fl_pid; | |
549 | conflock->fl.fl_type = lock->fl.fl_type; | |
550 | conflock->fl.fl_start = lock->fl.fl_start; | |
551 | conflock->fl.fl_end = lock->fl.fl_end; | |
552 | ret = nlm_lck_denied; | |
553 | out: | |
554 | if (block) | |
555 | nlmsvc_release_block(block); | |
556 | return ret; | |
1da177e4 LT |
557 | } |
558 | ||
559 | /* | |
560 | * Remove a lock. | |
561 | * This implies a CANCEL call: We send a GRANT_MSG, the client replies | |
562 | * with a GRANT_RES call which gets lost, and calls UNLOCK immediately | |
563 | * afterwards. In this case the block will still be there, and hence | |
564 | * must be removed. | |
565 | */ | |
52921e02 | 566 | __be32 |
1da177e4 LT |
567 | nlmsvc_unlock(struct nlm_file *file, struct nlm_lock *lock) |
568 | { | |
569 | int error; | |
570 | ||
571 | dprintk("lockd: nlmsvc_unlock(%s/%ld, pi=%d, %Ld-%Ld)\n", | |
225a719f JS |
572 | file->f_file->f_path.dentry->d_inode->i_sb->s_id, |
573 | file->f_file->f_path.dentry->d_inode->i_ino, | |
1da177e4 LT |
574 | lock->fl.fl_pid, |
575 | (long long)lock->fl.fl_start, | |
576 | (long long)lock->fl.fl_end); | |
577 | ||
578 | /* First, cancel any lock that might be there */ | |
579 | nlmsvc_cancel_blocked(file, lock); | |
580 | ||
581 | lock->fl.fl_type = F_UNLCK; | |
1a8322b2 | 582 | error = vfs_lock_file(file->f_file, F_SETLK, &lock->fl, NULL); |
1da177e4 LT |
583 | |
584 | return (error < 0)? nlm_lck_denied_nolocks : nlm_granted; | |
585 | } | |
586 | ||
587 | /* | |
588 | * Cancel a previously blocked request. | |
589 | * | |
590 | * A cancel request always overrides any grant that may currently | |
591 | * be in progress. | |
592 | * The calling procedure must check whether the file can be closed. | |
593 | */ | |
52921e02 | 594 | __be32 |
1da177e4 LT |
595 | nlmsvc_cancel_blocked(struct nlm_file *file, struct nlm_lock *lock) |
596 | { | |
597 | struct nlm_block *block; | |
64a318ee | 598 | int status = 0; |
1da177e4 LT |
599 | |
600 | dprintk("lockd: nlmsvc_cancel(%s/%ld, pi=%d, %Ld-%Ld)\n", | |
225a719f JS |
601 | file->f_file->f_path.dentry->d_inode->i_sb->s_id, |
602 | file->f_file->f_path.dentry->d_inode->i_ino, | |
1da177e4 LT |
603 | lock->fl.fl_pid, |
604 | (long long)lock->fl.fl_start, | |
605 | (long long)lock->fl.fl_end); | |
606 | ||
b2b50289 BF |
607 | if (locks_in_grace()) |
608 | return nlm_lck_denied_grace_period; | |
609 | ||
89e63ef6 | 610 | mutex_lock(&file->f_mutex); |
d9f6eb75 | 611 | block = nlmsvc_lookup_block(file, lock); |
89e63ef6 | 612 | mutex_unlock(&file->f_mutex); |
d9f6eb75 | 613 | if (block != NULL) { |
1a8322b2 ME |
614 | vfs_cancel_lock(block->b_file->f_file, |
615 | &block->b_call->a_args.lock.fl); | |
6849c0ca TM |
616 | status = nlmsvc_unlink_block(block); |
617 | nlmsvc_release_block(block); | |
618 | } | |
64a318ee | 619 | return status ? nlm_lck_denied : nlm_granted; |
1da177e4 LT |
620 | } |
621 | ||
0e4ac9d9 ME |
622 | /* |
623 | * This is a callback from the filesystem for VFS file lock requests. | |
624 | * It will be used if fl_grant is defined and the filesystem can not | |
625 | * respond to the request immediately. | |
626 | * For GETLK request it will copy the reply to the nlm_block. | |
627 | * For SETLK or SETLKW request it will get the local posix lock. | |
628 | * In all cases it will move the block to the head of nlm_blocked q where | |
629 | * nlmsvc_retry_blocked() can send back a reply for SETLKW or revisit the | |
630 | * deferred rpc for GETLK and SETLK. | |
631 | */ | |
632 | static void | |
633 | nlmsvc_update_deferred_block(struct nlm_block *block, struct file_lock *conf, | |
634 | int result) | |
635 | { | |
636 | block->b_flags |= B_GOT_CALLBACK; | |
637 | if (result == 0) | |
638 | block->b_granted = 1; | |
639 | else | |
640 | block->b_flags |= B_TIMED_OUT; | |
641 | if (conf) { | |
0e4ac9d9 | 642 | if (block->b_fl) |
1a747ee0 | 643 | __locks_copy_lock(block->b_fl, conf); |
0e4ac9d9 ME |
644 | } |
645 | } | |
646 | ||
647 | static int nlmsvc_grant_deferred(struct file_lock *fl, struct file_lock *conf, | |
648 | int result) | |
649 | { | |
650 | struct nlm_block *block; | |
651 | int rc = -ENOENT; | |
652 | ||
653 | lock_kernel(); | |
654 | list_for_each_entry(block, &nlm_blocked, b_list) { | |
655 | if (nlm_compare_locks(&block->b_call->a_args.lock.fl, fl)) { | |
656 | dprintk("lockd: nlmsvc_notify_blocked block %p flags %d\n", | |
657 | block, block->b_flags); | |
658 | if (block->b_flags & B_QUEUED) { | |
659 | if (block->b_flags & B_TIMED_OUT) { | |
660 | rc = -ENOLCK; | |
661 | break; | |
662 | } | |
663 | nlmsvc_update_deferred_block(block, conf, result); | |
664 | } else if (result == 0) | |
665 | block->b_granted = 1; | |
666 | ||
667 | nlmsvc_insert_block(block, 0); | |
668 | svc_wake_up(block->b_daemon); | |
669 | rc = 0; | |
670 | break; | |
671 | } | |
672 | } | |
673 | unlock_kernel(); | |
674 | if (rc == -ENOENT) | |
675 | printk(KERN_WARNING "lockd: grant for unknown block\n"); | |
676 | return rc; | |
677 | } | |
678 | ||
1da177e4 LT |
679 | /* |
680 | * Unblock a blocked lock request. This is a callback invoked from the | |
681 | * VFS layer when a lock on which we blocked is removed. | |
682 | * | |
683 | * This function doesn't grant the blocked lock instantly, but rather moves | |
684 | * the block to the head of nlm_blocked where it can be picked up by lockd. | |
685 | */ | |
686 | static void | |
687 | nlmsvc_notify_blocked(struct file_lock *fl) | |
688 | { | |
68a2d76c | 689 | struct nlm_block *block; |
1da177e4 LT |
690 | |
691 | dprintk("lockd: VFS unblock notification for block %p\n", fl); | |
68a2d76c | 692 | list_for_each_entry(block, &nlm_blocked, b_list) { |
92737230 | 693 | if (nlm_compare_locks(&block->b_call->a_args.lock.fl, fl)) { |
1da177e4 LT |
694 | nlmsvc_insert_block(block, 0); |
695 | svc_wake_up(block->b_daemon); | |
696 | return; | |
697 | } | |
698 | } | |
699 | ||
700 | printk(KERN_WARNING "lockd: notification for unknown block!\n"); | |
701 | } | |
702 | ||
703 | static int nlmsvc_same_owner(struct file_lock *fl1, struct file_lock *fl2) | |
704 | { | |
705 | return fl1->fl_owner == fl2->fl_owner && fl1->fl_pid == fl2->fl_pid; | |
706 | } | |
707 | ||
7b021967 | 708 | const struct lock_manager_operations nlmsvc_lock_operations = { |
1da177e4 LT |
709 | .fl_compare_owner = nlmsvc_same_owner, |
710 | .fl_notify = nlmsvc_notify_blocked, | |
0e4ac9d9 | 711 | .fl_grant = nlmsvc_grant_deferred, |
1da177e4 LT |
712 | }; |
713 | ||
714 | /* | |
715 | * Try to claim a lock that was previously blocked. | |
716 | * | |
717 | * Note that we use both the RPC_GRANTED_MSG call _and_ an async | |
718 | * RPC thread when notifying the client. This seems like overkill... | |
719 | * Here's why: | |
720 | * - we don't want to use a synchronous RPC thread, otherwise | |
721 | * we might find ourselves hanging on a dead portmapper. | |
722 | * - Some lockd implementations (e.g. HP) don't react to | |
723 | * RPC_GRANTED calls; they seem to insist on RPC_GRANTED_MSG calls. | |
724 | */ | |
725 | static void | |
726 | nlmsvc_grant_blocked(struct nlm_block *block) | |
727 | { | |
728 | struct nlm_file *file = block->b_file; | |
92737230 | 729 | struct nlm_lock *lock = &block->b_call->a_args.lock; |
1da177e4 LT |
730 | int error; |
731 | ||
732 | dprintk("lockd: grant blocked lock %p\n", block); | |
733 | ||
0e4ac9d9 ME |
734 | kref_get(&block->b_count); |
735 | ||
1da177e4 | 736 | /* Unlink block request from list */ |
6849c0ca | 737 | nlmsvc_unlink_block(block); |
1da177e4 LT |
738 | |
739 | /* If b_granted is true this means we've been here before. | |
740 | * Just retry the grant callback, possibly refreshing the RPC | |
741 | * binding */ | |
742 | if (block->b_granted) { | |
743 | nlm_rebind_host(block->b_host); | |
744 | goto callback; | |
745 | } | |
746 | ||
747 | /* Try the lock operation again */ | |
09c7938c | 748 | lock->fl.fl_flags |= FL_SLEEP; |
1a8322b2 | 749 | error = vfs_lock_file(file->f_file, F_SETLK, &lock->fl, NULL); |
09c7938c TM |
750 | lock->fl.fl_flags &= ~FL_SLEEP; |
751 | ||
5de0e502 AA |
752 | switch (error) { |
753 | case 0: | |
754 | break; | |
bde74e4b | 755 | case FILE_LOCK_DEFERRED: |
1a8322b2 | 756 | dprintk("lockd: lock still blocked error %d\n", error); |
1da177e4 | 757 | nlmsvc_insert_block(block, NLM_NEVER); |
0e4ac9d9 | 758 | nlmsvc_release_block(block); |
d9f6eb75 | 759 | return; |
5de0e502 | 760 | default: |
1da177e4 | 761 | printk(KERN_WARNING "lockd: unexpected error %d in %s!\n", |
8e24eea7 | 762 | -error, __func__); |
1da177e4 | 763 | nlmsvc_insert_block(block, 10 * HZ); |
0e4ac9d9 | 764 | nlmsvc_release_block(block); |
d9f6eb75 | 765 | return; |
1da177e4 LT |
766 | } |
767 | ||
768 | callback: | |
769 | /* Lock was granted by VFS. */ | |
770 | dprintk("lockd: GRANTing blocked lock.\n"); | |
771 | block->b_granted = 1; | |
1da177e4 | 772 | |
9706501e JL |
773 | /* keep block on the list, but don't reattempt until the RPC |
774 | * completes or the submission fails | |
775 | */ | |
776 | nlmsvc_insert_block(block, NLM_NEVER); | |
1da177e4 | 777 | |
9706501e JL |
778 | /* Call the client -- use a soft RPC task since nlmsvc_retry_blocked |
779 | * will queue up a new one if this one times out | |
780 | */ | |
781 | error = nlm_async_call(block->b_call, NLMPROC_GRANTED_MSG, | |
782 | &nlmsvc_grant_ops); | |
783 | ||
784 | /* RPC submission failed, wait a bit and retry */ | |
785 | if (error < 0) | |
786 | nlmsvc_insert_block(block, 10 * HZ); | |
1da177e4 LT |
787 | } |
788 | ||
789 | /* | |
790 | * This is the callback from the RPC layer when the NLM_GRANTED_MSG | |
791 | * RPC call has succeeded or timed out. | |
792 | * Like all RPC callbacks, it is invoked by the rpciod process, so it | |
793 | * better not sleep. Therefore, we put the blocked lock on the nlm_blocked | |
794 | * chain once more in order to have it removed by lockd itself (which can | |
795 | * then sleep on the file semaphore without disrupting e.g. the nfs client). | |
796 | */ | |
963d8fe5 | 797 | static void nlmsvc_grant_callback(struct rpc_task *task, void *data) |
1da177e4 | 798 | { |
963d8fe5 | 799 | struct nlm_rqst *call = data; |
92737230 | 800 | struct nlm_block *block = call->a_block; |
1da177e4 | 801 | unsigned long timeout; |
1da177e4 LT |
802 | |
803 | dprintk("lockd: GRANT_MSG RPC callback\n"); | |
1da177e4 | 804 | |
a86dc496 | 805 | lock_kernel(); |
c64e80d5 JL |
806 | /* if the block is not on a list at this point then it has |
807 | * been invalidated. Don't try to requeue it. | |
808 | * | |
809 | * FIXME: it's possible that the block is removed from the list | |
810 | * after this check but before the nlmsvc_insert_block. In that | |
811 | * case it will be added back. Perhaps we need better locking | |
812 | * for nlm_blocked? | |
813 | */ | |
814 | if (list_empty(&block->b_list)) | |
a86dc496 | 815 | goto out; |
c64e80d5 | 816 | |
1da177e4 LT |
817 | /* Technically, we should down the file semaphore here. Since we |
818 | * move the block towards the head of the queue only, no harm | |
819 | * can be done, though. */ | |
820 | if (task->tk_status < 0) { | |
821 | /* RPC error: Re-insert for retransmission */ | |
822 | timeout = 10 * HZ; | |
1da177e4 LT |
823 | } else { |
824 | /* Call was successful, now wait for client callback */ | |
825 | timeout = 60 * HZ; | |
826 | } | |
827 | nlmsvc_insert_block(block, timeout); | |
828 | svc_wake_up(block->b_daemon); | |
a86dc496 TM |
829 | out: |
830 | unlock_kernel(); | |
5e1abf8c TM |
831 | } |
832 | ||
ec535ce1 | 833 | static void nlmsvc_grant_release(void *data) |
5e1abf8c | 834 | { |
6041b791 TM |
835 | struct nlm_rqst *call = data; |
836 | ||
a86dc496 | 837 | lock_kernel(); |
6041b791 | 838 | nlmsvc_release_block(call->a_block); |
a86dc496 | 839 | unlock_kernel(); |
1da177e4 LT |
840 | } |
841 | ||
963d8fe5 TM |
842 | static const struct rpc_call_ops nlmsvc_grant_ops = { |
843 | .rpc_call_done = nlmsvc_grant_callback, | |
5e1abf8c | 844 | .rpc_release = nlmsvc_grant_release, |
963d8fe5 TM |
845 | }; |
846 | ||
1da177e4 LT |
847 | /* |
848 | * We received a GRANT_RES callback. Try to find the corresponding | |
849 | * block. | |
850 | */ | |
851 | void | |
e8c5c045 | 852 | nlmsvc_grant_reply(struct nlm_cookie *cookie, __be32 status) |
1da177e4 LT |
853 | { |
854 | struct nlm_block *block; | |
1da177e4 | 855 | |
39be4502 OK |
856 | dprintk("grant_reply: looking for cookie %x, s=%d \n", |
857 | *(unsigned int *)(cookie->data), status); | |
858 | if (!(block = nlmsvc_find_block(cookie))) | |
1da177e4 | 859 | return; |
1da177e4 | 860 | |
f232142c | 861 | if (block) { |
e8c5c045 | 862 | if (status == nlm_lck_denied_grace_period) { |
1da177e4 LT |
863 | /* Try again in a couple of seconds */ |
864 | nlmsvc_insert_block(block, 10 * HZ); | |
1da177e4 LT |
865 | } else { |
866 | /* Lock is now held by client, or has been rejected. | |
867 | * In both cases, the block should be removed. */ | |
6849c0ca | 868 | nlmsvc_unlink_block(block); |
1da177e4 LT |
869 | } |
870 | } | |
6849c0ca | 871 | nlmsvc_release_block(block); |
1da177e4 LT |
872 | } |
873 | ||
0e4ac9d9 ME |
874 | /* Helper function to handle retry of a deferred block. |
875 | * If it is a blocking lock, call grant_blocked. | |
876 | * For a non-blocking lock or test lock, revisit the request. | |
877 | */ | |
878 | static void | |
879 | retry_deferred_block(struct nlm_block *block) | |
880 | { | |
881 | if (!(block->b_flags & B_GOT_CALLBACK)) | |
882 | block->b_flags |= B_TIMED_OUT; | |
883 | nlmsvc_insert_block(block, NLM_TIMEOUT); | |
884 | dprintk("revisit block %p flags %d\n", block, block->b_flags); | |
885 | if (block->b_deferred_req) { | |
886 | block->b_deferred_req->revisit(block->b_deferred_req, 0); | |
887 | block->b_deferred_req = NULL; | |
888 | } | |
889 | } | |
890 | ||
1da177e4 LT |
891 | /* |
892 | * Retry all blocked locks that have been notified. This is where lockd | |
893 | * picks up locks that can be granted, or grant notifications that must | |
894 | * be retransmitted. | |
895 | */ | |
896 | unsigned long | |
897 | nlmsvc_retry_blocked(void) | |
898 | { | |
68a2d76c OK |
899 | unsigned long timeout = MAX_SCHEDULE_TIMEOUT; |
900 | struct nlm_block *block; | |
901 | ||
d751a7cd | 902 | while (!list_empty(&nlm_blocked) && !kthread_should_stop()) { |
68a2d76c | 903 | block = list_entry(nlm_blocked.next, struct nlm_block, b_list); |
1da177e4 | 904 | |
1da177e4 LT |
905 | if (block->b_when == NLM_NEVER) |
906 | break; | |
6d7bbbba | 907 | if (time_after(block->b_when, jiffies)) { |
68a2d76c | 908 | timeout = block->b_when - jiffies; |
1da177e4 | 909 | break; |
68a2d76c OK |
910 | } |
911 | ||
f3d43c76 BF |
912 | dprintk("nlmsvc_retry_blocked(%p, when=%ld)\n", |
913 | block, block->b_when); | |
0e4ac9d9 ME |
914 | if (block->b_flags & B_QUEUED) { |
915 | dprintk("nlmsvc_retry_blocked delete block (%p, granted=%d, flags=%d)\n", | |
916 | block, block->b_granted, block->b_flags); | |
917 | retry_deferred_block(block); | |
918 | } else | |
919 | nlmsvc_grant_blocked(block); | |
1da177e4 LT |
920 | } |
921 | ||
68a2d76c | 922 | return timeout; |
1da177e4 | 923 | } |