]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * linux/fs/locks.c | |
3 | * | |
4 | * Provide support for fcntl()'s F_GETLK, F_SETLK, and F_SETLKW calls. | |
5 | * Doug Evans (dje@spiff.uucp), August 07, 1992 | |
6 | * | |
7 | * Deadlock detection added. | |
8 | * FIXME: one thing isn't handled yet: | |
9 | * - mandatory locks (requires lots of changes elsewhere) | |
10 | * Kelly Carmichael (kelly@[142.24.8.65]), September 17, 1994. | |
11 | * | |
12 | * Miscellaneous edits, and a total rewrite of posix_lock_file() code. | |
13 | * Kai Petzke (wpp@marie.physik.tu-berlin.de), 1994 | |
14 | * | |
15 | * Converted file_lock_table to a linked list from an array, which eliminates | |
16 | * the limits on how many active file locks are open. | |
17 | * Chad Page (pageone@netcom.com), November 27, 1994 | |
18 | * | |
19 | * Removed dependency on file descriptors. dup()'ed file descriptors now | |
20 | * get the same locks as the original file descriptors, and a close() on | |
21 | * any file descriptor removes ALL the locks on the file for the current | |
22 | * process. Since locks still depend on the process id, locks are inherited | |
23 | * after an exec() but not after a fork(). This agrees with POSIX, and both | |
24 | * BSD and SVR4 practice. | |
25 | * Andy Walker (andy@lysaker.kvaerner.no), February 14, 1995 | |
26 | * | |
27 | * Scrapped free list which is redundant now that we allocate locks | |
28 | * dynamically with kmalloc()/kfree(). | |
29 | * Andy Walker (andy@lysaker.kvaerner.no), February 21, 1995 | |
30 | * | |
31 | * Implemented two lock personalities - FL_FLOCK and FL_POSIX. | |
32 | * | |
33 | * FL_POSIX locks are created with calls to fcntl() and lockf() through the | |
34 | * fcntl() system call. They have the semantics described above. | |
35 | * | |
36 | * FL_FLOCK locks are created with calls to flock(), through the flock() | |
37 | * system call, which is new. Old C libraries implement flock() via fcntl() | |
38 | * and will continue to use the old, broken implementation. | |
39 | * | |
40 | * FL_FLOCK locks follow the 4.4 BSD flock() semantics. They are associated | |
41 | * with a file pointer (filp). As a result they can be shared by a parent | |
42 | * process and its children after a fork(). They are removed when the last | |
43 | * file descriptor referring to the file pointer is closed (unless explicitly | |
44 | * unlocked). | |
45 | * | |
46 | * FL_FLOCK locks never deadlock, an existing lock is always removed before | |
47 | * upgrading from shared to exclusive (or vice versa). When this happens | |
48 | * any processes blocked by the current lock are woken up and allowed to | |
49 | * run before the new lock is applied. | |
50 | * Andy Walker (andy@lysaker.kvaerner.no), June 09, 1995 | |
51 | * | |
52 | * Removed some race conditions in flock_lock_file(), marked other possible | |
53 | * races. Just grep for FIXME to see them. | |
54 | * Dmitry Gorodchanin (pgmdsg@ibi.com), February 09, 1996. | |
55 | * | |
56 | * Addressed Dmitry's concerns. Deadlock checking no longer recursive. | |
57 | * Lock allocation changed to GFP_ATOMIC as we can't afford to sleep | |
58 | * once we've checked for blocking and deadlocking. | |
59 | * Andy Walker (andy@lysaker.kvaerner.no), April 03, 1996. | |
60 | * | |
61 | * Initial implementation of mandatory locks. SunOS turned out to be | |
62 | * a rotten model, so I implemented the "obvious" semantics. | |
63 | * See 'Documentation/mandatory.txt' for details. | |
64 | * Andy Walker (andy@lysaker.kvaerner.no), April 06, 1996. | |
65 | * | |
66 | * Don't allow mandatory locks on mmap()'ed files. Added simple functions to | |
67 | * check if a file has mandatory locks, used by mmap(), open() and creat() to | |
68 | * see if system call should be rejected. Ref. HP-UX/SunOS/Solaris Reference | |
69 | * Manual, Section 2. | |
70 | * Andy Walker (andy@lysaker.kvaerner.no), April 09, 1996. | |
71 | * | |
72 | * Tidied up block list handling. Added '/proc/locks' interface. | |
73 | * Andy Walker (andy@lysaker.kvaerner.no), April 24, 1996. | |
74 | * | |
75 | * Fixed deadlock condition for pathological code that mixes calls to | |
76 | * flock() and fcntl(). | |
77 | * Andy Walker (andy@lysaker.kvaerner.no), April 29, 1996. | |
78 | * | |
79 | * Allow only one type of locking scheme (FL_POSIX or FL_FLOCK) to be in use | |
80 | * for a given file at a time. Changed the CONFIG_LOCK_MANDATORY scheme to | |
81 | * guarantee sensible behaviour in the case where file system modules might | |
82 | * be compiled with different options than the kernel itself. | |
83 | * Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996. | |
84 | * | |
85 | * Added a couple of missing wake_up() calls. Thanks to Thomas Meckel | |
86 | * (Thomas.Meckel@mni.fh-giessen.de) for spotting this. | |
87 | * Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996. | |
88 | * | |
89 | * Changed FL_POSIX locks to use the block list in the same way as FL_FLOCK | |
90 | * locks. Changed process synchronisation to avoid dereferencing locks that | |
91 | * have already been freed. | |
92 | * Andy Walker (andy@lysaker.kvaerner.no), Sep 21, 1996. | |
93 | * | |
94 | * Made the block list a circular list to minimise searching in the list. | |
95 | * Andy Walker (andy@lysaker.kvaerner.no), Sep 25, 1996. | |
96 | * | |
97 | * Made mandatory locking a mount option. Default is not to allow mandatory | |
98 | * locking. | |
99 | * Andy Walker (andy@lysaker.kvaerner.no), Oct 04, 1996. | |
100 | * | |
101 | * Some adaptations for NFS support. | |
102 | * Olaf Kirch (okir@monad.swb.de), Dec 1996, | |
103 | * | |
104 | * Fixed /proc/locks interface so that we can't overrun the buffer we are handed. | |
105 | * Andy Walker (andy@lysaker.kvaerner.no), May 12, 1997. | |
106 | * | |
107 | * Use slab allocator instead of kmalloc/kfree. | |
108 | * Use generic list implementation from <linux/list.h>. | |
109 | * Sped up posix_locks_deadlock by only considering blocked locks. | |
110 | * Matthew Wilcox <willy@debian.org>, March, 2000. | |
111 | * | |
112 | * Leases and LOCK_MAND | |
113 | * Matthew Wilcox <willy@debian.org>, June, 2000. | |
114 | * Stephen Rothwell <sfr@canb.auug.org.au>, June, 2000. | |
115 | */ | |
116 | ||
117 | #include <linux/capability.h> | |
118 | #include <linux/file.h> | |
119 | #include <linux/fdtable.h> | |
120 | #include <linux/fs.h> | |
121 | #include <linux/init.h> | |
122 | #include <linux/module.h> | |
123 | #include <linux/security.h> | |
124 | #include <linux/slab.h> | |
125 | #include <linux/smp_lock.h> | |
126 | #include <linux/syscalls.h> | |
127 | #include <linux/time.h> | |
128 | #include <linux/rcupdate.h> | |
129 | #include <linux/pid_namespace.h> | |
130 | ||
131 | #include <asm/uaccess.h> | |
132 | ||
133 | #define IS_POSIX(fl) (fl->fl_flags & FL_POSIX) | |
134 | #define IS_FLOCK(fl) (fl->fl_flags & FL_FLOCK) | |
135 | #define IS_LEASE(fl) (fl->fl_flags & FL_LEASE) | |
136 | ||
137 | int leases_enable = 1; | |
138 | int lease_break_time = 45; | |
139 | ||
140 | #define for_each_lock(inode, lockp) \ | |
141 | for (lockp = &inode->i_flock; *lockp != NULL; lockp = &(*lockp)->fl_next) | |
142 | ||
143 | static LIST_HEAD(file_lock_list); | |
144 | static LIST_HEAD(blocked_list); | |
145 | static DEFINE_SPINLOCK(file_lock_lock); | |
146 | ||
147 | /* | |
148 | * Protects the two list heads above, plus the inode->i_flock list | |
149 | * FIXME: should use a spinlock, once lockd and ceph are ready. | |
150 | */ | |
151 | void lock_flocks(void) | |
152 | { | |
153 | spin_lock(&file_lock_lock); | |
154 | } | |
155 | EXPORT_SYMBOL_GPL(lock_flocks); | |
156 | ||
157 | void unlock_flocks(void) | |
158 | { | |
159 | spin_unlock(&file_lock_lock); | |
160 | } | |
161 | EXPORT_SYMBOL_GPL(unlock_flocks); | |
162 | ||
163 | static struct kmem_cache *filelock_cache __read_mostly; | |
164 | ||
165 | /* Allocate an empty lock structure. */ | |
166 | struct file_lock *locks_alloc_lock(void) | |
167 | { | |
168 | return kmem_cache_alloc(filelock_cache, GFP_KERNEL); | |
169 | } | |
170 | EXPORT_SYMBOL_GPL(locks_alloc_lock); | |
171 | ||
172 | void locks_release_private(struct file_lock *fl) | |
173 | { | |
174 | if (fl->fl_ops) { | |
175 | if (fl->fl_ops->fl_release_private) | |
176 | fl->fl_ops->fl_release_private(fl); | |
177 | fl->fl_ops = NULL; | |
178 | } | |
179 | if (fl->fl_lmops) { | |
180 | if (fl->fl_lmops->fl_release_private) | |
181 | fl->fl_lmops->fl_release_private(fl); | |
182 | fl->fl_lmops = NULL; | |
183 | } | |
184 | ||
185 | } | |
186 | EXPORT_SYMBOL_GPL(locks_release_private); | |
187 | ||
188 | /* Free a lock which is not in use. */ | |
189 | static void locks_free_lock(struct file_lock *fl) | |
190 | { | |
191 | BUG_ON(waitqueue_active(&fl->fl_wait)); | |
192 | BUG_ON(!list_empty(&fl->fl_block)); | |
193 | BUG_ON(!list_empty(&fl->fl_link)); | |
194 | ||
195 | locks_release_private(fl); | |
196 | kmem_cache_free(filelock_cache, fl); | |
197 | } | |
198 | ||
199 | void locks_init_lock(struct file_lock *fl) | |
200 | { | |
201 | INIT_LIST_HEAD(&fl->fl_link); | |
202 | INIT_LIST_HEAD(&fl->fl_block); | |
203 | init_waitqueue_head(&fl->fl_wait); | |
204 | fl->fl_next = NULL; | |
205 | fl->fl_fasync = NULL; | |
206 | fl->fl_owner = NULL; | |
207 | fl->fl_pid = 0; | |
208 | fl->fl_nspid = NULL; | |
209 | fl->fl_file = NULL; | |
210 | fl->fl_flags = 0; | |
211 | fl->fl_type = 0; | |
212 | fl->fl_start = fl->fl_end = 0; | |
213 | fl->fl_ops = NULL; | |
214 | fl->fl_lmops = NULL; | |
215 | } | |
216 | ||
217 | EXPORT_SYMBOL(locks_init_lock); | |
218 | ||
219 | /* | |
220 | * Initialises the fields of the file lock which are invariant for | |
221 | * free file_locks. | |
222 | */ | |
223 | static void init_once(void *foo) | |
224 | { | |
225 | struct file_lock *lock = (struct file_lock *) foo; | |
226 | ||
227 | locks_init_lock(lock); | |
228 | } | |
229 | ||
230 | static void locks_copy_private(struct file_lock *new, struct file_lock *fl) | |
231 | { | |
232 | if (fl->fl_ops) { | |
233 | if (fl->fl_ops->fl_copy_lock) | |
234 | fl->fl_ops->fl_copy_lock(new, fl); | |
235 | new->fl_ops = fl->fl_ops; | |
236 | } | |
237 | if (fl->fl_lmops) { | |
238 | if (fl->fl_lmops->fl_copy_lock) | |
239 | fl->fl_lmops->fl_copy_lock(new, fl); | |
240 | new->fl_lmops = fl->fl_lmops; | |
241 | } | |
242 | } | |
243 | ||
244 | /* | |
245 | * Initialize a new lock from an existing file_lock structure. | |
246 | */ | |
247 | void __locks_copy_lock(struct file_lock *new, const struct file_lock *fl) | |
248 | { | |
249 | new->fl_owner = fl->fl_owner; | |
250 | new->fl_pid = fl->fl_pid; | |
251 | new->fl_file = NULL; | |
252 | new->fl_flags = fl->fl_flags; | |
253 | new->fl_type = fl->fl_type; | |
254 | new->fl_start = fl->fl_start; | |
255 | new->fl_end = fl->fl_end; | |
256 | new->fl_ops = NULL; | |
257 | new->fl_lmops = NULL; | |
258 | } | |
259 | EXPORT_SYMBOL(__locks_copy_lock); | |
260 | ||
261 | void locks_copy_lock(struct file_lock *new, struct file_lock *fl) | |
262 | { | |
263 | locks_release_private(new); | |
264 | ||
265 | __locks_copy_lock(new, fl); | |
266 | new->fl_file = fl->fl_file; | |
267 | new->fl_ops = fl->fl_ops; | |
268 | new->fl_lmops = fl->fl_lmops; | |
269 | ||
270 | locks_copy_private(new, fl); | |
271 | } | |
272 | ||
273 | EXPORT_SYMBOL(locks_copy_lock); | |
274 | ||
275 | static inline int flock_translate_cmd(int cmd) { | |
276 | if (cmd & LOCK_MAND) | |
277 | return cmd & (LOCK_MAND | LOCK_RW); | |
278 | switch (cmd) { | |
279 | case LOCK_SH: | |
280 | return F_RDLCK; | |
281 | case LOCK_EX: | |
282 | return F_WRLCK; | |
283 | case LOCK_UN: | |
284 | return F_UNLCK; | |
285 | } | |
286 | return -EINVAL; | |
287 | } | |
288 | ||
289 | /* Fill in a file_lock structure with an appropriate FLOCK lock. */ | |
290 | static int flock_make_lock(struct file *filp, struct file_lock **lock, | |
291 | unsigned int cmd) | |
292 | { | |
293 | struct file_lock *fl; | |
294 | int type = flock_translate_cmd(cmd); | |
295 | if (type < 0) | |
296 | return type; | |
297 | ||
298 | fl = locks_alloc_lock(); | |
299 | if (fl == NULL) | |
300 | return -ENOMEM; | |
301 | ||
302 | fl->fl_file = filp; | |
303 | fl->fl_pid = current->tgid; | |
304 | fl->fl_flags = FL_FLOCK; | |
305 | fl->fl_type = type; | |
306 | fl->fl_end = OFFSET_MAX; | |
307 | ||
308 | *lock = fl; | |
309 | return 0; | |
310 | } | |
311 | ||
312 | static int assign_type(struct file_lock *fl, int type) | |
313 | { | |
314 | switch (type) { | |
315 | case F_RDLCK: | |
316 | case F_WRLCK: | |
317 | case F_UNLCK: | |
318 | fl->fl_type = type; | |
319 | break; | |
320 | default: | |
321 | return -EINVAL; | |
322 | } | |
323 | return 0; | |
324 | } | |
325 | ||
326 | /* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX | |
327 | * style lock. | |
328 | */ | |
329 | static int flock_to_posix_lock(struct file *filp, struct file_lock *fl, | |
330 | struct flock *l) | |
331 | { | |
332 | off_t start, end; | |
333 | ||
334 | switch (l->l_whence) { | |
335 | case SEEK_SET: | |
336 | start = 0; | |
337 | break; | |
338 | case SEEK_CUR: | |
339 | start = filp->f_pos; | |
340 | break; | |
341 | case SEEK_END: | |
342 | start = i_size_read(filp->f_path.dentry->d_inode); | |
343 | break; | |
344 | default: | |
345 | return -EINVAL; | |
346 | } | |
347 | ||
348 | /* POSIX-1996 leaves the case l->l_len < 0 undefined; | |
349 | POSIX-2001 defines it. */ | |
350 | start += l->l_start; | |
351 | if (start < 0) | |
352 | return -EINVAL; | |
353 | fl->fl_end = OFFSET_MAX; | |
354 | if (l->l_len > 0) { | |
355 | end = start + l->l_len - 1; | |
356 | fl->fl_end = end; | |
357 | } else if (l->l_len < 0) { | |
358 | end = start - 1; | |
359 | fl->fl_end = end; | |
360 | start += l->l_len; | |
361 | if (start < 0) | |
362 | return -EINVAL; | |
363 | } | |
364 | fl->fl_start = start; /* we record the absolute position */ | |
365 | if (fl->fl_end < fl->fl_start) | |
366 | return -EOVERFLOW; | |
367 | ||
368 | fl->fl_owner = current->files; | |
369 | fl->fl_pid = current->tgid; | |
370 | fl->fl_file = filp; | |
371 | fl->fl_flags = FL_POSIX; | |
372 | fl->fl_ops = NULL; | |
373 | fl->fl_lmops = NULL; | |
374 | ||
375 | return assign_type(fl, l->l_type); | |
376 | } | |
377 | ||
378 | #if BITS_PER_LONG == 32 | |
379 | static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl, | |
380 | struct flock64 *l) | |
381 | { | |
382 | loff_t start; | |
383 | ||
384 | switch (l->l_whence) { | |
385 | case SEEK_SET: | |
386 | start = 0; | |
387 | break; | |
388 | case SEEK_CUR: | |
389 | start = filp->f_pos; | |
390 | break; | |
391 | case SEEK_END: | |
392 | start = i_size_read(filp->f_path.dentry->d_inode); | |
393 | break; | |
394 | default: | |
395 | return -EINVAL; | |
396 | } | |
397 | ||
398 | start += l->l_start; | |
399 | if (start < 0) | |
400 | return -EINVAL; | |
401 | fl->fl_end = OFFSET_MAX; | |
402 | if (l->l_len > 0) { | |
403 | fl->fl_end = start + l->l_len - 1; | |
404 | } else if (l->l_len < 0) { | |
405 | fl->fl_end = start - 1; | |
406 | start += l->l_len; | |
407 | if (start < 0) | |
408 | return -EINVAL; | |
409 | } | |
410 | fl->fl_start = start; /* we record the absolute position */ | |
411 | if (fl->fl_end < fl->fl_start) | |
412 | return -EOVERFLOW; | |
413 | ||
414 | fl->fl_owner = current->files; | |
415 | fl->fl_pid = current->tgid; | |
416 | fl->fl_file = filp; | |
417 | fl->fl_flags = FL_POSIX; | |
418 | fl->fl_ops = NULL; | |
419 | fl->fl_lmops = NULL; | |
420 | ||
421 | switch (l->l_type) { | |
422 | case F_RDLCK: | |
423 | case F_WRLCK: | |
424 | case F_UNLCK: | |
425 | fl->fl_type = l->l_type; | |
426 | break; | |
427 | default: | |
428 | return -EINVAL; | |
429 | } | |
430 | ||
431 | return (0); | |
432 | } | |
433 | #endif | |
434 | ||
435 | /* default lease lock manager operations */ | |
436 | static void lease_break_callback(struct file_lock *fl) | |
437 | { | |
438 | kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG); | |
439 | } | |
440 | ||
441 | static void lease_release_private_callback(struct file_lock *fl) | |
442 | { | |
443 | if (!fl->fl_file) | |
444 | return; | |
445 | ||
446 | f_delown(fl->fl_file); | |
447 | fl->fl_file->f_owner.signum = 0; | |
448 | } | |
449 | ||
450 | static int lease_mylease_callback(struct file_lock *fl, struct file_lock *try) | |
451 | { | |
452 | return fl->fl_file == try->fl_file; | |
453 | } | |
454 | ||
455 | static const struct lock_manager_operations lease_manager_ops = { | |
456 | .fl_break = lease_break_callback, | |
457 | .fl_release_private = lease_release_private_callback, | |
458 | .fl_mylease = lease_mylease_callback, | |
459 | .fl_change = lease_modify, | |
460 | }; | |
461 | ||
462 | /* | |
463 | * Initialize a lease, use the default lock manager operations | |
464 | */ | |
465 | static int lease_init(struct file *filp, int type, struct file_lock *fl) | |
466 | { | |
467 | if (assign_type(fl, type) != 0) | |
468 | return -EINVAL; | |
469 | ||
470 | fl->fl_owner = current->files; | |
471 | fl->fl_pid = current->tgid; | |
472 | ||
473 | fl->fl_file = filp; | |
474 | fl->fl_flags = FL_LEASE; | |
475 | fl->fl_start = 0; | |
476 | fl->fl_end = OFFSET_MAX; | |
477 | fl->fl_ops = NULL; | |
478 | fl->fl_lmops = &lease_manager_ops; | |
479 | return 0; | |
480 | } | |
481 | ||
482 | /* Allocate a file_lock initialised to this type of lease */ | |
483 | static struct file_lock *lease_alloc(struct file *filp, int type) | |
484 | { | |
485 | struct file_lock *fl = locks_alloc_lock(); | |
486 | int error = -ENOMEM; | |
487 | ||
488 | if (fl == NULL) | |
489 | return ERR_PTR(error); | |
490 | ||
491 | error = lease_init(filp, type, fl); | |
492 | if (error) { | |
493 | locks_free_lock(fl); | |
494 | return ERR_PTR(error); | |
495 | } | |
496 | return fl; | |
497 | } | |
498 | ||
499 | /* Check if two locks overlap each other. | |
500 | */ | |
501 | static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2) | |
502 | { | |
503 | return ((fl1->fl_end >= fl2->fl_start) && | |
504 | (fl2->fl_end >= fl1->fl_start)); | |
505 | } | |
506 | ||
507 | /* | |
508 | * Check whether two locks have the same owner. | |
509 | */ | |
510 | static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2) | |
511 | { | |
512 | if (fl1->fl_lmops && fl1->fl_lmops->fl_compare_owner) | |
513 | return fl2->fl_lmops == fl1->fl_lmops && | |
514 | fl1->fl_lmops->fl_compare_owner(fl1, fl2); | |
515 | return fl1->fl_owner == fl2->fl_owner; | |
516 | } | |
517 | ||
518 | /* Remove waiter from blocker's block list. | |
519 | * When blocker ends up pointing to itself then the list is empty. | |
520 | */ | |
521 | static void __locks_delete_block(struct file_lock *waiter) | |
522 | { | |
523 | list_del_init(&waiter->fl_block); | |
524 | list_del_init(&waiter->fl_link); | |
525 | waiter->fl_next = NULL; | |
526 | } | |
527 | ||
528 | /* | |
529 | */ | |
530 | static void locks_delete_block(struct file_lock *waiter) | |
531 | { | |
532 | lock_flocks(); | |
533 | __locks_delete_block(waiter); | |
534 | unlock_flocks(); | |
535 | } | |
536 | ||
537 | /* Insert waiter into blocker's block list. | |
538 | * We use a circular list so that processes can be easily woken up in | |
539 | * the order they blocked. The documentation doesn't require this but | |
540 | * it seems like the reasonable thing to do. | |
541 | */ | |
542 | static void locks_insert_block(struct file_lock *blocker, | |
543 | struct file_lock *waiter) | |
544 | { | |
545 | BUG_ON(!list_empty(&waiter->fl_block)); | |
546 | list_add_tail(&waiter->fl_block, &blocker->fl_block); | |
547 | waiter->fl_next = blocker; | |
548 | if (IS_POSIX(blocker)) | |
549 | list_add(&waiter->fl_link, &blocked_list); | |
550 | } | |
551 | ||
552 | /* Wake up processes blocked waiting for blocker. | |
553 | * If told to wait then schedule the processes until the block list | |
554 | * is empty, otherwise empty the block list ourselves. | |
555 | */ | |
556 | static void locks_wake_up_blocks(struct file_lock *blocker) | |
557 | { | |
558 | while (!list_empty(&blocker->fl_block)) { | |
559 | struct file_lock *waiter; | |
560 | ||
561 | waiter = list_first_entry(&blocker->fl_block, | |
562 | struct file_lock, fl_block); | |
563 | __locks_delete_block(waiter); | |
564 | if (waiter->fl_lmops && waiter->fl_lmops->fl_notify) | |
565 | waiter->fl_lmops->fl_notify(waiter); | |
566 | else | |
567 | wake_up(&waiter->fl_wait); | |
568 | } | |
569 | } | |
570 | ||
571 | /* Insert file lock fl into an inode's lock list at the position indicated | |
572 | * by pos. At the same time add the lock to the global file lock list. | |
573 | */ | |
574 | static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl) | |
575 | { | |
576 | list_add(&fl->fl_link, &file_lock_list); | |
577 | ||
578 | fl->fl_nspid = get_pid(task_tgid(current)); | |
579 | ||
580 | /* insert into file's list */ | |
581 | fl->fl_next = *pos; | |
582 | *pos = fl; | |
583 | } | |
584 | ||
585 | /* | |
586 | * Delete a lock and then free it. | |
587 | * Wake up processes that are blocked waiting for this lock, | |
588 | * notify the FS that the lock has been cleared and | |
589 | * finally free the lock. | |
590 | */ | |
591 | static void locks_delete_lock(struct file_lock **thisfl_p) | |
592 | { | |
593 | struct file_lock *fl = *thisfl_p; | |
594 | ||
595 | *thisfl_p = fl->fl_next; | |
596 | fl->fl_next = NULL; | |
597 | list_del_init(&fl->fl_link); | |
598 | ||
599 | fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync); | |
600 | if (fl->fl_fasync != NULL) { | |
601 | printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync); | |
602 | fl->fl_fasync = NULL; | |
603 | } | |
604 | ||
605 | if (fl->fl_nspid) { | |
606 | put_pid(fl->fl_nspid); | |
607 | fl->fl_nspid = NULL; | |
608 | } | |
609 | ||
610 | locks_wake_up_blocks(fl); | |
611 | locks_free_lock(fl); | |
612 | } | |
613 | ||
614 | /* Determine if lock sys_fl blocks lock caller_fl. Common functionality | |
615 | * checks for shared/exclusive status of overlapping locks. | |
616 | */ | |
617 | static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl) | |
618 | { | |
619 | if (sys_fl->fl_type == F_WRLCK) | |
620 | return 1; | |
621 | if (caller_fl->fl_type == F_WRLCK) | |
622 | return 1; | |
623 | return 0; | |
624 | } | |
625 | ||
626 | /* Determine if lock sys_fl blocks lock caller_fl. POSIX specific | |
627 | * checking before calling the locks_conflict(). | |
628 | */ | |
629 | static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl) | |
630 | { | |
631 | /* POSIX locks owned by the same process do not conflict with | |
632 | * each other. | |
633 | */ | |
634 | if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl)) | |
635 | return (0); | |
636 | ||
637 | /* Check whether they overlap */ | |
638 | if (!locks_overlap(caller_fl, sys_fl)) | |
639 | return 0; | |
640 | ||
641 | return (locks_conflict(caller_fl, sys_fl)); | |
642 | } | |
643 | ||
644 | /* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific | |
645 | * checking before calling the locks_conflict(). | |
646 | */ | |
647 | static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl) | |
648 | { | |
649 | /* FLOCK locks referring to the same filp do not conflict with | |
650 | * each other. | |
651 | */ | |
652 | if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file)) | |
653 | return (0); | |
654 | if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND)) | |
655 | return 0; | |
656 | ||
657 | return (locks_conflict(caller_fl, sys_fl)); | |
658 | } | |
659 | ||
660 | void | |
661 | posix_test_lock(struct file *filp, struct file_lock *fl) | |
662 | { | |
663 | struct file_lock *cfl; | |
664 | ||
665 | lock_flocks(); | |
666 | for (cfl = filp->f_path.dentry->d_inode->i_flock; cfl; cfl = cfl->fl_next) { | |
667 | if (!IS_POSIX(cfl)) | |
668 | continue; | |
669 | if (posix_locks_conflict(fl, cfl)) | |
670 | break; | |
671 | } | |
672 | if (cfl) { | |
673 | __locks_copy_lock(fl, cfl); | |
674 | if (cfl->fl_nspid) | |
675 | fl->fl_pid = pid_vnr(cfl->fl_nspid); | |
676 | } else | |
677 | fl->fl_type = F_UNLCK; | |
678 | unlock_flocks(); | |
679 | return; | |
680 | } | |
681 | EXPORT_SYMBOL(posix_test_lock); | |
682 | ||
683 | /* | |
684 | * Deadlock detection: | |
685 | * | |
686 | * We attempt to detect deadlocks that are due purely to posix file | |
687 | * locks. | |
688 | * | |
689 | * We assume that a task can be waiting for at most one lock at a time. | |
690 | * So for any acquired lock, the process holding that lock may be | |
691 | * waiting on at most one other lock. That lock in turns may be held by | |
692 | * someone waiting for at most one other lock. Given a requested lock | |
693 | * caller_fl which is about to wait for a conflicting lock block_fl, we | |
694 | * follow this chain of waiters to ensure we are not about to create a | |
695 | * cycle. | |
696 | * | |
697 | * Since we do this before we ever put a process to sleep on a lock, we | |
698 | * are ensured that there is never a cycle; that is what guarantees that | |
699 | * the while() loop in posix_locks_deadlock() eventually completes. | |
700 | * | |
701 | * Note: the above assumption may not be true when handling lock | |
702 | * requests from a broken NFS client. It may also fail in the presence | |
703 | * of tasks (such as posix threads) sharing the same open file table. | |
704 | * | |
705 | * To handle those cases, we just bail out after a few iterations. | |
706 | */ | |
707 | ||
708 | #define MAX_DEADLK_ITERATIONS 10 | |
709 | ||
710 | /* Find a lock that the owner of the given block_fl is blocking on. */ | |
711 | static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl) | |
712 | { | |
713 | struct file_lock *fl; | |
714 | ||
715 | list_for_each_entry(fl, &blocked_list, fl_link) { | |
716 | if (posix_same_owner(fl, block_fl)) | |
717 | return fl->fl_next; | |
718 | } | |
719 | return NULL; | |
720 | } | |
721 | ||
722 | static int posix_locks_deadlock(struct file_lock *caller_fl, | |
723 | struct file_lock *block_fl) | |
724 | { | |
725 | int i = 0; | |
726 | ||
727 | while ((block_fl = what_owner_is_waiting_for(block_fl))) { | |
728 | if (i++ > MAX_DEADLK_ITERATIONS) | |
729 | return 0; | |
730 | if (posix_same_owner(caller_fl, block_fl)) | |
731 | return 1; | |
732 | } | |
733 | return 0; | |
734 | } | |
735 | ||
736 | /* Try to create a FLOCK lock on filp. We always insert new FLOCK locks | |
737 | * after any leases, but before any posix locks. | |
738 | * | |
739 | * Note that if called with an FL_EXISTS argument, the caller may determine | |
740 | * whether or not a lock was successfully freed by testing the return | |
741 | * value for -ENOENT. | |
742 | */ | |
743 | static int flock_lock_file(struct file *filp, struct file_lock *request) | |
744 | { | |
745 | struct file_lock *new_fl = NULL; | |
746 | struct file_lock **before; | |
747 | struct inode * inode = filp->f_path.dentry->d_inode; | |
748 | int error = 0; | |
749 | int found = 0; | |
750 | ||
751 | if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) { | |
752 | new_fl = locks_alloc_lock(); | |
753 | if (!new_fl) | |
754 | return -ENOMEM; | |
755 | } | |
756 | ||
757 | lock_flocks(); | |
758 | if (request->fl_flags & FL_ACCESS) | |
759 | goto find_conflict; | |
760 | ||
761 | for_each_lock(inode, before) { | |
762 | struct file_lock *fl = *before; | |
763 | if (IS_POSIX(fl)) | |
764 | break; | |
765 | if (IS_LEASE(fl)) | |
766 | continue; | |
767 | if (filp != fl->fl_file) | |
768 | continue; | |
769 | if (request->fl_type == fl->fl_type) | |
770 | goto out; | |
771 | found = 1; | |
772 | locks_delete_lock(before); | |
773 | break; | |
774 | } | |
775 | ||
776 | if (request->fl_type == F_UNLCK) { | |
777 | if ((request->fl_flags & FL_EXISTS) && !found) | |
778 | error = -ENOENT; | |
779 | goto out; | |
780 | } | |
781 | ||
782 | /* | |
783 | * If a higher-priority process was blocked on the old file lock, | |
784 | * give it the opportunity to lock the file. | |
785 | */ | |
786 | if (found) { | |
787 | unlock_flocks(); | |
788 | cond_resched(); | |
789 | lock_flocks(); | |
790 | } | |
791 | ||
792 | find_conflict: | |
793 | for_each_lock(inode, before) { | |
794 | struct file_lock *fl = *before; | |
795 | if (IS_POSIX(fl)) | |
796 | break; | |
797 | if (IS_LEASE(fl)) | |
798 | continue; | |
799 | if (!flock_locks_conflict(request, fl)) | |
800 | continue; | |
801 | error = -EAGAIN; | |
802 | if (!(request->fl_flags & FL_SLEEP)) | |
803 | goto out; | |
804 | error = FILE_LOCK_DEFERRED; | |
805 | locks_insert_block(fl, request); | |
806 | goto out; | |
807 | } | |
808 | if (request->fl_flags & FL_ACCESS) | |
809 | goto out; | |
810 | locks_copy_lock(new_fl, request); | |
811 | locks_insert_lock(before, new_fl); | |
812 | new_fl = NULL; | |
813 | error = 0; | |
814 | ||
815 | out: | |
816 | unlock_flocks(); | |
817 | if (new_fl) | |
818 | locks_free_lock(new_fl); | |
819 | return error; | |
820 | } | |
821 | ||
822 | static int __posix_lock_file(struct inode *inode, struct file_lock *request, struct file_lock *conflock) | |
823 | { | |
824 | struct file_lock *fl; | |
825 | struct file_lock *new_fl = NULL; | |
826 | struct file_lock *new_fl2 = NULL; | |
827 | struct file_lock *left = NULL; | |
828 | struct file_lock *right = NULL; | |
829 | struct file_lock **before; | |
830 | int error, added = 0; | |
831 | ||
832 | /* | |
833 | * We may need two file_lock structures for this operation, | |
834 | * so we get them in advance to avoid races. | |
835 | * | |
836 | * In some cases we can be sure, that no new locks will be needed | |
837 | */ | |
838 | if (!(request->fl_flags & FL_ACCESS) && | |
839 | (request->fl_type != F_UNLCK || | |
840 | request->fl_start != 0 || request->fl_end != OFFSET_MAX)) { | |
841 | new_fl = locks_alloc_lock(); | |
842 | new_fl2 = locks_alloc_lock(); | |
843 | } | |
844 | ||
845 | lock_flocks(); | |
846 | if (request->fl_type != F_UNLCK) { | |
847 | for_each_lock(inode, before) { | |
848 | fl = *before; | |
849 | if (!IS_POSIX(fl)) | |
850 | continue; | |
851 | if (!posix_locks_conflict(request, fl)) | |
852 | continue; | |
853 | if (conflock) | |
854 | __locks_copy_lock(conflock, fl); | |
855 | error = -EAGAIN; | |
856 | if (!(request->fl_flags & FL_SLEEP)) | |
857 | goto out; | |
858 | error = -EDEADLK; | |
859 | if (posix_locks_deadlock(request, fl)) | |
860 | goto out; | |
861 | error = FILE_LOCK_DEFERRED; | |
862 | locks_insert_block(fl, request); | |
863 | goto out; | |
864 | } | |
865 | } | |
866 | ||
867 | /* If we're just looking for a conflict, we're done. */ | |
868 | error = 0; | |
869 | if (request->fl_flags & FL_ACCESS) | |
870 | goto out; | |
871 | ||
872 | /* | |
873 | * Find the first old lock with the same owner as the new lock. | |
874 | */ | |
875 | ||
876 | before = &inode->i_flock; | |
877 | ||
878 | /* First skip locks owned by other processes. */ | |
879 | while ((fl = *before) && (!IS_POSIX(fl) || | |
880 | !posix_same_owner(request, fl))) { | |
881 | before = &fl->fl_next; | |
882 | } | |
883 | ||
884 | /* Process locks with this owner. */ | |
885 | while ((fl = *before) && posix_same_owner(request, fl)) { | |
886 | /* Detect adjacent or overlapping regions (if same lock type) | |
887 | */ | |
888 | if (request->fl_type == fl->fl_type) { | |
889 | /* In all comparisons of start vs end, use | |
890 | * "start - 1" rather than "end + 1". If end | |
891 | * is OFFSET_MAX, end + 1 will become negative. | |
892 | */ | |
893 | if (fl->fl_end < request->fl_start - 1) | |
894 | goto next_lock; | |
895 | /* If the next lock in the list has entirely bigger | |
896 | * addresses than the new one, insert the lock here. | |
897 | */ | |
898 | if (fl->fl_start - 1 > request->fl_end) | |
899 | break; | |
900 | ||
901 | /* If we come here, the new and old lock are of the | |
902 | * same type and adjacent or overlapping. Make one | |
903 | * lock yielding from the lower start address of both | |
904 | * locks to the higher end address. | |
905 | */ | |
906 | if (fl->fl_start > request->fl_start) | |
907 | fl->fl_start = request->fl_start; | |
908 | else | |
909 | request->fl_start = fl->fl_start; | |
910 | if (fl->fl_end < request->fl_end) | |
911 | fl->fl_end = request->fl_end; | |
912 | else | |
913 | request->fl_end = fl->fl_end; | |
914 | if (added) { | |
915 | locks_delete_lock(before); | |
916 | continue; | |
917 | } | |
918 | request = fl; | |
919 | added = 1; | |
920 | } | |
921 | else { | |
922 | /* Processing for different lock types is a bit | |
923 | * more complex. | |
924 | */ | |
925 | if (fl->fl_end < request->fl_start) | |
926 | goto next_lock; | |
927 | if (fl->fl_start > request->fl_end) | |
928 | break; | |
929 | if (request->fl_type == F_UNLCK) | |
930 | added = 1; | |
931 | if (fl->fl_start < request->fl_start) | |
932 | left = fl; | |
933 | /* If the next lock in the list has a higher end | |
934 | * address than the new one, insert the new one here. | |
935 | */ | |
936 | if (fl->fl_end > request->fl_end) { | |
937 | right = fl; | |
938 | break; | |
939 | } | |
940 | if (fl->fl_start >= request->fl_start) { | |
941 | /* The new lock completely replaces an old | |
942 | * one (This may happen several times). | |
943 | */ | |
944 | if (added) { | |
945 | locks_delete_lock(before); | |
946 | continue; | |
947 | } | |
948 | /* Replace the old lock with the new one. | |
949 | * Wake up anybody waiting for the old one, | |
950 | * as the change in lock type might satisfy | |
951 | * their needs. | |
952 | */ | |
953 | locks_wake_up_blocks(fl); | |
954 | fl->fl_start = request->fl_start; | |
955 | fl->fl_end = request->fl_end; | |
956 | fl->fl_type = request->fl_type; | |
957 | locks_release_private(fl); | |
958 | locks_copy_private(fl, request); | |
959 | request = fl; | |
960 | added = 1; | |
961 | } | |
962 | } | |
963 | /* Go on to next lock. | |
964 | */ | |
965 | next_lock: | |
966 | before = &fl->fl_next; | |
967 | } | |
968 | ||
969 | /* | |
970 | * The above code only modifies existing locks in case of | |
971 | * merging or replacing. If new lock(s) need to be inserted | |
972 | * all modifications are done bellow this, so it's safe yet to | |
973 | * bail out. | |
974 | */ | |
975 | error = -ENOLCK; /* "no luck" */ | |
976 | if (right && left == right && !new_fl2) | |
977 | goto out; | |
978 | ||
979 | error = 0; | |
980 | if (!added) { | |
981 | if (request->fl_type == F_UNLCK) { | |
982 | if (request->fl_flags & FL_EXISTS) | |
983 | error = -ENOENT; | |
984 | goto out; | |
985 | } | |
986 | ||
987 | if (!new_fl) { | |
988 | error = -ENOLCK; | |
989 | goto out; | |
990 | } | |
991 | locks_copy_lock(new_fl, request); | |
992 | locks_insert_lock(before, new_fl); | |
993 | new_fl = NULL; | |
994 | } | |
995 | if (right) { | |
996 | if (left == right) { | |
997 | /* The new lock breaks the old one in two pieces, | |
998 | * so we have to use the second new lock. | |
999 | */ | |
1000 | left = new_fl2; | |
1001 | new_fl2 = NULL; | |
1002 | locks_copy_lock(left, right); | |
1003 | locks_insert_lock(before, left); | |
1004 | } | |
1005 | right->fl_start = request->fl_end + 1; | |
1006 | locks_wake_up_blocks(right); | |
1007 | } | |
1008 | if (left) { | |
1009 | left->fl_end = request->fl_start - 1; | |
1010 | locks_wake_up_blocks(left); | |
1011 | } | |
1012 | out: | |
1013 | unlock_flocks(); | |
1014 | /* | |
1015 | * Free any unused locks. | |
1016 | */ | |
1017 | if (new_fl) | |
1018 | locks_free_lock(new_fl); | |
1019 | if (new_fl2) | |
1020 | locks_free_lock(new_fl2); | |
1021 | return error; | |
1022 | } | |
1023 | ||
1024 | /** | |
1025 | * posix_lock_file - Apply a POSIX-style lock to a file | |
1026 | * @filp: The file to apply the lock to | |
1027 | * @fl: The lock to be applied | |
1028 | * @conflock: Place to return a copy of the conflicting lock, if found. | |
1029 | * | |
1030 | * Add a POSIX style lock to a file. | |
1031 | * We merge adjacent & overlapping locks whenever possible. | |
1032 | * POSIX locks are sorted by owner task, then by starting address | |
1033 | * | |
1034 | * Note that if called with an FL_EXISTS argument, the caller may determine | |
1035 | * whether or not a lock was successfully freed by testing the return | |
1036 | * value for -ENOENT. | |
1037 | */ | |
1038 | int posix_lock_file(struct file *filp, struct file_lock *fl, | |
1039 | struct file_lock *conflock) | |
1040 | { | |
1041 | return __posix_lock_file(filp->f_path.dentry->d_inode, fl, conflock); | |
1042 | } | |
1043 | EXPORT_SYMBOL(posix_lock_file); | |
1044 | ||
1045 | /** | |
1046 | * posix_lock_file_wait - Apply a POSIX-style lock to a file | |
1047 | * @filp: The file to apply the lock to | |
1048 | * @fl: The lock to be applied | |
1049 | * | |
1050 | * Add a POSIX style lock to a file. | |
1051 | * We merge adjacent & overlapping locks whenever possible. | |
1052 | * POSIX locks are sorted by owner task, then by starting address | |
1053 | */ | |
1054 | int posix_lock_file_wait(struct file *filp, struct file_lock *fl) | |
1055 | { | |
1056 | int error; | |
1057 | might_sleep (); | |
1058 | for (;;) { | |
1059 | error = posix_lock_file(filp, fl, NULL); | |
1060 | if (error != FILE_LOCK_DEFERRED) | |
1061 | break; | |
1062 | error = wait_event_interruptible(fl->fl_wait, !fl->fl_next); | |
1063 | if (!error) | |
1064 | continue; | |
1065 | ||
1066 | locks_delete_block(fl); | |
1067 | break; | |
1068 | } | |
1069 | return error; | |
1070 | } | |
1071 | EXPORT_SYMBOL(posix_lock_file_wait); | |
1072 | ||
1073 | /** | |
1074 | * locks_mandatory_locked - Check for an active lock | |
1075 | * @inode: the file to check | |
1076 | * | |
1077 | * Searches the inode's list of locks to find any POSIX locks which conflict. | |
1078 | * This function is called from locks_verify_locked() only. | |
1079 | */ | |
1080 | int locks_mandatory_locked(struct inode *inode) | |
1081 | { | |
1082 | fl_owner_t owner = current->files; | |
1083 | struct file_lock *fl; | |
1084 | ||
1085 | /* | |
1086 | * Search the lock list for this inode for any POSIX locks. | |
1087 | */ | |
1088 | lock_flocks(); | |
1089 | for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) { | |
1090 | if (!IS_POSIX(fl)) | |
1091 | continue; | |
1092 | if (fl->fl_owner != owner) | |
1093 | break; | |
1094 | } | |
1095 | unlock_flocks(); | |
1096 | return fl ? -EAGAIN : 0; | |
1097 | } | |
1098 | ||
1099 | /** | |
1100 | * locks_mandatory_area - Check for a conflicting lock | |
1101 | * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ | |
1102 | * for shared | |
1103 | * @inode: the file to check | |
1104 | * @filp: how the file was opened (if it was) | |
1105 | * @offset: start of area to check | |
1106 | * @count: length of area to check | |
1107 | * | |
1108 | * Searches the inode's list of locks to find any POSIX locks which conflict. | |
1109 | * This function is called from rw_verify_area() and | |
1110 | * locks_verify_truncate(). | |
1111 | */ | |
1112 | int locks_mandatory_area(int read_write, struct inode *inode, | |
1113 | struct file *filp, loff_t offset, | |
1114 | size_t count) | |
1115 | { | |
1116 | struct file_lock fl; | |
1117 | int error; | |
1118 | ||
1119 | locks_init_lock(&fl); | |
1120 | fl.fl_owner = current->files; | |
1121 | fl.fl_pid = current->tgid; | |
1122 | fl.fl_file = filp; | |
1123 | fl.fl_flags = FL_POSIX | FL_ACCESS; | |
1124 | if (filp && !(filp->f_flags & O_NONBLOCK)) | |
1125 | fl.fl_flags |= FL_SLEEP; | |
1126 | fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK; | |
1127 | fl.fl_start = offset; | |
1128 | fl.fl_end = offset + count - 1; | |
1129 | ||
1130 | for (;;) { | |
1131 | error = __posix_lock_file(inode, &fl, NULL); | |
1132 | if (error != FILE_LOCK_DEFERRED) | |
1133 | break; | |
1134 | error = wait_event_interruptible(fl.fl_wait, !fl.fl_next); | |
1135 | if (!error) { | |
1136 | /* | |
1137 | * If we've been sleeping someone might have | |
1138 | * changed the permissions behind our back. | |
1139 | */ | |
1140 | if (__mandatory_lock(inode)) | |
1141 | continue; | |
1142 | } | |
1143 | ||
1144 | locks_delete_block(&fl); | |
1145 | break; | |
1146 | } | |
1147 | ||
1148 | return error; | |
1149 | } | |
1150 | ||
1151 | EXPORT_SYMBOL(locks_mandatory_area); | |
1152 | ||
1153 | /* We already had a lease on this file; just change its type */ | |
1154 | int lease_modify(struct file_lock **before, int arg) | |
1155 | { | |
1156 | struct file_lock *fl = *before; | |
1157 | int error = assign_type(fl, arg); | |
1158 | ||
1159 | if (error) | |
1160 | return error; | |
1161 | locks_wake_up_blocks(fl); | |
1162 | if (arg == F_UNLCK) | |
1163 | locks_delete_lock(before); | |
1164 | return 0; | |
1165 | } | |
1166 | ||
1167 | EXPORT_SYMBOL(lease_modify); | |
1168 | ||
1169 | static void time_out_leases(struct inode *inode) | |
1170 | { | |
1171 | struct file_lock **before; | |
1172 | struct file_lock *fl; | |
1173 | ||
1174 | before = &inode->i_flock; | |
1175 | while ((fl = *before) && IS_LEASE(fl) && (fl->fl_type & F_INPROGRESS)) { | |
1176 | if ((fl->fl_break_time == 0) | |
1177 | || time_before(jiffies, fl->fl_break_time)) { | |
1178 | before = &fl->fl_next; | |
1179 | continue; | |
1180 | } | |
1181 | lease_modify(before, fl->fl_type & ~F_INPROGRESS); | |
1182 | if (fl == *before) /* lease_modify may have freed fl */ | |
1183 | before = &fl->fl_next; | |
1184 | } | |
1185 | } | |
1186 | ||
1187 | /** | |
1188 | * __break_lease - revoke all outstanding leases on file | |
1189 | * @inode: the inode of the file to return | |
1190 | * @mode: the open mode (read or write) | |
1191 | * | |
1192 | * break_lease (inlined for speed) has checked there already is at least | |
1193 | * some kind of lock (maybe a lease) on this file. Leases are broken on | |
1194 | * a call to open() or truncate(). This function can sleep unless you | |
1195 | * specified %O_NONBLOCK to your open(). | |
1196 | */ | |
1197 | int __break_lease(struct inode *inode, unsigned int mode) | |
1198 | { | |
1199 | int error = 0, future; | |
1200 | struct file_lock *new_fl, *flock; | |
1201 | struct file_lock *fl; | |
1202 | unsigned long break_time; | |
1203 | int i_have_this_lease = 0; | |
1204 | int want_write = (mode & O_ACCMODE) != O_RDONLY; | |
1205 | ||
1206 | new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK); | |
1207 | ||
1208 | lock_flocks(); | |
1209 | ||
1210 | time_out_leases(inode); | |
1211 | ||
1212 | flock = inode->i_flock; | |
1213 | if ((flock == NULL) || !IS_LEASE(flock)) | |
1214 | goto out; | |
1215 | ||
1216 | for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) | |
1217 | if (fl->fl_owner == current->files) | |
1218 | i_have_this_lease = 1; | |
1219 | ||
1220 | if (want_write) { | |
1221 | /* If we want write access, we have to revoke any lease. */ | |
1222 | future = F_UNLCK | F_INPROGRESS; | |
1223 | } else if (flock->fl_type & F_INPROGRESS) { | |
1224 | /* If the lease is already being broken, we just leave it */ | |
1225 | future = flock->fl_type; | |
1226 | } else if (flock->fl_type & F_WRLCK) { | |
1227 | /* Downgrade the exclusive lease to a read-only lease. */ | |
1228 | future = F_RDLCK | F_INPROGRESS; | |
1229 | } else { | |
1230 | /* the existing lease was read-only, so we can read too. */ | |
1231 | goto out; | |
1232 | } | |
1233 | ||
1234 | if (IS_ERR(new_fl) && !i_have_this_lease | |
1235 | && ((mode & O_NONBLOCK) == 0)) { | |
1236 | error = PTR_ERR(new_fl); | |
1237 | goto out; | |
1238 | } | |
1239 | ||
1240 | break_time = 0; | |
1241 | if (lease_break_time > 0) { | |
1242 | break_time = jiffies + lease_break_time * HZ; | |
1243 | if (break_time == 0) | |
1244 | break_time++; /* so that 0 means no break time */ | |
1245 | } | |
1246 | ||
1247 | for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) { | |
1248 | if (fl->fl_type != future) { | |
1249 | fl->fl_type = future; | |
1250 | fl->fl_break_time = break_time; | |
1251 | /* lease must have lmops break callback */ | |
1252 | fl->fl_lmops->fl_break(fl); | |
1253 | } | |
1254 | } | |
1255 | ||
1256 | if (i_have_this_lease || (mode & O_NONBLOCK)) { | |
1257 | error = -EWOULDBLOCK; | |
1258 | goto out; | |
1259 | } | |
1260 | ||
1261 | restart: | |
1262 | break_time = flock->fl_break_time; | |
1263 | if (break_time != 0) { | |
1264 | break_time -= jiffies; | |
1265 | if (break_time == 0) | |
1266 | break_time++; | |
1267 | } | |
1268 | locks_insert_block(flock, new_fl); | |
1269 | unlock_flocks(); | |
1270 | error = wait_event_interruptible_timeout(new_fl->fl_wait, | |
1271 | !new_fl->fl_next, break_time); | |
1272 | lock_flocks(); | |
1273 | __locks_delete_block(new_fl); | |
1274 | if (error >= 0) { | |
1275 | if (error == 0) | |
1276 | time_out_leases(inode); | |
1277 | /* Wait for the next lease that has not been broken yet */ | |
1278 | for (flock = inode->i_flock; flock && IS_LEASE(flock); | |
1279 | flock = flock->fl_next) { | |
1280 | if (flock->fl_type & F_INPROGRESS) | |
1281 | goto restart; | |
1282 | } | |
1283 | error = 0; | |
1284 | } | |
1285 | ||
1286 | out: | |
1287 | unlock_flocks(); | |
1288 | if (!IS_ERR(new_fl)) | |
1289 | locks_free_lock(new_fl); | |
1290 | return error; | |
1291 | } | |
1292 | ||
1293 | EXPORT_SYMBOL(__break_lease); | |
1294 | ||
1295 | /** | |
1296 | * lease_get_mtime - get the last modified time of an inode | |
1297 | * @inode: the inode | |
1298 | * @time: pointer to a timespec which will contain the last modified time | |
1299 | * | |
1300 | * This is to force NFS clients to flush their caches for files with | |
1301 | * exclusive leases. The justification is that if someone has an | |
1302 | * exclusive lease, then they could be modifying it. | |
1303 | */ | |
1304 | void lease_get_mtime(struct inode *inode, struct timespec *time) | |
1305 | { | |
1306 | struct file_lock *flock = inode->i_flock; | |
1307 | if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK)) | |
1308 | *time = current_fs_time(inode->i_sb); | |
1309 | else | |
1310 | *time = inode->i_mtime; | |
1311 | } | |
1312 | ||
1313 | EXPORT_SYMBOL(lease_get_mtime); | |
1314 | ||
1315 | /** | |
1316 | * fcntl_getlease - Enquire what lease is currently active | |
1317 | * @filp: the file | |
1318 | * | |
1319 | * The value returned by this function will be one of | |
1320 | * (if no lease break is pending): | |
1321 | * | |
1322 | * %F_RDLCK to indicate a shared lease is held. | |
1323 | * | |
1324 | * %F_WRLCK to indicate an exclusive lease is held. | |
1325 | * | |
1326 | * %F_UNLCK to indicate no lease is held. | |
1327 | * | |
1328 | * (if a lease break is pending): | |
1329 | * | |
1330 | * %F_RDLCK to indicate an exclusive lease needs to be | |
1331 | * changed to a shared lease (or removed). | |
1332 | * | |
1333 | * %F_UNLCK to indicate the lease needs to be removed. | |
1334 | * | |
1335 | * XXX: sfr & willy disagree over whether F_INPROGRESS | |
1336 | * should be returned to userspace. | |
1337 | */ | |
1338 | int fcntl_getlease(struct file *filp) | |
1339 | { | |
1340 | struct file_lock *fl; | |
1341 | int type = F_UNLCK; | |
1342 | ||
1343 | lock_flocks(); | |
1344 | time_out_leases(filp->f_path.dentry->d_inode); | |
1345 | for (fl = filp->f_path.dentry->d_inode->i_flock; fl && IS_LEASE(fl); | |
1346 | fl = fl->fl_next) { | |
1347 | if (fl->fl_file == filp) { | |
1348 | type = fl->fl_type & ~F_INPROGRESS; | |
1349 | break; | |
1350 | } | |
1351 | } | |
1352 | unlock_flocks(); | |
1353 | return type; | |
1354 | } | |
1355 | ||
1356 | /** | |
1357 | * generic_setlease - sets a lease on an open file | |
1358 | * @filp: file pointer | |
1359 | * @arg: type of lease to obtain | |
1360 | * @flp: input - file_lock to use, output - file_lock inserted | |
1361 | * | |
1362 | * The (input) flp->fl_lmops->fl_break function is required | |
1363 | * by break_lease(). | |
1364 | * | |
1365 | * Called with file_lock_lock held. | |
1366 | */ | |
1367 | int generic_setlease(struct file *filp, long arg, struct file_lock **flp) | |
1368 | { | |
1369 | struct file_lock *fl, **before, **my_before = NULL, *lease; | |
1370 | struct dentry *dentry = filp->f_path.dentry; | |
1371 | struct inode *inode = dentry->d_inode; | |
1372 | int error, rdlease_count = 0, wrlease_count = 0; | |
1373 | ||
1374 | if ((current_fsuid() != inode->i_uid) && !capable(CAP_LEASE)) | |
1375 | return -EACCES; | |
1376 | if (!S_ISREG(inode->i_mode)) | |
1377 | return -EINVAL; | |
1378 | error = security_file_lock(filp, arg); | |
1379 | if (error) | |
1380 | return error; | |
1381 | ||
1382 | time_out_leases(inode); | |
1383 | ||
1384 | BUG_ON(!(*flp)->fl_lmops->fl_break); | |
1385 | ||
1386 | lease = *flp; | |
1387 | ||
1388 | if (arg != F_UNLCK) { | |
1389 | error = -EAGAIN; | |
1390 | if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0)) | |
1391 | goto out; | |
1392 | if ((arg == F_WRLCK) | |
1393 | && ((atomic_read(&dentry->d_count) > 1) | |
1394 | || (atomic_read(&inode->i_count) > 1))) | |
1395 | goto out; | |
1396 | } | |
1397 | ||
1398 | /* | |
1399 | * At this point, we know that if there is an exclusive | |
1400 | * lease on this file, then we hold it on this filp | |
1401 | * (otherwise our open of this file would have blocked). | |
1402 | * And if we are trying to acquire an exclusive lease, | |
1403 | * then the file is not open by anyone (including us) | |
1404 | * except for this filp. | |
1405 | */ | |
1406 | for (before = &inode->i_flock; | |
1407 | ((fl = *before) != NULL) && IS_LEASE(fl); | |
1408 | before = &fl->fl_next) { | |
1409 | if (lease->fl_lmops->fl_mylease(fl, lease)) | |
1410 | my_before = before; | |
1411 | else if (fl->fl_type == (F_INPROGRESS | F_UNLCK)) | |
1412 | /* | |
1413 | * Someone is in the process of opening this | |
1414 | * file for writing so we may not take an | |
1415 | * exclusive lease on it. | |
1416 | */ | |
1417 | wrlease_count++; | |
1418 | else | |
1419 | rdlease_count++; | |
1420 | } | |
1421 | ||
1422 | error = -EAGAIN; | |
1423 | if ((arg == F_RDLCK && (wrlease_count > 0)) || | |
1424 | (arg == F_WRLCK && ((rdlease_count + wrlease_count) > 0))) | |
1425 | goto out; | |
1426 | ||
1427 | if (my_before != NULL) { | |
1428 | *flp = *my_before; | |
1429 | error = lease->fl_lmops->fl_change(my_before, arg); | |
1430 | goto out; | |
1431 | } | |
1432 | ||
1433 | if (arg == F_UNLCK) | |
1434 | goto out; | |
1435 | ||
1436 | error = -EINVAL; | |
1437 | if (!leases_enable) | |
1438 | goto out; | |
1439 | ||
1440 | locks_insert_lock(before, lease); | |
1441 | return 0; | |
1442 | ||
1443 | out: | |
1444 | locks_free_lock(lease); | |
1445 | return error; | |
1446 | } | |
1447 | EXPORT_SYMBOL(generic_setlease); | |
1448 | ||
1449 | static int __vfs_setlease(struct file *filp, long arg, struct file_lock **lease) | |
1450 | { | |
1451 | if (filp->f_op && filp->f_op->setlease) | |
1452 | return filp->f_op->setlease(filp, arg, lease); | |
1453 | else | |
1454 | return generic_setlease(filp, arg, lease); | |
1455 | } | |
1456 | ||
1457 | /** | |
1458 | * vfs_setlease - sets a lease on an open file | |
1459 | * @filp: file pointer | |
1460 | * @arg: type of lease to obtain | |
1461 | * @lease: file_lock to use | |
1462 | * | |
1463 | * Call this to establish a lease on the file. | |
1464 | * The (*lease)->fl_lmops->fl_break operation must be set; if not, | |
1465 | * break_lease will oops! | |
1466 | * | |
1467 | * This will call the filesystem's setlease file method, if | |
1468 | * defined. Note that there is no getlease method; instead, the | |
1469 | * filesystem setlease method should call back to setlease() to | |
1470 | * add a lease to the inode's lease list, where fcntl_getlease() can | |
1471 | * find it. Since fcntl_getlease() only reports whether the current | |
1472 | * task holds a lease, a cluster filesystem need only do this for | |
1473 | * leases held by processes on this node. | |
1474 | * | |
1475 | * There is also no break_lease method; filesystems that | |
1476 | * handle their own leases should break leases themselves from the | |
1477 | * filesystem's open, create, and (on truncate) setattr methods. | |
1478 | * | |
1479 | * Warning: the only current setlease methods exist only to disable | |
1480 | * leases in certain cases. More vfs changes may be required to | |
1481 | * allow a full filesystem lease implementation. | |
1482 | */ | |
1483 | ||
1484 | int vfs_setlease(struct file *filp, long arg, struct file_lock **lease) | |
1485 | { | |
1486 | int error; | |
1487 | ||
1488 | lock_flocks(); | |
1489 | error = __vfs_setlease(filp, arg, lease); | |
1490 | unlock_flocks(); | |
1491 | ||
1492 | return error; | |
1493 | } | |
1494 | EXPORT_SYMBOL_GPL(vfs_setlease); | |
1495 | ||
1496 | /** | |
1497 | * fcntl_setlease - sets a lease on an open file | |
1498 | * @fd: open file descriptor | |
1499 | * @filp: file pointer | |
1500 | * @arg: type of lease to obtain | |
1501 | * | |
1502 | * Call this fcntl to establish a lease on the file. | |
1503 | * Note that you also need to call %F_SETSIG to | |
1504 | * receive a signal when the lease is broken. | |
1505 | */ | |
1506 | int fcntl_setlease(unsigned int fd, struct file *filp, long arg) | |
1507 | { | |
1508 | struct file_lock *fl; | |
1509 | struct fasync_struct *new; | |
1510 | struct inode *inode = filp->f_path.dentry->d_inode; | |
1511 | int error; | |
1512 | ||
1513 | fl = lease_alloc(filp, arg); | |
1514 | if (IS_ERR(fl)) | |
1515 | return PTR_ERR(fl); | |
1516 | ||
1517 | new = fasync_alloc(); | |
1518 | if (!new) { | |
1519 | locks_free_lock(fl); | |
1520 | return -ENOMEM; | |
1521 | } | |
1522 | lock_flocks(); | |
1523 | error = __vfs_setlease(filp, arg, &fl); | |
1524 | if (error || arg == F_UNLCK) | |
1525 | goto out_unlock; | |
1526 | ||
1527 | /* | |
1528 | * fasync_insert_entry() returns the old entry if any. | |
1529 | * If there was no old entry, then it used 'new' and | |
1530 | * inserted it into the fasync list. Clear new so that | |
1531 | * we don't release it here. | |
1532 | */ | |
1533 | if (!fasync_insert_entry(fd, filp, &fl->fl_fasync, new)) | |
1534 | new = NULL; | |
1535 | ||
1536 | if (error < 0) { | |
1537 | /* remove lease just inserted by setlease */ | |
1538 | fl->fl_type = F_UNLCK | F_INPROGRESS; | |
1539 | fl->fl_break_time = jiffies - 10; | |
1540 | time_out_leases(inode); | |
1541 | goto out_unlock; | |
1542 | } | |
1543 | ||
1544 | error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0); | |
1545 | out_unlock: | |
1546 | unlock_flocks(); | |
1547 | if (new) | |
1548 | fasync_free(new); | |
1549 | return error; | |
1550 | } | |
1551 | ||
1552 | /** | |
1553 | * flock_lock_file_wait - Apply a FLOCK-style lock to a file | |
1554 | * @filp: The file to apply the lock to | |
1555 | * @fl: The lock to be applied | |
1556 | * | |
1557 | * Add a FLOCK style lock to a file. | |
1558 | */ | |
1559 | int flock_lock_file_wait(struct file *filp, struct file_lock *fl) | |
1560 | { | |
1561 | int error; | |
1562 | might_sleep(); | |
1563 | for (;;) { | |
1564 | error = flock_lock_file(filp, fl); | |
1565 | if (error != FILE_LOCK_DEFERRED) | |
1566 | break; | |
1567 | error = wait_event_interruptible(fl->fl_wait, !fl->fl_next); | |
1568 | if (!error) | |
1569 | continue; | |
1570 | ||
1571 | locks_delete_block(fl); | |
1572 | break; | |
1573 | } | |
1574 | return error; | |
1575 | } | |
1576 | ||
1577 | EXPORT_SYMBOL(flock_lock_file_wait); | |
1578 | ||
1579 | /** | |
1580 | * sys_flock: - flock() system call. | |
1581 | * @fd: the file descriptor to lock. | |
1582 | * @cmd: the type of lock to apply. | |
1583 | * | |
1584 | * Apply a %FL_FLOCK style lock to an open file descriptor. | |
1585 | * The @cmd can be one of | |
1586 | * | |
1587 | * %LOCK_SH -- a shared lock. | |
1588 | * | |
1589 | * %LOCK_EX -- an exclusive lock. | |
1590 | * | |
1591 | * %LOCK_UN -- remove an existing lock. | |
1592 | * | |
1593 | * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes. | |
1594 | * | |
1595 | * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other | |
1596 | * processes read and write access respectively. | |
1597 | */ | |
1598 | SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd) | |
1599 | { | |
1600 | struct file *filp; | |
1601 | struct file_lock *lock; | |
1602 | int can_sleep, unlock; | |
1603 | int error; | |
1604 | ||
1605 | error = -EBADF; | |
1606 | filp = fget(fd); | |
1607 | if (!filp) | |
1608 | goto out; | |
1609 | ||
1610 | can_sleep = !(cmd & LOCK_NB); | |
1611 | cmd &= ~LOCK_NB; | |
1612 | unlock = (cmd == LOCK_UN); | |
1613 | ||
1614 | if (!unlock && !(cmd & LOCK_MAND) && | |
1615 | !(filp->f_mode & (FMODE_READ|FMODE_WRITE))) | |
1616 | goto out_putf; | |
1617 | ||
1618 | error = flock_make_lock(filp, &lock, cmd); | |
1619 | if (error) | |
1620 | goto out_putf; | |
1621 | if (can_sleep) | |
1622 | lock->fl_flags |= FL_SLEEP; | |
1623 | ||
1624 | error = security_file_lock(filp, lock->fl_type); | |
1625 | if (error) | |
1626 | goto out_free; | |
1627 | ||
1628 | if (filp->f_op && filp->f_op->flock) | |
1629 | error = filp->f_op->flock(filp, | |
1630 | (can_sleep) ? F_SETLKW : F_SETLK, | |
1631 | lock); | |
1632 | else | |
1633 | error = flock_lock_file_wait(filp, lock); | |
1634 | ||
1635 | out_free: | |
1636 | locks_free_lock(lock); | |
1637 | ||
1638 | out_putf: | |
1639 | fput(filp); | |
1640 | out: | |
1641 | return error; | |
1642 | } | |
1643 | ||
1644 | /** | |
1645 | * vfs_test_lock - test file byte range lock | |
1646 | * @filp: The file to test lock for | |
1647 | * @fl: The lock to test; also used to hold result | |
1648 | * | |
1649 | * Returns -ERRNO on failure. Indicates presence of conflicting lock by | |
1650 | * setting conf->fl_type to something other than F_UNLCK. | |
1651 | */ | |
1652 | int vfs_test_lock(struct file *filp, struct file_lock *fl) | |
1653 | { | |
1654 | if (filp->f_op && filp->f_op->lock) | |
1655 | return filp->f_op->lock(filp, F_GETLK, fl); | |
1656 | posix_test_lock(filp, fl); | |
1657 | return 0; | |
1658 | } | |
1659 | EXPORT_SYMBOL_GPL(vfs_test_lock); | |
1660 | ||
1661 | static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl) | |
1662 | { | |
1663 | flock->l_pid = fl->fl_pid; | |
1664 | #if BITS_PER_LONG == 32 | |
1665 | /* | |
1666 | * Make sure we can represent the posix lock via | |
1667 | * legacy 32bit flock. | |
1668 | */ | |
1669 | if (fl->fl_start > OFFT_OFFSET_MAX) | |
1670 | return -EOVERFLOW; | |
1671 | if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX) | |
1672 | return -EOVERFLOW; | |
1673 | #endif | |
1674 | flock->l_start = fl->fl_start; | |
1675 | flock->l_len = fl->fl_end == OFFSET_MAX ? 0 : | |
1676 | fl->fl_end - fl->fl_start + 1; | |
1677 | flock->l_whence = 0; | |
1678 | flock->l_type = fl->fl_type; | |
1679 | return 0; | |
1680 | } | |
1681 | ||
1682 | #if BITS_PER_LONG == 32 | |
1683 | static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl) | |
1684 | { | |
1685 | flock->l_pid = fl->fl_pid; | |
1686 | flock->l_start = fl->fl_start; | |
1687 | flock->l_len = fl->fl_end == OFFSET_MAX ? 0 : | |
1688 | fl->fl_end - fl->fl_start + 1; | |
1689 | flock->l_whence = 0; | |
1690 | flock->l_type = fl->fl_type; | |
1691 | } | |
1692 | #endif | |
1693 | ||
1694 | /* Report the first existing lock that would conflict with l. | |
1695 | * This implements the F_GETLK command of fcntl(). | |
1696 | */ | |
1697 | int fcntl_getlk(struct file *filp, struct flock __user *l) | |
1698 | { | |
1699 | struct file_lock file_lock; | |
1700 | struct flock flock; | |
1701 | int error; | |
1702 | ||
1703 | error = -EFAULT; | |
1704 | if (copy_from_user(&flock, l, sizeof(flock))) | |
1705 | goto out; | |
1706 | error = -EINVAL; | |
1707 | if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK)) | |
1708 | goto out; | |
1709 | ||
1710 | error = flock_to_posix_lock(filp, &file_lock, &flock); | |
1711 | if (error) | |
1712 | goto out; | |
1713 | ||
1714 | error = vfs_test_lock(filp, &file_lock); | |
1715 | if (error) | |
1716 | goto out; | |
1717 | ||
1718 | flock.l_type = file_lock.fl_type; | |
1719 | if (file_lock.fl_type != F_UNLCK) { | |
1720 | error = posix_lock_to_flock(&flock, &file_lock); | |
1721 | if (error) | |
1722 | goto out; | |
1723 | } | |
1724 | error = -EFAULT; | |
1725 | if (!copy_to_user(l, &flock, sizeof(flock))) | |
1726 | error = 0; | |
1727 | out: | |
1728 | return error; | |
1729 | } | |
1730 | ||
1731 | /** | |
1732 | * vfs_lock_file - file byte range lock | |
1733 | * @filp: The file to apply the lock to | |
1734 | * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.) | |
1735 | * @fl: The lock to be applied | |
1736 | * @conf: Place to return a copy of the conflicting lock, if found. | |
1737 | * | |
1738 | * A caller that doesn't care about the conflicting lock may pass NULL | |
1739 | * as the final argument. | |
1740 | * | |
1741 | * If the filesystem defines a private ->lock() method, then @conf will | |
1742 | * be left unchanged; so a caller that cares should initialize it to | |
1743 | * some acceptable default. | |
1744 | * | |
1745 | * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX | |
1746 | * locks, the ->lock() interface may return asynchronously, before the lock has | |
1747 | * been granted or denied by the underlying filesystem, if (and only if) | |
1748 | * fl_grant is set. Callers expecting ->lock() to return asynchronously | |
1749 | * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if) | |
1750 | * the request is for a blocking lock. When ->lock() does return asynchronously, | |
1751 | * it must return FILE_LOCK_DEFERRED, and call ->fl_grant() when the lock | |
1752 | * request completes. | |
1753 | * If the request is for non-blocking lock the file system should return | |
1754 | * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine | |
1755 | * with the result. If the request timed out the callback routine will return a | |
1756 | * nonzero return code and the file system should release the lock. The file | |
1757 | * system is also responsible to keep a corresponding posix lock when it | |
1758 | * grants a lock so the VFS can find out which locks are locally held and do | |
1759 | * the correct lock cleanup when required. | |
1760 | * The underlying filesystem must not drop the kernel lock or call | |
1761 | * ->fl_grant() before returning to the caller with a FILE_LOCK_DEFERRED | |
1762 | * return code. | |
1763 | */ | |
1764 | int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf) | |
1765 | { | |
1766 | if (filp->f_op && filp->f_op->lock) | |
1767 | return filp->f_op->lock(filp, cmd, fl); | |
1768 | else | |
1769 | return posix_lock_file(filp, fl, conf); | |
1770 | } | |
1771 | EXPORT_SYMBOL_GPL(vfs_lock_file); | |
1772 | ||
1773 | static int do_lock_file_wait(struct file *filp, unsigned int cmd, | |
1774 | struct file_lock *fl) | |
1775 | { | |
1776 | int error; | |
1777 | ||
1778 | error = security_file_lock(filp, fl->fl_type); | |
1779 | if (error) | |
1780 | return error; | |
1781 | ||
1782 | for (;;) { | |
1783 | error = vfs_lock_file(filp, cmd, fl, NULL); | |
1784 | if (error != FILE_LOCK_DEFERRED) | |
1785 | break; | |
1786 | error = wait_event_interruptible(fl->fl_wait, !fl->fl_next); | |
1787 | if (!error) | |
1788 | continue; | |
1789 | ||
1790 | locks_delete_block(fl); | |
1791 | break; | |
1792 | } | |
1793 | ||
1794 | return error; | |
1795 | } | |
1796 | ||
1797 | /* Apply the lock described by l to an open file descriptor. | |
1798 | * This implements both the F_SETLK and F_SETLKW commands of fcntl(). | |
1799 | */ | |
1800 | int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd, | |
1801 | struct flock __user *l) | |
1802 | { | |
1803 | struct file_lock *file_lock = locks_alloc_lock(); | |
1804 | struct flock flock; | |
1805 | struct inode *inode; | |
1806 | struct file *f; | |
1807 | int error; | |
1808 | ||
1809 | if (file_lock == NULL) | |
1810 | return -ENOLCK; | |
1811 | ||
1812 | /* | |
1813 | * This might block, so we do it before checking the inode. | |
1814 | */ | |
1815 | error = -EFAULT; | |
1816 | if (copy_from_user(&flock, l, sizeof(flock))) | |
1817 | goto out; | |
1818 | ||
1819 | inode = filp->f_path.dentry->d_inode; | |
1820 | ||
1821 | /* Don't allow mandatory locks on files that may be memory mapped | |
1822 | * and shared. | |
1823 | */ | |
1824 | if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) { | |
1825 | error = -EAGAIN; | |
1826 | goto out; | |
1827 | } | |
1828 | ||
1829 | again: | |
1830 | error = flock_to_posix_lock(filp, file_lock, &flock); | |
1831 | if (error) | |
1832 | goto out; | |
1833 | if (cmd == F_SETLKW) { | |
1834 | file_lock->fl_flags |= FL_SLEEP; | |
1835 | } | |
1836 | ||
1837 | error = -EBADF; | |
1838 | switch (flock.l_type) { | |
1839 | case F_RDLCK: | |
1840 | if (!(filp->f_mode & FMODE_READ)) | |
1841 | goto out; | |
1842 | break; | |
1843 | case F_WRLCK: | |
1844 | if (!(filp->f_mode & FMODE_WRITE)) | |
1845 | goto out; | |
1846 | break; | |
1847 | case F_UNLCK: | |
1848 | break; | |
1849 | default: | |
1850 | error = -EINVAL; | |
1851 | goto out; | |
1852 | } | |
1853 | ||
1854 | error = do_lock_file_wait(filp, cmd, file_lock); | |
1855 | ||
1856 | /* | |
1857 | * Attempt to detect a close/fcntl race and recover by | |
1858 | * releasing the lock that was just acquired. | |
1859 | */ | |
1860 | /* | |
1861 | * we need that spin_lock here - it prevents reordering between | |
1862 | * update of inode->i_flock and check for it done in close(). | |
1863 | * rcu_read_lock() wouldn't do. | |
1864 | */ | |
1865 | spin_lock(¤t->files->file_lock); | |
1866 | f = fcheck(fd); | |
1867 | spin_unlock(¤t->files->file_lock); | |
1868 | if (!error && f != filp && flock.l_type != F_UNLCK) { | |
1869 | flock.l_type = F_UNLCK; | |
1870 | goto again; | |
1871 | } | |
1872 | ||
1873 | out: | |
1874 | locks_free_lock(file_lock); | |
1875 | return error; | |
1876 | } | |
1877 | ||
1878 | #if BITS_PER_LONG == 32 | |
1879 | /* Report the first existing lock that would conflict with l. | |
1880 | * This implements the F_GETLK command of fcntl(). | |
1881 | */ | |
1882 | int fcntl_getlk64(struct file *filp, struct flock64 __user *l) | |
1883 | { | |
1884 | struct file_lock file_lock; | |
1885 | struct flock64 flock; | |
1886 | int error; | |
1887 | ||
1888 | error = -EFAULT; | |
1889 | if (copy_from_user(&flock, l, sizeof(flock))) | |
1890 | goto out; | |
1891 | error = -EINVAL; | |
1892 | if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK)) | |
1893 | goto out; | |
1894 | ||
1895 | error = flock64_to_posix_lock(filp, &file_lock, &flock); | |
1896 | if (error) | |
1897 | goto out; | |
1898 | ||
1899 | error = vfs_test_lock(filp, &file_lock); | |
1900 | if (error) | |
1901 | goto out; | |
1902 | ||
1903 | flock.l_type = file_lock.fl_type; | |
1904 | if (file_lock.fl_type != F_UNLCK) | |
1905 | posix_lock_to_flock64(&flock, &file_lock); | |
1906 | ||
1907 | error = -EFAULT; | |
1908 | if (!copy_to_user(l, &flock, sizeof(flock))) | |
1909 | error = 0; | |
1910 | ||
1911 | out: | |
1912 | return error; | |
1913 | } | |
1914 | ||
1915 | /* Apply the lock described by l to an open file descriptor. | |
1916 | * This implements both the F_SETLK and F_SETLKW commands of fcntl(). | |
1917 | */ | |
1918 | int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd, | |
1919 | struct flock64 __user *l) | |
1920 | { | |
1921 | struct file_lock *file_lock = locks_alloc_lock(); | |
1922 | struct flock64 flock; | |
1923 | struct inode *inode; | |
1924 | struct file *f; | |
1925 | int error; | |
1926 | ||
1927 | if (file_lock == NULL) | |
1928 | return -ENOLCK; | |
1929 | ||
1930 | /* | |
1931 | * This might block, so we do it before checking the inode. | |
1932 | */ | |
1933 | error = -EFAULT; | |
1934 | if (copy_from_user(&flock, l, sizeof(flock))) | |
1935 | goto out; | |
1936 | ||
1937 | inode = filp->f_path.dentry->d_inode; | |
1938 | ||
1939 | /* Don't allow mandatory locks on files that may be memory mapped | |
1940 | * and shared. | |
1941 | */ | |
1942 | if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) { | |
1943 | error = -EAGAIN; | |
1944 | goto out; | |
1945 | } | |
1946 | ||
1947 | again: | |
1948 | error = flock64_to_posix_lock(filp, file_lock, &flock); | |
1949 | if (error) | |
1950 | goto out; | |
1951 | if (cmd == F_SETLKW64) { | |
1952 | file_lock->fl_flags |= FL_SLEEP; | |
1953 | } | |
1954 | ||
1955 | error = -EBADF; | |
1956 | switch (flock.l_type) { | |
1957 | case F_RDLCK: | |
1958 | if (!(filp->f_mode & FMODE_READ)) | |
1959 | goto out; | |
1960 | break; | |
1961 | case F_WRLCK: | |
1962 | if (!(filp->f_mode & FMODE_WRITE)) | |
1963 | goto out; | |
1964 | break; | |
1965 | case F_UNLCK: | |
1966 | break; | |
1967 | default: | |
1968 | error = -EINVAL; | |
1969 | goto out; | |
1970 | } | |
1971 | ||
1972 | error = do_lock_file_wait(filp, cmd, file_lock); | |
1973 | ||
1974 | /* | |
1975 | * Attempt to detect a close/fcntl race and recover by | |
1976 | * releasing the lock that was just acquired. | |
1977 | */ | |
1978 | spin_lock(¤t->files->file_lock); | |
1979 | f = fcheck(fd); | |
1980 | spin_unlock(¤t->files->file_lock); | |
1981 | if (!error && f != filp && flock.l_type != F_UNLCK) { | |
1982 | flock.l_type = F_UNLCK; | |
1983 | goto again; | |
1984 | } | |
1985 | ||
1986 | out: | |
1987 | locks_free_lock(file_lock); | |
1988 | return error; | |
1989 | } | |
1990 | #endif /* BITS_PER_LONG == 32 */ | |
1991 | ||
1992 | /* | |
1993 | * This function is called when the file is being removed | |
1994 | * from the task's fd array. POSIX locks belonging to this task | |
1995 | * are deleted at this time. | |
1996 | */ | |
1997 | void locks_remove_posix(struct file *filp, fl_owner_t owner) | |
1998 | { | |
1999 | struct file_lock lock; | |
2000 | ||
2001 | /* | |
2002 | * If there are no locks held on this file, we don't need to call | |
2003 | * posix_lock_file(). Another process could be setting a lock on this | |
2004 | * file at the same time, but we wouldn't remove that lock anyway. | |
2005 | */ | |
2006 | if (!filp->f_path.dentry->d_inode->i_flock) | |
2007 | return; | |
2008 | ||
2009 | lock.fl_type = F_UNLCK; | |
2010 | lock.fl_flags = FL_POSIX | FL_CLOSE; | |
2011 | lock.fl_start = 0; | |
2012 | lock.fl_end = OFFSET_MAX; | |
2013 | lock.fl_owner = owner; | |
2014 | lock.fl_pid = current->tgid; | |
2015 | lock.fl_file = filp; | |
2016 | lock.fl_ops = NULL; | |
2017 | lock.fl_lmops = NULL; | |
2018 | ||
2019 | vfs_lock_file(filp, F_SETLK, &lock, NULL); | |
2020 | ||
2021 | if (lock.fl_ops && lock.fl_ops->fl_release_private) | |
2022 | lock.fl_ops->fl_release_private(&lock); | |
2023 | } | |
2024 | ||
2025 | EXPORT_SYMBOL(locks_remove_posix); | |
2026 | ||
2027 | /* | |
2028 | * This function is called on the last close of an open file. | |
2029 | */ | |
2030 | void locks_remove_flock(struct file *filp) | |
2031 | { | |
2032 | struct inode * inode = filp->f_path.dentry->d_inode; | |
2033 | struct file_lock *fl; | |
2034 | struct file_lock **before; | |
2035 | ||
2036 | if (!inode->i_flock) | |
2037 | return; | |
2038 | ||
2039 | if (filp->f_op && filp->f_op->flock) { | |
2040 | struct file_lock fl = { | |
2041 | .fl_pid = current->tgid, | |
2042 | .fl_file = filp, | |
2043 | .fl_flags = FL_FLOCK, | |
2044 | .fl_type = F_UNLCK, | |
2045 | .fl_end = OFFSET_MAX, | |
2046 | }; | |
2047 | filp->f_op->flock(filp, F_SETLKW, &fl); | |
2048 | if (fl.fl_ops && fl.fl_ops->fl_release_private) | |
2049 | fl.fl_ops->fl_release_private(&fl); | |
2050 | } | |
2051 | ||
2052 | lock_flocks(); | |
2053 | before = &inode->i_flock; | |
2054 | ||
2055 | while ((fl = *before) != NULL) { | |
2056 | if (fl->fl_file == filp) { | |
2057 | if (IS_FLOCK(fl)) { | |
2058 | locks_delete_lock(before); | |
2059 | continue; | |
2060 | } | |
2061 | if (IS_LEASE(fl)) { | |
2062 | lease_modify(before, F_UNLCK); | |
2063 | continue; | |
2064 | } | |
2065 | /* What? */ | |
2066 | BUG(); | |
2067 | } | |
2068 | before = &fl->fl_next; | |
2069 | } | |
2070 | unlock_flocks(); | |
2071 | } | |
2072 | ||
2073 | /** | |
2074 | * posix_unblock_lock - stop waiting for a file lock | |
2075 | * @filp: how the file was opened | |
2076 | * @waiter: the lock which was waiting | |
2077 | * | |
2078 | * lockd needs to block waiting for locks. | |
2079 | */ | |
2080 | int | |
2081 | posix_unblock_lock(struct file *filp, struct file_lock *waiter) | |
2082 | { | |
2083 | int status = 0; | |
2084 | ||
2085 | lock_flocks(); | |
2086 | if (waiter->fl_next) | |
2087 | __locks_delete_block(waiter); | |
2088 | else | |
2089 | status = -ENOENT; | |
2090 | unlock_flocks(); | |
2091 | return status; | |
2092 | } | |
2093 | ||
2094 | EXPORT_SYMBOL(posix_unblock_lock); | |
2095 | ||
2096 | /** | |
2097 | * vfs_cancel_lock - file byte range unblock lock | |
2098 | * @filp: The file to apply the unblock to | |
2099 | * @fl: The lock to be unblocked | |
2100 | * | |
2101 | * Used by lock managers to cancel blocked requests | |
2102 | */ | |
2103 | int vfs_cancel_lock(struct file *filp, struct file_lock *fl) | |
2104 | { | |
2105 | if (filp->f_op && filp->f_op->lock) | |
2106 | return filp->f_op->lock(filp, F_CANCELLK, fl); | |
2107 | return 0; | |
2108 | } | |
2109 | ||
2110 | EXPORT_SYMBOL_GPL(vfs_cancel_lock); | |
2111 | ||
2112 | #ifdef CONFIG_PROC_FS | |
2113 | #include <linux/proc_fs.h> | |
2114 | #include <linux/seq_file.h> | |
2115 | ||
2116 | static void lock_get_status(struct seq_file *f, struct file_lock *fl, | |
2117 | loff_t id, char *pfx) | |
2118 | { | |
2119 | struct inode *inode = NULL; | |
2120 | unsigned int fl_pid; | |
2121 | ||
2122 | if (fl->fl_nspid) | |
2123 | fl_pid = pid_vnr(fl->fl_nspid); | |
2124 | else | |
2125 | fl_pid = fl->fl_pid; | |
2126 | ||
2127 | if (fl->fl_file != NULL) | |
2128 | inode = fl->fl_file->f_path.dentry->d_inode; | |
2129 | ||
2130 | seq_printf(f, "%lld:%s ", id, pfx); | |
2131 | if (IS_POSIX(fl)) { | |
2132 | seq_printf(f, "%6s %s ", | |
2133 | (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ", | |
2134 | (inode == NULL) ? "*NOINODE*" : | |
2135 | mandatory_lock(inode) ? "MANDATORY" : "ADVISORY "); | |
2136 | } else if (IS_FLOCK(fl)) { | |
2137 | if (fl->fl_type & LOCK_MAND) { | |
2138 | seq_printf(f, "FLOCK MSNFS "); | |
2139 | } else { | |
2140 | seq_printf(f, "FLOCK ADVISORY "); | |
2141 | } | |
2142 | } else if (IS_LEASE(fl)) { | |
2143 | seq_printf(f, "LEASE "); | |
2144 | if (fl->fl_type & F_INPROGRESS) | |
2145 | seq_printf(f, "BREAKING "); | |
2146 | else if (fl->fl_file) | |
2147 | seq_printf(f, "ACTIVE "); | |
2148 | else | |
2149 | seq_printf(f, "BREAKER "); | |
2150 | } else { | |
2151 | seq_printf(f, "UNKNOWN UNKNOWN "); | |
2152 | } | |
2153 | if (fl->fl_type & LOCK_MAND) { | |
2154 | seq_printf(f, "%s ", | |
2155 | (fl->fl_type & LOCK_READ) | |
2156 | ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ " | |
2157 | : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE "); | |
2158 | } else { | |
2159 | seq_printf(f, "%s ", | |
2160 | (fl->fl_type & F_INPROGRESS) | |
2161 | ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ " | |
2162 | : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ "); | |
2163 | } | |
2164 | if (inode) { | |
2165 | #ifdef WE_CAN_BREAK_LSLK_NOW | |
2166 | seq_printf(f, "%d %s:%ld ", fl_pid, | |
2167 | inode->i_sb->s_id, inode->i_ino); | |
2168 | #else | |
2169 | /* userspace relies on this representation of dev_t ;-( */ | |
2170 | seq_printf(f, "%d %02x:%02x:%ld ", fl_pid, | |
2171 | MAJOR(inode->i_sb->s_dev), | |
2172 | MINOR(inode->i_sb->s_dev), inode->i_ino); | |
2173 | #endif | |
2174 | } else { | |
2175 | seq_printf(f, "%d <none>:0 ", fl_pid); | |
2176 | } | |
2177 | if (IS_POSIX(fl)) { | |
2178 | if (fl->fl_end == OFFSET_MAX) | |
2179 | seq_printf(f, "%Ld EOF\n", fl->fl_start); | |
2180 | else | |
2181 | seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end); | |
2182 | } else { | |
2183 | seq_printf(f, "0 EOF\n"); | |
2184 | } | |
2185 | } | |
2186 | ||
2187 | static int locks_show(struct seq_file *f, void *v) | |
2188 | { | |
2189 | struct file_lock *fl, *bfl; | |
2190 | ||
2191 | fl = list_entry(v, struct file_lock, fl_link); | |
2192 | ||
2193 | lock_get_status(f, fl, *((loff_t *)f->private), ""); | |
2194 | ||
2195 | list_for_each_entry(bfl, &fl->fl_block, fl_block) | |
2196 | lock_get_status(f, bfl, *((loff_t *)f->private), " ->"); | |
2197 | ||
2198 | return 0; | |
2199 | } | |
2200 | ||
2201 | static void *locks_start(struct seq_file *f, loff_t *pos) | |
2202 | { | |
2203 | loff_t *p = f->private; | |
2204 | ||
2205 | lock_flocks(); | |
2206 | *p = (*pos + 1); | |
2207 | return seq_list_start(&file_lock_list, *pos); | |
2208 | } | |
2209 | ||
2210 | static void *locks_next(struct seq_file *f, void *v, loff_t *pos) | |
2211 | { | |
2212 | loff_t *p = f->private; | |
2213 | ++*p; | |
2214 | return seq_list_next(v, &file_lock_list, pos); | |
2215 | } | |
2216 | ||
2217 | static void locks_stop(struct seq_file *f, void *v) | |
2218 | { | |
2219 | unlock_flocks(); | |
2220 | } | |
2221 | ||
2222 | static const struct seq_operations locks_seq_operations = { | |
2223 | .start = locks_start, | |
2224 | .next = locks_next, | |
2225 | .stop = locks_stop, | |
2226 | .show = locks_show, | |
2227 | }; | |
2228 | ||
2229 | static int locks_open(struct inode *inode, struct file *filp) | |
2230 | { | |
2231 | return seq_open_private(filp, &locks_seq_operations, sizeof(loff_t)); | |
2232 | } | |
2233 | ||
2234 | static const struct file_operations proc_locks_operations = { | |
2235 | .open = locks_open, | |
2236 | .read = seq_read, | |
2237 | .llseek = seq_lseek, | |
2238 | .release = seq_release_private, | |
2239 | }; | |
2240 | ||
2241 | static int __init proc_locks_init(void) | |
2242 | { | |
2243 | proc_create("locks", 0, NULL, &proc_locks_operations); | |
2244 | return 0; | |
2245 | } | |
2246 | module_init(proc_locks_init); | |
2247 | #endif | |
2248 | ||
2249 | /** | |
2250 | * lock_may_read - checks that the region is free of locks | |
2251 | * @inode: the inode that is being read | |
2252 | * @start: the first byte to read | |
2253 | * @len: the number of bytes to read | |
2254 | * | |
2255 | * Emulates Windows locking requirements. Whole-file | |
2256 | * mandatory locks (share modes) can prohibit a read and | |
2257 | * byte-range POSIX locks can prohibit a read if they overlap. | |
2258 | * | |
2259 | * N.B. this function is only ever called | |
2260 | * from knfsd and ownership of locks is never checked. | |
2261 | */ | |
2262 | int lock_may_read(struct inode *inode, loff_t start, unsigned long len) | |
2263 | { | |
2264 | struct file_lock *fl; | |
2265 | int result = 1; | |
2266 | lock_flocks(); | |
2267 | for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) { | |
2268 | if (IS_POSIX(fl)) { | |
2269 | if (fl->fl_type == F_RDLCK) | |
2270 | continue; | |
2271 | if ((fl->fl_end < start) || (fl->fl_start > (start + len))) | |
2272 | continue; | |
2273 | } else if (IS_FLOCK(fl)) { | |
2274 | if (!(fl->fl_type & LOCK_MAND)) | |
2275 | continue; | |
2276 | if (fl->fl_type & LOCK_READ) | |
2277 | continue; | |
2278 | } else | |
2279 | continue; | |
2280 | result = 0; | |
2281 | break; | |
2282 | } | |
2283 | unlock_flocks(); | |
2284 | return result; | |
2285 | } | |
2286 | ||
2287 | EXPORT_SYMBOL(lock_may_read); | |
2288 | ||
2289 | /** | |
2290 | * lock_may_write - checks that the region is free of locks | |
2291 | * @inode: the inode that is being written | |
2292 | * @start: the first byte to write | |
2293 | * @len: the number of bytes to write | |
2294 | * | |
2295 | * Emulates Windows locking requirements. Whole-file | |
2296 | * mandatory locks (share modes) can prohibit a write and | |
2297 | * byte-range POSIX locks can prohibit a write if they overlap. | |
2298 | * | |
2299 | * N.B. this function is only ever called | |
2300 | * from knfsd and ownership of locks is never checked. | |
2301 | */ | |
2302 | int lock_may_write(struct inode *inode, loff_t start, unsigned long len) | |
2303 | { | |
2304 | struct file_lock *fl; | |
2305 | int result = 1; | |
2306 | lock_flocks(); | |
2307 | for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) { | |
2308 | if (IS_POSIX(fl)) { | |
2309 | if ((fl->fl_end < start) || (fl->fl_start > (start + len))) | |
2310 | continue; | |
2311 | } else if (IS_FLOCK(fl)) { | |
2312 | if (!(fl->fl_type & LOCK_MAND)) | |
2313 | continue; | |
2314 | if (fl->fl_type & LOCK_WRITE) | |
2315 | continue; | |
2316 | } else | |
2317 | continue; | |
2318 | result = 0; | |
2319 | break; | |
2320 | } | |
2321 | unlock_flocks(); | |
2322 | return result; | |
2323 | } | |
2324 | ||
2325 | EXPORT_SYMBOL(lock_may_write); | |
2326 | ||
2327 | static int __init filelock_init(void) | |
2328 | { | |
2329 | filelock_cache = kmem_cache_create("file_lock_cache", | |
2330 | sizeof(struct file_lock), 0, SLAB_PANIC, | |
2331 | init_once); | |
2332 | return 0; | |
2333 | } | |
2334 | ||
2335 | core_initcall(filelock_init); |