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; |
d8367c50 | 49 | static const char *nlmdbg_cookie2a(const struct nlm_cookie *cookie); |
1da177e4 LT |
50 | |
51 | /* | |
52 | * The list of blocked locks to retry | |
53 | */ | |
68a2d76c | 54 | static LIST_HEAD(nlm_blocked); |
f904be9c | 55 | static DEFINE_SPINLOCK(nlm_blocked_lock); |
1da177e4 LT |
56 | |
57 | /* | |
58 | * Insert a blocked lock into the global list | |
59 | */ | |
60 | static void | |
f904be9c | 61 | nlmsvc_insert_block_locked(struct nlm_block *block, unsigned long when) |
1da177e4 | 62 | { |
68a2d76c OK |
63 | struct nlm_block *b; |
64 | struct list_head *pos; | |
1da177e4 LT |
65 | |
66 | dprintk("lockd: nlmsvc_insert_block(%p, %ld)\n", block, when); | |
68a2d76c OK |
67 | if (list_empty(&block->b_list)) { |
68 | kref_get(&block->b_count); | |
69 | } else { | |
70 | list_del_init(&block->b_list); | |
71 | } | |
72 | ||
73 | pos = &nlm_blocked; | |
1da177e4 LT |
74 | if (when != NLM_NEVER) { |
75 | if ((when += jiffies) == NLM_NEVER) | |
76 | when ++; | |
68a2d76c OK |
77 | list_for_each(pos, &nlm_blocked) { |
78 | b = list_entry(pos, struct nlm_block, b_list); | |
79 | if (time_after(b->b_when,when) || b->b_when == NLM_NEVER) | |
80 | break; | |
81 | } | |
82 | /* On normal exit from the loop, pos == &nlm_blocked, | |
83 | * so we will be adding to the end of the list - good | |
84 | */ | |
85 | } | |
1da177e4 | 86 | |
68a2d76c | 87 | list_add_tail(&block->b_list, pos); |
1da177e4 | 88 | block->b_when = when; |
1da177e4 LT |
89 | } |
90 | ||
f904be9c BS |
91 | static void nlmsvc_insert_block(struct nlm_block *block, unsigned long when) |
92 | { | |
93 | spin_lock(&nlm_blocked_lock); | |
94 | nlmsvc_insert_block_locked(block, when); | |
95 | spin_unlock(&nlm_blocked_lock); | |
96 | } | |
97 | ||
1da177e4 LT |
98 | /* |
99 | * Remove a block from the global list | |
100 | */ | |
68a2d76c | 101 | static inline void |
1da177e4 LT |
102 | nlmsvc_remove_block(struct nlm_block *block) |
103 | { | |
68a2d76c | 104 | if (!list_empty(&block->b_list)) { |
f904be9c | 105 | spin_lock(&nlm_blocked_lock); |
68a2d76c | 106 | list_del_init(&block->b_list); |
f904be9c | 107 | spin_unlock(&nlm_blocked_lock); |
68a2d76c | 108 | nlmsvc_release_block(block); |
1da177e4 | 109 | } |
1da177e4 LT |
110 | } |
111 | ||
112 | /* | |
d9f6eb75 | 113 | * Find a block for a given lock |
1da177e4 LT |
114 | */ |
115 | static struct nlm_block * | |
d9f6eb75 | 116 | nlmsvc_lookup_block(struct nlm_file *file, struct nlm_lock *lock) |
1da177e4 | 117 | { |
68a2d76c | 118 | struct nlm_block *block; |
1da177e4 LT |
119 | struct file_lock *fl; |
120 | ||
121 | dprintk("lockd: nlmsvc_lookup_block f=%p pd=%d %Ld-%Ld ty=%d\n", | |
122 | file, lock->fl.fl_pid, | |
123 | (long long)lock->fl.fl_start, | |
124 | (long long)lock->fl.fl_end, lock->fl.fl_type); | |
68a2d76c | 125 | list_for_each_entry(block, &nlm_blocked, b_list) { |
92737230 | 126 | fl = &block->b_call->a_args.lock.fl; |
1da177e4 LT |
127 | dprintk("lockd: check f=%p pd=%d %Ld-%Ld ty=%d cookie=%s\n", |
128 | block->b_file, fl->fl_pid, | |
129 | (long long)fl->fl_start, | |
130 | (long long)fl->fl_end, fl->fl_type, | |
92737230 | 131 | nlmdbg_cookie2a(&block->b_call->a_args.cookie)); |
1da177e4 | 132 | if (block->b_file == file && nlm_compare_locks(fl, &lock->fl)) { |
6849c0ca | 133 | kref_get(&block->b_count); |
1da177e4 LT |
134 | return block; |
135 | } | |
136 | } | |
137 | ||
138 | return NULL; | |
139 | } | |
140 | ||
141 | static inline int nlm_cookie_match(struct nlm_cookie *a, struct nlm_cookie *b) | |
142 | { | |
6d7bbbba | 143 | if (a->len != b->len) |
1da177e4 | 144 | return 0; |
6d7bbbba | 145 | if (memcmp(a->data, b->data, a->len)) |
1da177e4 LT |
146 | return 0; |
147 | return 1; | |
148 | } | |
149 | ||
150 | /* | |
151 | * Find a block with a given NLM cookie. | |
152 | */ | |
153 | static inline struct nlm_block * | |
39be4502 | 154 | nlmsvc_find_block(struct nlm_cookie *cookie) |
1da177e4 LT |
155 | { |
156 | struct nlm_block *block; | |
157 | ||
68a2d76c | 158 | list_for_each_entry(block, &nlm_blocked, b_list) { |
39be4502 | 159 | if (nlm_cookie_match(&block->b_call->a_args.cookie,cookie)) |
68a2d76c | 160 | goto found; |
1da177e4 LT |
161 | } |
162 | ||
68a2d76c OK |
163 | return NULL; |
164 | ||
165 | found: | |
39be4502 | 166 | dprintk("nlmsvc_find_block(%s): block=%p\n", nlmdbg_cookie2a(cookie), block); |
68a2d76c | 167 | kref_get(&block->b_count); |
1da177e4 LT |
168 | return block; |
169 | } | |
170 | ||
171 | /* | |
172 | * Create a block and initialize it. | |
173 | * | |
174 | * Note: we explicitly set the cookie of the grant reply to that of | |
175 | * the blocked lock request. The spec explicitly mentions that the client | |
176 | * should _not_ rely on the callback containing the same cookie as the | |
177 | * request, but (as I found out later) that's because some implementations | |
178 | * do just this. Never mind the standards comittees, they support our | |
179 | * logging industries. | |
39be4502 OK |
180 | * |
181 | * 10 years later: I hope we can safely ignore these old and broken | |
182 | * clients by now. Let's fix this so we can uniquely identify an incoming | |
183 | * GRANTED_RES message by cookie, without having to rely on the client's IP | |
184 | * address. --okir | |
1da177e4 | 185 | */ |
255129d1 TM |
186 | static struct nlm_block * |
187 | nlmsvc_create_block(struct svc_rqst *rqstp, struct nlm_host *host, | |
188 | struct nlm_file *file, struct nlm_lock *lock, | |
189 | struct nlm_cookie *cookie) | |
1da177e4 LT |
190 | { |
191 | struct nlm_block *block; | |
92737230 | 192 | struct nlm_rqst *call = NULL; |
1da177e4 | 193 | |
560de0e6 | 194 | nlm_get_host(host); |
92737230 TM |
195 | call = nlm_alloc_call(host); |
196 | if (call == NULL) | |
197 | return NULL; | |
198 | ||
1da177e4 | 199 | /* Allocate memory for block, and initialize arguments */ |
92737230 TM |
200 | block = kzalloc(sizeof(*block), GFP_KERNEL); |
201 | if (block == NULL) | |
1da177e4 | 202 | goto failed; |
6849c0ca | 203 | kref_init(&block->b_count); |
68a2d76c OK |
204 | INIT_LIST_HEAD(&block->b_list); |
205 | INIT_LIST_HEAD(&block->b_flist); | |
1da177e4 | 206 | |
92737230 | 207 | if (!nlmsvc_setgrantargs(call, lock)) |
1da177e4 LT |
208 | goto failed_free; |
209 | ||
210 | /* Set notifier function for VFS, and init args */ | |
92737230 TM |
211 | call->a_args.lock.fl.fl_flags |= FL_SLEEP; |
212 | call->a_args.lock.fl.fl_lmops = &nlmsvc_lock_operations; | |
39be4502 | 213 | nlmclnt_next_cookie(&call->a_args.cookie); |
1da177e4 LT |
214 | |
215 | dprintk("lockd: created block %p...\n", block); | |
216 | ||
217 | /* Create and initialize the block */ | |
218 | block->b_daemon = rqstp->rq_server; | |
219 | block->b_host = host; | |
220 | block->b_file = file; | |
5ea0d750 | 221 | block->b_fl = NULL; |
d9f6eb75 | 222 | file->f_count++; |
1da177e4 LT |
223 | |
224 | /* Add to file's list of blocks */ | |
68a2d76c | 225 | list_add(&block->b_flist, &file->f_blocks); |
1da177e4 LT |
226 | |
227 | /* Set up RPC arguments for callback */ | |
92737230 | 228 | block->b_call = call; |
1da177e4 | 229 | call->a_flags = RPC_TASK_ASYNC; |
92737230 | 230 | call->a_block = block; |
1da177e4 LT |
231 | |
232 | return block; | |
233 | ||
234 | failed_free: | |
235 | kfree(block); | |
236 | failed: | |
7db836d4 | 237 | nlmsvc_release_call(call); |
1da177e4 LT |
238 | return NULL; |
239 | } | |
240 | ||
241 | /* | |
3c61eecb | 242 | * Delete a block. |
1da177e4 LT |
243 | * It is the caller's responsibility to check whether the file |
244 | * can be closed hereafter. | |
245 | */ | |
6849c0ca | 246 | static int nlmsvc_unlink_block(struct nlm_block *block) |
1da177e4 | 247 | { |
09c7938c | 248 | int status; |
6849c0ca | 249 | dprintk("lockd: unlinking block %p...\n", block); |
1da177e4 LT |
250 | |
251 | /* Remove block from list */ | |
92737230 | 252 | status = posix_unblock_lock(block->b_file->f_file, &block->b_call->a_args.lock.fl); |
1da177e4 | 253 | nlmsvc_remove_block(block); |
6849c0ca TM |
254 | return status; |
255 | } | |
1da177e4 | 256 | |
6849c0ca TM |
257 | static void nlmsvc_free_block(struct kref *kref) |
258 | { | |
259 | struct nlm_block *block = container_of(kref, struct nlm_block, b_count); | |
260 | struct nlm_file *file = block->b_file; | |
6849c0ca TM |
261 | |
262 | dprintk("lockd: freeing block %p...\n", block); | |
1da177e4 LT |
263 | |
264 | /* Remove block from file's list of blocks */ | |
89e63ef6 | 265 | mutex_lock(&file->f_mutex); |
68a2d76c | 266 | list_del_init(&block->b_flist); |
89e63ef6 | 267 | mutex_unlock(&file->f_mutex); |
1da177e4 | 268 | |
92737230 | 269 | nlmsvc_freegrantargs(block->b_call); |
7db836d4 | 270 | nlmsvc_release_call(block->b_call); |
d9f6eb75 | 271 | nlm_release_file(block->b_file); |
0e4ac9d9 | 272 | kfree(block->b_fl); |
1da177e4 | 273 | kfree(block); |
6849c0ca TM |
274 | } |
275 | ||
276 | static void nlmsvc_release_block(struct nlm_block *block) | |
277 | { | |
278 | if (block != NULL) | |
279 | kref_put(&block->b_count, nlmsvc_free_block); | |
1da177e4 LT |
280 | } |
281 | ||
f2af793d OK |
282 | /* |
283 | * Loop over all blocks and delete blocks held by | |
284 | * a matching host. | |
285 | */ | |
286 | void nlmsvc_traverse_blocks(struct nlm_host *host, | |
287 | struct nlm_file *file, | |
288 | nlm_host_match_fn_t match) | |
d9f6eb75 | 289 | { |
68a2d76c | 290 | struct nlm_block *block, *next; |
d9f6eb75 TM |
291 | |
292 | restart: | |
89e63ef6 | 293 | mutex_lock(&file->f_mutex); |
68a2d76c | 294 | list_for_each_entry_safe(block, next, &file->f_blocks, b_flist) { |
f2af793d | 295 | if (!match(block->b_host, host)) |
d9f6eb75 | 296 | continue; |
68a2d76c OK |
297 | /* Do not destroy blocks that are not on |
298 | * the global retry list - why? */ | |
299 | if (list_empty(&block->b_list)) | |
d9f6eb75 TM |
300 | continue; |
301 | kref_get(&block->b_count); | |
89e63ef6 | 302 | mutex_unlock(&file->f_mutex); |
d9f6eb75 TM |
303 | nlmsvc_unlink_block(block); |
304 | nlmsvc_release_block(block); | |
305 | goto restart; | |
306 | } | |
89e63ef6 | 307 | mutex_unlock(&file->f_mutex); |
d9f6eb75 TM |
308 | } |
309 | ||
5e1abf8c TM |
310 | /* |
311 | * Initialize arguments for GRANTED call. The nlm_rqst structure | |
312 | * has been cleared already. | |
313 | */ | |
314 | static int nlmsvc_setgrantargs(struct nlm_rqst *call, struct nlm_lock *lock) | |
315 | { | |
316 | locks_copy_lock(&call->a_args.lock.fl, &lock->fl); | |
317 | memcpy(&call->a_args.lock.fh, &lock->fh, sizeof(call->a_args.lock.fh)); | |
e9ff3990 | 318 | call->a_args.lock.caller = utsname()->nodename; |
5e1abf8c TM |
319 | call->a_args.lock.oh.len = lock->oh.len; |
320 | ||
321 | /* set default data area */ | |
322 | call->a_args.lock.oh.data = call->a_owner; | |
323 | call->a_args.lock.svid = lock->fl.fl_pid; | |
324 | ||
325 | if (lock->oh.len > NLMCLNT_OHSIZE) { | |
326 | void *data = kmalloc(lock->oh.len, GFP_KERNEL); | |
92737230 | 327 | if (!data) |
5e1abf8c | 328 | return 0; |
5e1abf8c TM |
329 | call->a_args.lock.oh.data = (u8 *) data; |
330 | } | |
331 | ||
332 | memcpy(call->a_args.lock.oh.data, lock->oh.data, lock->oh.len); | |
333 | return 1; | |
334 | } | |
335 | ||
336 | static void nlmsvc_freegrantargs(struct nlm_rqst *call) | |
337 | { | |
92737230 | 338 | if (call->a_args.lock.oh.data != call->a_owner) |
5e1abf8c | 339 | kfree(call->a_args.lock.oh.data); |
a9e61e25 FB |
340 | |
341 | locks_release_private(&call->a_args.lock.fl); | |
5e1abf8c TM |
342 | } |
343 | ||
2b36f412 ME |
344 | /* |
345 | * Deferred lock request handling for non-blocking lock | |
346 | */ | |
ca5c8cde | 347 | static __be32 |
2b36f412 ME |
348 | nlmsvc_defer_lock_rqst(struct svc_rqst *rqstp, struct nlm_block *block) |
349 | { | |
ca5c8cde | 350 | __be32 status = nlm_lck_denied_nolocks; |
2b36f412 ME |
351 | |
352 | block->b_flags |= B_QUEUED; | |
353 | ||
354 | nlmsvc_insert_block(block, NLM_TIMEOUT); | |
355 | ||
356 | block->b_cache_req = &rqstp->rq_chandle; | |
357 | if (rqstp->rq_chandle.defer) { | |
358 | block->b_deferred_req = | |
359 | rqstp->rq_chandle.defer(block->b_cache_req); | |
360 | if (block->b_deferred_req != NULL) | |
361 | status = nlm_drop_reply; | |
362 | } | |
363 | dprintk("lockd: nlmsvc_defer_lock_rqst block %p flags %d status %d\n", | |
ca5c8cde | 364 | block, block->b_flags, ntohl(status)); |
2b36f412 ME |
365 | |
366 | return status; | |
367 | } | |
368 | ||
1da177e4 LT |
369 | /* |
370 | * Attempt to establish a lock, and if it can't be granted, block it | |
371 | * if required. | |
372 | */ | |
52921e02 | 373 | __be32 |
1da177e4 | 374 | nlmsvc_lock(struct svc_rqst *rqstp, struct nlm_file *file, |
6cde4de8 | 375 | struct nlm_host *host, struct nlm_lock *lock, int wait, |
b2b50289 | 376 | struct nlm_cookie *cookie, int reclaim) |
1da177e4 | 377 | { |
f8120480 | 378 | struct nlm_block *block = NULL; |
1da177e4 | 379 | int error; |
52921e02 | 380 | __be32 ret; |
1da177e4 LT |
381 | |
382 | dprintk("lockd: nlmsvc_lock(%s/%ld, ty=%d, pi=%d, %Ld-%Ld, bl=%d)\n", | |
225a719f JS |
383 | file->f_file->f_path.dentry->d_inode->i_sb->s_id, |
384 | file->f_file->f_path.dentry->d_inode->i_ino, | |
1da177e4 LT |
385 | lock->fl.fl_type, lock->fl.fl_pid, |
386 | (long long)lock->fl.fl_start, | |
387 | (long long)lock->fl.fl_end, | |
388 | wait); | |
389 | ||
1da177e4 | 390 | /* Lock file against concurrent access */ |
89e63ef6 | 391 | mutex_lock(&file->f_mutex); |
f8120480 ME |
392 | /* Get existing block (in case client is busy-waiting) |
393 | * or create new block | |
394 | */ | |
d9f6eb75 | 395 | block = nlmsvc_lookup_block(file, lock); |
09c7938c | 396 | if (block == NULL) { |
560de0e6 | 397 | block = nlmsvc_create_block(rqstp, host, file, lock, cookie); |
f8120480 ME |
398 | ret = nlm_lck_denied_nolocks; |
399 | if (block == NULL) | |
400 | goto out; | |
92737230 | 401 | lock = &block->b_call->a_args.lock; |
f8120480 ME |
402 | } else |
403 | lock->fl.fl_flags &= ~FL_SLEEP; | |
1da177e4 | 404 | |
1a8322b2 ME |
405 | if (block->b_flags & B_QUEUED) { |
406 | dprintk("lockd: nlmsvc_lock deferred block %p flags %d\n", | |
407 | block, block->b_flags); | |
408 | if (block->b_granted) { | |
409 | nlmsvc_unlink_block(block); | |
410 | ret = nlm_granted; | |
411 | goto out; | |
412 | } | |
413 | if (block->b_flags & B_TIMED_OUT) { | |
414 | nlmsvc_unlink_block(block); | |
415 | ret = nlm_lck_denied; | |
416 | goto out; | |
417 | } | |
418 | ret = nlm_drop_reply; | |
419 | goto out; | |
420 | } | |
1da177e4 | 421 | |
b2b50289 BF |
422 | if (locks_in_grace() && !reclaim) { |
423 | ret = nlm_lck_denied_grace_period; | |
424 | goto out; | |
425 | } | |
d22b1cff BF |
426 | if (reclaim && !locks_in_grace()) { |
427 | ret = nlm_lck_denied_grace_period; | |
428 | goto out; | |
429 | } | |
b2b50289 | 430 | |
1a8322b2 ME |
431 | if (!wait) |
432 | lock->fl.fl_flags &= ~FL_SLEEP; | |
433 | error = vfs_lock_file(file->f_file, F_SETLK, &lock->fl, NULL); | |
434 | lock->fl.fl_flags &= ~FL_SLEEP; | |
a85f193e | 435 | |
1a8322b2 | 436 | dprintk("lockd: vfs_lock_file returned %d\n", error); |
6d7bbbba | 437 | switch (error) { |
1da177e4 | 438 | case 0: |
15dadef9 AA |
439 | ret = nlm_granted; |
440 | goto out; | |
09c7938c | 441 | case -EAGAIN: |
e33d1ea6 MS |
442 | /* |
443 | * If this is a blocking request for an | |
444 | * already pending lock request then we need | |
445 | * to put it back on lockd's block list | |
446 | */ | |
447 | if (wait) | |
448 | break; | |
1a8322b2 | 449 | ret = nlm_lck_denied; |
e33d1ea6 | 450 | goto out; |
bde74e4b | 451 | case FILE_LOCK_DEFERRED: |
1a8322b2 ME |
452 | if (wait) |
453 | break; | |
454 | /* Filesystem lock operation is in progress | |
455 | Add it to the queue waiting for callback */ | |
456 | ret = nlmsvc_defer_lock_rqst(rqstp, block); | |
457 | goto out; | |
09c7938c | 458 | case -EDEADLK: |
15dadef9 AA |
459 | ret = nlm_deadlock; |
460 | goto out; | |
1da177e4 | 461 | default: /* includes ENOLCK */ |
15dadef9 AA |
462 | ret = nlm_lck_denied_nolocks; |
463 | goto out; | |
1da177e4 LT |
464 | } |
465 | ||
09c7938c | 466 | ret = nlm_lck_blocked; |
1da177e4 LT |
467 | |
468 | /* Append to list of blocked */ | |
f8120480 | 469 | nlmsvc_insert_block(block, NLM_NEVER); |
15dadef9 | 470 | out: |
89e63ef6 | 471 | mutex_unlock(&file->f_mutex); |
6849c0ca | 472 | nlmsvc_release_block(block); |
15dadef9 AA |
473 | dprintk("lockd: nlmsvc_lock returned %u\n", ret); |
474 | return ret; | |
1da177e4 LT |
475 | } |
476 | ||
477 | /* | |
478 | * Test for presence of a conflicting lock. | |
479 | */ | |
52921e02 | 480 | __be32 |
85f3f1b3 | 481 | nlmsvc_testlock(struct svc_rqst *rqstp, struct nlm_file *file, |
8f920d5e JL |
482 | struct nlm_host *host, struct nlm_lock *lock, |
483 | struct nlm_lock *conflock, struct nlm_cookie *cookie) | |
1da177e4 | 484 | { |
5ea0d750 ME |
485 | struct nlm_block *block = NULL; |
486 | int error; | |
487 | __be32 ret; | |
488 | ||
1da177e4 | 489 | dprintk("lockd: nlmsvc_testlock(%s/%ld, ty=%d, %Ld-%Ld)\n", |
225a719f JS |
490 | file->f_file->f_path.dentry->d_inode->i_sb->s_id, |
491 | file->f_file->f_path.dentry->d_inode->i_ino, | |
1da177e4 LT |
492 | lock->fl.fl_type, |
493 | (long long)lock->fl.fl_start, | |
494 | (long long)lock->fl.fl_end); | |
495 | ||
5ea0d750 ME |
496 | /* Get existing block (in case client is busy-waiting) */ |
497 | block = nlmsvc_lookup_block(file, lock); | |
498 | ||
499 | if (block == NULL) { | |
500 | struct file_lock *conf = kzalloc(sizeof(*conf), GFP_KERNEL); | |
501 | ||
502 | if (conf == NULL) | |
503 | return nlm_granted; | |
255129d1 | 504 | block = nlmsvc_create_block(rqstp, host, file, lock, cookie); |
5ea0d750 ME |
505 | if (block == NULL) { |
506 | kfree(conf); | |
507 | return nlm_granted; | |
508 | } | |
509 | block->b_fl = conf; | |
510 | } | |
511 | if (block->b_flags & B_QUEUED) { | |
512 | dprintk("lockd: nlmsvc_testlock deferred block %p flags %d fl %p\n", | |
513 | block, block->b_flags, block->b_fl); | |
514 | if (block->b_flags & B_TIMED_OUT) { | |
515 | nlmsvc_unlink_block(block); | |
29dbf546 OD |
516 | ret = nlm_lck_denied; |
517 | goto out; | |
5ea0d750 ME |
518 | } |
519 | if (block->b_flags & B_GOT_CALLBACK) { | |
54ca95eb | 520 | nlmsvc_unlink_block(block); |
5ea0d750 ME |
521 | if (block->b_fl != NULL |
522 | && block->b_fl->fl_type != F_UNLCK) { | |
523 | lock->fl = *block->b_fl; | |
524 | goto conf_lock; | |
29dbf546 | 525 | } else { |
29dbf546 OD |
526 | ret = nlm_granted; |
527 | goto out; | |
5ea0d750 ME |
528 | } |
529 | } | |
29dbf546 OD |
530 | ret = nlm_drop_reply; |
531 | goto out; | |
1da177e4 LT |
532 | } |
533 | ||
b2b50289 BF |
534 | if (locks_in_grace()) { |
535 | ret = nlm_lck_denied_grace_period; | |
536 | goto out; | |
537 | } | |
5ea0d750 | 538 | error = vfs_test_lock(file->f_file, &lock->fl); |
bde74e4b | 539 | if (error == FILE_LOCK_DEFERRED) { |
29dbf546 OD |
540 | ret = nlmsvc_defer_lock_rqst(rqstp, block); |
541 | goto out; | |
542 | } | |
5ea0d750 ME |
543 | if (error) { |
544 | ret = nlm_lck_denied_nolocks; | |
545 | goto out; | |
546 | } | |
547 | if (lock->fl.fl_type == F_UNLCK) { | |
548 | ret = nlm_granted; | |
549 | goto out; | |
550 | } | |
551 | ||
552 | conf_lock: | |
553 | dprintk("lockd: conflicting lock(ty=%d, %Ld-%Ld)\n", | |
554 | lock->fl.fl_type, (long long)lock->fl.fl_start, | |
555 | (long long)lock->fl.fl_end); | |
556 | conflock->caller = "somehost"; /* FIXME */ | |
557 | conflock->len = strlen(conflock->caller); | |
558 | conflock->oh.len = 0; /* don't return OH info */ | |
559 | conflock->svid = lock->fl.fl_pid; | |
560 | conflock->fl.fl_type = lock->fl.fl_type; | |
561 | conflock->fl.fl_start = lock->fl.fl_start; | |
562 | conflock->fl.fl_end = lock->fl.fl_end; | |
563 | ret = nlm_lck_denied; | |
564 | out: | |
565 | if (block) | |
566 | nlmsvc_release_block(block); | |
567 | return ret; | |
1da177e4 LT |
568 | } |
569 | ||
570 | /* | |
571 | * Remove a lock. | |
572 | * This implies a CANCEL call: We send a GRANT_MSG, the client replies | |
573 | * with a GRANT_RES call which gets lost, and calls UNLOCK immediately | |
574 | * afterwards. In this case the block will still be there, and hence | |
575 | * must be removed. | |
576 | */ | |
52921e02 | 577 | __be32 |
1da177e4 LT |
578 | nlmsvc_unlock(struct nlm_file *file, struct nlm_lock *lock) |
579 | { | |
580 | int error; | |
581 | ||
582 | dprintk("lockd: nlmsvc_unlock(%s/%ld, pi=%d, %Ld-%Ld)\n", | |
225a719f JS |
583 | file->f_file->f_path.dentry->d_inode->i_sb->s_id, |
584 | file->f_file->f_path.dentry->d_inode->i_ino, | |
1da177e4 LT |
585 | lock->fl.fl_pid, |
586 | (long long)lock->fl.fl_start, | |
587 | (long long)lock->fl.fl_end); | |
588 | ||
589 | /* First, cancel any lock that might be there */ | |
590 | nlmsvc_cancel_blocked(file, lock); | |
591 | ||
592 | lock->fl.fl_type = F_UNLCK; | |
1a8322b2 | 593 | error = vfs_lock_file(file->f_file, F_SETLK, &lock->fl, NULL); |
1da177e4 LT |
594 | |
595 | return (error < 0)? nlm_lck_denied_nolocks : nlm_granted; | |
596 | } | |
597 | ||
598 | /* | |
599 | * Cancel a previously blocked request. | |
600 | * | |
601 | * A cancel request always overrides any grant that may currently | |
602 | * be in progress. | |
603 | * The calling procedure must check whether the file can be closed. | |
604 | */ | |
52921e02 | 605 | __be32 |
1da177e4 LT |
606 | nlmsvc_cancel_blocked(struct nlm_file *file, struct nlm_lock *lock) |
607 | { | |
608 | struct nlm_block *block; | |
64a318ee | 609 | int status = 0; |
1da177e4 LT |
610 | |
611 | dprintk("lockd: nlmsvc_cancel(%s/%ld, pi=%d, %Ld-%Ld)\n", | |
225a719f JS |
612 | file->f_file->f_path.dentry->d_inode->i_sb->s_id, |
613 | file->f_file->f_path.dentry->d_inode->i_ino, | |
1da177e4 LT |
614 | lock->fl.fl_pid, |
615 | (long long)lock->fl.fl_start, | |
616 | (long long)lock->fl.fl_end); | |
617 | ||
b2b50289 BF |
618 | if (locks_in_grace()) |
619 | return nlm_lck_denied_grace_period; | |
620 | ||
89e63ef6 | 621 | mutex_lock(&file->f_mutex); |
d9f6eb75 | 622 | block = nlmsvc_lookup_block(file, lock); |
89e63ef6 | 623 | mutex_unlock(&file->f_mutex); |
d9f6eb75 | 624 | if (block != NULL) { |
1a8322b2 ME |
625 | vfs_cancel_lock(block->b_file->f_file, |
626 | &block->b_call->a_args.lock.fl); | |
6849c0ca TM |
627 | status = nlmsvc_unlink_block(block); |
628 | nlmsvc_release_block(block); | |
629 | } | |
64a318ee | 630 | return status ? nlm_lck_denied : nlm_granted; |
1da177e4 LT |
631 | } |
632 | ||
0e4ac9d9 ME |
633 | /* |
634 | * This is a callback from the filesystem for VFS file lock requests. | |
8fb47a4f | 635 | * It will be used if lm_grant is defined and the filesystem can not |
0e4ac9d9 ME |
636 | * respond to the request immediately. |
637 | * For GETLK request it will copy the reply to the nlm_block. | |
638 | * For SETLK or SETLKW request it will get the local posix lock. | |
639 | * In all cases it will move the block to the head of nlm_blocked q where | |
640 | * nlmsvc_retry_blocked() can send back a reply for SETLKW or revisit the | |
641 | * deferred rpc for GETLK and SETLK. | |
642 | */ | |
643 | static void | |
644 | nlmsvc_update_deferred_block(struct nlm_block *block, struct file_lock *conf, | |
645 | int result) | |
646 | { | |
647 | block->b_flags |= B_GOT_CALLBACK; | |
648 | if (result == 0) | |
649 | block->b_granted = 1; | |
650 | else | |
651 | block->b_flags |= B_TIMED_OUT; | |
652 | if (conf) { | |
0e4ac9d9 | 653 | if (block->b_fl) |
1a747ee0 | 654 | __locks_copy_lock(block->b_fl, conf); |
0e4ac9d9 ME |
655 | } |
656 | } | |
657 | ||
658 | static int nlmsvc_grant_deferred(struct file_lock *fl, struct file_lock *conf, | |
659 | int result) | |
660 | { | |
661 | struct nlm_block *block; | |
662 | int rc = -ENOENT; | |
663 | ||
f904be9c | 664 | spin_lock(&nlm_blocked_lock); |
0e4ac9d9 ME |
665 | list_for_each_entry(block, &nlm_blocked, b_list) { |
666 | if (nlm_compare_locks(&block->b_call->a_args.lock.fl, fl)) { | |
667 | dprintk("lockd: nlmsvc_notify_blocked block %p flags %d\n", | |
668 | block, block->b_flags); | |
669 | if (block->b_flags & B_QUEUED) { | |
670 | if (block->b_flags & B_TIMED_OUT) { | |
671 | rc = -ENOLCK; | |
672 | break; | |
673 | } | |
674 | nlmsvc_update_deferred_block(block, conf, result); | |
675 | } else if (result == 0) | |
676 | block->b_granted = 1; | |
677 | ||
f904be9c | 678 | nlmsvc_insert_block_locked(block, 0); |
0e4ac9d9 ME |
679 | svc_wake_up(block->b_daemon); |
680 | rc = 0; | |
681 | break; | |
682 | } | |
683 | } | |
f904be9c | 684 | spin_unlock(&nlm_blocked_lock); |
0e4ac9d9 ME |
685 | if (rc == -ENOENT) |
686 | printk(KERN_WARNING "lockd: grant for unknown block\n"); | |
687 | return rc; | |
688 | } | |
689 | ||
1da177e4 LT |
690 | /* |
691 | * Unblock a blocked lock request. This is a callback invoked from the | |
692 | * VFS layer when a lock on which we blocked is removed. | |
693 | * | |
694 | * This function doesn't grant the blocked lock instantly, but rather moves | |
695 | * the block to the head of nlm_blocked where it can be picked up by lockd. | |
696 | */ | |
697 | static void | |
698 | nlmsvc_notify_blocked(struct file_lock *fl) | |
699 | { | |
68a2d76c | 700 | struct nlm_block *block; |
1da177e4 LT |
701 | |
702 | dprintk("lockd: VFS unblock notification for block %p\n", fl); | |
a282a1fa | 703 | spin_lock(&nlm_blocked_lock); |
68a2d76c | 704 | list_for_each_entry(block, &nlm_blocked, b_list) { |
92737230 | 705 | if (nlm_compare_locks(&block->b_call->a_args.lock.fl, fl)) { |
a282a1fa BF |
706 | nlmsvc_insert_block_locked(block, 0); |
707 | spin_unlock(&nlm_blocked_lock); | |
1da177e4 LT |
708 | svc_wake_up(block->b_daemon); |
709 | return; | |
710 | } | |
711 | } | |
a282a1fa | 712 | spin_unlock(&nlm_blocked_lock); |
1da177e4 LT |
713 | printk(KERN_WARNING "lockd: notification for unknown block!\n"); |
714 | } | |
715 | ||
716 | static int nlmsvc_same_owner(struct file_lock *fl1, struct file_lock *fl2) | |
717 | { | |
718 | return fl1->fl_owner == fl2->fl_owner && fl1->fl_pid == fl2->fl_pid; | |
719 | } | |
720 | ||
7b021967 | 721 | const struct lock_manager_operations nlmsvc_lock_operations = { |
8fb47a4f BF |
722 | .lm_compare_owner = nlmsvc_same_owner, |
723 | .lm_notify = nlmsvc_notify_blocked, | |
724 | .lm_grant = nlmsvc_grant_deferred, | |
1da177e4 LT |
725 | }; |
726 | ||
727 | /* | |
728 | * Try to claim a lock that was previously blocked. | |
729 | * | |
730 | * Note that we use both the RPC_GRANTED_MSG call _and_ an async | |
731 | * RPC thread when notifying the client. This seems like overkill... | |
732 | * Here's why: | |
733 | * - we don't want to use a synchronous RPC thread, otherwise | |
734 | * we might find ourselves hanging on a dead portmapper. | |
735 | * - Some lockd implementations (e.g. HP) don't react to | |
736 | * RPC_GRANTED calls; they seem to insist on RPC_GRANTED_MSG calls. | |
737 | */ | |
738 | static void | |
739 | nlmsvc_grant_blocked(struct nlm_block *block) | |
740 | { | |
741 | struct nlm_file *file = block->b_file; | |
92737230 | 742 | struct nlm_lock *lock = &block->b_call->a_args.lock; |
1da177e4 LT |
743 | int error; |
744 | ||
745 | dprintk("lockd: grant blocked lock %p\n", block); | |
746 | ||
0e4ac9d9 ME |
747 | kref_get(&block->b_count); |
748 | ||
1da177e4 | 749 | /* Unlink block request from list */ |
6849c0ca | 750 | nlmsvc_unlink_block(block); |
1da177e4 LT |
751 | |
752 | /* If b_granted is true this means we've been here before. | |
753 | * Just retry the grant callback, possibly refreshing the RPC | |
754 | * binding */ | |
755 | if (block->b_granted) { | |
756 | nlm_rebind_host(block->b_host); | |
757 | goto callback; | |
758 | } | |
759 | ||
760 | /* Try the lock operation again */ | |
09c7938c | 761 | lock->fl.fl_flags |= FL_SLEEP; |
1a8322b2 | 762 | error = vfs_lock_file(file->f_file, F_SETLK, &lock->fl, NULL); |
09c7938c TM |
763 | lock->fl.fl_flags &= ~FL_SLEEP; |
764 | ||
5de0e502 AA |
765 | switch (error) { |
766 | case 0: | |
767 | break; | |
bde74e4b | 768 | case FILE_LOCK_DEFERRED: |
1a8322b2 | 769 | dprintk("lockd: lock still blocked error %d\n", error); |
1da177e4 | 770 | nlmsvc_insert_block(block, NLM_NEVER); |
0e4ac9d9 | 771 | nlmsvc_release_block(block); |
d9f6eb75 | 772 | return; |
5de0e502 | 773 | default: |
1da177e4 | 774 | printk(KERN_WARNING "lockd: unexpected error %d in %s!\n", |
8e24eea7 | 775 | -error, __func__); |
1da177e4 | 776 | nlmsvc_insert_block(block, 10 * HZ); |
0e4ac9d9 | 777 | nlmsvc_release_block(block); |
d9f6eb75 | 778 | return; |
1da177e4 LT |
779 | } |
780 | ||
781 | callback: | |
782 | /* Lock was granted by VFS. */ | |
783 | dprintk("lockd: GRANTing blocked lock.\n"); | |
784 | block->b_granted = 1; | |
1da177e4 | 785 | |
9706501e JL |
786 | /* keep block on the list, but don't reattempt until the RPC |
787 | * completes or the submission fails | |
788 | */ | |
789 | nlmsvc_insert_block(block, NLM_NEVER); | |
1da177e4 | 790 | |
9706501e JL |
791 | /* Call the client -- use a soft RPC task since nlmsvc_retry_blocked |
792 | * will queue up a new one if this one times out | |
793 | */ | |
794 | error = nlm_async_call(block->b_call, NLMPROC_GRANTED_MSG, | |
795 | &nlmsvc_grant_ops); | |
796 | ||
797 | /* RPC submission failed, wait a bit and retry */ | |
798 | if (error < 0) | |
799 | nlmsvc_insert_block(block, 10 * HZ); | |
1da177e4 LT |
800 | } |
801 | ||
802 | /* | |
803 | * This is the callback from the RPC layer when the NLM_GRANTED_MSG | |
804 | * RPC call has succeeded or timed out. | |
805 | * Like all RPC callbacks, it is invoked by the rpciod process, so it | |
806 | * better not sleep. Therefore, we put the blocked lock on the nlm_blocked | |
807 | * chain once more in order to have it removed by lockd itself (which can | |
808 | * then sleep on the file semaphore without disrupting e.g. the nfs client). | |
809 | */ | |
963d8fe5 | 810 | static void nlmsvc_grant_callback(struct rpc_task *task, void *data) |
1da177e4 | 811 | { |
963d8fe5 | 812 | struct nlm_rqst *call = data; |
92737230 | 813 | struct nlm_block *block = call->a_block; |
1da177e4 | 814 | unsigned long timeout; |
1da177e4 LT |
815 | |
816 | dprintk("lockd: GRANT_MSG RPC callback\n"); | |
1da177e4 | 817 | |
f904be9c | 818 | spin_lock(&nlm_blocked_lock); |
c64e80d5 JL |
819 | /* if the block is not on a list at this point then it has |
820 | * been invalidated. Don't try to requeue it. | |
821 | * | |
822 | * FIXME: it's possible that the block is removed from the list | |
823 | * after this check but before the nlmsvc_insert_block. In that | |
824 | * case it will be added back. Perhaps we need better locking | |
825 | * for nlm_blocked? | |
826 | */ | |
827 | if (list_empty(&block->b_list)) | |
a86dc496 | 828 | goto out; |
c64e80d5 | 829 | |
1da177e4 LT |
830 | /* Technically, we should down the file semaphore here. Since we |
831 | * move the block towards the head of the queue only, no harm | |
832 | * can be done, though. */ | |
833 | if (task->tk_status < 0) { | |
834 | /* RPC error: Re-insert for retransmission */ | |
835 | timeout = 10 * HZ; | |
1da177e4 LT |
836 | } else { |
837 | /* Call was successful, now wait for client callback */ | |
838 | timeout = 60 * HZ; | |
839 | } | |
f904be9c | 840 | nlmsvc_insert_block_locked(block, timeout); |
1da177e4 | 841 | svc_wake_up(block->b_daemon); |
a86dc496 | 842 | out: |
f904be9c | 843 | spin_unlock(&nlm_blocked_lock); |
5e1abf8c TM |
844 | } |
845 | ||
f904be9c BS |
846 | /* |
847 | * FIXME: nlmsvc_release_block() grabs a mutex. This is not allowed for an | |
848 | * .rpc_release rpc_call_op | |
849 | */ | |
ec535ce1 | 850 | static void nlmsvc_grant_release(void *data) |
5e1abf8c | 851 | { |
6041b791 | 852 | struct nlm_rqst *call = data; |
6041b791 | 853 | nlmsvc_release_block(call->a_block); |
1da177e4 LT |
854 | } |
855 | ||
963d8fe5 TM |
856 | static const struct rpc_call_ops nlmsvc_grant_ops = { |
857 | .rpc_call_done = nlmsvc_grant_callback, | |
5e1abf8c | 858 | .rpc_release = nlmsvc_grant_release, |
963d8fe5 TM |
859 | }; |
860 | ||
1da177e4 LT |
861 | /* |
862 | * We received a GRANT_RES callback. Try to find the corresponding | |
863 | * block. | |
864 | */ | |
865 | void | |
e8c5c045 | 866 | nlmsvc_grant_reply(struct nlm_cookie *cookie, __be32 status) |
1da177e4 LT |
867 | { |
868 | struct nlm_block *block; | |
1da177e4 | 869 | |
39be4502 OK |
870 | dprintk("grant_reply: looking for cookie %x, s=%d \n", |
871 | *(unsigned int *)(cookie->data), status); | |
872 | if (!(block = nlmsvc_find_block(cookie))) | |
1da177e4 | 873 | return; |
1da177e4 | 874 | |
f232142c | 875 | if (block) { |
e8c5c045 | 876 | if (status == nlm_lck_denied_grace_period) { |
1da177e4 LT |
877 | /* Try again in a couple of seconds */ |
878 | nlmsvc_insert_block(block, 10 * HZ); | |
1da177e4 LT |
879 | } else { |
880 | /* Lock is now held by client, or has been rejected. | |
881 | * In both cases, the block should be removed. */ | |
6849c0ca | 882 | nlmsvc_unlink_block(block); |
1da177e4 LT |
883 | } |
884 | } | |
6849c0ca | 885 | nlmsvc_release_block(block); |
1da177e4 LT |
886 | } |
887 | ||
0e4ac9d9 ME |
888 | /* Helper function to handle retry of a deferred block. |
889 | * If it is a blocking lock, call grant_blocked. | |
890 | * For a non-blocking lock or test lock, revisit the request. | |
891 | */ | |
892 | static void | |
893 | retry_deferred_block(struct nlm_block *block) | |
894 | { | |
895 | if (!(block->b_flags & B_GOT_CALLBACK)) | |
896 | block->b_flags |= B_TIMED_OUT; | |
897 | nlmsvc_insert_block(block, NLM_TIMEOUT); | |
898 | dprintk("revisit block %p flags %d\n", block, block->b_flags); | |
899 | if (block->b_deferred_req) { | |
900 | block->b_deferred_req->revisit(block->b_deferred_req, 0); | |
901 | block->b_deferred_req = NULL; | |
902 | } | |
903 | } | |
904 | ||
1da177e4 LT |
905 | /* |
906 | * Retry all blocked locks that have been notified. This is where lockd | |
907 | * picks up locks that can be granted, or grant notifications that must | |
908 | * be retransmitted. | |
909 | */ | |
910 | unsigned long | |
911 | nlmsvc_retry_blocked(void) | |
912 | { | |
68a2d76c OK |
913 | unsigned long timeout = MAX_SCHEDULE_TIMEOUT; |
914 | struct nlm_block *block; | |
915 | ||
d751a7cd | 916 | while (!list_empty(&nlm_blocked) && !kthread_should_stop()) { |
68a2d76c | 917 | block = list_entry(nlm_blocked.next, struct nlm_block, b_list); |
1da177e4 | 918 | |
1da177e4 LT |
919 | if (block->b_when == NLM_NEVER) |
920 | break; | |
6d7bbbba | 921 | if (time_after(block->b_when, jiffies)) { |
68a2d76c | 922 | timeout = block->b_when - jiffies; |
1da177e4 | 923 | break; |
68a2d76c OK |
924 | } |
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); | |
1da177e4 LT |
934 | } |
935 | ||
68a2d76c | 936 | return timeout; |
1da177e4 | 937 | } |
d8367c50 CL |
938 | |
939 | #ifdef RPC_DEBUG | |
940 | static const char *nlmdbg_cookie2a(const struct nlm_cookie *cookie) | |
941 | { | |
942 | /* | |
943 | * We can get away with a static buffer because we're only | |
944 | * called with BKL held. | |
945 | */ | |
946 | static char buf[2*NLM_MAXCOOKIELEN+1]; | |
947 | unsigned int i, len = sizeof(buf); | |
948 | char *p = buf; | |
949 | ||
950 | len--; /* allow for trailing \0 */ | |
951 | if (len < 3) | |
952 | return "???"; | |
953 | for (i = 0 ; i < cookie->len ; i++) { | |
954 | if (len < 2) { | |
955 | strcpy(p-3, "..."); | |
956 | break; | |
957 | } | |
958 | sprintf(p, "%02x", cookie->data[i]); | |
959 | p += 2; | |
960 | len -= 2; | |
961 | } | |
962 | *p = '\0'; | |
963 | ||
964 | return buf; | |
965 | } | |
966 | #endif |