]>
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> | |
32 | ||
33 | #define NLMDBG_FACILITY NLMDBG_SVCLOCK | |
34 | ||
35 | #ifdef CONFIG_LOCKD_V4 | |
36 | #define nlm_deadlock nlm4_deadlock | |
37 | #else | |
38 | #define nlm_deadlock nlm_lck_denied | |
39 | #endif | |
40 | ||
6849c0ca | 41 | static void nlmsvc_release_block(struct nlm_block *block); |
1da177e4 | 42 | static void nlmsvc_insert_block(struct nlm_block *block, unsigned long); |
68a2d76c | 43 | static void nlmsvc_remove_block(struct nlm_block *block); |
963d8fe5 | 44 | |
5e1abf8c TM |
45 | static int nlmsvc_setgrantargs(struct nlm_rqst *call, struct nlm_lock *lock); |
46 | static void nlmsvc_freegrantargs(struct nlm_rqst *call); | |
963d8fe5 | 47 | static const struct rpc_call_ops nlmsvc_grant_ops; |
1da177e4 LT |
48 | |
49 | /* | |
50 | * The list of blocked locks to retry | |
51 | */ | |
68a2d76c | 52 | static LIST_HEAD(nlm_blocked); |
1da177e4 LT |
53 | |
54 | /* | |
55 | * Insert a blocked lock into the global list | |
56 | */ | |
57 | static void | |
58 | nlmsvc_insert_block(struct nlm_block *block, unsigned long when) | |
59 | { | |
68a2d76c OK |
60 | struct nlm_block *b; |
61 | struct list_head *pos; | |
1da177e4 LT |
62 | |
63 | dprintk("lockd: nlmsvc_insert_block(%p, %ld)\n", block, when); | |
68a2d76c OK |
64 | if (list_empty(&block->b_list)) { |
65 | kref_get(&block->b_count); | |
66 | } else { | |
67 | list_del_init(&block->b_list); | |
68 | } | |
69 | ||
70 | pos = &nlm_blocked; | |
1da177e4 LT |
71 | if (when != NLM_NEVER) { |
72 | if ((when += jiffies) == NLM_NEVER) | |
73 | when ++; | |
68a2d76c OK |
74 | list_for_each(pos, &nlm_blocked) { |
75 | b = list_entry(pos, struct nlm_block, b_list); | |
76 | if (time_after(b->b_when,when) || b->b_when == NLM_NEVER) | |
77 | break; | |
78 | } | |
79 | /* On normal exit from the loop, pos == &nlm_blocked, | |
80 | * so we will be adding to the end of the list - good | |
81 | */ | |
82 | } | |
1da177e4 | 83 | |
68a2d76c | 84 | list_add_tail(&block->b_list, pos); |
1da177e4 | 85 | block->b_when = when; |
1da177e4 LT |
86 | } |
87 | ||
88 | /* | |
89 | * Remove a block from the global list | |
90 | */ | |
68a2d76c | 91 | static inline void |
1da177e4 LT |
92 | nlmsvc_remove_block(struct nlm_block *block) |
93 | { | |
68a2d76c OK |
94 | if (!list_empty(&block->b_list)) { |
95 | list_del_init(&block->b_list); | |
96 | nlmsvc_release_block(block); | |
1da177e4 | 97 | } |
1da177e4 LT |
98 | } |
99 | ||
100 | /* | |
d9f6eb75 | 101 | * Find a block for a given lock |
1da177e4 LT |
102 | */ |
103 | static struct nlm_block * | |
d9f6eb75 | 104 | nlmsvc_lookup_block(struct nlm_file *file, struct nlm_lock *lock) |
1da177e4 | 105 | { |
68a2d76c | 106 | struct nlm_block *block; |
1da177e4 LT |
107 | struct file_lock *fl; |
108 | ||
109 | dprintk("lockd: nlmsvc_lookup_block f=%p pd=%d %Ld-%Ld ty=%d\n", | |
110 | file, lock->fl.fl_pid, | |
111 | (long long)lock->fl.fl_start, | |
112 | (long long)lock->fl.fl_end, lock->fl.fl_type); | |
68a2d76c | 113 | list_for_each_entry(block, &nlm_blocked, b_list) { |
92737230 | 114 | fl = &block->b_call->a_args.lock.fl; |
1da177e4 LT |
115 | dprintk("lockd: check f=%p pd=%d %Ld-%Ld ty=%d cookie=%s\n", |
116 | block->b_file, fl->fl_pid, | |
117 | (long long)fl->fl_start, | |
118 | (long long)fl->fl_end, fl->fl_type, | |
92737230 | 119 | nlmdbg_cookie2a(&block->b_call->a_args.cookie)); |
1da177e4 | 120 | if (block->b_file == file && nlm_compare_locks(fl, &lock->fl)) { |
6849c0ca | 121 | kref_get(&block->b_count); |
1da177e4 LT |
122 | return block; |
123 | } | |
124 | } | |
125 | ||
126 | return NULL; | |
127 | } | |
128 | ||
129 | static inline int nlm_cookie_match(struct nlm_cookie *a, struct nlm_cookie *b) | |
130 | { | |
131 | if(a->len != b->len) | |
132 | return 0; | |
133 | if(memcmp(a->data,b->data,a->len)) | |
134 | return 0; | |
135 | return 1; | |
136 | } | |
137 | ||
138 | /* | |
139 | * Find a block with a given NLM cookie. | |
140 | */ | |
141 | static inline struct nlm_block * | |
39be4502 | 142 | nlmsvc_find_block(struct nlm_cookie *cookie) |
1da177e4 LT |
143 | { |
144 | struct nlm_block *block; | |
145 | ||
68a2d76c | 146 | list_for_each_entry(block, &nlm_blocked, b_list) { |
39be4502 | 147 | if (nlm_cookie_match(&block->b_call->a_args.cookie,cookie)) |
68a2d76c | 148 | goto found; |
1da177e4 LT |
149 | } |
150 | ||
68a2d76c OK |
151 | return NULL; |
152 | ||
153 | found: | |
39be4502 | 154 | dprintk("nlmsvc_find_block(%s): block=%p\n", nlmdbg_cookie2a(cookie), block); |
68a2d76c | 155 | kref_get(&block->b_count); |
1da177e4 LT |
156 | return block; |
157 | } | |
158 | ||
159 | /* | |
160 | * Create a block and initialize it. | |
161 | * | |
162 | * Note: we explicitly set the cookie of the grant reply to that of | |
163 | * the blocked lock request. The spec explicitly mentions that the client | |
164 | * should _not_ rely on the callback containing the same cookie as the | |
165 | * request, but (as I found out later) that's because some implementations | |
166 | * do just this. Never mind the standards comittees, they support our | |
167 | * logging industries. | |
39be4502 OK |
168 | * |
169 | * 10 years later: I hope we can safely ignore these old and broken | |
170 | * clients by now. Let's fix this so we can uniquely identify an incoming | |
171 | * GRANTED_RES message by cookie, without having to rely on the client's IP | |
172 | * address. --okir | |
1da177e4 LT |
173 | */ |
174 | static inline struct nlm_block * | |
175 | nlmsvc_create_block(struct svc_rqst *rqstp, struct nlm_file *file, | |
176 | struct nlm_lock *lock, struct nlm_cookie *cookie) | |
177 | { | |
178 | struct nlm_block *block; | |
179 | struct nlm_host *host; | |
92737230 | 180 | struct nlm_rqst *call = NULL; |
1da177e4 LT |
181 | |
182 | /* Create host handle for callback */ | |
db4e4c9a | 183 | host = nlmsvc_lookup_host(rqstp, lock->caller, lock->len); |
1da177e4 LT |
184 | if (host == NULL) |
185 | return NULL; | |
186 | ||
92737230 TM |
187 | call = nlm_alloc_call(host); |
188 | if (call == NULL) | |
189 | return NULL; | |
190 | ||
1da177e4 | 191 | /* Allocate memory for block, and initialize arguments */ |
92737230 TM |
192 | block = kzalloc(sizeof(*block), GFP_KERNEL); |
193 | if (block == NULL) | |
1da177e4 | 194 | goto failed; |
6849c0ca | 195 | kref_init(&block->b_count); |
68a2d76c OK |
196 | INIT_LIST_HEAD(&block->b_list); |
197 | INIT_LIST_HEAD(&block->b_flist); | |
1da177e4 | 198 | |
92737230 | 199 | if (!nlmsvc_setgrantargs(call, lock)) |
1da177e4 LT |
200 | goto failed_free; |
201 | ||
202 | /* Set notifier function for VFS, and init args */ | |
92737230 TM |
203 | call->a_args.lock.fl.fl_flags |= FL_SLEEP; |
204 | call->a_args.lock.fl.fl_lmops = &nlmsvc_lock_operations; | |
39be4502 | 205 | nlmclnt_next_cookie(&call->a_args.cookie); |
1da177e4 LT |
206 | |
207 | dprintk("lockd: created block %p...\n", block); | |
208 | ||
209 | /* Create and initialize the block */ | |
210 | block->b_daemon = rqstp->rq_server; | |
211 | block->b_host = host; | |
212 | block->b_file = file; | |
d9f6eb75 | 213 | file->f_count++; |
1da177e4 LT |
214 | |
215 | /* Add to file's list of blocks */ | |
68a2d76c | 216 | list_add(&block->b_flist, &file->f_blocks); |
1da177e4 LT |
217 | |
218 | /* Set up RPC arguments for callback */ | |
92737230 | 219 | block->b_call = call; |
1da177e4 | 220 | call->a_flags = RPC_TASK_ASYNC; |
92737230 | 221 | call->a_block = block; |
1da177e4 LT |
222 | |
223 | return block; | |
224 | ||
225 | failed_free: | |
226 | kfree(block); | |
227 | failed: | |
92737230 | 228 | nlm_release_call(call); |
1da177e4 LT |
229 | return NULL; |
230 | } | |
231 | ||
232 | /* | |
233 | * Delete a block. If the lock was cancelled or the grant callback | |
234 | * failed, unlock is set to 1. | |
235 | * It is the caller's responsibility to check whether the file | |
236 | * can be closed hereafter. | |
237 | */ | |
6849c0ca | 238 | static int nlmsvc_unlink_block(struct nlm_block *block) |
1da177e4 | 239 | { |
09c7938c | 240 | int status; |
6849c0ca | 241 | dprintk("lockd: unlinking block %p...\n", block); |
1da177e4 LT |
242 | |
243 | /* Remove block from list */ | |
92737230 | 244 | status = posix_unblock_lock(block->b_file->f_file, &block->b_call->a_args.lock.fl); |
1da177e4 | 245 | nlmsvc_remove_block(block); |
6849c0ca TM |
246 | return status; |
247 | } | |
1da177e4 | 248 | |
6849c0ca TM |
249 | static void nlmsvc_free_block(struct kref *kref) |
250 | { | |
251 | struct nlm_block *block = container_of(kref, struct nlm_block, b_count); | |
252 | struct nlm_file *file = block->b_file; | |
6849c0ca TM |
253 | |
254 | dprintk("lockd: freeing block %p...\n", block); | |
1da177e4 LT |
255 | |
256 | /* Remove block from file's list of blocks */ | |
89e63ef6 | 257 | mutex_lock(&file->f_mutex); |
68a2d76c | 258 | list_del_init(&block->b_flist); |
89e63ef6 | 259 | mutex_unlock(&file->f_mutex); |
1da177e4 | 260 | |
92737230 TM |
261 | nlmsvc_freegrantargs(block->b_call); |
262 | nlm_release_call(block->b_call); | |
d9f6eb75 | 263 | nlm_release_file(block->b_file); |
0e4ac9d9 | 264 | kfree(block->b_fl); |
1da177e4 | 265 | kfree(block); |
6849c0ca TM |
266 | } |
267 | ||
268 | static void nlmsvc_release_block(struct nlm_block *block) | |
269 | { | |
270 | if (block != NULL) | |
271 | kref_put(&block->b_count, nlmsvc_free_block); | |
1da177e4 LT |
272 | } |
273 | ||
f2af793d OK |
274 | /* |
275 | * Loop over all blocks and delete blocks held by | |
276 | * a matching host. | |
277 | */ | |
278 | void nlmsvc_traverse_blocks(struct nlm_host *host, | |
279 | struct nlm_file *file, | |
280 | nlm_host_match_fn_t match) | |
d9f6eb75 | 281 | { |
68a2d76c | 282 | struct nlm_block *block, *next; |
d9f6eb75 TM |
283 | |
284 | restart: | |
89e63ef6 | 285 | mutex_lock(&file->f_mutex); |
68a2d76c | 286 | list_for_each_entry_safe(block, next, &file->f_blocks, b_flist) { |
f2af793d | 287 | if (!match(block->b_host, host)) |
d9f6eb75 | 288 | continue; |
68a2d76c OK |
289 | /* Do not destroy blocks that are not on |
290 | * the global retry list - why? */ | |
291 | if (list_empty(&block->b_list)) | |
d9f6eb75 TM |
292 | continue; |
293 | kref_get(&block->b_count); | |
89e63ef6 | 294 | mutex_unlock(&file->f_mutex); |
d9f6eb75 TM |
295 | nlmsvc_unlink_block(block); |
296 | nlmsvc_release_block(block); | |
297 | goto restart; | |
298 | } | |
89e63ef6 | 299 | mutex_unlock(&file->f_mutex); |
d9f6eb75 TM |
300 | } |
301 | ||
5e1abf8c TM |
302 | /* |
303 | * Initialize arguments for GRANTED call. The nlm_rqst structure | |
304 | * has been cleared already. | |
305 | */ | |
306 | static int nlmsvc_setgrantargs(struct nlm_rqst *call, struct nlm_lock *lock) | |
307 | { | |
308 | locks_copy_lock(&call->a_args.lock.fl, &lock->fl); | |
309 | memcpy(&call->a_args.lock.fh, &lock->fh, sizeof(call->a_args.lock.fh)); | |
e9ff3990 | 310 | call->a_args.lock.caller = utsname()->nodename; |
5e1abf8c TM |
311 | call->a_args.lock.oh.len = lock->oh.len; |
312 | ||
313 | /* set default data area */ | |
314 | call->a_args.lock.oh.data = call->a_owner; | |
315 | call->a_args.lock.svid = lock->fl.fl_pid; | |
316 | ||
317 | if (lock->oh.len > NLMCLNT_OHSIZE) { | |
318 | void *data = kmalloc(lock->oh.len, GFP_KERNEL); | |
92737230 | 319 | if (!data) |
5e1abf8c | 320 | return 0; |
5e1abf8c TM |
321 | call->a_args.lock.oh.data = (u8 *) data; |
322 | } | |
323 | ||
324 | memcpy(call->a_args.lock.oh.data, lock->oh.data, lock->oh.len); | |
325 | return 1; | |
326 | } | |
327 | ||
328 | static void nlmsvc_freegrantargs(struct nlm_rqst *call) | |
329 | { | |
92737230 | 330 | if (call->a_args.lock.oh.data != call->a_owner) |
5e1abf8c | 331 | kfree(call->a_args.lock.oh.data); |
5e1abf8c TM |
332 | } |
333 | ||
2b36f412 ME |
334 | /* |
335 | * Deferred lock request handling for non-blocking lock | |
336 | */ | |
337 | static u32 | |
338 | nlmsvc_defer_lock_rqst(struct svc_rqst *rqstp, struct nlm_block *block) | |
339 | { | |
340 | u32 status = nlm_lck_denied_nolocks; | |
341 | ||
342 | block->b_flags |= B_QUEUED; | |
343 | ||
344 | nlmsvc_insert_block(block, NLM_TIMEOUT); | |
345 | ||
346 | block->b_cache_req = &rqstp->rq_chandle; | |
347 | if (rqstp->rq_chandle.defer) { | |
348 | block->b_deferred_req = | |
349 | rqstp->rq_chandle.defer(block->b_cache_req); | |
350 | if (block->b_deferred_req != NULL) | |
351 | status = nlm_drop_reply; | |
352 | } | |
353 | dprintk("lockd: nlmsvc_defer_lock_rqst block %p flags %d status %d\n", | |
354 | block, block->b_flags, status); | |
355 | ||
356 | return status; | |
357 | } | |
358 | ||
1da177e4 LT |
359 | /* |
360 | * Attempt to establish a lock, and if it can't be granted, block it | |
361 | * if required. | |
362 | */ | |
52921e02 | 363 | __be32 |
1da177e4 LT |
364 | nlmsvc_lock(struct svc_rqst *rqstp, struct nlm_file *file, |
365 | struct nlm_lock *lock, int wait, struct nlm_cookie *cookie) | |
366 | { | |
09c7938c | 367 | struct nlm_block *block, *newblock = 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 | ||
379 | ||
09c7938c | 380 | lock->fl.fl_flags &= ~FL_SLEEP; |
1da177e4 LT |
381 | again: |
382 | /* Lock file against concurrent access */ | |
89e63ef6 | 383 | mutex_lock(&file->f_mutex); |
09c7938c | 384 | /* Get existing block (in case client is busy-waiting) */ |
d9f6eb75 | 385 | block = nlmsvc_lookup_block(file, lock); |
09c7938c TM |
386 | if (block == NULL) { |
387 | if (newblock != NULL) | |
92737230 | 388 | lock = &newblock->b_call->a_args.lock; |
09c7938c | 389 | } else |
92737230 | 390 | lock = &block->b_call->a_args.lock; |
1da177e4 | 391 | |
150b3934 | 392 | error = posix_lock_file(file->f_file, &lock->fl, NULL); |
09c7938c | 393 | lock->fl.fl_flags &= ~FL_SLEEP; |
1da177e4 | 394 | |
a85f193e AA |
395 | dprintk("lockd: posix_lock_file returned %d\n", error); |
396 | ||
09c7938c | 397 | switch(error) { |
1da177e4 | 398 | case 0: |
15dadef9 AA |
399 | ret = nlm_granted; |
400 | goto out; | |
09c7938c TM |
401 | case -EAGAIN: |
402 | break; | |
403 | case -EDEADLK: | |
15dadef9 AA |
404 | ret = nlm_deadlock; |
405 | goto out; | |
1da177e4 | 406 | default: /* includes ENOLCK */ |
15dadef9 AA |
407 | ret = nlm_lck_denied_nolocks; |
408 | goto out; | |
1da177e4 LT |
409 | } |
410 | ||
09c7938c TM |
411 | ret = nlm_lck_denied; |
412 | if (!wait) | |
413 | goto out; | |
414 | ||
415 | ret = nlm_lck_blocked; | |
416 | if (block != NULL) | |
417 | goto out; | |
1da177e4 | 418 | |
1da177e4 LT |
419 | /* If we don't have a block, create and initialize it. Then |
420 | * retry because we may have slept in kmalloc. */ | |
89e63ef6 | 421 | /* We have to release f_mutex as nlmsvc_create_block may try to |
1da177e4 | 422 | * to claim it while doing host garbage collection */ |
09c7938c | 423 | if (newblock == NULL) { |
89e63ef6 | 424 | mutex_unlock(&file->f_mutex); |
1da177e4 | 425 | dprintk("lockd: blocking on this lock (allocating).\n"); |
09c7938c | 426 | if (!(newblock = nlmsvc_create_block(rqstp, file, lock, cookie))) |
1da177e4 LT |
427 | return nlm_lck_denied_nolocks; |
428 | goto again; | |
429 | } | |
430 | ||
431 | /* Append to list of blocked */ | |
09c7938c | 432 | nlmsvc_insert_block(newblock, NLM_NEVER); |
15dadef9 | 433 | out: |
89e63ef6 | 434 | mutex_unlock(&file->f_mutex); |
6849c0ca TM |
435 | nlmsvc_release_block(newblock); |
436 | nlmsvc_release_block(block); | |
15dadef9 AA |
437 | dprintk("lockd: nlmsvc_lock returned %u\n", ret); |
438 | return ret; | |
1da177e4 LT |
439 | } |
440 | ||
441 | /* | |
442 | * Test for presence of a conflicting lock. | |
443 | */ | |
52921e02 | 444 | __be32 |
85f3f1b3 ME |
445 | nlmsvc_testlock(struct svc_rqst *rqstp, struct nlm_file *file, |
446 | struct nlm_lock *lock, struct nlm_lock *conflock, | |
447 | struct nlm_cookie *cookie) | |
1da177e4 | 448 | { |
1da177e4 | 449 | dprintk("lockd: nlmsvc_testlock(%s/%ld, ty=%d, %Ld-%Ld)\n", |
225a719f JS |
450 | file->f_file->f_path.dentry->d_inode->i_sb->s_id, |
451 | file->f_file->f_path.dentry->d_inode->i_ino, | |
1da177e4 LT |
452 | lock->fl.fl_type, |
453 | (long long)lock->fl.fl_start, | |
454 | (long long)lock->fl.fl_end); | |
455 | ||
9d6a8c5c | 456 | if (posix_test_lock(file->f_file, &lock->fl)) { |
1da177e4 | 457 | dprintk("lockd: conflicting lock(ty=%d, %Ld-%Ld)\n", |
9d6a8c5c ME |
458 | lock->fl.fl_type, |
459 | (long long)lock->fl.fl_start, | |
460 | (long long)lock->fl.fl_end); | |
1da177e4 | 461 | conflock->caller = "somehost"; /* FIXME */ |
db4e4c9a | 462 | conflock->len = strlen(conflock->caller); |
1da177e4 | 463 | conflock->oh.len = 0; /* don't return OH info */ |
9d6a8c5c ME |
464 | conflock->svid = lock->fl.fl_pid; |
465 | conflock->fl.fl_type = lock->fl.fl_type; | |
466 | conflock->fl.fl_start = lock->fl.fl_start; | |
467 | conflock->fl.fl_end = lock->fl.fl_end; | |
1da177e4 LT |
468 | return nlm_lck_denied; |
469 | } | |
470 | ||
471 | return nlm_granted; | |
472 | } | |
473 | ||
474 | /* | |
475 | * Remove a lock. | |
476 | * This implies a CANCEL call: We send a GRANT_MSG, the client replies | |
477 | * with a GRANT_RES call which gets lost, and calls UNLOCK immediately | |
478 | * afterwards. In this case the block will still be there, and hence | |
479 | * must be removed. | |
480 | */ | |
52921e02 | 481 | __be32 |
1da177e4 LT |
482 | nlmsvc_unlock(struct nlm_file *file, struct nlm_lock *lock) |
483 | { | |
484 | int error; | |
485 | ||
486 | dprintk("lockd: nlmsvc_unlock(%s/%ld, pi=%d, %Ld-%Ld)\n", | |
225a719f JS |
487 | file->f_file->f_path.dentry->d_inode->i_sb->s_id, |
488 | file->f_file->f_path.dentry->d_inode->i_ino, | |
1da177e4 LT |
489 | lock->fl.fl_pid, |
490 | (long long)lock->fl.fl_start, | |
491 | (long long)lock->fl.fl_end); | |
492 | ||
493 | /* First, cancel any lock that might be there */ | |
494 | nlmsvc_cancel_blocked(file, lock); | |
495 | ||
496 | lock->fl.fl_type = F_UNLCK; | |
150b3934 | 497 | error = posix_lock_file(file->f_file, &lock->fl, NULL); |
1da177e4 LT |
498 | |
499 | return (error < 0)? nlm_lck_denied_nolocks : nlm_granted; | |
500 | } | |
501 | ||
502 | /* | |
503 | * Cancel a previously blocked request. | |
504 | * | |
505 | * A cancel request always overrides any grant that may currently | |
506 | * be in progress. | |
507 | * The calling procedure must check whether the file can be closed. | |
508 | */ | |
52921e02 | 509 | __be32 |
1da177e4 LT |
510 | nlmsvc_cancel_blocked(struct nlm_file *file, struct nlm_lock *lock) |
511 | { | |
512 | struct nlm_block *block; | |
64a318ee | 513 | int status = 0; |
1da177e4 LT |
514 | |
515 | dprintk("lockd: nlmsvc_cancel(%s/%ld, pi=%d, %Ld-%Ld)\n", | |
225a719f JS |
516 | file->f_file->f_path.dentry->d_inode->i_sb->s_id, |
517 | file->f_file->f_path.dentry->d_inode->i_ino, | |
1da177e4 LT |
518 | lock->fl.fl_pid, |
519 | (long long)lock->fl.fl_start, | |
520 | (long long)lock->fl.fl_end); | |
521 | ||
89e63ef6 | 522 | mutex_lock(&file->f_mutex); |
d9f6eb75 | 523 | block = nlmsvc_lookup_block(file, lock); |
89e63ef6 | 524 | mutex_unlock(&file->f_mutex); |
d9f6eb75 | 525 | if (block != NULL) { |
6849c0ca TM |
526 | status = nlmsvc_unlink_block(block); |
527 | nlmsvc_release_block(block); | |
528 | } | |
64a318ee | 529 | return status ? nlm_lck_denied : nlm_granted; |
1da177e4 LT |
530 | } |
531 | ||
0e4ac9d9 ME |
532 | /* |
533 | * This is a callback from the filesystem for VFS file lock requests. | |
534 | * It will be used if fl_grant is defined and the filesystem can not | |
535 | * respond to the request immediately. | |
536 | * For GETLK request it will copy the reply to the nlm_block. | |
537 | * For SETLK or SETLKW request it will get the local posix lock. | |
538 | * In all cases it will move the block to the head of nlm_blocked q where | |
539 | * nlmsvc_retry_blocked() can send back a reply for SETLKW or revisit the | |
540 | * deferred rpc for GETLK and SETLK. | |
541 | */ | |
542 | static void | |
543 | nlmsvc_update_deferred_block(struct nlm_block *block, struct file_lock *conf, | |
544 | int result) | |
545 | { | |
546 | block->b_flags |= B_GOT_CALLBACK; | |
547 | if (result == 0) | |
548 | block->b_granted = 1; | |
549 | else | |
550 | block->b_flags |= B_TIMED_OUT; | |
551 | if (conf) { | |
552 | block->b_fl = kzalloc(sizeof(struct file_lock), GFP_KERNEL); | |
553 | if (block->b_fl) | |
554 | locks_copy_lock(block->b_fl, conf); | |
555 | } | |
556 | } | |
557 | ||
558 | static int nlmsvc_grant_deferred(struct file_lock *fl, struct file_lock *conf, | |
559 | int result) | |
560 | { | |
561 | struct nlm_block *block; | |
562 | int rc = -ENOENT; | |
563 | ||
564 | lock_kernel(); | |
565 | list_for_each_entry(block, &nlm_blocked, b_list) { | |
566 | if (nlm_compare_locks(&block->b_call->a_args.lock.fl, fl)) { | |
567 | dprintk("lockd: nlmsvc_notify_blocked block %p flags %d\n", | |
568 | block, block->b_flags); | |
569 | if (block->b_flags & B_QUEUED) { | |
570 | if (block->b_flags & B_TIMED_OUT) { | |
571 | rc = -ENOLCK; | |
572 | break; | |
573 | } | |
574 | nlmsvc_update_deferred_block(block, conf, result); | |
575 | } else if (result == 0) | |
576 | block->b_granted = 1; | |
577 | ||
578 | nlmsvc_insert_block(block, 0); | |
579 | svc_wake_up(block->b_daemon); | |
580 | rc = 0; | |
581 | break; | |
582 | } | |
583 | } | |
584 | unlock_kernel(); | |
585 | if (rc == -ENOENT) | |
586 | printk(KERN_WARNING "lockd: grant for unknown block\n"); | |
587 | return rc; | |
588 | } | |
589 | ||
1da177e4 LT |
590 | /* |
591 | * Unblock a blocked lock request. This is a callback invoked from the | |
592 | * VFS layer when a lock on which we blocked is removed. | |
593 | * | |
594 | * This function doesn't grant the blocked lock instantly, but rather moves | |
595 | * the block to the head of nlm_blocked where it can be picked up by lockd. | |
596 | */ | |
597 | static void | |
598 | nlmsvc_notify_blocked(struct file_lock *fl) | |
599 | { | |
68a2d76c | 600 | struct nlm_block *block; |
1da177e4 LT |
601 | |
602 | dprintk("lockd: VFS unblock notification for block %p\n", fl); | |
68a2d76c | 603 | list_for_each_entry(block, &nlm_blocked, b_list) { |
92737230 | 604 | if (nlm_compare_locks(&block->b_call->a_args.lock.fl, fl)) { |
1da177e4 LT |
605 | nlmsvc_insert_block(block, 0); |
606 | svc_wake_up(block->b_daemon); | |
607 | return; | |
608 | } | |
609 | } | |
610 | ||
611 | printk(KERN_WARNING "lockd: notification for unknown block!\n"); | |
612 | } | |
613 | ||
614 | static int nlmsvc_same_owner(struct file_lock *fl1, struct file_lock *fl2) | |
615 | { | |
616 | return fl1->fl_owner == fl2->fl_owner && fl1->fl_pid == fl2->fl_pid; | |
617 | } | |
618 | ||
619 | struct lock_manager_operations nlmsvc_lock_operations = { | |
620 | .fl_compare_owner = nlmsvc_same_owner, | |
621 | .fl_notify = nlmsvc_notify_blocked, | |
0e4ac9d9 | 622 | .fl_grant = nlmsvc_grant_deferred, |
1da177e4 LT |
623 | }; |
624 | ||
625 | /* | |
626 | * Try to claim a lock that was previously blocked. | |
627 | * | |
628 | * Note that we use both the RPC_GRANTED_MSG call _and_ an async | |
629 | * RPC thread when notifying the client. This seems like overkill... | |
630 | * Here's why: | |
631 | * - we don't want to use a synchronous RPC thread, otherwise | |
632 | * we might find ourselves hanging on a dead portmapper. | |
633 | * - Some lockd implementations (e.g. HP) don't react to | |
634 | * RPC_GRANTED calls; they seem to insist on RPC_GRANTED_MSG calls. | |
635 | */ | |
636 | static void | |
637 | nlmsvc_grant_blocked(struct nlm_block *block) | |
638 | { | |
639 | struct nlm_file *file = block->b_file; | |
92737230 | 640 | struct nlm_lock *lock = &block->b_call->a_args.lock; |
1da177e4 LT |
641 | int error; |
642 | ||
643 | dprintk("lockd: grant blocked lock %p\n", block); | |
644 | ||
0e4ac9d9 ME |
645 | kref_get(&block->b_count); |
646 | ||
1da177e4 | 647 | /* Unlink block request from list */ |
6849c0ca | 648 | nlmsvc_unlink_block(block); |
1da177e4 LT |
649 | |
650 | /* If b_granted is true this means we've been here before. | |
651 | * Just retry the grant callback, possibly refreshing the RPC | |
652 | * binding */ | |
653 | if (block->b_granted) { | |
654 | nlm_rebind_host(block->b_host); | |
655 | goto callback; | |
656 | } | |
657 | ||
658 | /* Try the lock operation again */ | |
09c7938c | 659 | lock->fl.fl_flags |= FL_SLEEP; |
150b3934 | 660 | error = posix_lock_file(file->f_file, &lock->fl, NULL); |
09c7938c TM |
661 | lock->fl.fl_flags &= ~FL_SLEEP; |
662 | ||
5de0e502 AA |
663 | switch (error) { |
664 | case 0: | |
665 | break; | |
666 | case -EAGAIN: | |
1da177e4 LT |
667 | dprintk("lockd: lock still blocked\n"); |
668 | nlmsvc_insert_block(block, NLM_NEVER); | |
0e4ac9d9 | 669 | nlmsvc_release_block(block); |
d9f6eb75 | 670 | return; |
5de0e502 | 671 | default: |
1da177e4 LT |
672 | printk(KERN_WARNING "lockd: unexpected error %d in %s!\n", |
673 | -error, __FUNCTION__); | |
674 | nlmsvc_insert_block(block, 10 * HZ); | |
0e4ac9d9 | 675 | nlmsvc_release_block(block); |
d9f6eb75 | 676 | return; |
1da177e4 LT |
677 | } |
678 | ||
679 | callback: | |
680 | /* Lock was granted by VFS. */ | |
681 | dprintk("lockd: GRANTing blocked lock.\n"); | |
682 | block->b_granted = 1; | |
1da177e4 LT |
683 | |
684 | /* Schedule next grant callback in 30 seconds */ | |
685 | nlmsvc_insert_block(block, 30 * HZ); | |
686 | ||
687 | /* Call the client */ | |
a995e9eb | 688 | nlm_async_call(block->b_call, NLMPROC_GRANTED_MSG, &nlmsvc_grant_ops); |
1da177e4 LT |
689 | } |
690 | ||
691 | /* | |
692 | * This is the callback from the RPC layer when the NLM_GRANTED_MSG | |
693 | * RPC call has succeeded or timed out. | |
694 | * Like all RPC callbacks, it is invoked by the rpciod process, so it | |
695 | * better not sleep. Therefore, we put the blocked lock on the nlm_blocked | |
696 | * chain once more in order to have it removed by lockd itself (which can | |
697 | * then sleep on the file semaphore without disrupting e.g. the nfs client). | |
698 | */ | |
963d8fe5 | 699 | static void nlmsvc_grant_callback(struct rpc_task *task, void *data) |
1da177e4 | 700 | { |
963d8fe5 | 701 | struct nlm_rqst *call = data; |
92737230 | 702 | struct nlm_block *block = call->a_block; |
1da177e4 | 703 | unsigned long timeout; |
1da177e4 LT |
704 | |
705 | dprintk("lockd: GRANT_MSG RPC callback\n"); | |
1da177e4 LT |
706 | |
707 | /* Technically, we should down the file semaphore here. Since we | |
708 | * move the block towards the head of the queue only, no harm | |
709 | * can be done, though. */ | |
710 | if (task->tk_status < 0) { | |
711 | /* RPC error: Re-insert for retransmission */ | |
712 | timeout = 10 * HZ; | |
1da177e4 LT |
713 | } else { |
714 | /* Call was successful, now wait for client callback */ | |
715 | timeout = 60 * HZ; | |
716 | } | |
717 | nlmsvc_insert_block(block, timeout); | |
718 | svc_wake_up(block->b_daemon); | |
5e1abf8c TM |
719 | } |
720 | ||
ec535ce1 | 721 | static void nlmsvc_grant_release(void *data) |
5e1abf8c | 722 | { |
6041b791 TM |
723 | struct nlm_rqst *call = data; |
724 | ||
725 | nlmsvc_release_block(call->a_block); | |
1da177e4 LT |
726 | } |
727 | ||
963d8fe5 TM |
728 | static const struct rpc_call_ops nlmsvc_grant_ops = { |
729 | .rpc_call_done = nlmsvc_grant_callback, | |
5e1abf8c | 730 | .rpc_release = nlmsvc_grant_release, |
963d8fe5 TM |
731 | }; |
732 | ||
1da177e4 LT |
733 | /* |
734 | * We received a GRANT_RES callback. Try to find the corresponding | |
735 | * block. | |
736 | */ | |
737 | void | |
e8c5c045 | 738 | nlmsvc_grant_reply(struct nlm_cookie *cookie, __be32 status) |
1da177e4 LT |
739 | { |
740 | struct nlm_block *block; | |
1da177e4 | 741 | |
39be4502 OK |
742 | dprintk("grant_reply: looking for cookie %x, s=%d \n", |
743 | *(unsigned int *)(cookie->data), status); | |
744 | if (!(block = nlmsvc_find_block(cookie))) | |
1da177e4 | 745 | return; |
1da177e4 | 746 | |
f232142c | 747 | if (block) { |
e8c5c045 | 748 | if (status == nlm_lck_denied_grace_period) { |
1da177e4 LT |
749 | /* Try again in a couple of seconds */ |
750 | nlmsvc_insert_block(block, 10 * HZ); | |
1da177e4 LT |
751 | } else { |
752 | /* Lock is now held by client, or has been rejected. | |
753 | * In both cases, the block should be removed. */ | |
6849c0ca | 754 | nlmsvc_unlink_block(block); |
1da177e4 LT |
755 | } |
756 | } | |
6849c0ca | 757 | nlmsvc_release_block(block); |
1da177e4 LT |
758 | } |
759 | ||
0e4ac9d9 ME |
760 | /* Helper function to handle retry of a deferred block. |
761 | * If it is a blocking lock, call grant_blocked. | |
762 | * For a non-blocking lock or test lock, revisit the request. | |
763 | */ | |
764 | static void | |
765 | retry_deferred_block(struct nlm_block *block) | |
766 | { | |
767 | if (!(block->b_flags & B_GOT_CALLBACK)) | |
768 | block->b_flags |= B_TIMED_OUT; | |
769 | nlmsvc_insert_block(block, NLM_TIMEOUT); | |
770 | dprintk("revisit block %p flags %d\n", block, block->b_flags); | |
771 | if (block->b_deferred_req) { | |
772 | block->b_deferred_req->revisit(block->b_deferred_req, 0); | |
773 | block->b_deferred_req = NULL; | |
774 | } | |
775 | } | |
776 | ||
1da177e4 LT |
777 | /* |
778 | * Retry all blocked locks that have been notified. This is where lockd | |
779 | * picks up locks that can be granted, or grant notifications that must | |
780 | * be retransmitted. | |
781 | */ | |
782 | unsigned long | |
783 | nlmsvc_retry_blocked(void) | |
784 | { | |
68a2d76c OK |
785 | unsigned long timeout = MAX_SCHEDULE_TIMEOUT; |
786 | struct nlm_block *block; | |
787 | ||
788 | while (!list_empty(&nlm_blocked)) { | |
789 | block = list_entry(nlm_blocked.next, struct nlm_block, b_list); | |
1da177e4 | 790 | |
1da177e4 LT |
791 | if (block->b_when == NLM_NEVER) |
792 | break; | |
68a2d76c OK |
793 | if (time_after(block->b_when,jiffies)) { |
794 | timeout = block->b_when - jiffies; | |
1da177e4 | 795 | break; |
68a2d76c OK |
796 | } |
797 | ||
f3d43c76 BF |
798 | dprintk("nlmsvc_retry_blocked(%p, when=%ld)\n", |
799 | block, block->b_when); | |
0e4ac9d9 ME |
800 | if (block->b_flags & B_QUEUED) { |
801 | dprintk("nlmsvc_retry_blocked delete block (%p, granted=%d, flags=%d)\n", | |
802 | block, block->b_granted, block->b_flags); | |
803 | retry_deferred_block(block); | |
804 | } else | |
805 | nlmsvc_grant_blocked(block); | |
1da177e4 LT |
806 | } |
807 | ||
68a2d76c | 808 | return timeout; |
1da177e4 | 809 | } |