]>
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 | 23 | #include <linux/types.h> |
5a0e3ad6 | 24 | #include <linux/slab.h> |
1da177e4 LT |
25 | #include <linux/errno.h> |
26 | #include <linux/kernel.h> | |
27 | #include <linux/sched.h> | |
1da177e4 LT |
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); |
f904be9c | 54 | static DEFINE_SPINLOCK(nlm_blocked_lock); |
1da177e4 LT |
55 | |
56 | /* | |
57 | * Insert a blocked lock into the global list | |
58 | */ | |
59 | static void | |
f904be9c | 60 | nlmsvc_insert_block_locked(struct nlm_block *block, unsigned long when) |
1da177e4 | 61 | { |
68a2d76c OK |
62 | struct nlm_block *b; |
63 | struct list_head *pos; | |
1da177e4 LT |
64 | |
65 | dprintk("lockd: nlmsvc_insert_block(%p, %ld)\n", block, when); | |
68a2d76c OK |
66 | if (list_empty(&block->b_list)) { |
67 | kref_get(&block->b_count); | |
68 | } else { | |
69 | list_del_init(&block->b_list); | |
70 | } | |
71 | ||
72 | pos = &nlm_blocked; | |
1da177e4 LT |
73 | if (when != NLM_NEVER) { |
74 | if ((when += jiffies) == NLM_NEVER) | |
75 | when ++; | |
68a2d76c OK |
76 | list_for_each(pos, &nlm_blocked) { |
77 | b = list_entry(pos, struct nlm_block, b_list); | |
78 | if (time_after(b->b_when,when) || b->b_when == NLM_NEVER) | |
79 | break; | |
80 | } | |
81 | /* On normal exit from the loop, pos == &nlm_blocked, | |
82 | * so we will be adding to the end of the list - good | |
83 | */ | |
84 | } | |
1da177e4 | 85 | |
68a2d76c | 86 | list_add_tail(&block->b_list, pos); |
1da177e4 | 87 | block->b_when = when; |
1da177e4 LT |
88 | } |
89 | ||
f904be9c BS |
90 | static void nlmsvc_insert_block(struct nlm_block *block, unsigned long when) |
91 | { | |
92 | spin_lock(&nlm_blocked_lock); | |
93 | nlmsvc_insert_block_locked(block, when); | |
94 | spin_unlock(&nlm_blocked_lock); | |
95 | } | |
96 | ||
1da177e4 LT |
97 | /* |
98 | * Remove a block from the global list | |
99 | */ | |
68a2d76c | 100 | static inline void |
1da177e4 LT |
101 | nlmsvc_remove_block(struct nlm_block *block) |
102 | { | |
68a2d76c | 103 | if (!list_empty(&block->b_list)) { |
f904be9c | 104 | spin_lock(&nlm_blocked_lock); |
68a2d76c | 105 | list_del_init(&block->b_list); |
f904be9c | 106 | spin_unlock(&nlm_blocked_lock); |
68a2d76c | 107 | nlmsvc_release_block(block); |
1da177e4 | 108 | } |
1da177e4 LT |
109 | } |
110 | ||
111 | /* | |
d9f6eb75 | 112 | * Find a block for a given lock |
1da177e4 LT |
113 | */ |
114 | static struct nlm_block * | |
d9f6eb75 | 115 | nlmsvc_lookup_block(struct nlm_file *file, struct nlm_lock *lock) |
1da177e4 | 116 | { |
68a2d76c | 117 | struct nlm_block *block; |
1da177e4 LT |
118 | struct file_lock *fl; |
119 | ||
120 | dprintk("lockd: nlmsvc_lookup_block f=%p pd=%d %Ld-%Ld ty=%d\n", | |
121 | file, lock->fl.fl_pid, | |
122 | (long long)lock->fl.fl_start, | |
123 | (long long)lock->fl.fl_end, lock->fl.fl_type); | |
68a2d76c | 124 | list_for_each_entry(block, &nlm_blocked, b_list) { |
92737230 | 125 | fl = &block->b_call->a_args.lock.fl; |
1da177e4 LT |
126 | dprintk("lockd: check f=%p pd=%d %Ld-%Ld ty=%d cookie=%s\n", |
127 | block->b_file, fl->fl_pid, | |
128 | (long long)fl->fl_start, | |
129 | (long long)fl->fl_end, fl->fl_type, | |
92737230 | 130 | nlmdbg_cookie2a(&block->b_call->a_args.cookie)); |
1da177e4 | 131 | if (block->b_file == file && nlm_compare_locks(fl, &lock->fl)) { |
6849c0ca | 132 | kref_get(&block->b_count); |
1da177e4 LT |
133 | return block; |
134 | } | |
135 | } | |
136 | ||
137 | return NULL; | |
138 | } | |
139 | ||
140 | static inline int nlm_cookie_match(struct nlm_cookie *a, struct nlm_cookie *b) | |
141 | { | |
6d7bbbba | 142 | if (a->len != b->len) |
1da177e4 | 143 | return 0; |
6d7bbbba | 144 | if (memcmp(a->data, b->data, a->len)) |
1da177e4 LT |
145 | return 0; |
146 | return 1; | |
147 | } | |
148 | ||
149 | /* | |
150 | * Find a block with a given NLM cookie. | |
151 | */ | |
152 | static inline struct nlm_block * | |
39be4502 | 153 | nlmsvc_find_block(struct nlm_cookie *cookie) |
1da177e4 LT |
154 | { |
155 | struct nlm_block *block; | |
156 | ||
68a2d76c | 157 | list_for_each_entry(block, &nlm_blocked, b_list) { |
39be4502 | 158 | if (nlm_cookie_match(&block->b_call->a_args.cookie,cookie)) |
68a2d76c | 159 | goto found; |
1da177e4 LT |
160 | } |
161 | ||
68a2d76c OK |
162 | return NULL; |
163 | ||
164 | found: | |
39be4502 | 165 | dprintk("nlmsvc_find_block(%s): block=%p\n", nlmdbg_cookie2a(cookie), block); |
68a2d76c | 166 | kref_get(&block->b_count); |
1da177e4 LT |
167 | return block; |
168 | } | |
169 | ||
170 | /* | |
171 | * Create a block and initialize it. | |
172 | * | |
173 | * Note: we explicitly set the cookie of the grant reply to that of | |
174 | * the blocked lock request. The spec explicitly mentions that the client | |
175 | * should _not_ rely on the callback containing the same cookie as the | |
176 | * request, but (as I found out later) that's because some implementations | |
177 | * do just this. Never mind the standards comittees, they support our | |
178 | * logging industries. | |
39be4502 OK |
179 | * |
180 | * 10 years later: I hope we can safely ignore these old and broken | |
181 | * clients by now. Let's fix this so we can uniquely identify an incoming | |
182 | * GRANTED_RES message by cookie, without having to rely on the client's IP | |
183 | * address. --okir | |
1da177e4 | 184 | */ |
255129d1 TM |
185 | static struct nlm_block * |
186 | nlmsvc_create_block(struct svc_rqst *rqstp, struct nlm_host *host, | |
187 | struct nlm_file *file, struct nlm_lock *lock, | |
188 | struct nlm_cookie *cookie) | |
1da177e4 LT |
189 | { |
190 | struct nlm_block *block; | |
92737230 | 191 | struct nlm_rqst *call = NULL; |
1da177e4 | 192 | |
560de0e6 | 193 | nlm_get_host(host); |
92737230 TM |
194 | call = nlm_alloc_call(host); |
195 | if (call == NULL) | |
196 | return NULL; | |
197 | ||
1da177e4 | 198 | /* Allocate memory for block, and initialize arguments */ |
92737230 TM |
199 | block = kzalloc(sizeof(*block), GFP_KERNEL); |
200 | if (block == NULL) | |
1da177e4 | 201 | goto failed; |
6849c0ca | 202 | kref_init(&block->b_count); |
68a2d76c OK |
203 | INIT_LIST_HEAD(&block->b_list); |
204 | INIT_LIST_HEAD(&block->b_flist); | |
1da177e4 | 205 | |
92737230 | 206 | if (!nlmsvc_setgrantargs(call, lock)) |
1da177e4 LT |
207 | goto failed_free; |
208 | ||
209 | /* Set notifier function for VFS, and init args */ | |
92737230 TM |
210 | call->a_args.lock.fl.fl_flags |= FL_SLEEP; |
211 | call->a_args.lock.fl.fl_lmops = &nlmsvc_lock_operations; | |
39be4502 | 212 | nlmclnt_next_cookie(&call->a_args.cookie); |
1da177e4 LT |
213 | |
214 | dprintk("lockd: created block %p...\n", block); | |
215 | ||
216 | /* Create and initialize the block */ | |
217 | block->b_daemon = rqstp->rq_server; | |
218 | block->b_host = host; | |
219 | block->b_file = file; | |
5ea0d750 | 220 | block->b_fl = NULL; |
d9f6eb75 | 221 | file->f_count++; |
1da177e4 LT |
222 | |
223 | /* Add to file's list of blocks */ | |
68a2d76c | 224 | list_add(&block->b_flist, &file->f_blocks); |
1da177e4 LT |
225 | |
226 | /* Set up RPC arguments for callback */ | |
92737230 | 227 | block->b_call = call; |
1da177e4 | 228 | call->a_flags = RPC_TASK_ASYNC; |
92737230 | 229 | call->a_block = block; |
1da177e4 LT |
230 | |
231 | return block; | |
232 | ||
233 | failed_free: | |
234 | kfree(block); | |
235 | failed: | |
92737230 | 236 | nlm_release_call(call); |
1da177e4 LT |
237 | return NULL; |
238 | } | |
239 | ||
240 | /* | |
3c61eecb | 241 | * Delete a block. |
1da177e4 LT |
242 | * It is the caller's responsibility to check whether the file |
243 | * can be closed hereafter. | |
244 | */ | |
6849c0ca | 245 | static int nlmsvc_unlink_block(struct nlm_block *block) |
1da177e4 | 246 | { |
09c7938c | 247 | int status; |
6849c0ca | 248 | dprintk("lockd: unlinking block %p...\n", block); |
1da177e4 LT |
249 | |
250 | /* Remove block from list */ | |
92737230 | 251 | status = posix_unblock_lock(block->b_file->f_file, &block->b_call->a_args.lock.fl); |
1da177e4 | 252 | nlmsvc_remove_block(block); |
6849c0ca TM |
253 | return status; |
254 | } | |
1da177e4 | 255 | |
6849c0ca TM |
256 | static void nlmsvc_free_block(struct kref *kref) |
257 | { | |
258 | struct nlm_block *block = container_of(kref, struct nlm_block, b_count); | |
259 | struct nlm_file *file = block->b_file; | |
6849c0ca TM |
260 | |
261 | dprintk("lockd: freeing block %p...\n", block); | |
1da177e4 LT |
262 | |
263 | /* Remove block from file's list of blocks */ | |
89e63ef6 | 264 | mutex_lock(&file->f_mutex); |
68a2d76c | 265 | list_del_init(&block->b_flist); |
89e63ef6 | 266 | mutex_unlock(&file->f_mutex); |
1da177e4 | 267 | |
92737230 TM |
268 | nlmsvc_freegrantargs(block->b_call); |
269 | nlm_release_call(block->b_call); | |
d9f6eb75 | 270 | nlm_release_file(block->b_file); |
0e4ac9d9 | 271 | kfree(block->b_fl); |
1da177e4 | 272 | kfree(block); |
6849c0ca TM |
273 | } |
274 | ||
275 | static void nlmsvc_release_block(struct nlm_block *block) | |
276 | { | |
277 | if (block != NULL) | |
278 | kref_put(&block->b_count, nlmsvc_free_block); | |
1da177e4 LT |
279 | } |
280 | ||
f2af793d OK |
281 | /* |
282 | * Loop over all blocks and delete blocks held by | |
283 | * a matching host. | |
284 | */ | |
285 | void nlmsvc_traverse_blocks(struct nlm_host *host, | |
286 | struct nlm_file *file, | |
287 | nlm_host_match_fn_t match) | |
d9f6eb75 | 288 | { |
68a2d76c | 289 | struct nlm_block *block, *next; |
d9f6eb75 TM |
290 | |
291 | restart: | |
89e63ef6 | 292 | mutex_lock(&file->f_mutex); |
68a2d76c | 293 | list_for_each_entry_safe(block, next, &file->f_blocks, b_flist) { |
f2af793d | 294 | if (!match(block->b_host, host)) |
d9f6eb75 | 295 | continue; |
68a2d76c OK |
296 | /* Do not destroy blocks that are not on |
297 | * the global retry list - why? */ | |
298 | if (list_empty(&block->b_list)) | |
d9f6eb75 TM |
299 | continue; |
300 | kref_get(&block->b_count); | |
89e63ef6 | 301 | mutex_unlock(&file->f_mutex); |
d9f6eb75 TM |
302 | nlmsvc_unlink_block(block); |
303 | nlmsvc_release_block(block); | |
304 | goto restart; | |
305 | } | |
89e63ef6 | 306 | mutex_unlock(&file->f_mutex); |
d9f6eb75 TM |
307 | } |
308 | ||
5e1abf8c TM |
309 | /* |
310 | * Initialize arguments for GRANTED call. The nlm_rqst structure | |
311 | * has been cleared already. | |
312 | */ | |
313 | static int nlmsvc_setgrantargs(struct nlm_rqst *call, struct nlm_lock *lock) | |
314 | { | |
315 | locks_copy_lock(&call->a_args.lock.fl, &lock->fl); | |
316 | memcpy(&call->a_args.lock.fh, &lock->fh, sizeof(call->a_args.lock.fh)); | |
e9ff3990 | 317 | call->a_args.lock.caller = utsname()->nodename; |
5e1abf8c TM |
318 | call->a_args.lock.oh.len = lock->oh.len; |
319 | ||
320 | /* set default data area */ | |
321 | call->a_args.lock.oh.data = call->a_owner; | |
322 | call->a_args.lock.svid = lock->fl.fl_pid; | |
323 | ||
324 | if (lock->oh.len > NLMCLNT_OHSIZE) { | |
325 | void *data = kmalloc(lock->oh.len, GFP_KERNEL); | |
92737230 | 326 | if (!data) |
5e1abf8c | 327 | return 0; |
5e1abf8c TM |
328 | call->a_args.lock.oh.data = (u8 *) data; |
329 | } | |
330 | ||
331 | memcpy(call->a_args.lock.oh.data, lock->oh.data, lock->oh.len); | |
332 | return 1; | |
333 | } | |
334 | ||
335 | static void nlmsvc_freegrantargs(struct nlm_rqst *call) | |
336 | { | |
92737230 | 337 | if (call->a_args.lock.oh.data != call->a_owner) |
5e1abf8c | 338 | kfree(call->a_args.lock.oh.data); |
a9e61e25 FB |
339 | |
340 | locks_release_private(&call->a_args.lock.fl); | |
5e1abf8c TM |
341 | } |
342 | ||
2b36f412 ME |
343 | /* |
344 | * Deferred lock request handling for non-blocking lock | |
345 | */ | |
ca5c8cde | 346 | static __be32 |
2b36f412 ME |
347 | nlmsvc_defer_lock_rqst(struct svc_rqst *rqstp, struct nlm_block *block) |
348 | { | |
ca5c8cde | 349 | __be32 status = nlm_lck_denied_nolocks; |
2b36f412 ME |
350 | |
351 | block->b_flags |= B_QUEUED; | |
352 | ||
353 | nlmsvc_insert_block(block, NLM_TIMEOUT); | |
354 | ||
355 | block->b_cache_req = &rqstp->rq_chandle; | |
356 | if (rqstp->rq_chandle.defer) { | |
357 | block->b_deferred_req = | |
358 | rqstp->rq_chandle.defer(block->b_cache_req); | |
359 | if (block->b_deferred_req != NULL) | |
360 | status = nlm_drop_reply; | |
361 | } | |
362 | dprintk("lockd: nlmsvc_defer_lock_rqst block %p flags %d status %d\n", | |
ca5c8cde | 363 | block, block->b_flags, ntohl(status)); |
2b36f412 ME |
364 | |
365 | return status; | |
366 | } | |
367 | ||
1da177e4 LT |
368 | /* |
369 | * Attempt to establish a lock, and if it can't be granted, block it | |
370 | * if required. | |
371 | */ | |
52921e02 | 372 | __be32 |
1da177e4 | 373 | nlmsvc_lock(struct svc_rqst *rqstp, struct nlm_file *file, |
6cde4de8 | 374 | struct nlm_host *host, struct nlm_lock *lock, int wait, |
b2b50289 | 375 | struct nlm_cookie *cookie, int reclaim) |
1da177e4 | 376 | { |
f8120480 | 377 | struct nlm_block *block = NULL; |
1da177e4 | 378 | int error; |
52921e02 | 379 | __be32 ret; |
1da177e4 LT |
380 | |
381 | dprintk("lockd: nlmsvc_lock(%s/%ld, ty=%d, pi=%d, %Ld-%Ld, bl=%d)\n", | |
225a719f JS |
382 | file->f_file->f_path.dentry->d_inode->i_sb->s_id, |
383 | file->f_file->f_path.dentry->d_inode->i_ino, | |
1da177e4 LT |
384 | lock->fl.fl_type, lock->fl.fl_pid, |
385 | (long long)lock->fl.fl_start, | |
386 | (long long)lock->fl.fl_end, | |
387 | wait); | |
388 | ||
1da177e4 | 389 | /* Lock file against concurrent access */ |
89e63ef6 | 390 | mutex_lock(&file->f_mutex); |
f8120480 ME |
391 | /* Get existing block (in case client is busy-waiting) |
392 | * or create new block | |
393 | */ | |
d9f6eb75 | 394 | block = nlmsvc_lookup_block(file, lock); |
09c7938c | 395 | if (block == NULL) { |
560de0e6 | 396 | block = nlmsvc_create_block(rqstp, host, file, lock, cookie); |
f8120480 ME |
397 | ret = nlm_lck_denied_nolocks; |
398 | if (block == NULL) | |
399 | goto out; | |
92737230 | 400 | lock = &block->b_call->a_args.lock; |
f8120480 ME |
401 | } else |
402 | lock->fl.fl_flags &= ~FL_SLEEP; | |
1da177e4 | 403 | |
1a8322b2 ME |
404 | if (block->b_flags & B_QUEUED) { |
405 | dprintk("lockd: nlmsvc_lock deferred block %p flags %d\n", | |
406 | block, block->b_flags); | |
407 | if (block->b_granted) { | |
408 | nlmsvc_unlink_block(block); | |
409 | ret = nlm_granted; | |
410 | goto out; | |
411 | } | |
412 | if (block->b_flags & B_TIMED_OUT) { | |
413 | nlmsvc_unlink_block(block); | |
414 | ret = nlm_lck_denied; | |
415 | goto out; | |
416 | } | |
417 | ret = nlm_drop_reply; | |
418 | goto out; | |
419 | } | |
1da177e4 | 420 | |
b2b50289 BF |
421 | if (locks_in_grace() && !reclaim) { |
422 | ret = nlm_lck_denied_grace_period; | |
423 | goto out; | |
424 | } | |
d22b1cff BF |
425 | if (reclaim && !locks_in_grace()) { |
426 | ret = nlm_lck_denied_grace_period; | |
427 | goto out; | |
428 | } | |
b2b50289 | 429 | |
1a8322b2 ME |
430 | if (!wait) |
431 | lock->fl.fl_flags &= ~FL_SLEEP; | |
432 | error = vfs_lock_file(file->f_file, F_SETLK, &lock->fl, NULL); | |
433 | lock->fl.fl_flags &= ~FL_SLEEP; | |
a85f193e | 434 | |
1a8322b2 | 435 | dprintk("lockd: vfs_lock_file returned %d\n", error); |
6d7bbbba | 436 | switch (error) { |
1da177e4 | 437 | case 0: |
15dadef9 AA |
438 | ret = nlm_granted; |
439 | goto out; | |
09c7938c | 440 | case -EAGAIN: |
e33d1ea6 MS |
441 | /* |
442 | * If this is a blocking request for an | |
443 | * already pending lock request then we need | |
444 | * to put it back on lockd's block list | |
445 | */ | |
446 | if (wait) | |
447 | break; | |
1a8322b2 | 448 | ret = nlm_lck_denied; |
e33d1ea6 | 449 | goto out; |
bde74e4b | 450 | case FILE_LOCK_DEFERRED: |
1a8322b2 ME |
451 | if (wait) |
452 | break; | |
453 | /* Filesystem lock operation is in progress | |
454 | Add it to the queue waiting for callback */ | |
455 | ret = nlmsvc_defer_lock_rqst(rqstp, block); | |
456 | goto out; | |
09c7938c | 457 | case -EDEADLK: |
15dadef9 AA |
458 | ret = nlm_deadlock; |
459 | goto out; | |
1da177e4 | 460 | default: /* includes ENOLCK */ |
15dadef9 AA |
461 | ret = nlm_lck_denied_nolocks; |
462 | goto out; | |
1da177e4 LT |
463 | } |
464 | ||
09c7938c | 465 | ret = nlm_lck_blocked; |
1da177e4 LT |
466 | |
467 | /* Append to list of blocked */ | |
f8120480 | 468 | nlmsvc_insert_block(block, NLM_NEVER); |
15dadef9 | 469 | out: |
89e63ef6 | 470 | mutex_unlock(&file->f_mutex); |
6849c0ca | 471 | nlmsvc_release_block(block); |
15dadef9 AA |
472 | dprintk("lockd: nlmsvc_lock returned %u\n", ret); |
473 | return ret; | |
1da177e4 LT |
474 | } |
475 | ||
476 | /* | |
477 | * Test for presence of a conflicting lock. | |
478 | */ | |
52921e02 | 479 | __be32 |
85f3f1b3 | 480 | nlmsvc_testlock(struct svc_rqst *rqstp, struct nlm_file *file, |
8f920d5e JL |
481 | struct nlm_host *host, struct nlm_lock *lock, |
482 | struct nlm_lock *conflock, struct nlm_cookie *cookie) | |
1da177e4 | 483 | { |
5ea0d750 ME |
484 | struct nlm_block *block = NULL; |
485 | int error; | |
486 | __be32 ret; | |
487 | ||
1da177e4 | 488 | dprintk("lockd: nlmsvc_testlock(%s/%ld, ty=%d, %Ld-%Ld)\n", |
225a719f JS |
489 | file->f_file->f_path.dentry->d_inode->i_sb->s_id, |
490 | file->f_file->f_path.dentry->d_inode->i_ino, | |
1da177e4 LT |
491 | lock->fl.fl_type, |
492 | (long long)lock->fl.fl_start, | |
493 | (long long)lock->fl.fl_end); | |
494 | ||
5ea0d750 ME |
495 | /* Get existing block (in case client is busy-waiting) */ |
496 | block = nlmsvc_lookup_block(file, lock); | |
497 | ||
498 | if (block == NULL) { | |
499 | struct file_lock *conf = kzalloc(sizeof(*conf), GFP_KERNEL); | |
500 | ||
501 | if (conf == NULL) | |
502 | return nlm_granted; | |
255129d1 | 503 | block = nlmsvc_create_block(rqstp, host, file, lock, cookie); |
5ea0d750 ME |
504 | if (block == NULL) { |
505 | kfree(conf); | |
506 | return nlm_granted; | |
507 | } | |
508 | block->b_fl = conf; | |
509 | } | |
510 | if (block->b_flags & B_QUEUED) { | |
511 | dprintk("lockd: nlmsvc_testlock deferred block %p flags %d fl %p\n", | |
512 | block, block->b_flags, block->b_fl); | |
513 | if (block->b_flags & B_TIMED_OUT) { | |
514 | nlmsvc_unlink_block(block); | |
29dbf546 OD |
515 | ret = nlm_lck_denied; |
516 | goto out; | |
5ea0d750 ME |
517 | } |
518 | if (block->b_flags & B_GOT_CALLBACK) { | |
54ca95eb | 519 | nlmsvc_unlink_block(block); |
5ea0d750 ME |
520 | if (block->b_fl != NULL |
521 | && block->b_fl->fl_type != F_UNLCK) { | |
522 | lock->fl = *block->b_fl; | |
523 | goto conf_lock; | |
29dbf546 | 524 | } else { |
29dbf546 OD |
525 | ret = nlm_granted; |
526 | goto out; | |
5ea0d750 ME |
527 | } |
528 | } | |
29dbf546 OD |
529 | ret = nlm_drop_reply; |
530 | goto out; | |
1da177e4 LT |
531 | } |
532 | ||
b2b50289 BF |
533 | if (locks_in_grace()) { |
534 | ret = nlm_lck_denied_grace_period; | |
535 | goto out; | |
536 | } | |
5ea0d750 | 537 | error = vfs_test_lock(file->f_file, &lock->fl); |
bde74e4b | 538 | if (error == FILE_LOCK_DEFERRED) { |
29dbf546 OD |
539 | ret = nlmsvc_defer_lock_rqst(rqstp, block); |
540 | goto out; | |
541 | } | |
5ea0d750 ME |
542 | if (error) { |
543 | ret = nlm_lck_denied_nolocks; | |
544 | goto out; | |
545 | } | |
546 | if (lock->fl.fl_type == F_UNLCK) { | |
547 | ret = nlm_granted; | |
548 | goto out; | |
549 | } | |
550 | ||
551 | conf_lock: | |
552 | dprintk("lockd: conflicting lock(ty=%d, %Ld-%Ld)\n", | |
553 | lock->fl.fl_type, (long long)lock->fl.fl_start, | |
554 | (long long)lock->fl.fl_end); | |
555 | conflock->caller = "somehost"; /* FIXME */ | |
556 | conflock->len = strlen(conflock->caller); | |
557 | conflock->oh.len = 0; /* don't return OH info */ | |
558 | conflock->svid = lock->fl.fl_pid; | |
559 | conflock->fl.fl_type = lock->fl.fl_type; | |
560 | conflock->fl.fl_start = lock->fl.fl_start; | |
561 | conflock->fl.fl_end = lock->fl.fl_end; | |
562 | ret = nlm_lck_denied; | |
563 | out: | |
564 | if (block) | |
565 | nlmsvc_release_block(block); | |
566 | return ret; | |
1da177e4 LT |
567 | } |
568 | ||
569 | /* | |
570 | * Remove a lock. | |
571 | * This implies a CANCEL call: We send a GRANT_MSG, the client replies | |
572 | * with a GRANT_RES call which gets lost, and calls UNLOCK immediately | |
573 | * afterwards. In this case the block will still be there, and hence | |
574 | * must be removed. | |
575 | */ | |
52921e02 | 576 | __be32 |
1da177e4 LT |
577 | nlmsvc_unlock(struct nlm_file *file, struct nlm_lock *lock) |
578 | { | |
579 | int error; | |
580 | ||
581 | dprintk("lockd: nlmsvc_unlock(%s/%ld, pi=%d, %Ld-%Ld)\n", | |
225a719f JS |
582 | file->f_file->f_path.dentry->d_inode->i_sb->s_id, |
583 | file->f_file->f_path.dentry->d_inode->i_ino, | |
1da177e4 LT |
584 | lock->fl.fl_pid, |
585 | (long long)lock->fl.fl_start, | |
586 | (long long)lock->fl.fl_end); | |
587 | ||
588 | /* First, cancel any lock that might be there */ | |
589 | nlmsvc_cancel_blocked(file, lock); | |
590 | ||
591 | lock->fl.fl_type = F_UNLCK; | |
1a8322b2 | 592 | error = vfs_lock_file(file->f_file, F_SETLK, &lock->fl, NULL); |
1da177e4 LT |
593 | |
594 | return (error < 0)? nlm_lck_denied_nolocks : nlm_granted; | |
595 | } | |
596 | ||
597 | /* | |
598 | * Cancel a previously blocked request. | |
599 | * | |
600 | * A cancel request always overrides any grant that may currently | |
601 | * be in progress. | |
602 | * The calling procedure must check whether the file can be closed. | |
603 | */ | |
52921e02 | 604 | __be32 |
1da177e4 LT |
605 | nlmsvc_cancel_blocked(struct nlm_file *file, struct nlm_lock *lock) |
606 | { | |
607 | struct nlm_block *block; | |
64a318ee | 608 | int status = 0; |
1da177e4 LT |
609 | |
610 | dprintk("lockd: nlmsvc_cancel(%s/%ld, pi=%d, %Ld-%Ld)\n", | |
225a719f JS |
611 | file->f_file->f_path.dentry->d_inode->i_sb->s_id, |
612 | file->f_file->f_path.dentry->d_inode->i_ino, | |
1da177e4 LT |
613 | lock->fl.fl_pid, |
614 | (long long)lock->fl.fl_start, | |
615 | (long long)lock->fl.fl_end); | |
616 | ||
b2b50289 BF |
617 | if (locks_in_grace()) |
618 | return nlm_lck_denied_grace_period; | |
619 | ||
89e63ef6 | 620 | mutex_lock(&file->f_mutex); |
d9f6eb75 | 621 | block = nlmsvc_lookup_block(file, lock); |
89e63ef6 | 622 | mutex_unlock(&file->f_mutex); |
d9f6eb75 | 623 | if (block != NULL) { |
1a8322b2 ME |
624 | vfs_cancel_lock(block->b_file->f_file, |
625 | &block->b_call->a_args.lock.fl); | |
6849c0ca TM |
626 | status = nlmsvc_unlink_block(block); |
627 | nlmsvc_release_block(block); | |
628 | } | |
64a318ee | 629 | return status ? nlm_lck_denied : nlm_granted; |
1da177e4 LT |
630 | } |
631 | ||
0e4ac9d9 ME |
632 | /* |
633 | * This is a callback from the filesystem for VFS file lock requests. | |
634 | * It will be used if fl_grant is defined and the filesystem can not | |
635 | * respond to the request immediately. | |
636 | * For GETLK request it will copy the reply to the nlm_block. | |
637 | * For SETLK or SETLKW request it will get the local posix lock. | |
638 | * In all cases it will move the block to the head of nlm_blocked q where | |
639 | * nlmsvc_retry_blocked() can send back a reply for SETLKW or revisit the | |
640 | * deferred rpc for GETLK and SETLK. | |
641 | */ | |
642 | static void | |
643 | nlmsvc_update_deferred_block(struct nlm_block *block, struct file_lock *conf, | |
644 | int result) | |
645 | { | |
646 | block->b_flags |= B_GOT_CALLBACK; | |
647 | if (result == 0) | |
648 | block->b_granted = 1; | |
649 | else | |
650 | block->b_flags |= B_TIMED_OUT; | |
651 | if (conf) { | |
0e4ac9d9 | 652 | if (block->b_fl) |
1a747ee0 | 653 | __locks_copy_lock(block->b_fl, conf); |
0e4ac9d9 ME |
654 | } |
655 | } | |
656 | ||
657 | static int nlmsvc_grant_deferred(struct file_lock *fl, struct file_lock *conf, | |
658 | int result) | |
659 | { | |
660 | struct nlm_block *block; | |
661 | int rc = -ENOENT; | |
662 | ||
f904be9c | 663 | spin_lock(&nlm_blocked_lock); |
0e4ac9d9 ME |
664 | list_for_each_entry(block, &nlm_blocked, b_list) { |
665 | if (nlm_compare_locks(&block->b_call->a_args.lock.fl, fl)) { | |
666 | dprintk("lockd: nlmsvc_notify_blocked block %p flags %d\n", | |
667 | block, block->b_flags); | |
668 | if (block->b_flags & B_QUEUED) { | |
669 | if (block->b_flags & B_TIMED_OUT) { | |
670 | rc = -ENOLCK; | |
671 | break; | |
672 | } | |
673 | nlmsvc_update_deferred_block(block, conf, result); | |
674 | } else if (result == 0) | |
675 | block->b_granted = 1; | |
676 | ||
f904be9c | 677 | nlmsvc_insert_block_locked(block, 0); |
0e4ac9d9 ME |
678 | svc_wake_up(block->b_daemon); |
679 | rc = 0; | |
680 | break; | |
681 | } | |
682 | } | |
f904be9c | 683 | spin_unlock(&nlm_blocked_lock); |
0e4ac9d9 ME |
684 | if (rc == -ENOENT) |
685 | printk(KERN_WARNING "lockd: grant for unknown block\n"); | |
686 | return rc; | |
687 | } | |
688 | ||
1da177e4 LT |
689 | /* |
690 | * Unblock a blocked lock request. This is a callback invoked from the | |
691 | * VFS layer when a lock on which we blocked is removed. | |
692 | * | |
693 | * This function doesn't grant the blocked lock instantly, but rather moves | |
694 | * the block to the head of nlm_blocked where it can be picked up by lockd. | |
695 | */ | |
696 | static void | |
697 | nlmsvc_notify_blocked(struct file_lock *fl) | |
698 | { | |
68a2d76c | 699 | struct nlm_block *block; |
1da177e4 LT |
700 | |
701 | dprintk("lockd: VFS unblock notification for block %p\n", fl); | |
a282a1fa | 702 | spin_lock(&nlm_blocked_lock); |
68a2d76c | 703 | list_for_each_entry(block, &nlm_blocked, b_list) { |
92737230 | 704 | if (nlm_compare_locks(&block->b_call->a_args.lock.fl, fl)) { |
a282a1fa BF |
705 | nlmsvc_insert_block_locked(block, 0); |
706 | spin_unlock(&nlm_blocked_lock); | |
1da177e4 LT |
707 | svc_wake_up(block->b_daemon); |
708 | return; | |
709 | } | |
710 | } | |
a282a1fa | 711 | spin_unlock(&nlm_blocked_lock); |
1da177e4 LT |
712 | printk(KERN_WARNING "lockd: notification for unknown block!\n"); |
713 | } | |
714 | ||
715 | static int nlmsvc_same_owner(struct file_lock *fl1, struct file_lock *fl2) | |
716 | { | |
717 | return fl1->fl_owner == fl2->fl_owner && fl1->fl_pid == fl2->fl_pid; | |
718 | } | |
719 | ||
7b021967 | 720 | const struct lock_manager_operations nlmsvc_lock_operations = { |
1da177e4 LT |
721 | .fl_compare_owner = nlmsvc_same_owner, |
722 | .fl_notify = nlmsvc_notify_blocked, | |
0e4ac9d9 | 723 | .fl_grant = nlmsvc_grant_deferred, |
1da177e4 LT |
724 | }; |
725 | ||
726 | /* | |
727 | * Try to claim a lock that was previously blocked. | |
728 | * | |
729 | * Note that we use both the RPC_GRANTED_MSG call _and_ an async | |
730 | * RPC thread when notifying the client. This seems like overkill... | |
731 | * Here's why: | |
732 | * - we don't want to use a synchronous RPC thread, otherwise | |
733 | * we might find ourselves hanging on a dead portmapper. | |
734 | * - Some lockd implementations (e.g. HP) don't react to | |
735 | * RPC_GRANTED calls; they seem to insist on RPC_GRANTED_MSG calls. | |
736 | */ | |
737 | static void | |
738 | nlmsvc_grant_blocked(struct nlm_block *block) | |
739 | { | |
740 | struct nlm_file *file = block->b_file; | |
92737230 | 741 | struct nlm_lock *lock = &block->b_call->a_args.lock; |
1da177e4 LT |
742 | int error; |
743 | ||
744 | dprintk("lockd: grant blocked lock %p\n", block); | |
745 | ||
0e4ac9d9 ME |
746 | kref_get(&block->b_count); |
747 | ||
1da177e4 | 748 | /* Unlink block request from list */ |
6849c0ca | 749 | nlmsvc_unlink_block(block); |
1da177e4 LT |
750 | |
751 | /* If b_granted is true this means we've been here before. | |
752 | * Just retry the grant callback, possibly refreshing the RPC | |
753 | * binding */ | |
754 | if (block->b_granted) { | |
755 | nlm_rebind_host(block->b_host); | |
756 | goto callback; | |
757 | } | |
758 | ||
759 | /* Try the lock operation again */ | |
09c7938c | 760 | lock->fl.fl_flags |= FL_SLEEP; |
1a8322b2 | 761 | error = vfs_lock_file(file->f_file, F_SETLK, &lock->fl, NULL); |
09c7938c TM |
762 | lock->fl.fl_flags &= ~FL_SLEEP; |
763 | ||
5de0e502 AA |
764 | switch (error) { |
765 | case 0: | |
766 | break; | |
bde74e4b | 767 | case FILE_LOCK_DEFERRED: |
1a8322b2 | 768 | dprintk("lockd: lock still blocked error %d\n", error); |
1da177e4 | 769 | nlmsvc_insert_block(block, NLM_NEVER); |
0e4ac9d9 | 770 | nlmsvc_release_block(block); |
d9f6eb75 | 771 | return; |
5de0e502 | 772 | default: |
1da177e4 | 773 | printk(KERN_WARNING "lockd: unexpected error %d in %s!\n", |
8e24eea7 | 774 | -error, __func__); |
1da177e4 | 775 | nlmsvc_insert_block(block, 10 * HZ); |
0e4ac9d9 | 776 | nlmsvc_release_block(block); |
d9f6eb75 | 777 | return; |
1da177e4 LT |
778 | } |
779 | ||
780 | callback: | |
781 | /* Lock was granted by VFS. */ | |
782 | dprintk("lockd: GRANTing blocked lock.\n"); | |
783 | block->b_granted = 1; | |
1da177e4 | 784 | |
9706501e JL |
785 | /* keep block on the list, but don't reattempt until the RPC |
786 | * completes or the submission fails | |
787 | */ | |
788 | nlmsvc_insert_block(block, NLM_NEVER); | |
1da177e4 | 789 | |
9706501e JL |
790 | /* Call the client -- use a soft RPC task since nlmsvc_retry_blocked |
791 | * will queue up a new one if this one times out | |
792 | */ | |
793 | error = nlm_async_call(block->b_call, NLMPROC_GRANTED_MSG, | |
794 | &nlmsvc_grant_ops); | |
795 | ||
796 | /* RPC submission failed, wait a bit and retry */ | |
797 | if (error < 0) | |
798 | nlmsvc_insert_block(block, 10 * HZ); | |
1da177e4 LT |
799 | } |
800 | ||
801 | /* | |
802 | * This is the callback from the RPC layer when the NLM_GRANTED_MSG | |
803 | * RPC call has succeeded or timed out. | |
804 | * Like all RPC callbacks, it is invoked by the rpciod process, so it | |
805 | * better not sleep. Therefore, we put the blocked lock on the nlm_blocked | |
806 | * chain once more in order to have it removed by lockd itself (which can | |
807 | * then sleep on the file semaphore without disrupting e.g. the nfs client). | |
808 | */ | |
963d8fe5 | 809 | static void nlmsvc_grant_callback(struct rpc_task *task, void *data) |
1da177e4 | 810 | { |
963d8fe5 | 811 | struct nlm_rqst *call = data; |
92737230 | 812 | struct nlm_block *block = call->a_block; |
1da177e4 | 813 | unsigned long timeout; |
1da177e4 LT |
814 | |
815 | dprintk("lockd: GRANT_MSG RPC callback\n"); | |
1da177e4 | 816 | |
f904be9c | 817 | spin_lock(&nlm_blocked_lock); |
c64e80d5 JL |
818 | /* if the block is not on a list at this point then it has |
819 | * been invalidated. Don't try to requeue it. | |
820 | * | |
821 | * FIXME: it's possible that the block is removed from the list | |
822 | * after this check but before the nlmsvc_insert_block. In that | |
823 | * case it will be added back. Perhaps we need better locking | |
824 | * for nlm_blocked? | |
825 | */ | |
826 | if (list_empty(&block->b_list)) | |
a86dc496 | 827 | goto out; |
c64e80d5 | 828 | |
1da177e4 LT |
829 | /* Technically, we should down the file semaphore here. Since we |
830 | * move the block towards the head of the queue only, no harm | |
831 | * can be done, though. */ | |
832 | if (task->tk_status < 0) { | |
833 | /* RPC error: Re-insert for retransmission */ | |
834 | timeout = 10 * HZ; | |
1da177e4 LT |
835 | } else { |
836 | /* Call was successful, now wait for client callback */ | |
837 | timeout = 60 * HZ; | |
838 | } | |
f904be9c | 839 | nlmsvc_insert_block_locked(block, timeout); |
1da177e4 | 840 | svc_wake_up(block->b_daemon); |
a86dc496 | 841 | out: |
f904be9c | 842 | spin_unlock(&nlm_blocked_lock); |
5e1abf8c TM |
843 | } |
844 | ||
f904be9c BS |
845 | /* |
846 | * FIXME: nlmsvc_release_block() grabs a mutex. This is not allowed for an | |
847 | * .rpc_release rpc_call_op | |
848 | */ | |
ec535ce1 | 849 | static void nlmsvc_grant_release(void *data) |
5e1abf8c | 850 | { |
6041b791 | 851 | struct nlm_rqst *call = data; |
6041b791 | 852 | nlmsvc_release_block(call->a_block); |
1da177e4 LT |
853 | } |
854 | ||
963d8fe5 TM |
855 | static const struct rpc_call_ops nlmsvc_grant_ops = { |
856 | .rpc_call_done = nlmsvc_grant_callback, | |
5e1abf8c | 857 | .rpc_release = nlmsvc_grant_release, |
963d8fe5 TM |
858 | }; |
859 | ||
1da177e4 LT |
860 | /* |
861 | * We received a GRANT_RES callback. Try to find the corresponding | |
862 | * block. | |
863 | */ | |
864 | void | |
e8c5c045 | 865 | nlmsvc_grant_reply(struct nlm_cookie *cookie, __be32 status) |
1da177e4 LT |
866 | { |
867 | struct nlm_block *block; | |
1da177e4 | 868 | |
39be4502 OK |
869 | dprintk("grant_reply: looking for cookie %x, s=%d \n", |
870 | *(unsigned int *)(cookie->data), status); | |
871 | if (!(block = nlmsvc_find_block(cookie))) | |
1da177e4 | 872 | return; |
1da177e4 | 873 | |
f232142c | 874 | if (block) { |
e8c5c045 | 875 | if (status == nlm_lck_denied_grace_period) { |
1da177e4 LT |
876 | /* Try again in a couple of seconds */ |
877 | nlmsvc_insert_block(block, 10 * HZ); | |
1da177e4 LT |
878 | } else { |
879 | /* Lock is now held by client, or has been rejected. | |
880 | * In both cases, the block should be removed. */ | |
6849c0ca | 881 | nlmsvc_unlink_block(block); |
1da177e4 LT |
882 | } |
883 | } | |
6849c0ca | 884 | nlmsvc_release_block(block); |
1da177e4 LT |
885 | } |
886 | ||
0e4ac9d9 ME |
887 | /* Helper function to handle retry of a deferred block. |
888 | * If it is a blocking lock, call grant_blocked. | |
889 | * For a non-blocking lock or test lock, revisit the request. | |
890 | */ | |
891 | static void | |
892 | retry_deferred_block(struct nlm_block *block) | |
893 | { | |
894 | if (!(block->b_flags & B_GOT_CALLBACK)) | |
895 | block->b_flags |= B_TIMED_OUT; | |
896 | nlmsvc_insert_block(block, NLM_TIMEOUT); | |
897 | dprintk("revisit block %p flags %d\n", block, block->b_flags); | |
898 | if (block->b_deferred_req) { | |
899 | block->b_deferred_req->revisit(block->b_deferred_req, 0); | |
900 | block->b_deferred_req = NULL; | |
901 | } | |
902 | } | |
903 | ||
1da177e4 LT |
904 | /* |
905 | * Retry all blocked locks that have been notified. This is where lockd | |
906 | * picks up locks that can be granted, or grant notifications that must | |
907 | * be retransmitted. | |
908 | */ | |
909 | unsigned long | |
910 | nlmsvc_retry_blocked(void) | |
911 | { | |
68a2d76c OK |
912 | unsigned long timeout = MAX_SCHEDULE_TIMEOUT; |
913 | struct nlm_block *block; | |
914 | ||
d751a7cd | 915 | while (!list_empty(&nlm_blocked) && !kthread_should_stop()) { |
68a2d76c | 916 | block = list_entry(nlm_blocked.next, struct nlm_block, b_list); |
1da177e4 | 917 | |
1da177e4 LT |
918 | if (block->b_when == NLM_NEVER) |
919 | break; | |
6d7bbbba | 920 | if (time_after(block->b_when, jiffies)) { |
68a2d76c | 921 | timeout = block->b_when - jiffies; |
1da177e4 | 922 | break; |
68a2d76c OK |
923 | } |
924 | ||
f3d43c76 BF |
925 | dprintk("nlmsvc_retry_blocked(%p, when=%ld)\n", |
926 | block, block->b_when); | |
0e4ac9d9 ME |
927 | if (block->b_flags & B_QUEUED) { |
928 | dprintk("nlmsvc_retry_blocked delete block (%p, granted=%d, flags=%d)\n", | |
929 | block, block->b_granted, block->b_flags); | |
930 | retry_deferred_block(block); | |
931 | } else | |
932 | nlmsvc_grant_blocked(block); | |
1da177e4 LT |
933 | } |
934 | ||
68a2d76c | 935 | return timeout; |
1da177e4 | 936 | } |