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 | 28 | #include <linux/sunrpc/clnt.h> |
5ccb0066 | 29 | #include <linux/sunrpc/svc_xprt.h> |
1da177e4 LT |
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 | 55 | |
10b89567 | 56 | #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) |
ffa94db6 TM |
57 | static const char *nlmdbg_cookie2a(const struct nlm_cookie *cookie) |
58 | { | |
59 | /* | |
3c519914 JL |
60 | * We can get away with a static buffer because this is only called |
61 | * from lockd, which is single-threaded. | |
ffa94db6 TM |
62 | */ |
63 | static char buf[2*NLM_MAXCOOKIELEN+1]; | |
64 | unsigned int i, len = sizeof(buf); | |
65 | char *p = buf; | |
66 | ||
67 | len--; /* allow for trailing \0 */ | |
68 | if (len < 3) | |
69 | return "???"; | |
70 | for (i = 0 ; i < cookie->len ; i++) { | |
71 | if (len < 2) { | |
72 | strcpy(p-3, "..."); | |
73 | break; | |
74 | } | |
75 | sprintf(p, "%02x", cookie->data[i]); | |
76 | p += 2; | |
77 | len -= 2; | |
78 | } | |
79 | *p = '\0'; | |
80 | ||
81 | return buf; | |
82 | } | |
83 | #endif | |
84 | ||
1da177e4 LT |
85 | /* |
86 | * Insert a blocked lock into the global list | |
87 | */ | |
88 | static void | |
f904be9c | 89 | nlmsvc_insert_block_locked(struct nlm_block *block, unsigned long when) |
1da177e4 | 90 | { |
68a2d76c OK |
91 | struct nlm_block *b; |
92 | struct list_head *pos; | |
1da177e4 LT |
93 | |
94 | dprintk("lockd: nlmsvc_insert_block(%p, %ld)\n", block, when); | |
68a2d76c OK |
95 | if (list_empty(&block->b_list)) { |
96 | kref_get(&block->b_count); | |
97 | } else { | |
98 | list_del_init(&block->b_list); | |
99 | } | |
100 | ||
101 | pos = &nlm_blocked; | |
1da177e4 LT |
102 | if (when != NLM_NEVER) { |
103 | if ((when += jiffies) == NLM_NEVER) | |
104 | when ++; | |
68a2d76c OK |
105 | list_for_each(pos, &nlm_blocked) { |
106 | b = list_entry(pos, struct nlm_block, b_list); | |
107 | if (time_after(b->b_when,when) || b->b_when == NLM_NEVER) | |
108 | break; | |
109 | } | |
110 | /* On normal exit from the loop, pos == &nlm_blocked, | |
111 | * so we will be adding to the end of the list - good | |
112 | */ | |
113 | } | |
1da177e4 | 114 | |
68a2d76c | 115 | list_add_tail(&block->b_list, pos); |
1da177e4 | 116 | block->b_when = when; |
1da177e4 LT |
117 | } |
118 | ||
f904be9c BS |
119 | static void nlmsvc_insert_block(struct nlm_block *block, unsigned long when) |
120 | { | |
121 | spin_lock(&nlm_blocked_lock); | |
122 | nlmsvc_insert_block_locked(block, when); | |
123 | spin_unlock(&nlm_blocked_lock); | |
124 | } | |
125 | ||
1da177e4 LT |
126 | /* |
127 | * Remove a block from the global list | |
128 | */ | |
68a2d76c | 129 | static inline void |
1da177e4 LT |
130 | nlmsvc_remove_block(struct nlm_block *block) |
131 | { | |
68a2d76c | 132 | if (!list_empty(&block->b_list)) { |
f904be9c | 133 | spin_lock(&nlm_blocked_lock); |
68a2d76c | 134 | list_del_init(&block->b_list); |
f904be9c | 135 | spin_unlock(&nlm_blocked_lock); |
68a2d76c | 136 | nlmsvc_release_block(block); |
1da177e4 | 137 | } |
1da177e4 LT |
138 | } |
139 | ||
140 | /* | |
d9f6eb75 | 141 | * Find a block for a given lock |
1da177e4 LT |
142 | */ |
143 | static struct nlm_block * | |
d9f6eb75 | 144 | nlmsvc_lookup_block(struct nlm_file *file, struct nlm_lock *lock) |
1da177e4 | 145 | { |
68a2d76c | 146 | struct nlm_block *block; |
1da177e4 LT |
147 | struct file_lock *fl; |
148 | ||
149 | dprintk("lockd: nlmsvc_lookup_block f=%p pd=%d %Ld-%Ld ty=%d\n", | |
150 | file, lock->fl.fl_pid, | |
151 | (long long)lock->fl.fl_start, | |
152 | (long long)lock->fl.fl_end, lock->fl.fl_type); | |
68a2d76c | 153 | list_for_each_entry(block, &nlm_blocked, b_list) { |
92737230 | 154 | fl = &block->b_call->a_args.lock.fl; |
1da177e4 LT |
155 | dprintk("lockd: check f=%p pd=%d %Ld-%Ld ty=%d cookie=%s\n", |
156 | block->b_file, fl->fl_pid, | |
157 | (long long)fl->fl_start, | |
158 | (long long)fl->fl_end, fl->fl_type, | |
92737230 | 159 | nlmdbg_cookie2a(&block->b_call->a_args.cookie)); |
1da177e4 | 160 | if (block->b_file == file && nlm_compare_locks(fl, &lock->fl)) { |
6849c0ca | 161 | kref_get(&block->b_count); |
1da177e4 LT |
162 | return block; |
163 | } | |
164 | } | |
165 | ||
166 | return NULL; | |
167 | } | |
168 | ||
169 | static inline int nlm_cookie_match(struct nlm_cookie *a, struct nlm_cookie *b) | |
170 | { | |
6d7bbbba | 171 | if (a->len != b->len) |
1da177e4 | 172 | return 0; |
6d7bbbba | 173 | if (memcmp(a->data, b->data, a->len)) |
1da177e4 LT |
174 | return 0; |
175 | return 1; | |
176 | } | |
177 | ||
178 | /* | |
179 | * Find a block with a given NLM cookie. | |
180 | */ | |
181 | static inline struct nlm_block * | |
39be4502 | 182 | nlmsvc_find_block(struct nlm_cookie *cookie) |
1da177e4 LT |
183 | { |
184 | struct nlm_block *block; | |
185 | ||
68a2d76c | 186 | list_for_each_entry(block, &nlm_blocked, b_list) { |
39be4502 | 187 | if (nlm_cookie_match(&block->b_call->a_args.cookie,cookie)) |
68a2d76c | 188 | goto found; |
1da177e4 LT |
189 | } |
190 | ||
68a2d76c OK |
191 | return NULL; |
192 | ||
193 | found: | |
39be4502 | 194 | dprintk("nlmsvc_find_block(%s): block=%p\n", nlmdbg_cookie2a(cookie), block); |
68a2d76c | 195 | kref_get(&block->b_count); |
1da177e4 LT |
196 | return block; |
197 | } | |
198 | ||
199 | /* | |
200 | * Create a block and initialize it. | |
201 | * | |
202 | * Note: we explicitly set the cookie of the grant reply to that of | |
203 | * the blocked lock request. The spec explicitly mentions that the client | |
204 | * should _not_ rely on the callback containing the same cookie as the | |
205 | * request, but (as I found out later) that's because some implementations | |
206 | * do just this. Never mind the standards comittees, they support our | |
207 | * logging industries. | |
39be4502 OK |
208 | * |
209 | * 10 years later: I hope we can safely ignore these old and broken | |
210 | * clients by now. Let's fix this so we can uniquely identify an incoming | |
211 | * GRANTED_RES message by cookie, without having to rely on the client's IP | |
212 | * address. --okir | |
1da177e4 | 213 | */ |
255129d1 TM |
214 | static struct nlm_block * |
215 | nlmsvc_create_block(struct svc_rqst *rqstp, struct nlm_host *host, | |
216 | struct nlm_file *file, struct nlm_lock *lock, | |
217 | struct nlm_cookie *cookie) | |
1da177e4 LT |
218 | { |
219 | struct nlm_block *block; | |
92737230 | 220 | struct nlm_rqst *call = NULL; |
1da177e4 | 221 | |
92737230 TM |
222 | call = nlm_alloc_call(host); |
223 | if (call == NULL) | |
224 | return NULL; | |
225 | ||
1da177e4 | 226 | /* Allocate memory for block, and initialize arguments */ |
92737230 TM |
227 | block = kzalloc(sizeof(*block), GFP_KERNEL); |
228 | if (block == NULL) | |
1da177e4 | 229 | goto failed; |
6849c0ca | 230 | kref_init(&block->b_count); |
68a2d76c OK |
231 | INIT_LIST_HEAD(&block->b_list); |
232 | INIT_LIST_HEAD(&block->b_flist); | |
1da177e4 | 233 | |
92737230 | 234 | if (!nlmsvc_setgrantargs(call, lock)) |
1da177e4 LT |
235 | goto failed_free; |
236 | ||
237 | /* Set notifier function for VFS, and init args */ | |
92737230 TM |
238 | call->a_args.lock.fl.fl_flags |= FL_SLEEP; |
239 | call->a_args.lock.fl.fl_lmops = &nlmsvc_lock_operations; | |
39be4502 | 240 | nlmclnt_next_cookie(&call->a_args.cookie); |
1da177e4 LT |
241 | |
242 | dprintk("lockd: created block %p...\n", block); | |
243 | ||
244 | /* Create and initialize the block */ | |
245 | block->b_daemon = rqstp->rq_server; | |
246 | block->b_host = host; | |
247 | block->b_file = file; | |
d9f6eb75 | 248 | file->f_count++; |
1da177e4 LT |
249 | |
250 | /* Add to file's list of blocks */ | |
68a2d76c | 251 | list_add(&block->b_flist, &file->f_blocks); |
1da177e4 LT |
252 | |
253 | /* Set up RPC arguments for callback */ | |
92737230 | 254 | block->b_call = call; |
1da177e4 | 255 | call->a_flags = RPC_TASK_ASYNC; |
92737230 | 256 | call->a_block = block; |
1da177e4 LT |
257 | |
258 | return block; | |
259 | ||
260 | failed_free: | |
261 | kfree(block); | |
262 | failed: | |
7db836d4 | 263 | nlmsvc_release_call(call); |
1da177e4 LT |
264 | return NULL; |
265 | } | |
266 | ||
267 | /* | |
3c61eecb | 268 | * Delete a block. |
1da177e4 LT |
269 | * It is the caller's responsibility to check whether the file |
270 | * can be closed hereafter. | |
271 | */ | |
6849c0ca | 272 | static int nlmsvc_unlink_block(struct nlm_block *block) |
1da177e4 | 273 | { |
09c7938c | 274 | int status; |
6849c0ca | 275 | dprintk("lockd: unlinking block %p...\n", block); |
1da177e4 LT |
276 | |
277 | /* Remove block from list */ | |
f891a29f | 278 | status = posix_unblock_lock(&block->b_call->a_args.lock.fl); |
1da177e4 | 279 | nlmsvc_remove_block(block); |
6849c0ca TM |
280 | return status; |
281 | } | |
1da177e4 | 282 | |
6849c0ca TM |
283 | static void nlmsvc_free_block(struct kref *kref) |
284 | { | |
285 | struct nlm_block *block = container_of(kref, struct nlm_block, b_count); | |
286 | struct nlm_file *file = block->b_file; | |
6849c0ca TM |
287 | |
288 | dprintk("lockd: freeing block %p...\n", block); | |
1da177e4 LT |
289 | |
290 | /* Remove block from file's list of blocks */ | |
68a2d76c | 291 | list_del_init(&block->b_flist); |
89e63ef6 | 292 | mutex_unlock(&file->f_mutex); |
1da177e4 | 293 | |
92737230 | 294 | nlmsvc_freegrantargs(block->b_call); |
7db836d4 | 295 | nlmsvc_release_call(block->b_call); |
d9f6eb75 | 296 | nlm_release_file(block->b_file); |
1da177e4 | 297 | kfree(block); |
6849c0ca TM |
298 | } |
299 | ||
300 | static void nlmsvc_release_block(struct nlm_block *block) | |
301 | { | |
302 | if (block != NULL) | |
c5aa1e55 | 303 | kref_put_mutex(&block->b_count, nlmsvc_free_block, &block->b_file->f_mutex); |
1da177e4 LT |
304 | } |
305 | ||
f2af793d OK |
306 | /* |
307 | * Loop over all blocks and delete blocks held by | |
308 | * a matching host. | |
309 | */ | |
310 | void nlmsvc_traverse_blocks(struct nlm_host *host, | |
311 | struct nlm_file *file, | |
312 | nlm_host_match_fn_t match) | |
d9f6eb75 | 313 | { |
68a2d76c | 314 | struct nlm_block *block, *next; |
d9f6eb75 TM |
315 | |
316 | restart: | |
89e63ef6 | 317 | mutex_lock(&file->f_mutex); |
68a2d76c | 318 | list_for_each_entry_safe(block, next, &file->f_blocks, b_flist) { |
f2af793d | 319 | if (!match(block->b_host, host)) |
d9f6eb75 | 320 | continue; |
68a2d76c OK |
321 | /* Do not destroy blocks that are not on |
322 | * the global retry list - why? */ | |
323 | if (list_empty(&block->b_list)) | |
d9f6eb75 TM |
324 | continue; |
325 | kref_get(&block->b_count); | |
89e63ef6 | 326 | mutex_unlock(&file->f_mutex); |
d9f6eb75 TM |
327 | nlmsvc_unlink_block(block); |
328 | nlmsvc_release_block(block); | |
329 | goto restart; | |
330 | } | |
89e63ef6 | 331 | mutex_unlock(&file->f_mutex); |
d9f6eb75 TM |
332 | } |
333 | ||
5e1abf8c TM |
334 | /* |
335 | * Initialize arguments for GRANTED call. The nlm_rqst structure | |
336 | * has been cleared already. | |
337 | */ | |
338 | static int nlmsvc_setgrantargs(struct nlm_rqst *call, struct nlm_lock *lock) | |
339 | { | |
340 | locks_copy_lock(&call->a_args.lock.fl, &lock->fl); | |
341 | memcpy(&call->a_args.lock.fh, &lock->fh, sizeof(call->a_args.lock.fh)); | |
e9ff3990 | 342 | call->a_args.lock.caller = utsname()->nodename; |
5e1abf8c TM |
343 | call->a_args.lock.oh.len = lock->oh.len; |
344 | ||
345 | /* set default data area */ | |
346 | call->a_args.lock.oh.data = call->a_owner; | |
347 | call->a_args.lock.svid = lock->fl.fl_pid; | |
348 | ||
349 | if (lock->oh.len > NLMCLNT_OHSIZE) { | |
350 | void *data = kmalloc(lock->oh.len, GFP_KERNEL); | |
92737230 | 351 | if (!data) |
5e1abf8c | 352 | return 0; |
5e1abf8c TM |
353 | call->a_args.lock.oh.data = (u8 *) data; |
354 | } | |
355 | ||
356 | memcpy(call->a_args.lock.oh.data, lock->oh.data, lock->oh.len); | |
357 | return 1; | |
358 | } | |
359 | ||
360 | static void nlmsvc_freegrantargs(struct nlm_rqst *call) | |
361 | { | |
92737230 | 362 | if (call->a_args.lock.oh.data != call->a_owner) |
5e1abf8c | 363 | kfree(call->a_args.lock.oh.data); |
a9e61e25 FB |
364 | |
365 | locks_release_private(&call->a_args.lock.fl); | |
5e1abf8c TM |
366 | } |
367 | ||
2b36f412 ME |
368 | /* |
369 | * Deferred lock request handling for non-blocking lock | |
370 | */ | |
ca5c8cde | 371 | static __be32 |
2b36f412 ME |
372 | nlmsvc_defer_lock_rqst(struct svc_rqst *rqstp, struct nlm_block *block) |
373 | { | |
ca5c8cde | 374 | __be32 status = nlm_lck_denied_nolocks; |
2b36f412 ME |
375 | |
376 | block->b_flags |= B_QUEUED; | |
377 | ||
378 | nlmsvc_insert_block(block, NLM_TIMEOUT); | |
379 | ||
380 | block->b_cache_req = &rqstp->rq_chandle; | |
381 | if (rqstp->rq_chandle.defer) { | |
382 | block->b_deferred_req = | |
383 | rqstp->rq_chandle.defer(block->b_cache_req); | |
384 | if (block->b_deferred_req != NULL) | |
385 | status = nlm_drop_reply; | |
386 | } | |
387 | dprintk("lockd: nlmsvc_defer_lock_rqst block %p flags %d status %d\n", | |
ca5c8cde | 388 | block, block->b_flags, ntohl(status)); |
2b36f412 ME |
389 | |
390 | return status; | |
391 | } | |
392 | ||
1da177e4 LT |
393 | /* |
394 | * Attempt to establish a lock, and if it can't be granted, block it | |
395 | * if required. | |
396 | */ | |
52921e02 | 397 | __be32 |
1da177e4 | 398 | nlmsvc_lock(struct svc_rqst *rqstp, struct nlm_file *file, |
6cde4de8 | 399 | struct nlm_host *host, struct nlm_lock *lock, int wait, |
b2b50289 | 400 | struct nlm_cookie *cookie, int reclaim) |
1da177e4 | 401 | { |
f8120480 | 402 | struct nlm_block *block = NULL; |
1da177e4 | 403 | int error; |
52921e02 | 404 | __be32 ret; |
1da177e4 LT |
405 | |
406 | dprintk("lockd: nlmsvc_lock(%s/%ld, ty=%d, pi=%d, %Ld-%Ld, bl=%d)\n", | |
496ad9aa AV |
407 | file_inode(file->f_file)->i_sb->s_id, |
408 | file_inode(file->f_file)->i_ino, | |
1da177e4 LT |
409 | lock->fl.fl_type, lock->fl.fl_pid, |
410 | (long long)lock->fl.fl_start, | |
411 | (long long)lock->fl.fl_end, | |
412 | wait); | |
413 | ||
1da177e4 | 414 | /* Lock file against concurrent access */ |
89e63ef6 | 415 | mutex_lock(&file->f_mutex); |
f8120480 ME |
416 | /* Get existing block (in case client is busy-waiting) |
417 | * or create new block | |
418 | */ | |
d9f6eb75 | 419 | block = nlmsvc_lookup_block(file, lock); |
09c7938c | 420 | if (block == NULL) { |
560de0e6 | 421 | block = nlmsvc_create_block(rqstp, host, file, lock, cookie); |
f8120480 ME |
422 | ret = nlm_lck_denied_nolocks; |
423 | if (block == NULL) | |
424 | goto out; | |
92737230 | 425 | lock = &block->b_call->a_args.lock; |
f8120480 ME |
426 | } else |
427 | lock->fl.fl_flags &= ~FL_SLEEP; | |
1da177e4 | 428 | |
1a8322b2 ME |
429 | if (block->b_flags & B_QUEUED) { |
430 | dprintk("lockd: nlmsvc_lock deferred block %p flags %d\n", | |
431 | block, block->b_flags); | |
432 | if (block->b_granted) { | |
433 | nlmsvc_unlink_block(block); | |
434 | ret = nlm_granted; | |
435 | goto out; | |
436 | } | |
437 | if (block->b_flags & B_TIMED_OUT) { | |
438 | nlmsvc_unlink_block(block); | |
439 | ret = nlm_lck_denied; | |
440 | goto out; | |
441 | } | |
442 | ret = nlm_drop_reply; | |
443 | goto out; | |
444 | } | |
1da177e4 | 445 | |
5ccb0066 | 446 | if (locks_in_grace(SVC_NET(rqstp)) && !reclaim) { |
b2b50289 BF |
447 | ret = nlm_lck_denied_grace_period; |
448 | goto out; | |
449 | } | |
5ccb0066 | 450 | if (reclaim && !locks_in_grace(SVC_NET(rqstp))) { |
d22b1cff BF |
451 | ret = nlm_lck_denied_grace_period; |
452 | goto out; | |
453 | } | |
b2b50289 | 454 | |
1a8322b2 ME |
455 | if (!wait) |
456 | lock->fl.fl_flags &= ~FL_SLEEP; | |
457 | error = vfs_lock_file(file->f_file, F_SETLK, &lock->fl, NULL); | |
458 | lock->fl.fl_flags &= ~FL_SLEEP; | |
a85f193e | 459 | |
1a8322b2 | 460 | dprintk("lockd: vfs_lock_file returned %d\n", error); |
6d7bbbba | 461 | switch (error) { |
1da177e4 | 462 | case 0: |
15dadef9 AA |
463 | ret = nlm_granted; |
464 | goto out; | |
09c7938c | 465 | case -EAGAIN: |
e33d1ea6 MS |
466 | /* |
467 | * If this is a blocking request for an | |
468 | * already pending lock request then we need | |
469 | * to put it back on lockd's block list | |
470 | */ | |
471 | if (wait) | |
472 | break; | |
1a8322b2 | 473 | ret = nlm_lck_denied; |
e33d1ea6 | 474 | goto out; |
bde74e4b | 475 | case FILE_LOCK_DEFERRED: |
1a8322b2 ME |
476 | if (wait) |
477 | break; | |
478 | /* Filesystem lock operation is in progress | |
479 | Add it to the queue waiting for callback */ | |
480 | ret = nlmsvc_defer_lock_rqst(rqstp, block); | |
481 | goto out; | |
09c7938c | 482 | case -EDEADLK: |
15dadef9 AA |
483 | ret = nlm_deadlock; |
484 | goto out; | |
1da177e4 | 485 | default: /* includes ENOLCK */ |
15dadef9 AA |
486 | ret = nlm_lck_denied_nolocks; |
487 | goto out; | |
1da177e4 LT |
488 | } |
489 | ||
09c7938c | 490 | ret = nlm_lck_blocked; |
1da177e4 LT |
491 | |
492 | /* Append to list of blocked */ | |
f8120480 | 493 | nlmsvc_insert_block(block, NLM_NEVER); |
15dadef9 | 494 | out: |
89e63ef6 | 495 | mutex_unlock(&file->f_mutex); |
6849c0ca | 496 | nlmsvc_release_block(block); |
15dadef9 AA |
497 | dprintk("lockd: nlmsvc_lock returned %u\n", ret); |
498 | return ret; | |
1da177e4 LT |
499 | } |
500 | ||
501 | /* | |
502 | * Test for presence of a conflicting lock. | |
503 | */ | |
52921e02 | 504 | __be32 |
85f3f1b3 | 505 | nlmsvc_testlock(struct svc_rqst *rqstp, struct nlm_file *file, |
8f920d5e JL |
506 | struct nlm_host *host, struct nlm_lock *lock, |
507 | struct nlm_lock *conflock, struct nlm_cookie *cookie) | |
1da177e4 | 508 | { |
5ea0d750 ME |
509 | int error; |
510 | __be32 ret; | |
511 | ||
1da177e4 | 512 | dprintk("lockd: nlmsvc_testlock(%s/%ld, ty=%d, %Ld-%Ld)\n", |
496ad9aa AV |
513 | file_inode(file->f_file)->i_sb->s_id, |
514 | file_inode(file->f_file)->i_ino, | |
1da177e4 LT |
515 | lock->fl.fl_type, |
516 | (long long)lock->fl.fl_start, | |
517 | (long long)lock->fl.fl_end); | |
518 | ||
5ccb0066 | 519 | if (locks_in_grace(SVC_NET(rqstp))) { |
b2b50289 BF |
520 | ret = nlm_lck_denied_grace_period; |
521 | goto out; | |
522 | } | |
09802fd2 | 523 | |
5ea0d750 | 524 | error = vfs_test_lock(file->f_file, &lock->fl); |
5ea0d750 | 525 | if (error) { |
09802fd2 JL |
526 | /* We can't currently deal with deferred test requests */ |
527 | if (error == FILE_LOCK_DEFERRED) | |
528 | WARN_ON_ONCE(1); | |
529 | ||
5ea0d750 ME |
530 | ret = nlm_lck_denied_nolocks; |
531 | goto out; | |
532 | } | |
09802fd2 | 533 | |
5ea0d750 ME |
534 | if (lock->fl.fl_type == F_UNLCK) { |
535 | ret = nlm_granted; | |
536 | goto out; | |
537 | } | |
538 | ||
5ea0d750 ME |
539 | dprintk("lockd: conflicting lock(ty=%d, %Ld-%Ld)\n", |
540 | lock->fl.fl_type, (long long)lock->fl.fl_start, | |
541 | (long long)lock->fl.fl_end); | |
542 | conflock->caller = "somehost"; /* FIXME */ | |
543 | conflock->len = strlen(conflock->caller); | |
544 | conflock->oh.len = 0; /* don't return OH info */ | |
545 | conflock->svid = lock->fl.fl_pid; | |
546 | conflock->fl.fl_type = lock->fl.fl_type; | |
547 | conflock->fl.fl_start = lock->fl.fl_start; | |
548 | conflock->fl.fl_end = lock->fl.fl_end; | |
f328296e | 549 | locks_release_private(&lock->fl); |
5ea0d750 ME |
550 | ret = nlm_lck_denied; |
551 | out: | |
5ea0d750 | 552 | return ret; |
1da177e4 LT |
553 | } |
554 | ||
555 | /* | |
556 | * Remove a lock. | |
557 | * This implies a CANCEL call: We send a GRANT_MSG, the client replies | |
558 | * with a GRANT_RES call which gets lost, and calls UNLOCK immediately | |
559 | * afterwards. In this case the block will still be there, and hence | |
560 | * must be removed. | |
561 | */ | |
52921e02 | 562 | __be32 |
5ccb0066 | 563 | nlmsvc_unlock(struct net *net, struct nlm_file *file, struct nlm_lock *lock) |
1da177e4 LT |
564 | { |
565 | int error; | |
566 | ||
567 | dprintk("lockd: nlmsvc_unlock(%s/%ld, pi=%d, %Ld-%Ld)\n", | |
496ad9aa AV |
568 | file_inode(file->f_file)->i_sb->s_id, |
569 | file_inode(file->f_file)->i_ino, | |
1da177e4 LT |
570 | lock->fl.fl_pid, |
571 | (long long)lock->fl.fl_start, | |
572 | (long long)lock->fl.fl_end); | |
573 | ||
574 | /* First, cancel any lock that might be there */ | |
5ccb0066 | 575 | nlmsvc_cancel_blocked(net, file, lock); |
1da177e4 LT |
576 | |
577 | lock->fl.fl_type = F_UNLCK; | |
1a8322b2 | 578 | error = vfs_lock_file(file->f_file, F_SETLK, &lock->fl, NULL); |
1da177e4 LT |
579 | |
580 | return (error < 0)? nlm_lck_denied_nolocks : nlm_granted; | |
581 | } | |
582 | ||
583 | /* | |
584 | * Cancel a previously blocked request. | |
585 | * | |
586 | * A cancel request always overrides any grant that may currently | |
587 | * be in progress. | |
588 | * The calling procedure must check whether the file can be closed. | |
589 | */ | |
52921e02 | 590 | __be32 |
5ccb0066 | 591 | nlmsvc_cancel_blocked(struct net *net, struct nlm_file *file, struct nlm_lock *lock) |
1da177e4 LT |
592 | { |
593 | struct nlm_block *block; | |
64a318ee | 594 | int status = 0; |
1da177e4 LT |
595 | |
596 | dprintk("lockd: nlmsvc_cancel(%s/%ld, pi=%d, %Ld-%Ld)\n", | |
496ad9aa AV |
597 | file_inode(file->f_file)->i_sb->s_id, |
598 | file_inode(file->f_file)->i_ino, | |
1da177e4 LT |
599 | lock->fl.fl_pid, |
600 | (long long)lock->fl.fl_start, | |
601 | (long long)lock->fl.fl_end); | |
602 | ||
5ccb0066 | 603 | if (locks_in_grace(net)) |
b2b50289 BF |
604 | return nlm_lck_denied_grace_period; |
605 | ||
89e63ef6 | 606 | mutex_lock(&file->f_mutex); |
d9f6eb75 | 607 | block = nlmsvc_lookup_block(file, lock); |
89e63ef6 | 608 | mutex_unlock(&file->f_mutex); |
d9f6eb75 | 609 | if (block != NULL) { |
1a8322b2 ME |
610 | vfs_cancel_lock(block->b_file->f_file, |
611 | &block->b_call->a_args.lock.fl); | |
6849c0ca TM |
612 | status = nlmsvc_unlink_block(block); |
613 | nlmsvc_release_block(block); | |
614 | } | |
64a318ee | 615 | return status ? nlm_lck_denied : nlm_granted; |
1da177e4 LT |
616 | } |
617 | ||
0e4ac9d9 ME |
618 | /* |
619 | * This is a callback from the filesystem for VFS file lock requests. | |
8fb47a4f | 620 | * It will be used if lm_grant is defined and the filesystem can not |
0e4ac9d9 | 621 | * respond to the request immediately. |
0e4ac9d9 ME |
622 | * For SETLK or SETLKW request it will get the local posix lock. |
623 | * In all cases it will move the block to the head of nlm_blocked q where | |
624 | * nlmsvc_retry_blocked() can send back a reply for SETLKW or revisit the | |
625 | * deferred rpc for GETLK and SETLK. | |
626 | */ | |
627 | static void | |
d0449b90 | 628 | nlmsvc_update_deferred_block(struct nlm_block *block, int result) |
0e4ac9d9 ME |
629 | { |
630 | block->b_flags |= B_GOT_CALLBACK; | |
631 | if (result == 0) | |
632 | block->b_granted = 1; | |
633 | else | |
634 | block->b_flags |= B_TIMED_OUT; | |
0e4ac9d9 ME |
635 | } |
636 | ||
d0449b90 | 637 | static int nlmsvc_grant_deferred(struct file_lock *fl, int result) |
0e4ac9d9 ME |
638 | { |
639 | struct nlm_block *block; | |
640 | int rc = -ENOENT; | |
641 | ||
f904be9c | 642 | spin_lock(&nlm_blocked_lock); |
0e4ac9d9 ME |
643 | list_for_each_entry(block, &nlm_blocked, b_list) { |
644 | if (nlm_compare_locks(&block->b_call->a_args.lock.fl, fl)) { | |
645 | dprintk("lockd: nlmsvc_notify_blocked block %p flags %d\n", | |
646 | block, block->b_flags); | |
647 | if (block->b_flags & B_QUEUED) { | |
648 | if (block->b_flags & B_TIMED_OUT) { | |
649 | rc = -ENOLCK; | |
650 | break; | |
651 | } | |
d0449b90 | 652 | nlmsvc_update_deferred_block(block, result); |
0e4ac9d9 ME |
653 | } else if (result == 0) |
654 | block->b_granted = 1; | |
655 | ||
f904be9c | 656 | nlmsvc_insert_block_locked(block, 0); |
0e4ac9d9 ME |
657 | svc_wake_up(block->b_daemon); |
658 | rc = 0; | |
659 | break; | |
660 | } | |
661 | } | |
f904be9c | 662 | spin_unlock(&nlm_blocked_lock); |
0e4ac9d9 ME |
663 | if (rc == -ENOENT) |
664 | printk(KERN_WARNING "lockd: grant for unknown block\n"); | |
665 | return rc; | |
666 | } | |
667 | ||
1da177e4 LT |
668 | /* |
669 | * Unblock a blocked lock request. This is a callback invoked from the | |
670 | * VFS layer when a lock on which we blocked is removed. | |
671 | * | |
672 | * This function doesn't grant the blocked lock instantly, but rather moves | |
673 | * the block to the head of nlm_blocked where it can be picked up by lockd. | |
674 | */ | |
675 | static void | |
676 | nlmsvc_notify_blocked(struct file_lock *fl) | |
677 | { | |
68a2d76c | 678 | struct nlm_block *block; |
1da177e4 LT |
679 | |
680 | dprintk("lockd: VFS unblock notification for block %p\n", fl); | |
a282a1fa | 681 | spin_lock(&nlm_blocked_lock); |
68a2d76c | 682 | list_for_each_entry(block, &nlm_blocked, b_list) { |
92737230 | 683 | if (nlm_compare_locks(&block->b_call->a_args.lock.fl, fl)) { |
a282a1fa BF |
684 | nlmsvc_insert_block_locked(block, 0); |
685 | spin_unlock(&nlm_blocked_lock); | |
1da177e4 LT |
686 | svc_wake_up(block->b_daemon); |
687 | return; | |
688 | } | |
689 | } | |
a282a1fa | 690 | spin_unlock(&nlm_blocked_lock); |
1da177e4 LT |
691 | printk(KERN_WARNING "lockd: notification for unknown block!\n"); |
692 | } | |
693 | ||
694 | static int nlmsvc_same_owner(struct file_lock *fl1, struct file_lock *fl2) | |
695 | { | |
696 | return fl1->fl_owner == fl2->fl_owner && fl1->fl_pid == fl2->fl_pid; | |
697 | } | |
698 | ||
3999e493 JL |
699 | /* |
700 | * Since NLM uses two "keys" for tracking locks, we need to hash them down | |
701 | * to one for the blocked_hash. Here, we're just xor'ing the host address | |
702 | * with the pid in order to create a key value for picking a hash bucket. | |
703 | */ | |
704 | static unsigned long | |
705 | nlmsvc_owner_key(struct file_lock *fl) | |
706 | { | |
707 | return (unsigned long)fl->fl_owner ^ (unsigned long)fl->fl_pid; | |
708 | } | |
709 | ||
7b021967 | 710 | const struct lock_manager_operations nlmsvc_lock_operations = { |
8fb47a4f | 711 | .lm_compare_owner = nlmsvc_same_owner, |
3999e493 | 712 | .lm_owner_key = nlmsvc_owner_key, |
8fb47a4f BF |
713 | .lm_notify = nlmsvc_notify_blocked, |
714 | .lm_grant = nlmsvc_grant_deferred, | |
1da177e4 LT |
715 | }; |
716 | ||
717 | /* | |
718 | * Try to claim a lock that was previously blocked. | |
719 | * | |
720 | * Note that we use both the RPC_GRANTED_MSG call _and_ an async | |
721 | * RPC thread when notifying the client. This seems like overkill... | |
722 | * Here's why: | |
723 | * - we don't want to use a synchronous RPC thread, otherwise | |
724 | * we might find ourselves hanging on a dead portmapper. | |
725 | * - Some lockd implementations (e.g. HP) don't react to | |
726 | * RPC_GRANTED calls; they seem to insist on RPC_GRANTED_MSG calls. | |
727 | */ | |
728 | static void | |
729 | nlmsvc_grant_blocked(struct nlm_block *block) | |
730 | { | |
731 | struct nlm_file *file = block->b_file; | |
92737230 | 732 | struct nlm_lock *lock = &block->b_call->a_args.lock; |
1da177e4 | 733 | int error; |
2ec197db | 734 | loff_t fl_start, fl_end; |
1da177e4 LT |
735 | |
736 | dprintk("lockd: grant blocked lock %p\n", block); | |
737 | ||
0e4ac9d9 ME |
738 | kref_get(&block->b_count); |
739 | ||
1da177e4 | 740 | /* Unlink block request from list */ |
6849c0ca | 741 | nlmsvc_unlink_block(block); |
1da177e4 LT |
742 | |
743 | /* If b_granted is true this means we've been here before. | |
744 | * Just retry the grant callback, possibly refreshing the RPC | |
745 | * binding */ | |
746 | if (block->b_granted) { | |
747 | nlm_rebind_host(block->b_host); | |
748 | goto callback; | |
749 | } | |
750 | ||
751 | /* Try the lock operation again */ | |
2ec197db N |
752 | /* vfs_lock_file() can mangle fl_start and fl_end, but we need |
753 | * them unchanged for the GRANT_MSG | |
754 | */ | |
09c7938c | 755 | lock->fl.fl_flags |= FL_SLEEP; |
2ec197db N |
756 | fl_start = lock->fl.fl_start; |
757 | fl_end = lock->fl.fl_end; | |
1a8322b2 | 758 | error = vfs_lock_file(file->f_file, F_SETLK, &lock->fl, NULL); |
09c7938c | 759 | lock->fl.fl_flags &= ~FL_SLEEP; |
2ec197db N |
760 | lock->fl.fl_start = fl_start; |
761 | lock->fl.fl_end = fl_end; | |
09c7938c | 762 | |
5de0e502 AA |
763 | switch (error) { |
764 | case 0: | |
765 | break; | |
bde74e4b | 766 | case FILE_LOCK_DEFERRED: |
1a8322b2 | 767 | dprintk("lockd: lock still blocked error %d\n", error); |
1da177e4 | 768 | nlmsvc_insert_block(block, NLM_NEVER); |
0e4ac9d9 | 769 | nlmsvc_release_block(block); |
d9f6eb75 | 770 | return; |
5de0e502 | 771 | default: |
1da177e4 | 772 | printk(KERN_WARNING "lockd: unexpected error %d in %s!\n", |
8e24eea7 | 773 | -error, __func__); |
1da177e4 | 774 | nlmsvc_insert_block(block, 10 * HZ); |
0e4ac9d9 | 775 | nlmsvc_release_block(block); |
d9f6eb75 | 776 | return; |
1da177e4 LT |
777 | } |
778 | ||
779 | callback: | |
780 | /* Lock was granted by VFS. */ | |
781 | dprintk("lockd: GRANTing blocked lock.\n"); | |
782 | block->b_granted = 1; | |
1da177e4 | 783 | |
9706501e JL |
784 | /* keep block on the list, but don't reattempt until the RPC |
785 | * completes or the submission fails | |
786 | */ | |
787 | nlmsvc_insert_block(block, NLM_NEVER); | |
1da177e4 | 788 | |
9706501e JL |
789 | /* Call the client -- use a soft RPC task since nlmsvc_retry_blocked |
790 | * will queue up a new one if this one times out | |
791 | */ | |
792 | error = nlm_async_call(block->b_call, NLMPROC_GRANTED_MSG, | |
793 | &nlmsvc_grant_ops); | |
794 | ||
795 | /* RPC submission failed, wait a bit and retry */ | |
796 | if (error < 0) | |
797 | nlmsvc_insert_block(block, 10 * HZ); | |
1da177e4 LT |
798 | } |
799 | ||
800 | /* | |
801 | * This is the callback from the RPC layer when the NLM_GRANTED_MSG | |
802 | * RPC call has succeeded or timed out. | |
803 | * Like all RPC callbacks, it is invoked by the rpciod process, so it | |
804 | * better not sleep. Therefore, we put the blocked lock on the nlm_blocked | |
805 | * chain once more in order to have it removed by lockd itself (which can | |
806 | * then sleep on the file semaphore without disrupting e.g. the nfs client). | |
807 | */ | |
963d8fe5 | 808 | static void nlmsvc_grant_callback(struct rpc_task *task, void *data) |
1da177e4 | 809 | { |
963d8fe5 | 810 | struct nlm_rqst *call = data; |
92737230 | 811 | struct nlm_block *block = call->a_block; |
1da177e4 | 812 | unsigned long timeout; |
1da177e4 LT |
813 | |
814 | dprintk("lockd: GRANT_MSG RPC callback\n"); | |
1da177e4 | 815 | |
f904be9c | 816 | spin_lock(&nlm_blocked_lock); |
c64e80d5 JL |
817 | /* if the block is not on a list at this point then it has |
818 | * been invalidated. Don't try to requeue it. | |
819 | * | |
820 | * FIXME: it's possible that the block is removed from the list | |
821 | * after this check but before the nlmsvc_insert_block. In that | |
822 | * case it will be added back. Perhaps we need better locking | |
823 | * for nlm_blocked? | |
824 | */ | |
825 | if (list_empty(&block->b_list)) | |
a86dc496 | 826 | goto out; |
c64e80d5 | 827 | |
1da177e4 LT |
828 | /* Technically, we should down the file semaphore here. Since we |
829 | * move the block towards the head of the queue only, no harm | |
830 | * can be done, though. */ | |
831 | if (task->tk_status < 0) { | |
832 | /* RPC error: Re-insert for retransmission */ | |
833 | timeout = 10 * HZ; | |
1da177e4 LT |
834 | } else { |
835 | /* Call was successful, now wait for client callback */ | |
836 | timeout = 60 * HZ; | |
837 | } | |
f904be9c | 838 | nlmsvc_insert_block_locked(block, timeout); |
1da177e4 | 839 | svc_wake_up(block->b_daemon); |
a86dc496 | 840 | out: |
f904be9c | 841 | spin_unlock(&nlm_blocked_lock); |
5e1abf8c TM |
842 | } |
843 | ||
f904be9c BS |
844 | /* |
845 | * FIXME: nlmsvc_release_block() grabs a mutex. This is not allowed for an | |
846 | * .rpc_release rpc_call_op | |
847 | */ | |
ec535ce1 | 848 | static void nlmsvc_grant_release(void *data) |
5e1abf8c | 849 | { |
6041b791 | 850 | struct nlm_rqst *call = data; |
6041b791 | 851 | nlmsvc_release_block(call->a_block); |
1da177e4 LT |
852 | } |
853 | ||
963d8fe5 TM |
854 | static const struct rpc_call_ops nlmsvc_grant_ops = { |
855 | .rpc_call_done = nlmsvc_grant_callback, | |
5e1abf8c | 856 | .rpc_release = nlmsvc_grant_release, |
963d8fe5 TM |
857 | }; |
858 | ||
1da177e4 LT |
859 | /* |
860 | * We received a GRANT_RES callback. Try to find the corresponding | |
861 | * block. | |
862 | */ | |
863 | void | |
e8c5c045 | 864 | nlmsvc_grant_reply(struct nlm_cookie *cookie, __be32 status) |
1da177e4 LT |
865 | { |
866 | struct nlm_block *block; | |
1da177e4 | 867 | |
39be4502 OK |
868 | dprintk("grant_reply: looking for cookie %x, s=%d \n", |
869 | *(unsigned int *)(cookie->data), status); | |
870 | if (!(block = nlmsvc_find_block(cookie))) | |
1da177e4 | 871 | return; |
1da177e4 | 872 | |
f232142c | 873 | if (block) { |
e8c5c045 | 874 | if (status == nlm_lck_denied_grace_period) { |
1da177e4 LT |
875 | /* Try again in a couple of seconds */ |
876 | nlmsvc_insert_block(block, 10 * HZ); | |
1da177e4 LT |
877 | } else { |
878 | /* Lock is now held by client, or has been rejected. | |
879 | * In both cases, the block should be removed. */ | |
6849c0ca | 880 | nlmsvc_unlink_block(block); |
1da177e4 LT |
881 | } |
882 | } | |
6849c0ca | 883 | nlmsvc_release_block(block); |
1da177e4 LT |
884 | } |
885 | ||
0e4ac9d9 ME |
886 | /* Helper function to handle retry of a deferred block. |
887 | * If it is a blocking lock, call grant_blocked. | |
888 | * For a non-blocking lock or test lock, revisit the request. | |
889 | */ | |
890 | static void | |
891 | retry_deferred_block(struct nlm_block *block) | |
892 | { | |
893 | if (!(block->b_flags & B_GOT_CALLBACK)) | |
894 | block->b_flags |= B_TIMED_OUT; | |
895 | nlmsvc_insert_block(block, NLM_TIMEOUT); | |
896 | dprintk("revisit block %p flags %d\n", block, block->b_flags); | |
897 | if (block->b_deferred_req) { | |
898 | block->b_deferred_req->revisit(block->b_deferred_req, 0); | |
899 | block->b_deferred_req = NULL; | |
900 | } | |
901 | } | |
902 | ||
1da177e4 LT |
903 | /* |
904 | * Retry all blocked locks that have been notified. This is where lockd | |
905 | * picks up locks that can be granted, or grant notifications that must | |
906 | * be retransmitted. | |
907 | */ | |
908 | unsigned long | |
909 | nlmsvc_retry_blocked(void) | |
910 | { | |
68a2d76c OK |
911 | unsigned long timeout = MAX_SCHEDULE_TIMEOUT; |
912 | struct nlm_block *block; | |
913 | ||
1c327d96 | 914 | spin_lock(&nlm_blocked_lock); |
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 | 923 | } |
1c327d96 | 924 | spin_unlock(&nlm_blocked_lock); |
68a2d76c | 925 | |
f3d43c76 BF |
926 | dprintk("nlmsvc_retry_blocked(%p, when=%ld)\n", |
927 | block, block->b_when); | |
0e4ac9d9 ME |
928 | if (block->b_flags & B_QUEUED) { |
929 | dprintk("nlmsvc_retry_blocked delete block (%p, granted=%d, flags=%d)\n", | |
930 | block, block->b_granted, block->b_flags); | |
931 | retry_deferred_block(block); | |
932 | } else | |
933 | nlmsvc_grant_blocked(block); | |
1c327d96 | 934 | spin_lock(&nlm_blocked_lock); |
1da177e4 | 935 | } |
1c327d96 | 936 | spin_unlock(&nlm_blocked_lock); |
1da177e4 | 937 | |
68a2d76c | 938 | return timeout; |
1da177e4 | 939 | } |