]>
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/filesystems/mandatory-locking.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/security.h> | |
123 | #include <linux/slab.h> | |
124 | #include <linux/syscalls.h> | |
125 | #include <linux/time.h> | |
126 | #include <linux/rcupdate.h> | |
127 | #include <linux/pid_namespace.h> | |
128 | #include <linux/hashtable.h> | |
129 | #include <linux/percpu.h> | |
130 | #include <linux/lglock.h> | |
131 | ||
132 | #define CREATE_TRACE_POINTS | |
133 | #include <trace/events/filelock.h> | |
134 | ||
135 | #include <asm/uaccess.h> | |
136 | ||
137 | #define IS_POSIX(fl) (fl->fl_flags & FL_POSIX) | |
138 | #define IS_FLOCK(fl) (fl->fl_flags & FL_FLOCK) | |
139 | #define IS_LEASE(fl) (fl->fl_flags & (FL_LEASE|FL_DELEG|FL_LAYOUT)) | |
140 | #define IS_OFDLCK(fl) (fl->fl_flags & FL_OFDLCK) | |
141 | ||
142 | static bool lease_breaking(struct file_lock *fl) | |
143 | { | |
144 | return fl->fl_flags & (FL_UNLOCK_PENDING | FL_DOWNGRADE_PENDING); | |
145 | } | |
146 | ||
147 | static int target_leasetype(struct file_lock *fl) | |
148 | { | |
149 | if (fl->fl_flags & FL_UNLOCK_PENDING) | |
150 | return F_UNLCK; | |
151 | if (fl->fl_flags & FL_DOWNGRADE_PENDING) | |
152 | return F_RDLCK; | |
153 | return fl->fl_type; | |
154 | } | |
155 | ||
156 | int leases_enable = 1; | |
157 | int lease_break_time = 45; | |
158 | ||
159 | /* | |
160 | * The global file_lock_list is only used for displaying /proc/locks, so we | |
161 | * keep a list on each CPU, with each list protected by its own spinlock via | |
162 | * the file_lock_lglock. Note that alterations to the list also require that | |
163 | * the relevant flc_lock is held. | |
164 | */ | |
165 | DEFINE_STATIC_LGLOCK(file_lock_lglock); | |
166 | static DEFINE_PER_CPU(struct hlist_head, file_lock_list); | |
167 | ||
168 | /* | |
169 | * The blocked_hash is used to find POSIX lock loops for deadlock detection. | |
170 | * It is protected by blocked_lock_lock. | |
171 | * | |
172 | * We hash locks by lockowner in order to optimize searching for the lock a | |
173 | * particular lockowner is waiting on. | |
174 | * | |
175 | * FIXME: make this value scale via some heuristic? We generally will want more | |
176 | * buckets when we have more lockowners holding locks, but that's a little | |
177 | * difficult to determine without knowing what the workload will look like. | |
178 | */ | |
179 | #define BLOCKED_HASH_BITS 7 | |
180 | static DEFINE_HASHTABLE(blocked_hash, BLOCKED_HASH_BITS); | |
181 | ||
182 | /* | |
183 | * This lock protects the blocked_hash. Generally, if you're accessing it, you | |
184 | * want to be holding this lock. | |
185 | * | |
186 | * In addition, it also protects the fl->fl_block list, and the fl->fl_next | |
187 | * pointer for file_lock structures that are acting as lock requests (in | |
188 | * contrast to those that are acting as records of acquired locks). | |
189 | * | |
190 | * Note that when we acquire this lock in order to change the above fields, | |
191 | * we often hold the flc_lock as well. In certain cases, when reading the fields | |
192 | * protected by this lock, we can skip acquiring it iff we already hold the | |
193 | * flc_lock. | |
194 | * | |
195 | * In particular, adding an entry to the fl_block list requires that you hold | |
196 | * both the flc_lock and the blocked_lock_lock (acquired in that order). | |
197 | * Deleting an entry from the list however only requires the file_lock_lock. | |
198 | */ | |
199 | static DEFINE_SPINLOCK(blocked_lock_lock); | |
200 | ||
201 | static struct kmem_cache *flctx_cache __read_mostly; | |
202 | static struct kmem_cache *filelock_cache __read_mostly; | |
203 | ||
204 | static struct file_lock_context * | |
205 | locks_get_lock_context(struct inode *inode, int type) | |
206 | { | |
207 | struct file_lock_context *ctx; | |
208 | ||
209 | /* paired with cmpxchg() below */ | |
210 | ctx = smp_load_acquire(&inode->i_flctx); | |
211 | if (likely(ctx) || type == F_UNLCK) | |
212 | goto out; | |
213 | ||
214 | ctx = kmem_cache_alloc(flctx_cache, GFP_KERNEL); | |
215 | if (!ctx) | |
216 | goto out; | |
217 | ||
218 | spin_lock_init(&ctx->flc_lock); | |
219 | INIT_LIST_HEAD(&ctx->flc_flock); | |
220 | INIT_LIST_HEAD(&ctx->flc_posix); | |
221 | INIT_LIST_HEAD(&ctx->flc_lease); | |
222 | ||
223 | /* | |
224 | * Assign the pointer if it's not already assigned. If it is, then | |
225 | * free the context we just allocated. | |
226 | */ | |
227 | if (cmpxchg(&inode->i_flctx, NULL, ctx)) { | |
228 | kmem_cache_free(flctx_cache, ctx); | |
229 | ctx = smp_load_acquire(&inode->i_flctx); | |
230 | } | |
231 | out: | |
232 | trace_locks_get_lock_context(inode, type, ctx); | |
233 | return ctx; | |
234 | } | |
235 | ||
236 | void | |
237 | locks_free_lock_context(struct file_lock_context *ctx) | |
238 | { | |
239 | if (ctx) { | |
240 | WARN_ON_ONCE(!list_empty(&ctx->flc_flock)); | |
241 | WARN_ON_ONCE(!list_empty(&ctx->flc_posix)); | |
242 | WARN_ON_ONCE(!list_empty(&ctx->flc_lease)); | |
243 | kmem_cache_free(flctx_cache, ctx); | |
244 | } | |
245 | } | |
246 | ||
247 | static void locks_init_lock_heads(struct file_lock *fl) | |
248 | { | |
249 | INIT_HLIST_NODE(&fl->fl_link); | |
250 | INIT_LIST_HEAD(&fl->fl_list); | |
251 | INIT_LIST_HEAD(&fl->fl_block); | |
252 | init_waitqueue_head(&fl->fl_wait); | |
253 | } | |
254 | ||
255 | /* Allocate an empty lock structure. */ | |
256 | struct file_lock *locks_alloc_lock(void) | |
257 | { | |
258 | struct file_lock *fl = kmem_cache_zalloc(filelock_cache, GFP_KERNEL); | |
259 | ||
260 | if (fl) | |
261 | locks_init_lock_heads(fl); | |
262 | ||
263 | return fl; | |
264 | } | |
265 | EXPORT_SYMBOL_GPL(locks_alloc_lock); | |
266 | ||
267 | void locks_release_private(struct file_lock *fl) | |
268 | { | |
269 | if (fl->fl_ops) { | |
270 | if (fl->fl_ops->fl_release_private) | |
271 | fl->fl_ops->fl_release_private(fl); | |
272 | fl->fl_ops = NULL; | |
273 | } | |
274 | ||
275 | if (fl->fl_lmops) { | |
276 | if (fl->fl_lmops->lm_put_owner) { | |
277 | fl->fl_lmops->lm_put_owner(fl->fl_owner); | |
278 | fl->fl_owner = NULL; | |
279 | } | |
280 | fl->fl_lmops = NULL; | |
281 | } | |
282 | } | |
283 | EXPORT_SYMBOL_GPL(locks_release_private); | |
284 | ||
285 | /* Free a lock which is not in use. */ | |
286 | void locks_free_lock(struct file_lock *fl) | |
287 | { | |
288 | BUG_ON(waitqueue_active(&fl->fl_wait)); | |
289 | BUG_ON(!list_empty(&fl->fl_list)); | |
290 | BUG_ON(!list_empty(&fl->fl_block)); | |
291 | BUG_ON(!hlist_unhashed(&fl->fl_link)); | |
292 | ||
293 | locks_release_private(fl); | |
294 | kmem_cache_free(filelock_cache, fl); | |
295 | } | |
296 | EXPORT_SYMBOL(locks_free_lock); | |
297 | ||
298 | static void | |
299 | locks_dispose_list(struct list_head *dispose) | |
300 | { | |
301 | struct file_lock *fl; | |
302 | ||
303 | while (!list_empty(dispose)) { | |
304 | fl = list_first_entry(dispose, struct file_lock, fl_list); | |
305 | list_del_init(&fl->fl_list); | |
306 | locks_free_lock(fl); | |
307 | } | |
308 | } | |
309 | ||
310 | void locks_init_lock(struct file_lock *fl) | |
311 | { | |
312 | memset(fl, 0, sizeof(struct file_lock)); | |
313 | locks_init_lock_heads(fl); | |
314 | } | |
315 | ||
316 | EXPORT_SYMBOL(locks_init_lock); | |
317 | ||
318 | /* | |
319 | * Initialize a new lock from an existing file_lock structure. | |
320 | */ | |
321 | void locks_copy_conflock(struct file_lock *new, struct file_lock *fl) | |
322 | { | |
323 | new->fl_owner = fl->fl_owner; | |
324 | new->fl_pid = fl->fl_pid; | |
325 | new->fl_file = NULL; | |
326 | new->fl_flags = fl->fl_flags; | |
327 | new->fl_type = fl->fl_type; | |
328 | new->fl_start = fl->fl_start; | |
329 | new->fl_end = fl->fl_end; | |
330 | new->fl_lmops = fl->fl_lmops; | |
331 | new->fl_ops = NULL; | |
332 | ||
333 | if (fl->fl_lmops) { | |
334 | if (fl->fl_lmops->lm_get_owner) | |
335 | fl->fl_lmops->lm_get_owner(fl->fl_owner); | |
336 | } | |
337 | } | |
338 | EXPORT_SYMBOL(locks_copy_conflock); | |
339 | ||
340 | void locks_copy_lock(struct file_lock *new, struct file_lock *fl) | |
341 | { | |
342 | /* "new" must be a freshly-initialized lock */ | |
343 | WARN_ON_ONCE(new->fl_ops); | |
344 | ||
345 | locks_copy_conflock(new, fl); | |
346 | ||
347 | new->fl_file = fl->fl_file; | |
348 | new->fl_ops = fl->fl_ops; | |
349 | ||
350 | if (fl->fl_ops) { | |
351 | if (fl->fl_ops->fl_copy_lock) | |
352 | fl->fl_ops->fl_copy_lock(new, fl); | |
353 | } | |
354 | } | |
355 | ||
356 | EXPORT_SYMBOL(locks_copy_lock); | |
357 | ||
358 | static inline int flock_translate_cmd(int cmd) { | |
359 | if (cmd & LOCK_MAND) | |
360 | return cmd & (LOCK_MAND | LOCK_RW); | |
361 | switch (cmd) { | |
362 | case LOCK_SH: | |
363 | return F_RDLCK; | |
364 | case LOCK_EX: | |
365 | return F_WRLCK; | |
366 | case LOCK_UN: | |
367 | return F_UNLCK; | |
368 | } | |
369 | return -EINVAL; | |
370 | } | |
371 | ||
372 | /* Fill in a file_lock structure with an appropriate FLOCK lock. */ | |
373 | static struct file_lock * | |
374 | flock_make_lock(struct file *filp, unsigned int cmd) | |
375 | { | |
376 | struct file_lock *fl; | |
377 | int type = flock_translate_cmd(cmd); | |
378 | ||
379 | if (type < 0) | |
380 | return ERR_PTR(type); | |
381 | ||
382 | fl = locks_alloc_lock(); | |
383 | if (fl == NULL) | |
384 | return ERR_PTR(-ENOMEM); | |
385 | ||
386 | fl->fl_file = filp; | |
387 | fl->fl_owner = filp; | |
388 | fl->fl_pid = current->tgid; | |
389 | fl->fl_flags = FL_FLOCK; | |
390 | fl->fl_type = type; | |
391 | fl->fl_end = OFFSET_MAX; | |
392 | ||
393 | return fl; | |
394 | } | |
395 | ||
396 | static int assign_type(struct file_lock *fl, long type) | |
397 | { | |
398 | switch (type) { | |
399 | case F_RDLCK: | |
400 | case F_WRLCK: | |
401 | case F_UNLCK: | |
402 | fl->fl_type = type; | |
403 | break; | |
404 | default: | |
405 | return -EINVAL; | |
406 | } | |
407 | return 0; | |
408 | } | |
409 | ||
410 | static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl, | |
411 | struct flock64 *l) | |
412 | { | |
413 | switch (l->l_whence) { | |
414 | case SEEK_SET: | |
415 | fl->fl_start = 0; | |
416 | break; | |
417 | case SEEK_CUR: | |
418 | fl->fl_start = filp->f_pos; | |
419 | break; | |
420 | case SEEK_END: | |
421 | fl->fl_start = i_size_read(file_inode(filp)); | |
422 | break; | |
423 | default: | |
424 | return -EINVAL; | |
425 | } | |
426 | if (l->l_start > OFFSET_MAX - fl->fl_start) | |
427 | return -EOVERFLOW; | |
428 | fl->fl_start += l->l_start; | |
429 | if (fl->fl_start < 0) | |
430 | return -EINVAL; | |
431 | ||
432 | /* POSIX-1996 leaves the case l->l_len < 0 undefined; | |
433 | POSIX-2001 defines it. */ | |
434 | if (l->l_len > 0) { | |
435 | if (l->l_len - 1 > OFFSET_MAX - fl->fl_start) | |
436 | return -EOVERFLOW; | |
437 | fl->fl_end = fl->fl_start + l->l_len - 1; | |
438 | ||
439 | } else if (l->l_len < 0) { | |
440 | if (fl->fl_start + l->l_len < 0) | |
441 | return -EINVAL; | |
442 | fl->fl_end = fl->fl_start - 1; | |
443 | fl->fl_start += l->l_len; | |
444 | } else | |
445 | fl->fl_end = OFFSET_MAX; | |
446 | ||
447 | fl->fl_owner = current->files; | |
448 | fl->fl_pid = current->tgid; | |
449 | fl->fl_file = filp; | |
450 | fl->fl_flags = FL_POSIX; | |
451 | fl->fl_ops = NULL; | |
452 | fl->fl_lmops = NULL; | |
453 | ||
454 | return assign_type(fl, l->l_type); | |
455 | } | |
456 | ||
457 | /* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX | |
458 | * style lock. | |
459 | */ | |
460 | static int flock_to_posix_lock(struct file *filp, struct file_lock *fl, | |
461 | struct flock *l) | |
462 | { | |
463 | struct flock64 ll = { | |
464 | .l_type = l->l_type, | |
465 | .l_whence = l->l_whence, | |
466 | .l_start = l->l_start, | |
467 | .l_len = l->l_len, | |
468 | }; | |
469 | ||
470 | return flock64_to_posix_lock(filp, fl, &ll); | |
471 | } | |
472 | ||
473 | /* default lease lock manager operations */ | |
474 | static bool | |
475 | lease_break_callback(struct file_lock *fl) | |
476 | { | |
477 | kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG); | |
478 | return false; | |
479 | } | |
480 | ||
481 | static void | |
482 | lease_setup(struct file_lock *fl, void **priv) | |
483 | { | |
484 | struct file *filp = fl->fl_file; | |
485 | struct fasync_struct *fa = *priv; | |
486 | ||
487 | /* | |
488 | * fasync_insert_entry() returns the old entry if any. If there was no | |
489 | * old entry, then it used "priv" and inserted it into the fasync list. | |
490 | * Clear the pointer to indicate that it shouldn't be freed. | |
491 | */ | |
492 | if (!fasync_insert_entry(fa->fa_fd, filp, &fl->fl_fasync, fa)) | |
493 | *priv = NULL; | |
494 | ||
495 | __f_setown(filp, task_pid(current), PIDTYPE_PID, 0); | |
496 | } | |
497 | ||
498 | static const struct lock_manager_operations lease_manager_ops = { | |
499 | .lm_break = lease_break_callback, | |
500 | .lm_change = lease_modify, | |
501 | .lm_setup = lease_setup, | |
502 | }; | |
503 | ||
504 | /* | |
505 | * Initialize a lease, use the default lock manager operations | |
506 | */ | |
507 | static int lease_init(struct file *filp, long type, struct file_lock *fl) | |
508 | { | |
509 | if (assign_type(fl, type) != 0) | |
510 | return -EINVAL; | |
511 | ||
512 | fl->fl_owner = filp; | |
513 | fl->fl_pid = current->tgid; | |
514 | ||
515 | fl->fl_file = filp; | |
516 | fl->fl_flags = FL_LEASE; | |
517 | fl->fl_start = 0; | |
518 | fl->fl_end = OFFSET_MAX; | |
519 | fl->fl_ops = NULL; | |
520 | fl->fl_lmops = &lease_manager_ops; | |
521 | return 0; | |
522 | } | |
523 | ||
524 | /* Allocate a file_lock initialised to this type of lease */ | |
525 | static struct file_lock *lease_alloc(struct file *filp, long type) | |
526 | { | |
527 | struct file_lock *fl = locks_alloc_lock(); | |
528 | int error = -ENOMEM; | |
529 | ||
530 | if (fl == NULL) | |
531 | return ERR_PTR(error); | |
532 | ||
533 | error = lease_init(filp, type, fl); | |
534 | if (error) { | |
535 | locks_free_lock(fl); | |
536 | return ERR_PTR(error); | |
537 | } | |
538 | return fl; | |
539 | } | |
540 | ||
541 | /* Check if two locks overlap each other. | |
542 | */ | |
543 | static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2) | |
544 | { | |
545 | return ((fl1->fl_end >= fl2->fl_start) && | |
546 | (fl2->fl_end >= fl1->fl_start)); | |
547 | } | |
548 | ||
549 | /* | |
550 | * Check whether two locks have the same owner. | |
551 | */ | |
552 | static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2) | |
553 | { | |
554 | if (fl1->fl_lmops && fl1->fl_lmops->lm_compare_owner) | |
555 | return fl2->fl_lmops == fl1->fl_lmops && | |
556 | fl1->fl_lmops->lm_compare_owner(fl1, fl2); | |
557 | return fl1->fl_owner == fl2->fl_owner; | |
558 | } | |
559 | ||
560 | /* Must be called with the flc_lock held! */ | |
561 | static void locks_insert_global_locks(struct file_lock *fl) | |
562 | { | |
563 | lg_local_lock(&file_lock_lglock); | |
564 | fl->fl_link_cpu = smp_processor_id(); | |
565 | hlist_add_head(&fl->fl_link, this_cpu_ptr(&file_lock_list)); | |
566 | lg_local_unlock(&file_lock_lglock); | |
567 | } | |
568 | ||
569 | /* Must be called with the flc_lock held! */ | |
570 | static void locks_delete_global_locks(struct file_lock *fl) | |
571 | { | |
572 | /* | |
573 | * Avoid taking lock if already unhashed. This is safe since this check | |
574 | * is done while holding the flc_lock, and new insertions into the list | |
575 | * also require that it be held. | |
576 | */ | |
577 | if (hlist_unhashed(&fl->fl_link)) | |
578 | return; | |
579 | lg_local_lock_cpu(&file_lock_lglock, fl->fl_link_cpu); | |
580 | hlist_del_init(&fl->fl_link); | |
581 | lg_local_unlock_cpu(&file_lock_lglock, fl->fl_link_cpu); | |
582 | } | |
583 | ||
584 | static unsigned long | |
585 | posix_owner_key(struct file_lock *fl) | |
586 | { | |
587 | if (fl->fl_lmops && fl->fl_lmops->lm_owner_key) | |
588 | return fl->fl_lmops->lm_owner_key(fl); | |
589 | return (unsigned long)fl->fl_owner; | |
590 | } | |
591 | ||
592 | static void locks_insert_global_blocked(struct file_lock *waiter) | |
593 | { | |
594 | lockdep_assert_held(&blocked_lock_lock); | |
595 | ||
596 | hash_add(blocked_hash, &waiter->fl_link, posix_owner_key(waiter)); | |
597 | } | |
598 | ||
599 | static void locks_delete_global_blocked(struct file_lock *waiter) | |
600 | { | |
601 | lockdep_assert_held(&blocked_lock_lock); | |
602 | ||
603 | hash_del(&waiter->fl_link); | |
604 | } | |
605 | ||
606 | /* Remove waiter from blocker's block list. | |
607 | * When blocker ends up pointing to itself then the list is empty. | |
608 | * | |
609 | * Must be called with blocked_lock_lock held. | |
610 | */ | |
611 | static void __locks_delete_block(struct file_lock *waiter) | |
612 | { | |
613 | locks_delete_global_blocked(waiter); | |
614 | list_del_init(&waiter->fl_block); | |
615 | waiter->fl_next = NULL; | |
616 | } | |
617 | ||
618 | static void locks_delete_block(struct file_lock *waiter) | |
619 | { | |
620 | spin_lock(&blocked_lock_lock); | |
621 | __locks_delete_block(waiter); | |
622 | spin_unlock(&blocked_lock_lock); | |
623 | } | |
624 | ||
625 | /* Insert waiter into blocker's block list. | |
626 | * We use a circular list so that processes can be easily woken up in | |
627 | * the order they blocked. The documentation doesn't require this but | |
628 | * it seems like the reasonable thing to do. | |
629 | * | |
630 | * Must be called with both the flc_lock and blocked_lock_lock held. The | |
631 | * fl_block list itself is protected by the blocked_lock_lock, but by ensuring | |
632 | * that the flc_lock is also held on insertions we can avoid taking the | |
633 | * blocked_lock_lock in some cases when we see that the fl_block list is empty. | |
634 | */ | |
635 | static void __locks_insert_block(struct file_lock *blocker, | |
636 | struct file_lock *waiter) | |
637 | { | |
638 | BUG_ON(!list_empty(&waiter->fl_block)); | |
639 | waiter->fl_next = blocker; | |
640 | list_add_tail(&waiter->fl_block, &blocker->fl_block); | |
641 | if (IS_POSIX(blocker) && !IS_OFDLCK(blocker)) | |
642 | locks_insert_global_blocked(waiter); | |
643 | } | |
644 | ||
645 | /* Must be called with flc_lock held. */ | |
646 | static void locks_insert_block(struct file_lock *blocker, | |
647 | struct file_lock *waiter) | |
648 | { | |
649 | spin_lock(&blocked_lock_lock); | |
650 | __locks_insert_block(blocker, waiter); | |
651 | spin_unlock(&blocked_lock_lock); | |
652 | } | |
653 | ||
654 | /* | |
655 | * Wake up processes blocked waiting for blocker. | |
656 | * | |
657 | * Must be called with the inode->flc_lock held! | |
658 | */ | |
659 | static void locks_wake_up_blocks(struct file_lock *blocker) | |
660 | { | |
661 | /* | |
662 | * Avoid taking global lock if list is empty. This is safe since new | |
663 | * blocked requests are only added to the list under the flc_lock, and | |
664 | * the flc_lock is always held here. Note that removal from the fl_block | |
665 | * list does not require the flc_lock, so we must recheck list_empty() | |
666 | * after acquiring the blocked_lock_lock. | |
667 | */ | |
668 | if (list_empty(&blocker->fl_block)) | |
669 | return; | |
670 | ||
671 | spin_lock(&blocked_lock_lock); | |
672 | while (!list_empty(&blocker->fl_block)) { | |
673 | struct file_lock *waiter; | |
674 | ||
675 | waiter = list_first_entry(&blocker->fl_block, | |
676 | struct file_lock, fl_block); | |
677 | __locks_delete_block(waiter); | |
678 | if (waiter->fl_lmops && waiter->fl_lmops->lm_notify) | |
679 | waiter->fl_lmops->lm_notify(waiter); | |
680 | else | |
681 | wake_up(&waiter->fl_wait); | |
682 | } | |
683 | spin_unlock(&blocked_lock_lock); | |
684 | } | |
685 | ||
686 | static void | |
687 | locks_insert_lock_ctx(struct file_lock *fl, struct list_head *before) | |
688 | { | |
689 | fl->fl_nspid = get_pid(task_tgid(current)); | |
690 | list_add_tail(&fl->fl_list, before); | |
691 | locks_insert_global_locks(fl); | |
692 | } | |
693 | ||
694 | static void | |
695 | locks_unlink_lock_ctx(struct file_lock *fl) | |
696 | { | |
697 | locks_delete_global_locks(fl); | |
698 | list_del_init(&fl->fl_list); | |
699 | if (fl->fl_nspid) { | |
700 | put_pid(fl->fl_nspid); | |
701 | fl->fl_nspid = NULL; | |
702 | } | |
703 | locks_wake_up_blocks(fl); | |
704 | } | |
705 | ||
706 | static void | |
707 | locks_delete_lock_ctx(struct file_lock *fl, struct list_head *dispose) | |
708 | { | |
709 | locks_unlink_lock_ctx(fl); | |
710 | if (dispose) | |
711 | list_add(&fl->fl_list, dispose); | |
712 | else | |
713 | locks_free_lock(fl); | |
714 | } | |
715 | ||
716 | /* Determine if lock sys_fl blocks lock caller_fl. Common functionality | |
717 | * checks for shared/exclusive status of overlapping locks. | |
718 | */ | |
719 | static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl) | |
720 | { | |
721 | if (sys_fl->fl_type == F_WRLCK) | |
722 | return 1; | |
723 | if (caller_fl->fl_type == F_WRLCK) | |
724 | return 1; | |
725 | return 0; | |
726 | } | |
727 | ||
728 | /* Determine if lock sys_fl blocks lock caller_fl. POSIX specific | |
729 | * checking before calling the locks_conflict(). | |
730 | */ | |
731 | static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl) | |
732 | { | |
733 | /* POSIX locks owned by the same process do not conflict with | |
734 | * each other. | |
735 | */ | |
736 | if (posix_same_owner(caller_fl, sys_fl)) | |
737 | return (0); | |
738 | ||
739 | /* Check whether they overlap */ | |
740 | if (!locks_overlap(caller_fl, sys_fl)) | |
741 | return 0; | |
742 | ||
743 | return (locks_conflict(caller_fl, sys_fl)); | |
744 | } | |
745 | ||
746 | /* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific | |
747 | * checking before calling the locks_conflict(). | |
748 | */ | |
749 | static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl) | |
750 | { | |
751 | /* FLOCK locks referring to the same filp do not conflict with | |
752 | * each other. | |
753 | */ | |
754 | if (caller_fl->fl_file == sys_fl->fl_file) | |
755 | return (0); | |
756 | if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND)) | |
757 | return 0; | |
758 | ||
759 | return (locks_conflict(caller_fl, sys_fl)); | |
760 | } | |
761 | ||
762 | void | |
763 | posix_test_lock(struct file *filp, struct file_lock *fl) | |
764 | { | |
765 | struct file_lock *cfl; | |
766 | struct file_lock_context *ctx; | |
767 | struct inode *inode = file_inode(filp); | |
768 | ||
769 | ctx = smp_load_acquire(&inode->i_flctx); | |
770 | if (!ctx || list_empty_careful(&ctx->flc_posix)) { | |
771 | fl->fl_type = F_UNLCK; | |
772 | return; | |
773 | } | |
774 | ||
775 | spin_lock(&ctx->flc_lock); | |
776 | list_for_each_entry(cfl, &ctx->flc_posix, fl_list) { | |
777 | if (posix_locks_conflict(fl, cfl)) { | |
778 | locks_copy_conflock(fl, cfl); | |
779 | if (cfl->fl_nspid) | |
780 | fl->fl_pid = pid_vnr(cfl->fl_nspid); | |
781 | goto out; | |
782 | } | |
783 | } | |
784 | fl->fl_type = F_UNLCK; | |
785 | out: | |
786 | spin_unlock(&ctx->flc_lock); | |
787 | return; | |
788 | } | |
789 | EXPORT_SYMBOL(posix_test_lock); | |
790 | ||
791 | /* | |
792 | * Deadlock detection: | |
793 | * | |
794 | * We attempt to detect deadlocks that are due purely to posix file | |
795 | * locks. | |
796 | * | |
797 | * We assume that a task can be waiting for at most one lock at a time. | |
798 | * So for any acquired lock, the process holding that lock may be | |
799 | * waiting on at most one other lock. That lock in turns may be held by | |
800 | * someone waiting for at most one other lock. Given a requested lock | |
801 | * caller_fl which is about to wait for a conflicting lock block_fl, we | |
802 | * follow this chain of waiters to ensure we are not about to create a | |
803 | * cycle. | |
804 | * | |
805 | * Since we do this before we ever put a process to sleep on a lock, we | |
806 | * are ensured that there is never a cycle; that is what guarantees that | |
807 | * the while() loop in posix_locks_deadlock() eventually completes. | |
808 | * | |
809 | * Note: the above assumption may not be true when handling lock | |
810 | * requests from a broken NFS client. It may also fail in the presence | |
811 | * of tasks (such as posix threads) sharing the same open file table. | |
812 | * To handle those cases, we just bail out after a few iterations. | |
813 | * | |
814 | * For FL_OFDLCK locks, the owner is the filp, not the files_struct. | |
815 | * Because the owner is not even nominally tied to a thread of | |
816 | * execution, the deadlock detection below can't reasonably work well. Just | |
817 | * skip it for those. | |
818 | * | |
819 | * In principle, we could do a more limited deadlock detection on FL_OFDLCK | |
820 | * locks that just checks for the case where two tasks are attempting to | |
821 | * upgrade from read to write locks on the same inode. | |
822 | */ | |
823 | ||
824 | #define MAX_DEADLK_ITERATIONS 10 | |
825 | ||
826 | /* Find a lock that the owner of the given block_fl is blocking on. */ | |
827 | static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl) | |
828 | { | |
829 | struct file_lock *fl; | |
830 | ||
831 | hash_for_each_possible(blocked_hash, fl, fl_link, posix_owner_key(block_fl)) { | |
832 | if (posix_same_owner(fl, block_fl)) | |
833 | return fl->fl_next; | |
834 | } | |
835 | return NULL; | |
836 | } | |
837 | ||
838 | /* Must be called with the blocked_lock_lock held! */ | |
839 | static int posix_locks_deadlock(struct file_lock *caller_fl, | |
840 | struct file_lock *block_fl) | |
841 | { | |
842 | int i = 0; | |
843 | ||
844 | lockdep_assert_held(&blocked_lock_lock); | |
845 | ||
846 | /* | |
847 | * This deadlock detector can't reasonably detect deadlocks with | |
848 | * FL_OFDLCK locks, since they aren't owned by a process, per-se. | |
849 | */ | |
850 | if (IS_OFDLCK(caller_fl)) | |
851 | return 0; | |
852 | ||
853 | while ((block_fl = what_owner_is_waiting_for(block_fl))) { | |
854 | if (i++ > MAX_DEADLK_ITERATIONS) | |
855 | return 0; | |
856 | if (posix_same_owner(caller_fl, block_fl)) | |
857 | return 1; | |
858 | } | |
859 | return 0; | |
860 | } | |
861 | ||
862 | /* Try to create a FLOCK lock on filp. We always insert new FLOCK locks | |
863 | * after any leases, but before any posix locks. | |
864 | * | |
865 | * Note that if called with an FL_EXISTS argument, the caller may determine | |
866 | * whether or not a lock was successfully freed by testing the return | |
867 | * value for -ENOENT. | |
868 | */ | |
869 | static int flock_lock_inode(struct inode *inode, struct file_lock *request) | |
870 | { | |
871 | struct file_lock *new_fl = NULL; | |
872 | struct file_lock *fl; | |
873 | struct file_lock_context *ctx; | |
874 | int error = 0; | |
875 | bool found = false; | |
876 | LIST_HEAD(dispose); | |
877 | ||
878 | ctx = locks_get_lock_context(inode, request->fl_type); | |
879 | if (!ctx) { | |
880 | if (request->fl_type != F_UNLCK) | |
881 | return -ENOMEM; | |
882 | return (request->fl_flags & FL_EXISTS) ? -ENOENT : 0; | |
883 | } | |
884 | ||
885 | if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) { | |
886 | new_fl = locks_alloc_lock(); | |
887 | if (!new_fl) | |
888 | return -ENOMEM; | |
889 | } | |
890 | ||
891 | spin_lock(&ctx->flc_lock); | |
892 | if (request->fl_flags & FL_ACCESS) | |
893 | goto find_conflict; | |
894 | ||
895 | list_for_each_entry(fl, &ctx->flc_flock, fl_list) { | |
896 | if (request->fl_file != fl->fl_file) | |
897 | continue; | |
898 | if (request->fl_type == fl->fl_type) | |
899 | goto out; | |
900 | found = true; | |
901 | locks_delete_lock_ctx(fl, &dispose); | |
902 | break; | |
903 | } | |
904 | ||
905 | if (request->fl_type == F_UNLCK) { | |
906 | if ((request->fl_flags & FL_EXISTS) && !found) | |
907 | error = -ENOENT; | |
908 | goto out; | |
909 | } | |
910 | ||
911 | find_conflict: | |
912 | list_for_each_entry(fl, &ctx->flc_flock, fl_list) { | |
913 | if (!flock_locks_conflict(request, fl)) | |
914 | continue; | |
915 | error = -EAGAIN; | |
916 | if (!(request->fl_flags & FL_SLEEP)) | |
917 | goto out; | |
918 | error = FILE_LOCK_DEFERRED; | |
919 | locks_insert_block(fl, request); | |
920 | goto out; | |
921 | } | |
922 | if (request->fl_flags & FL_ACCESS) | |
923 | goto out; | |
924 | locks_copy_lock(new_fl, request); | |
925 | locks_insert_lock_ctx(new_fl, &ctx->flc_flock); | |
926 | new_fl = NULL; | |
927 | error = 0; | |
928 | ||
929 | out: | |
930 | spin_unlock(&ctx->flc_lock); | |
931 | if (new_fl) | |
932 | locks_free_lock(new_fl); | |
933 | locks_dispose_list(&dispose); | |
934 | return error; | |
935 | } | |
936 | ||
937 | static int __posix_lock_file(struct inode *inode, struct file_lock *request, struct file_lock *conflock) | |
938 | { | |
939 | struct file_lock *fl, *tmp; | |
940 | struct file_lock *new_fl = NULL; | |
941 | struct file_lock *new_fl2 = NULL; | |
942 | struct file_lock *left = NULL; | |
943 | struct file_lock *right = NULL; | |
944 | struct file_lock_context *ctx; | |
945 | int error; | |
946 | bool added = false; | |
947 | LIST_HEAD(dispose); | |
948 | ||
949 | ctx = locks_get_lock_context(inode, request->fl_type); | |
950 | if (!ctx) | |
951 | return (request->fl_type == F_UNLCK) ? 0 : -ENOMEM; | |
952 | ||
953 | /* | |
954 | * We may need two file_lock structures for this operation, | |
955 | * so we get them in advance to avoid races. | |
956 | * | |
957 | * In some cases we can be sure, that no new locks will be needed | |
958 | */ | |
959 | if (!(request->fl_flags & FL_ACCESS) && | |
960 | (request->fl_type != F_UNLCK || | |
961 | request->fl_start != 0 || request->fl_end != OFFSET_MAX)) { | |
962 | new_fl = locks_alloc_lock(); | |
963 | new_fl2 = locks_alloc_lock(); | |
964 | } | |
965 | ||
966 | spin_lock(&ctx->flc_lock); | |
967 | /* | |
968 | * New lock request. Walk all POSIX locks and look for conflicts. If | |
969 | * there are any, either return error or put the request on the | |
970 | * blocker's list of waiters and the global blocked_hash. | |
971 | */ | |
972 | if (request->fl_type != F_UNLCK) { | |
973 | list_for_each_entry(fl, &ctx->flc_posix, fl_list) { | |
974 | if (!posix_locks_conflict(request, fl)) | |
975 | continue; | |
976 | if (conflock) | |
977 | locks_copy_conflock(conflock, fl); | |
978 | error = -EAGAIN; | |
979 | if (!(request->fl_flags & FL_SLEEP)) | |
980 | goto out; | |
981 | /* | |
982 | * Deadlock detection and insertion into the blocked | |
983 | * locks list must be done while holding the same lock! | |
984 | */ | |
985 | error = -EDEADLK; | |
986 | spin_lock(&blocked_lock_lock); | |
987 | if (likely(!posix_locks_deadlock(request, fl))) { | |
988 | error = FILE_LOCK_DEFERRED; | |
989 | __locks_insert_block(fl, request); | |
990 | } | |
991 | spin_unlock(&blocked_lock_lock); | |
992 | goto out; | |
993 | } | |
994 | } | |
995 | ||
996 | /* If we're just looking for a conflict, we're done. */ | |
997 | error = 0; | |
998 | if (request->fl_flags & FL_ACCESS) | |
999 | goto out; | |
1000 | ||
1001 | /* Find the first old lock with the same owner as the new lock */ | |
1002 | list_for_each_entry(fl, &ctx->flc_posix, fl_list) { | |
1003 | if (posix_same_owner(request, fl)) | |
1004 | break; | |
1005 | } | |
1006 | ||
1007 | /* Process locks with this owner. */ | |
1008 | list_for_each_entry_safe_from(fl, tmp, &ctx->flc_posix, fl_list) { | |
1009 | if (!posix_same_owner(request, fl)) | |
1010 | break; | |
1011 | ||
1012 | /* Detect adjacent or overlapping regions (if same lock type) */ | |
1013 | if (request->fl_type == fl->fl_type) { | |
1014 | /* In all comparisons of start vs end, use | |
1015 | * "start - 1" rather than "end + 1". If end | |
1016 | * is OFFSET_MAX, end + 1 will become negative. | |
1017 | */ | |
1018 | if (fl->fl_end < request->fl_start - 1) | |
1019 | continue; | |
1020 | /* If the next lock in the list has entirely bigger | |
1021 | * addresses than the new one, insert the lock here. | |
1022 | */ | |
1023 | if (fl->fl_start - 1 > request->fl_end) | |
1024 | break; | |
1025 | ||
1026 | /* If we come here, the new and old lock are of the | |
1027 | * same type and adjacent or overlapping. Make one | |
1028 | * lock yielding from the lower start address of both | |
1029 | * locks to the higher end address. | |
1030 | */ | |
1031 | if (fl->fl_start > request->fl_start) | |
1032 | fl->fl_start = request->fl_start; | |
1033 | else | |
1034 | request->fl_start = fl->fl_start; | |
1035 | if (fl->fl_end < request->fl_end) | |
1036 | fl->fl_end = request->fl_end; | |
1037 | else | |
1038 | request->fl_end = fl->fl_end; | |
1039 | if (added) { | |
1040 | locks_delete_lock_ctx(fl, &dispose); | |
1041 | continue; | |
1042 | } | |
1043 | request = fl; | |
1044 | added = true; | |
1045 | } else { | |
1046 | /* Processing for different lock types is a bit | |
1047 | * more complex. | |
1048 | */ | |
1049 | if (fl->fl_end < request->fl_start) | |
1050 | continue; | |
1051 | if (fl->fl_start > request->fl_end) | |
1052 | break; | |
1053 | if (request->fl_type == F_UNLCK) | |
1054 | added = true; | |
1055 | if (fl->fl_start < request->fl_start) | |
1056 | left = fl; | |
1057 | /* If the next lock in the list has a higher end | |
1058 | * address than the new one, insert the new one here. | |
1059 | */ | |
1060 | if (fl->fl_end > request->fl_end) { | |
1061 | right = fl; | |
1062 | break; | |
1063 | } | |
1064 | if (fl->fl_start >= request->fl_start) { | |
1065 | /* The new lock completely replaces an old | |
1066 | * one (This may happen several times). | |
1067 | */ | |
1068 | if (added) { | |
1069 | locks_delete_lock_ctx(fl, &dispose); | |
1070 | continue; | |
1071 | } | |
1072 | /* | |
1073 | * Replace the old lock with new_fl, and | |
1074 | * remove the old one. It's safe to do the | |
1075 | * insert here since we know that we won't be | |
1076 | * using new_fl later, and that the lock is | |
1077 | * just replacing an existing lock. | |
1078 | */ | |
1079 | error = -ENOLCK; | |
1080 | if (!new_fl) | |
1081 | goto out; | |
1082 | locks_copy_lock(new_fl, request); | |
1083 | request = new_fl; | |
1084 | new_fl = NULL; | |
1085 | locks_insert_lock_ctx(request, &fl->fl_list); | |
1086 | locks_delete_lock_ctx(fl, &dispose); | |
1087 | added = true; | |
1088 | } | |
1089 | } | |
1090 | } | |
1091 | ||
1092 | /* | |
1093 | * The above code only modifies existing locks in case of merging or | |
1094 | * replacing. If new lock(s) need to be inserted all modifications are | |
1095 | * done below this, so it's safe yet to bail out. | |
1096 | */ | |
1097 | error = -ENOLCK; /* "no luck" */ | |
1098 | if (right && left == right && !new_fl2) | |
1099 | goto out; | |
1100 | ||
1101 | error = 0; | |
1102 | if (!added) { | |
1103 | if (request->fl_type == F_UNLCK) { | |
1104 | if (request->fl_flags & FL_EXISTS) | |
1105 | error = -ENOENT; | |
1106 | goto out; | |
1107 | } | |
1108 | ||
1109 | if (!new_fl) { | |
1110 | error = -ENOLCK; | |
1111 | goto out; | |
1112 | } | |
1113 | locks_copy_lock(new_fl, request); | |
1114 | locks_insert_lock_ctx(new_fl, &fl->fl_list); | |
1115 | fl = new_fl; | |
1116 | new_fl = NULL; | |
1117 | } | |
1118 | if (right) { | |
1119 | if (left == right) { | |
1120 | /* The new lock breaks the old one in two pieces, | |
1121 | * so we have to use the second new lock. | |
1122 | */ | |
1123 | left = new_fl2; | |
1124 | new_fl2 = NULL; | |
1125 | locks_copy_lock(left, right); | |
1126 | locks_insert_lock_ctx(left, &fl->fl_list); | |
1127 | } | |
1128 | right->fl_start = request->fl_end + 1; | |
1129 | locks_wake_up_blocks(right); | |
1130 | } | |
1131 | if (left) { | |
1132 | left->fl_end = request->fl_start - 1; | |
1133 | locks_wake_up_blocks(left); | |
1134 | } | |
1135 | out: | |
1136 | spin_unlock(&ctx->flc_lock); | |
1137 | /* | |
1138 | * Free any unused locks. | |
1139 | */ | |
1140 | if (new_fl) | |
1141 | locks_free_lock(new_fl); | |
1142 | if (new_fl2) | |
1143 | locks_free_lock(new_fl2); | |
1144 | locks_dispose_list(&dispose); | |
1145 | trace_posix_lock_inode(inode, request, error); | |
1146 | ||
1147 | return error; | |
1148 | } | |
1149 | ||
1150 | /** | |
1151 | * posix_lock_file - Apply a POSIX-style lock to a file | |
1152 | * @filp: The file to apply the lock to | |
1153 | * @fl: The lock to be applied | |
1154 | * @conflock: Place to return a copy of the conflicting lock, if found. | |
1155 | * | |
1156 | * Add a POSIX style lock to a file. | |
1157 | * We merge adjacent & overlapping locks whenever possible. | |
1158 | * POSIX locks are sorted by owner task, then by starting address | |
1159 | * | |
1160 | * Note that if called with an FL_EXISTS argument, the caller may determine | |
1161 | * whether or not a lock was successfully freed by testing the return | |
1162 | * value for -ENOENT. | |
1163 | */ | |
1164 | int posix_lock_file(struct file *filp, struct file_lock *fl, | |
1165 | struct file_lock *conflock) | |
1166 | { | |
1167 | return __posix_lock_file(file_inode(filp), fl, conflock); | |
1168 | } | |
1169 | EXPORT_SYMBOL(posix_lock_file); | |
1170 | ||
1171 | /** | |
1172 | * posix_lock_inode_wait - Apply a POSIX-style lock to a file | |
1173 | * @inode: inode of file to which lock request should be applied | |
1174 | * @fl: The lock to be applied | |
1175 | * | |
1176 | * Apply a POSIX style lock request to an inode. | |
1177 | */ | |
1178 | static int posix_lock_inode_wait(struct inode *inode, struct file_lock *fl) | |
1179 | { | |
1180 | int error; | |
1181 | might_sleep (); | |
1182 | for (;;) { | |
1183 | error = __posix_lock_file(inode, fl, NULL); | |
1184 | if (error != FILE_LOCK_DEFERRED) | |
1185 | break; | |
1186 | error = wait_event_interruptible(fl->fl_wait, !fl->fl_next); | |
1187 | if (!error) | |
1188 | continue; | |
1189 | ||
1190 | locks_delete_block(fl); | |
1191 | break; | |
1192 | } | |
1193 | return error; | |
1194 | } | |
1195 | ||
1196 | #ifdef CONFIG_MANDATORY_FILE_LOCKING | |
1197 | /** | |
1198 | * locks_mandatory_locked - Check for an active lock | |
1199 | * @file: the file to check | |
1200 | * | |
1201 | * Searches the inode's list of locks to find any POSIX locks which conflict. | |
1202 | * This function is called from locks_verify_locked() only. | |
1203 | */ | |
1204 | int locks_mandatory_locked(struct file *file) | |
1205 | { | |
1206 | int ret; | |
1207 | struct inode *inode = file_inode(file); | |
1208 | struct file_lock_context *ctx; | |
1209 | struct file_lock *fl; | |
1210 | ||
1211 | ctx = smp_load_acquire(&inode->i_flctx); | |
1212 | if (!ctx || list_empty_careful(&ctx->flc_posix)) | |
1213 | return 0; | |
1214 | ||
1215 | /* | |
1216 | * Search the lock list for this inode for any POSIX locks. | |
1217 | */ | |
1218 | spin_lock(&ctx->flc_lock); | |
1219 | ret = 0; | |
1220 | list_for_each_entry(fl, &ctx->flc_posix, fl_list) { | |
1221 | if (fl->fl_owner != current->files && | |
1222 | fl->fl_owner != file) { | |
1223 | ret = -EAGAIN; | |
1224 | break; | |
1225 | } | |
1226 | } | |
1227 | spin_unlock(&ctx->flc_lock); | |
1228 | return ret; | |
1229 | } | |
1230 | ||
1231 | /** | |
1232 | * locks_mandatory_area - Check for a conflicting lock | |
1233 | * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ | |
1234 | * for shared | |
1235 | * @inode: the file to check | |
1236 | * @filp: how the file was opened (if it was) | |
1237 | * @offset: start of area to check | |
1238 | * @count: length of area to check | |
1239 | * | |
1240 | * Searches the inode's list of locks to find any POSIX locks which conflict. | |
1241 | * This function is called from rw_verify_area() and | |
1242 | * locks_verify_truncate(). | |
1243 | */ | |
1244 | int locks_mandatory_area(int read_write, struct inode *inode, | |
1245 | struct file *filp, loff_t offset, | |
1246 | size_t count) | |
1247 | { | |
1248 | struct file_lock fl; | |
1249 | int error; | |
1250 | bool sleep = false; | |
1251 | ||
1252 | locks_init_lock(&fl); | |
1253 | fl.fl_pid = current->tgid; | |
1254 | fl.fl_file = filp; | |
1255 | fl.fl_flags = FL_POSIX | FL_ACCESS; | |
1256 | if (filp && !(filp->f_flags & O_NONBLOCK)) | |
1257 | sleep = true; | |
1258 | fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK; | |
1259 | fl.fl_start = offset; | |
1260 | fl.fl_end = offset + count - 1; | |
1261 | ||
1262 | for (;;) { | |
1263 | if (filp) { | |
1264 | fl.fl_owner = filp; | |
1265 | fl.fl_flags &= ~FL_SLEEP; | |
1266 | error = __posix_lock_file(inode, &fl, NULL); | |
1267 | if (!error) | |
1268 | break; | |
1269 | } | |
1270 | ||
1271 | if (sleep) | |
1272 | fl.fl_flags |= FL_SLEEP; | |
1273 | fl.fl_owner = current->files; | |
1274 | error = __posix_lock_file(inode, &fl, NULL); | |
1275 | if (error != FILE_LOCK_DEFERRED) | |
1276 | break; | |
1277 | error = wait_event_interruptible(fl.fl_wait, !fl.fl_next); | |
1278 | if (!error) { | |
1279 | /* | |
1280 | * If we've been sleeping someone might have | |
1281 | * changed the permissions behind our back. | |
1282 | */ | |
1283 | if (__mandatory_lock(inode)) | |
1284 | continue; | |
1285 | } | |
1286 | ||
1287 | locks_delete_block(&fl); | |
1288 | break; | |
1289 | } | |
1290 | ||
1291 | return error; | |
1292 | } | |
1293 | ||
1294 | EXPORT_SYMBOL(locks_mandatory_area); | |
1295 | #endif /* CONFIG_MANDATORY_FILE_LOCKING */ | |
1296 | ||
1297 | static void lease_clear_pending(struct file_lock *fl, int arg) | |
1298 | { | |
1299 | switch (arg) { | |
1300 | case F_UNLCK: | |
1301 | fl->fl_flags &= ~FL_UNLOCK_PENDING; | |
1302 | /* fall through: */ | |
1303 | case F_RDLCK: | |
1304 | fl->fl_flags &= ~FL_DOWNGRADE_PENDING; | |
1305 | } | |
1306 | } | |
1307 | ||
1308 | /* We already had a lease on this file; just change its type */ | |
1309 | int lease_modify(struct file_lock *fl, int arg, struct list_head *dispose) | |
1310 | { | |
1311 | int error = assign_type(fl, arg); | |
1312 | ||
1313 | if (error) | |
1314 | return error; | |
1315 | lease_clear_pending(fl, arg); | |
1316 | locks_wake_up_blocks(fl); | |
1317 | if (arg == F_UNLCK) { | |
1318 | struct file *filp = fl->fl_file; | |
1319 | ||
1320 | f_delown(filp); | |
1321 | filp->f_owner.signum = 0; | |
1322 | fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync); | |
1323 | if (fl->fl_fasync != NULL) { | |
1324 | printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync); | |
1325 | fl->fl_fasync = NULL; | |
1326 | } | |
1327 | locks_delete_lock_ctx(fl, dispose); | |
1328 | } | |
1329 | return 0; | |
1330 | } | |
1331 | EXPORT_SYMBOL(lease_modify); | |
1332 | ||
1333 | static bool past_time(unsigned long then) | |
1334 | { | |
1335 | if (!then) | |
1336 | /* 0 is a special value meaning "this never expires": */ | |
1337 | return false; | |
1338 | return time_after(jiffies, then); | |
1339 | } | |
1340 | ||
1341 | static void time_out_leases(struct inode *inode, struct list_head *dispose) | |
1342 | { | |
1343 | struct file_lock_context *ctx = inode->i_flctx; | |
1344 | struct file_lock *fl, *tmp; | |
1345 | ||
1346 | lockdep_assert_held(&ctx->flc_lock); | |
1347 | ||
1348 | list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list) { | |
1349 | trace_time_out_leases(inode, fl); | |
1350 | if (past_time(fl->fl_downgrade_time)) | |
1351 | lease_modify(fl, F_RDLCK, dispose); | |
1352 | if (past_time(fl->fl_break_time)) | |
1353 | lease_modify(fl, F_UNLCK, dispose); | |
1354 | } | |
1355 | } | |
1356 | ||
1357 | static bool leases_conflict(struct file_lock *lease, struct file_lock *breaker) | |
1358 | { | |
1359 | if ((breaker->fl_flags & FL_LAYOUT) != (lease->fl_flags & FL_LAYOUT)) | |
1360 | return false; | |
1361 | if ((breaker->fl_flags & FL_DELEG) && (lease->fl_flags & FL_LEASE)) | |
1362 | return false; | |
1363 | return locks_conflict(breaker, lease); | |
1364 | } | |
1365 | ||
1366 | static bool | |
1367 | any_leases_conflict(struct inode *inode, struct file_lock *breaker) | |
1368 | { | |
1369 | struct file_lock_context *ctx = inode->i_flctx; | |
1370 | struct file_lock *fl; | |
1371 | ||
1372 | lockdep_assert_held(&ctx->flc_lock); | |
1373 | ||
1374 | list_for_each_entry(fl, &ctx->flc_lease, fl_list) { | |
1375 | if (leases_conflict(fl, breaker)) | |
1376 | return true; | |
1377 | } | |
1378 | return false; | |
1379 | } | |
1380 | ||
1381 | /** | |
1382 | * __break_lease - revoke all outstanding leases on file | |
1383 | * @inode: the inode of the file to return | |
1384 | * @mode: O_RDONLY: break only write leases; O_WRONLY or O_RDWR: | |
1385 | * break all leases | |
1386 | * @type: FL_LEASE: break leases and delegations; FL_DELEG: break | |
1387 | * only delegations | |
1388 | * | |
1389 | * break_lease (inlined for speed) has checked there already is at least | |
1390 | * some kind of lock (maybe a lease) on this file. Leases are broken on | |
1391 | * a call to open() or truncate(). This function can sleep unless you | |
1392 | * specified %O_NONBLOCK to your open(). | |
1393 | */ | |
1394 | int __break_lease(struct inode *inode, unsigned int mode, unsigned int type) | |
1395 | { | |
1396 | int error = 0; | |
1397 | struct file_lock_context *ctx; | |
1398 | struct file_lock *new_fl, *fl, *tmp; | |
1399 | unsigned long break_time; | |
1400 | int want_write = (mode & O_ACCMODE) != O_RDONLY; | |
1401 | LIST_HEAD(dispose); | |
1402 | ||
1403 | new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK); | |
1404 | if (IS_ERR(new_fl)) | |
1405 | return PTR_ERR(new_fl); | |
1406 | new_fl->fl_flags = type; | |
1407 | ||
1408 | /* typically we will check that ctx is non-NULL before calling */ | |
1409 | ctx = smp_load_acquire(&inode->i_flctx); | |
1410 | if (!ctx) { | |
1411 | WARN_ON_ONCE(1); | |
1412 | return error; | |
1413 | } | |
1414 | ||
1415 | spin_lock(&ctx->flc_lock); | |
1416 | ||
1417 | time_out_leases(inode, &dispose); | |
1418 | ||
1419 | if (!any_leases_conflict(inode, new_fl)) | |
1420 | goto out; | |
1421 | ||
1422 | break_time = 0; | |
1423 | if (lease_break_time > 0) { | |
1424 | break_time = jiffies + lease_break_time * HZ; | |
1425 | if (break_time == 0) | |
1426 | break_time++; /* so that 0 means no break time */ | |
1427 | } | |
1428 | ||
1429 | list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list) { | |
1430 | if (!leases_conflict(fl, new_fl)) | |
1431 | continue; | |
1432 | if (want_write) { | |
1433 | if (fl->fl_flags & FL_UNLOCK_PENDING) | |
1434 | continue; | |
1435 | fl->fl_flags |= FL_UNLOCK_PENDING; | |
1436 | fl->fl_break_time = break_time; | |
1437 | } else { | |
1438 | if (lease_breaking(fl)) | |
1439 | continue; | |
1440 | fl->fl_flags |= FL_DOWNGRADE_PENDING; | |
1441 | fl->fl_downgrade_time = break_time; | |
1442 | } | |
1443 | if (fl->fl_lmops->lm_break(fl)) | |
1444 | locks_delete_lock_ctx(fl, &dispose); | |
1445 | } | |
1446 | ||
1447 | if (list_empty(&ctx->flc_lease)) | |
1448 | goto out; | |
1449 | ||
1450 | if (mode & O_NONBLOCK) { | |
1451 | trace_break_lease_noblock(inode, new_fl); | |
1452 | error = -EWOULDBLOCK; | |
1453 | goto out; | |
1454 | } | |
1455 | ||
1456 | restart: | |
1457 | fl = list_first_entry(&ctx->flc_lease, struct file_lock, fl_list); | |
1458 | break_time = fl->fl_break_time; | |
1459 | if (break_time != 0) | |
1460 | break_time -= jiffies; | |
1461 | if (break_time == 0) | |
1462 | break_time++; | |
1463 | locks_insert_block(fl, new_fl); | |
1464 | trace_break_lease_block(inode, new_fl); | |
1465 | spin_unlock(&ctx->flc_lock); | |
1466 | locks_dispose_list(&dispose); | |
1467 | error = wait_event_interruptible_timeout(new_fl->fl_wait, | |
1468 | !new_fl->fl_next, break_time); | |
1469 | spin_lock(&ctx->flc_lock); | |
1470 | trace_break_lease_unblock(inode, new_fl); | |
1471 | locks_delete_block(new_fl); | |
1472 | if (error >= 0) { | |
1473 | /* | |
1474 | * Wait for the next conflicting lease that has not been | |
1475 | * broken yet | |
1476 | */ | |
1477 | if (error == 0) | |
1478 | time_out_leases(inode, &dispose); | |
1479 | if (any_leases_conflict(inode, new_fl)) | |
1480 | goto restart; | |
1481 | error = 0; | |
1482 | } | |
1483 | out: | |
1484 | spin_unlock(&ctx->flc_lock); | |
1485 | locks_dispose_list(&dispose); | |
1486 | locks_free_lock(new_fl); | |
1487 | return error; | |
1488 | } | |
1489 | ||
1490 | EXPORT_SYMBOL(__break_lease); | |
1491 | ||
1492 | /** | |
1493 | * lease_get_mtime - get the last modified time of an inode | |
1494 | * @inode: the inode | |
1495 | * @time: pointer to a timespec which will contain the last modified time | |
1496 | * | |
1497 | * This is to force NFS clients to flush their caches for files with | |
1498 | * exclusive leases. The justification is that if someone has an | |
1499 | * exclusive lease, then they could be modifying it. | |
1500 | */ | |
1501 | void lease_get_mtime(struct inode *inode, struct timespec *time) | |
1502 | { | |
1503 | bool has_lease = false; | |
1504 | struct file_lock_context *ctx; | |
1505 | struct file_lock *fl; | |
1506 | ||
1507 | ctx = smp_load_acquire(&inode->i_flctx); | |
1508 | if (ctx && !list_empty_careful(&ctx->flc_lease)) { | |
1509 | spin_lock(&ctx->flc_lock); | |
1510 | fl = list_first_entry_or_null(&ctx->flc_lease, | |
1511 | struct file_lock, fl_list); | |
1512 | if (fl && (fl->fl_type == F_WRLCK)) | |
1513 | has_lease = true; | |
1514 | spin_unlock(&ctx->flc_lock); | |
1515 | } | |
1516 | ||
1517 | if (has_lease) | |
1518 | *time = current_fs_time(inode->i_sb); | |
1519 | else | |
1520 | *time = inode->i_mtime; | |
1521 | } | |
1522 | ||
1523 | EXPORT_SYMBOL(lease_get_mtime); | |
1524 | ||
1525 | /** | |
1526 | * fcntl_getlease - Enquire what lease is currently active | |
1527 | * @filp: the file | |
1528 | * | |
1529 | * The value returned by this function will be one of | |
1530 | * (if no lease break is pending): | |
1531 | * | |
1532 | * %F_RDLCK to indicate a shared lease is held. | |
1533 | * | |
1534 | * %F_WRLCK to indicate an exclusive lease is held. | |
1535 | * | |
1536 | * %F_UNLCK to indicate no lease is held. | |
1537 | * | |
1538 | * (if a lease break is pending): | |
1539 | * | |
1540 | * %F_RDLCK to indicate an exclusive lease needs to be | |
1541 | * changed to a shared lease (or removed). | |
1542 | * | |
1543 | * %F_UNLCK to indicate the lease needs to be removed. | |
1544 | * | |
1545 | * XXX: sfr & willy disagree over whether F_INPROGRESS | |
1546 | * should be returned to userspace. | |
1547 | */ | |
1548 | int fcntl_getlease(struct file *filp) | |
1549 | { | |
1550 | struct file_lock *fl; | |
1551 | struct inode *inode = file_inode(filp); | |
1552 | struct file_lock_context *ctx; | |
1553 | int type = F_UNLCK; | |
1554 | LIST_HEAD(dispose); | |
1555 | ||
1556 | ctx = smp_load_acquire(&inode->i_flctx); | |
1557 | if (ctx && !list_empty_careful(&ctx->flc_lease)) { | |
1558 | spin_lock(&ctx->flc_lock); | |
1559 | time_out_leases(file_inode(filp), &dispose); | |
1560 | list_for_each_entry(fl, &ctx->flc_lease, fl_list) { | |
1561 | if (fl->fl_file != filp) | |
1562 | continue; | |
1563 | type = target_leasetype(fl); | |
1564 | break; | |
1565 | } | |
1566 | spin_unlock(&ctx->flc_lock); | |
1567 | locks_dispose_list(&dispose); | |
1568 | } | |
1569 | return type; | |
1570 | } | |
1571 | ||
1572 | /** | |
1573 | * check_conflicting_open - see if the given dentry points to a file that has | |
1574 | * an existing open that would conflict with the | |
1575 | * desired lease. | |
1576 | * @dentry: dentry to check | |
1577 | * @arg: type of lease that we're trying to acquire | |
1578 | * @flags: current lock flags | |
1579 | * | |
1580 | * Check to see if there's an existing open fd on this file that would | |
1581 | * conflict with the lease we're trying to set. | |
1582 | */ | |
1583 | static int | |
1584 | check_conflicting_open(const struct dentry *dentry, const long arg, int flags) | |
1585 | { | |
1586 | int ret = 0; | |
1587 | struct inode *inode = dentry->d_inode; | |
1588 | ||
1589 | if (flags & FL_LAYOUT) | |
1590 | return 0; | |
1591 | ||
1592 | if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0)) | |
1593 | return -EAGAIN; | |
1594 | ||
1595 | if ((arg == F_WRLCK) && ((d_count(dentry) > 1) || | |
1596 | (atomic_read(&inode->i_count) > 1))) | |
1597 | ret = -EAGAIN; | |
1598 | ||
1599 | return ret; | |
1600 | } | |
1601 | ||
1602 | static int | |
1603 | generic_add_lease(struct file *filp, long arg, struct file_lock **flp, void **priv) | |
1604 | { | |
1605 | struct file_lock *fl, *my_fl = NULL, *lease; | |
1606 | struct dentry *dentry = filp->f_path.dentry; | |
1607 | struct inode *inode = dentry->d_inode; | |
1608 | struct file_lock_context *ctx; | |
1609 | bool is_deleg = (*flp)->fl_flags & FL_DELEG; | |
1610 | int error; | |
1611 | LIST_HEAD(dispose); | |
1612 | ||
1613 | lease = *flp; | |
1614 | trace_generic_add_lease(inode, lease); | |
1615 | ||
1616 | /* Note that arg is never F_UNLCK here */ | |
1617 | ctx = locks_get_lock_context(inode, arg); | |
1618 | if (!ctx) | |
1619 | return -ENOMEM; | |
1620 | ||
1621 | /* | |
1622 | * In the delegation case we need mutual exclusion with | |
1623 | * a number of operations that take the i_mutex. We trylock | |
1624 | * because delegations are an optional optimization, and if | |
1625 | * there's some chance of a conflict--we'd rather not | |
1626 | * bother, maybe that's a sign this just isn't a good file to | |
1627 | * hand out a delegation on. | |
1628 | */ | |
1629 | if (is_deleg && !mutex_trylock(&inode->i_mutex)) | |
1630 | return -EAGAIN; | |
1631 | ||
1632 | if (is_deleg && arg == F_WRLCK) { | |
1633 | /* Write delegations are not currently supported: */ | |
1634 | mutex_unlock(&inode->i_mutex); | |
1635 | WARN_ON_ONCE(1); | |
1636 | return -EINVAL; | |
1637 | } | |
1638 | ||
1639 | spin_lock(&ctx->flc_lock); | |
1640 | time_out_leases(inode, &dispose); | |
1641 | error = check_conflicting_open(dentry, arg, lease->fl_flags); | |
1642 | if (error) | |
1643 | goto out; | |
1644 | ||
1645 | /* | |
1646 | * At this point, we know that if there is an exclusive | |
1647 | * lease on this file, then we hold it on this filp | |
1648 | * (otherwise our open of this file would have blocked). | |
1649 | * And if we are trying to acquire an exclusive lease, | |
1650 | * then the file is not open by anyone (including us) | |
1651 | * except for this filp. | |
1652 | */ | |
1653 | error = -EAGAIN; | |
1654 | list_for_each_entry(fl, &ctx->flc_lease, fl_list) { | |
1655 | if (fl->fl_file == filp && | |
1656 | fl->fl_owner == lease->fl_owner) { | |
1657 | my_fl = fl; | |
1658 | continue; | |
1659 | } | |
1660 | ||
1661 | /* | |
1662 | * No exclusive leases if someone else has a lease on | |
1663 | * this file: | |
1664 | */ | |
1665 | if (arg == F_WRLCK) | |
1666 | goto out; | |
1667 | /* | |
1668 | * Modifying our existing lease is OK, but no getting a | |
1669 | * new lease if someone else is opening for write: | |
1670 | */ | |
1671 | if (fl->fl_flags & FL_UNLOCK_PENDING) | |
1672 | goto out; | |
1673 | } | |
1674 | ||
1675 | if (my_fl != NULL) { | |
1676 | lease = my_fl; | |
1677 | error = lease->fl_lmops->lm_change(lease, arg, &dispose); | |
1678 | if (error) | |
1679 | goto out; | |
1680 | goto out_setup; | |
1681 | } | |
1682 | ||
1683 | error = -EINVAL; | |
1684 | if (!leases_enable) | |
1685 | goto out; | |
1686 | ||
1687 | locks_insert_lock_ctx(lease, &ctx->flc_lease); | |
1688 | /* | |
1689 | * The check in break_lease() is lockless. It's possible for another | |
1690 | * open to race in after we did the earlier check for a conflicting | |
1691 | * open but before the lease was inserted. Check again for a | |
1692 | * conflicting open and cancel the lease if there is one. | |
1693 | * | |
1694 | * We also add a barrier here to ensure that the insertion of the lock | |
1695 | * precedes these checks. | |
1696 | */ | |
1697 | smp_mb(); | |
1698 | error = check_conflicting_open(dentry, arg, lease->fl_flags); | |
1699 | if (error) { | |
1700 | locks_unlink_lock_ctx(lease); | |
1701 | goto out; | |
1702 | } | |
1703 | ||
1704 | out_setup: | |
1705 | if (lease->fl_lmops->lm_setup) | |
1706 | lease->fl_lmops->lm_setup(lease, priv); | |
1707 | out: | |
1708 | spin_unlock(&ctx->flc_lock); | |
1709 | locks_dispose_list(&dispose); | |
1710 | if (is_deleg) | |
1711 | mutex_unlock(&inode->i_mutex); | |
1712 | if (!error && !my_fl) | |
1713 | *flp = NULL; | |
1714 | return error; | |
1715 | } | |
1716 | ||
1717 | static int generic_delete_lease(struct file *filp, void *owner) | |
1718 | { | |
1719 | int error = -EAGAIN; | |
1720 | struct file_lock *fl, *victim = NULL; | |
1721 | struct inode *inode = file_inode(filp); | |
1722 | struct file_lock_context *ctx; | |
1723 | LIST_HEAD(dispose); | |
1724 | ||
1725 | ctx = smp_load_acquire(&inode->i_flctx); | |
1726 | if (!ctx) { | |
1727 | trace_generic_delete_lease(inode, NULL); | |
1728 | return error; | |
1729 | } | |
1730 | ||
1731 | spin_lock(&ctx->flc_lock); | |
1732 | list_for_each_entry(fl, &ctx->flc_lease, fl_list) { | |
1733 | if (fl->fl_file == filp && | |
1734 | fl->fl_owner == owner) { | |
1735 | victim = fl; | |
1736 | break; | |
1737 | } | |
1738 | } | |
1739 | trace_generic_delete_lease(inode, victim); | |
1740 | if (victim) | |
1741 | error = fl->fl_lmops->lm_change(victim, F_UNLCK, &dispose); | |
1742 | spin_unlock(&ctx->flc_lock); | |
1743 | locks_dispose_list(&dispose); | |
1744 | return error; | |
1745 | } | |
1746 | ||
1747 | /** | |
1748 | * generic_setlease - sets a lease on an open file | |
1749 | * @filp: file pointer | |
1750 | * @arg: type of lease to obtain | |
1751 | * @flp: input - file_lock to use, output - file_lock inserted | |
1752 | * @priv: private data for lm_setup (may be NULL if lm_setup | |
1753 | * doesn't require it) | |
1754 | * | |
1755 | * The (input) flp->fl_lmops->lm_break function is required | |
1756 | * by break_lease(). | |
1757 | */ | |
1758 | int generic_setlease(struct file *filp, long arg, struct file_lock **flp, | |
1759 | void **priv) | |
1760 | { | |
1761 | struct inode *inode = file_inode(filp); | |
1762 | int error; | |
1763 | ||
1764 | if ((!uid_eq(current_fsuid(), inode->i_uid)) && !capable(CAP_LEASE)) | |
1765 | return -EACCES; | |
1766 | if (!S_ISREG(inode->i_mode)) | |
1767 | return -EINVAL; | |
1768 | error = security_file_lock(filp, arg); | |
1769 | if (error) | |
1770 | return error; | |
1771 | ||
1772 | switch (arg) { | |
1773 | case F_UNLCK: | |
1774 | return generic_delete_lease(filp, *priv); | |
1775 | case F_RDLCK: | |
1776 | case F_WRLCK: | |
1777 | if (!(*flp)->fl_lmops->lm_break) { | |
1778 | WARN_ON_ONCE(1); | |
1779 | return -ENOLCK; | |
1780 | } | |
1781 | ||
1782 | return generic_add_lease(filp, arg, flp, priv); | |
1783 | default: | |
1784 | return -EINVAL; | |
1785 | } | |
1786 | } | |
1787 | EXPORT_SYMBOL(generic_setlease); | |
1788 | ||
1789 | /** | |
1790 | * vfs_setlease - sets a lease on an open file | |
1791 | * @filp: file pointer | |
1792 | * @arg: type of lease to obtain | |
1793 | * @lease: file_lock to use when adding a lease | |
1794 | * @priv: private info for lm_setup when adding a lease (may be | |
1795 | * NULL if lm_setup doesn't require it) | |
1796 | * | |
1797 | * Call this to establish a lease on the file. The "lease" argument is not | |
1798 | * used for F_UNLCK requests and may be NULL. For commands that set or alter | |
1799 | * an existing lease, the (*lease)->fl_lmops->lm_break operation must be set; | |
1800 | * if not, this function will return -ENOLCK (and generate a scary-looking | |
1801 | * stack trace). | |
1802 | * | |
1803 | * The "priv" pointer is passed directly to the lm_setup function as-is. It | |
1804 | * may be NULL if the lm_setup operation doesn't require it. | |
1805 | */ | |
1806 | int | |
1807 | vfs_setlease(struct file *filp, long arg, struct file_lock **lease, void **priv) | |
1808 | { | |
1809 | if (filp->f_op->setlease) | |
1810 | return filp->f_op->setlease(filp, arg, lease, priv); | |
1811 | else | |
1812 | return generic_setlease(filp, arg, lease, priv); | |
1813 | } | |
1814 | EXPORT_SYMBOL_GPL(vfs_setlease); | |
1815 | ||
1816 | static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg) | |
1817 | { | |
1818 | struct file_lock *fl; | |
1819 | struct fasync_struct *new; | |
1820 | int error; | |
1821 | ||
1822 | fl = lease_alloc(filp, arg); | |
1823 | if (IS_ERR(fl)) | |
1824 | return PTR_ERR(fl); | |
1825 | ||
1826 | new = fasync_alloc(); | |
1827 | if (!new) { | |
1828 | locks_free_lock(fl); | |
1829 | return -ENOMEM; | |
1830 | } | |
1831 | new->fa_fd = fd; | |
1832 | ||
1833 | error = vfs_setlease(filp, arg, &fl, (void **)&new); | |
1834 | if (fl) | |
1835 | locks_free_lock(fl); | |
1836 | if (new) | |
1837 | fasync_free(new); | |
1838 | return error; | |
1839 | } | |
1840 | ||
1841 | /** | |
1842 | * fcntl_setlease - sets a lease on an open file | |
1843 | * @fd: open file descriptor | |
1844 | * @filp: file pointer | |
1845 | * @arg: type of lease to obtain | |
1846 | * | |
1847 | * Call this fcntl to establish a lease on the file. | |
1848 | * Note that you also need to call %F_SETSIG to | |
1849 | * receive a signal when the lease is broken. | |
1850 | */ | |
1851 | int fcntl_setlease(unsigned int fd, struct file *filp, long arg) | |
1852 | { | |
1853 | if (arg == F_UNLCK) | |
1854 | return vfs_setlease(filp, F_UNLCK, NULL, (void **)&filp); | |
1855 | return do_fcntl_add_lease(fd, filp, arg); | |
1856 | } | |
1857 | ||
1858 | /** | |
1859 | * flock_lock_inode_wait - Apply a FLOCK-style lock to a file | |
1860 | * @inode: inode of the file to apply to | |
1861 | * @fl: The lock to be applied | |
1862 | * | |
1863 | * Apply a FLOCK style lock request to an inode. | |
1864 | */ | |
1865 | static int flock_lock_inode_wait(struct inode *inode, struct file_lock *fl) | |
1866 | { | |
1867 | int error; | |
1868 | might_sleep(); | |
1869 | for (;;) { | |
1870 | error = flock_lock_inode(inode, fl); | |
1871 | if (error != FILE_LOCK_DEFERRED) | |
1872 | break; | |
1873 | error = wait_event_interruptible(fl->fl_wait, !fl->fl_next); | |
1874 | if (!error) | |
1875 | continue; | |
1876 | ||
1877 | locks_delete_block(fl); | |
1878 | break; | |
1879 | } | |
1880 | return error; | |
1881 | } | |
1882 | ||
1883 | /** | |
1884 | * locks_lock_inode_wait - Apply a lock to an inode | |
1885 | * @inode: inode of the file to apply to | |
1886 | * @fl: The lock to be applied | |
1887 | * | |
1888 | * Apply a POSIX or FLOCK style lock request to an inode. | |
1889 | */ | |
1890 | int locks_lock_inode_wait(struct inode *inode, struct file_lock *fl) | |
1891 | { | |
1892 | int res = 0; | |
1893 | switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) { | |
1894 | case FL_POSIX: | |
1895 | res = posix_lock_inode_wait(inode, fl); | |
1896 | break; | |
1897 | case FL_FLOCK: | |
1898 | res = flock_lock_inode_wait(inode, fl); | |
1899 | break; | |
1900 | default: | |
1901 | BUG(); | |
1902 | } | |
1903 | return res; | |
1904 | } | |
1905 | EXPORT_SYMBOL(locks_lock_inode_wait); | |
1906 | ||
1907 | /** | |
1908 | * sys_flock: - flock() system call. | |
1909 | * @fd: the file descriptor to lock. | |
1910 | * @cmd: the type of lock to apply. | |
1911 | * | |
1912 | * Apply a %FL_FLOCK style lock to an open file descriptor. | |
1913 | * The @cmd can be one of | |
1914 | * | |
1915 | * %LOCK_SH -- a shared lock. | |
1916 | * | |
1917 | * %LOCK_EX -- an exclusive lock. | |
1918 | * | |
1919 | * %LOCK_UN -- remove an existing lock. | |
1920 | * | |
1921 | * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes. | |
1922 | * | |
1923 | * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other | |
1924 | * processes read and write access respectively. | |
1925 | */ | |
1926 | SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd) | |
1927 | { | |
1928 | struct fd f = fdget(fd); | |
1929 | struct file_lock *lock; | |
1930 | int can_sleep, unlock; | |
1931 | int error; | |
1932 | ||
1933 | error = -EBADF; | |
1934 | if (!f.file) | |
1935 | goto out; | |
1936 | ||
1937 | can_sleep = !(cmd & LOCK_NB); | |
1938 | cmd &= ~LOCK_NB; | |
1939 | unlock = (cmd == LOCK_UN); | |
1940 | ||
1941 | if (!unlock && !(cmd & LOCK_MAND) && | |
1942 | !(f.file->f_mode & (FMODE_READ|FMODE_WRITE))) | |
1943 | goto out_putf; | |
1944 | ||
1945 | lock = flock_make_lock(f.file, cmd); | |
1946 | if (IS_ERR(lock)) { | |
1947 | error = PTR_ERR(lock); | |
1948 | goto out_putf; | |
1949 | } | |
1950 | ||
1951 | if (can_sleep) | |
1952 | lock->fl_flags |= FL_SLEEP; | |
1953 | ||
1954 | error = security_file_lock(f.file, lock->fl_type); | |
1955 | if (error) | |
1956 | goto out_free; | |
1957 | ||
1958 | if (f.file->f_op->flock) | |
1959 | error = f.file->f_op->flock(f.file, | |
1960 | (can_sleep) ? F_SETLKW : F_SETLK, | |
1961 | lock); | |
1962 | else | |
1963 | error = locks_lock_file_wait(f.file, lock); | |
1964 | ||
1965 | out_free: | |
1966 | locks_free_lock(lock); | |
1967 | ||
1968 | out_putf: | |
1969 | fdput(f); | |
1970 | out: | |
1971 | return error; | |
1972 | } | |
1973 | ||
1974 | /** | |
1975 | * vfs_test_lock - test file byte range lock | |
1976 | * @filp: The file to test lock for | |
1977 | * @fl: The lock to test; also used to hold result | |
1978 | * | |
1979 | * Returns -ERRNO on failure. Indicates presence of conflicting lock by | |
1980 | * setting conf->fl_type to something other than F_UNLCK. | |
1981 | */ | |
1982 | int vfs_test_lock(struct file *filp, struct file_lock *fl) | |
1983 | { | |
1984 | if (filp->f_op->lock) | |
1985 | return filp->f_op->lock(filp, F_GETLK, fl); | |
1986 | posix_test_lock(filp, fl); | |
1987 | return 0; | |
1988 | } | |
1989 | EXPORT_SYMBOL_GPL(vfs_test_lock); | |
1990 | ||
1991 | static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl) | |
1992 | { | |
1993 | flock->l_pid = IS_OFDLCK(fl) ? -1 : fl->fl_pid; | |
1994 | #if BITS_PER_LONG == 32 | |
1995 | /* | |
1996 | * Make sure we can represent the posix lock via | |
1997 | * legacy 32bit flock. | |
1998 | */ | |
1999 | if (fl->fl_start > OFFT_OFFSET_MAX) | |
2000 | return -EOVERFLOW; | |
2001 | if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX) | |
2002 | return -EOVERFLOW; | |
2003 | #endif | |
2004 | flock->l_start = fl->fl_start; | |
2005 | flock->l_len = fl->fl_end == OFFSET_MAX ? 0 : | |
2006 | fl->fl_end - fl->fl_start + 1; | |
2007 | flock->l_whence = 0; | |
2008 | flock->l_type = fl->fl_type; | |
2009 | return 0; | |
2010 | } | |
2011 | ||
2012 | #if BITS_PER_LONG == 32 | |
2013 | static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl) | |
2014 | { | |
2015 | flock->l_pid = IS_OFDLCK(fl) ? -1 : fl->fl_pid; | |
2016 | flock->l_start = fl->fl_start; | |
2017 | flock->l_len = fl->fl_end == OFFSET_MAX ? 0 : | |
2018 | fl->fl_end - fl->fl_start + 1; | |
2019 | flock->l_whence = 0; | |
2020 | flock->l_type = fl->fl_type; | |
2021 | } | |
2022 | #endif | |
2023 | ||
2024 | /* Report the first existing lock that would conflict with l. | |
2025 | * This implements the F_GETLK command of fcntl(). | |
2026 | */ | |
2027 | int fcntl_getlk(struct file *filp, unsigned int cmd, struct flock __user *l) | |
2028 | { | |
2029 | struct file_lock file_lock; | |
2030 | struct flock flock; | |
2031 | int error; | |
2032 | ||
2033 | error = -EFAULT; | |
2034 | if (copy_from_user(&flock, l, sizeof(flock))) | |
2035 | goto out; | |
2036 | error = -EINVAL; | |
2037 | if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK)) | |
2038 | goto out; | |
2039 | ||
2040 | error = flock_to_posix_lock(filp, &file_lock, &flock); | |
2041 | if (error) | |
2042 | goto out; | |
2043 | ||
2044 | if (cmd == F_OFD_GETLK) { | |
2045 | error = -EINVAL; | |
2046 | if (flock.l_pid != 0) | |
2047 | goto out; | |
2048 | ||
2049 | cmd = F_GETLK; | |
2050 | file_lock.fl_flags |= FL_OFDLCK; | |
2051 | file_lock.fl_owner = filp; | |
2052 | } | |
2053 | ||
2054 | error = vfs_test_lock(filp, &file_lock); | |
2055 | if (error) | |
2056 | goto out; | |
2057 | ||
2058 | flock.l_type = file_lock.fl_type; | |
2059 | if (file_lock.fl_type != F_UNLCK) { | |
2060 | error = posix_lock_to_flock(&flock, &file_lock); | |
2061 | if (error) | |
2062 | goto rel_priv; | |
2063 | } | |
2064 | error = -EFAULT; | |
2065 | if (!copy_to_user(l, &flock, sizeof(flock))) | |
2066 | error = 0; | |
2067 | rel_priv: | |
2068 | locks_release_private(&file_lock); | |
2069 | out: | |
2070 | return error; | |
2071 | } | |
2072 | ||
2073 | /** | |
2074 | * vfs_lock_file - file byte range lock | |
2075 | * @filp: The file to apply the lock to | |
2076 | * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.) | |
2077 | * @fl: The lock to be applied | |
2078 | * @conf: Place to return a copy of the conflicting lock, if found. | |
2079 | * | |
2080 | * A caller that doesn't care about the conflicting lock may pass NULL | |
2081 | * as the final argument. | |
2082 | * | |
2083 | * If the filesystem defines a private ->lock() method, then @conf will | |
2084 | * be left unchanged; so a caller that cares should initialize it to | |
2085 | * some acceptable default. | |
2086 | * | |
2087 | * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX | |
2088 | * locks, the ->lock() interface may return asynchronously, before the lock has | |
2089 | * been granted or denied by the underlying filesystem, if (and only if) | |
2090 | * lm_grant is set. Callers expecting ->lock() to return asynchronously | |
2091 | * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if) | |
2092 | * the request is for a blocking lock. When ->lock() does return asynchronously, | |
2093 | * it must return FILE_LOCK_DEFERRED, and call ->lm_grant() when the lock | |
2094 | * request completes. | |
2095 | * If the request is for non-blocking lock the file system should return | |
2096 | * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine | |
2097 | * with the result. If the request timed out the callback routine will return a | |
2098 | * nonzero return code and the file system should release the lock. The file | |
2099 | * system is also responsible to keep a corresponding posix lock when it | |
2100 | * grants a lock so the VFS can find out which locks are locally held and do | |
2101 | * the correct lock cleanup when required. | |
2102 | * The underlying filesystem must not drop the kernel lock or call | |
2103 | * ->lm_grant() before returning to the caller with a FILE_LOCK_DEFERRED | |
2104 | * return code. | |
2105 | */ | |
2106 | int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf) | |
2107 | { | |
2108 | if (filp->f_op->lock) | |
2109 | return filp->f_op->lock(filp, cmd, fl); | |
2110 | else | |
2111 | return posix_lock_file(filp, fl, conf); | |
2112 | } | |
2113 | EXPORT_SYMBOL_GPL(vfs_lock_file); | |
2114 | ||
2115 | static int do_lock_file_wait(struct file *filp, unsigned int cmd, | |
2116 | struct file_lock *fl) | |
2117 | { | |
2118 | int error; | |
2119 | ||
2120 | error = security_file_lock(filp, fl->fl_type); | |
2121 | if (error) | |
2122 | return error; | |
2123 | ||
2124 | for (;;) { | |
2125 | error = vfs_lock_file(filp, cmd, fl, NULL); | |
2126 | if (error != FILE_LOCK_DEFERRED) | |
2127 | break; | |
2128 | error = wait_event_interruptible(fl->fl_wait, !fl->fl_next); | |
2129 | if (!error) | |
2130 | continue; | |
2131 | ||
2132 | locks_delete_block(fl); | |
2133 | break; | |
2134 | } | |
2135 | ||
2136 | return error; | |
2137 | } | |
2138 | ||
2139 | /* Ensure that fl->fl_file has compatible f_mode for F_SETLK calls */ | |
2140 | static int | |
2141 | check_fmode_for_setlk(struct file_lock *fl) | |
2142 | { | |
2143 | switch (fl->fl_type) { | |
2144 | case F_RDLCK: | |
2145 | if (!(fl->fl_file->f_mode & FMODE_READ)) | |
2146 | return -EBADF; | |
2147 | break; | |
2148 | case F_WRLCK: | |
2149 | if (!(fl->fl_file->f_mode & FMODE_WRITE)) | |
2150 | return -EBADF; | |
2151 | } | |
2152 | return 0; | |
2153 | } | |
2154 | ||
2155 | /* Apply the lock described by l to an open file descriptor. | |
2156 | * This implements both the F_SETLK and F_SETLKW commands of fcntl(). | |
2157 | */ | |
2158 | int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd, | |
2159 | struct flock __user *l) | |
2160 | { | |
2161 | struct file_lock *file_lock = locks_alloc_lock(); | |
2162 | struct flock flock; | |
2163 | struct inode *inode; | |
2164 | struct file *f; | |
2165 | int error; | |
2166 | ||
2167 | if (file_lock == NULL) | |
2168 | return -ENOLCK; | |
2169 | ||
2170 | inode = file_inode(filp); | |
2171 | ||
2172 | /* | |
2173 | * This might block, so we do it before checking the inode. | |
2174 | */ | |
2175 | error = -EFAULT; | |
2176 | if (copy_from_user(&flock, l, sizeof(flock))) | |
2177 | goto out; | |
2178 | ||
2179 | /* Don't allow mandatory locks on files that may be memory mapped | |
2180 | * and shared. | |
2181 | */ | |
2182 | if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) { | |
2183 | error = -EAGAIN; | |
2184 | goto out; | |
2185 | } | |
2186 | ||
2187 | error = flock_to_posix_lock(filp, file_lock, &flock); | |
2188 | if (error) | |
2189 | goto out; | |
2190 | ||
2191 | error = check_fmode_for_setlk(file_lock); | |
2192 | if (error) | |
2193 | goto out; | |
2194 | ||
2195 | /* | |
2196 | * If the cmd is requesting file-private locks, then set the | |
2197 | * FL_OFDLCK flag and override the owner. | |
2198 | */ | |
2199 | switch (cmd) { | |
2200 | case F_OFD_SETLK: | |
2201 | error = -EINVAL; | |
2202 | if (flock.l_pid != 0) | |
2203 | goto out; | |
2204 | ||
2205 | cmd = F_SETLK; | |
2206 | file_lock->fl_flags |= FL_OFDLCK; | |
2207 | file_lock->fl_owner = filp; | |
2208 | break; | |
2209 | case F_OFD_SETLKW: | |
2210 | error = -EINVAL; | |
2211 | if (flock.l_pid != 0) | |
2212 | goto out; | |
2213 | ||
2214 | cmd = F_SETLKW; | |
2215 | file_lock->fl_flags |= FL_OFDLCK; | |
2216 | file_lock->fl_owner = filp; | |
2217 | /* Fallthrough */ | |
2218 | case F_SETLKW: | |
2219 | file_lock->fl_flags |= FL_SLEEP; | |
2220 | } | |
2221 | ||
2222 | error = do_lock_file_wait(filp, cmd, file_lock); | |
2223 | ||
2224 | /* | |
2225 | * Attempt to detect a close/fcntl race and recover by releasing the | |
2226 | * lock that was just acquired. There is no need to do that when we're | |
2227 | * unlocking though, or for OFD locks. | |
2228 | */ | |
2229 | if (!error && file_lock->fl_type != F_UNLCK && | |
2230 | !(file_lock->fl_flags & FL_OFDLCK)) { | |
2231 | /* | |
2232 | * We need that spin_lock here - it prevents reordering between | |
2233 | * update of i_flctx->flc_posix and check for it done in | |
2234 | * close(). rcu_read_lock() wouldn't do. | |
2235 | */ | |
2236 | spin_lock(¤t->files->file_lock); | |
2237 | f = fcheck(fd); | |
2238 | spin_unlock(¤t->files->file_lock); | |
2239 | if (f != filp) { | |
2240 | file_lock->fl_type = F_UNLCK; | |
2241 | error = do_lock_file_wait(filp, cmd, file_lock); | |
2242 | WARN_ON_ONCE(error); | |
2243 | error = -EBADF; | |
2244 | } | |
2245 | } | |
2246 | out: | |
2247 | trace_fcntl_setlk(inode, file_lock, error); | |
2248 | locks_free_lock(file_lock); | |
2249 | return error; | |
2250 | } | |
2251 | ||
2252 | #if BITS_PER_LONG == 32 | |
2253 | /* Report the first existing lock that would conflict with l. | |
2254 | * This implements the F_GETLK command of fcntl(). | |
2255 | */ | |
2256 | int fcntl_getlk64(struct file *filp, unsigned int cmd, struct flock64 __user *l) | |
2257 | { | |
2258 | struct file_lock file_lock; | |
2259 | struct flock64 flock; | |
2260 | int error; | |
2261 | ||
2262 | error = -EFAULT; | |
2263 | if (copy_from_user(&flock, l, sizeof(flock))) | |
2264 | goto out; | |
2265 | error = -EINVAL; | |
2266 | if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK)) | |
2267 | goto out; | |
2268 | ||
2269 | error = flock64_to_posix_lock(filp, &file_lock, &flock); | |
2270 | if (error) | |
2271 | goto out; | |
2272 | ||
2273 | if (cmd == F_OFD_GETLK) { | |
2274 | error = -EINVAL; | |
2275 | if (flock.l_pid != 0) | |
2276 | goto out; | |
2277 | ||
2278 | cmd = F_GETLK64; | |
2279 | file_lock.fl_flags |= FL_OFDLCK; | |
2280 | file_lock.fl_owner = filp; | |
2281 | } | |
2282 | ||
2283 | error = vfs_test_lock(filp, &file_lock); | |
2284 | if (error) | |
2285 | goto out; | |
2286 | ||
2287 | flock.l_type = file_lock.fl_type; | |
2288 | if (file_lock.fl_type != F_UNLCK) | |
2289 | posix_lock_to_flock64(&flock, &file_lock); | |
2290 | ||
2291 | error = -EFAULT; | |
2292 | if (!copy_to_user(l, &flock, sizeof(flock))) | |
2293 | error = 0; | |
2294 | ||
2295 | locks_release_private(&file_lock); | |
2296 | out: | |
2297 | return error; | |
2298 | } | |
2299 | ||
2300 | /* Apply the lock described by l to an open file descriptor. | |
2301 | * This implements both the F_SETLK and F_SETLKW commands of fcntl(). | |
2302 | */ | |
2303 | int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd, | |
2304 | struct flock64 __user *l) | |
2305 | { | |
2306 | struct file_lock *file_lock = locks_alloc_lock(); | |
2307 | struct flock64 flock; | |
2308 | struct inode *inode; | |
2309 | struct file *f; | |
2310 | int error; | |
2311 | ||
2312 | if (file_lock == NULL) | |
2313 | return -ENOLCK; | |
2314 | ||
2315 | /* | |
2316 | * This might block, so we do it before checking the inode. | |
2317 | */ | |
2318 | error = -EFAULT; | |
2319 | if (copy_from_user(&flock, l, sizeof(flock))) | |
2320 | goto out; | |
2321 | ||
2322 | inode = file_inode(filp); | |
2323 | ||
2324 | /* Don't allow mandatory locks on files that may be memory mapped | |
2325 | * and shared. | |
2326 | */ | |
2327 | if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) { | |
2328 | error = -EAGAIN; | |
2329 | goto out; | |
2330 | } | |
2331 | ||
2332 | error = flock64_to_posix_lock(filp, file_lock, &flock); | |
2333 | if (error) | |
2334 | goto out; | |
2335 | ||
2336 | error = check_fmode_for_setlk(file_lock); | |
2337 | if (error) | |
2338 | goto out; | |
2339 | ||
2340 | /* | |
2341 | * If the cmd is requesting file-private locks, then set the | |
2342 | * FL_OFDLCK flag and override the owner. | |
2343 | */ | |
2344 | switch (cmd) { | |
2345 | case F_OFD_SETLK: | |
2346 | error = -EINVAL; | |
2347 | if (flock.l_pid != 0) | |
2348 | goto out; | |
2349 | ||
2350 | cmd = F_SETLK64; | |
2351 | file_lock->fl_flags |= FL_OFDLCK; | |
2352 | file_lock->fl_owner = filp; | |
2353 | break; | |
2354 | case F_OFD_SETLKW: | |
2355 | error = -EINVAL; | |
2356 | if (flock.l_pid != 0) | |
2357 | goto out; | |
2358 | ||
2359 | cmd = F_SETLKW64; | |
2360 | file_lock->fl_flags |= FL_OFDLCK; | |
2361 | file_lock->fl_owner = filp; | |
2362 | /* Fallthrough */ | |
2363 | case F_SETLKW64: | |
2364 | file_lock->fl_flags |= FL_SLEEP; | |
2365 | } | |
2366 | ||
2367 | error = do_lock_file_wait(filp, cmd, file_lock); | |
2368 | ||
2369 | /* | |
2370 | * Attempt to detect a close/fcntl race and recover by releasing the | |
2371 | * lock that was just acquired. There is no need to do that when we're | |
2372 | * unlocking though, or for OFD locks. | |
2373 | */ | |
2374 | if (!error && file_lock->fl_type != F_UNLCK && | |
2375 | !(file_lock->fl_flags & FL_OFDLCK)) { | |
2376 | /* | |
2377 | * We need that spin_lock here - it prevents reordering between | |
2378 | * update of i_flctx->flc_posix and check for it done in | |
2379 | * close(). rcu_read_lock() wouldn't do. | |
2380 | */ | |
2381 | spin_lock(¤t->files->file_lock); | |
2382 | f = fcheck(fd); | |
2383 | spin_unlock(¤t->files->file_lock); | |
2384 | if (f != filp) { | |
2385 | file_lock->fl_type = F_UNLCK; | |
2386 | error = do_lock_file_wait(filp, cmd, file_lock); | |
2387 | WARN_ON_ONCE(error); | |
2388 | error = -EBADF; | |
2389 | } | |
2390 | } | |
2391 | out: | |
2392 | locks_free_lock(file_lock); | |
2393 | return error; | |
2394 | } | |
2395 | #endif /* BITS_PER_LONG == 32 */ | |
2396 | ||
2397 | /* | |
2398 | * This function is called when the file is being removed | |
2399 | * from the task's fd array. POSIX locks belonging to this task | |
2400 | * are deleted at this time. | |
2401 | */ | |
2402 | void locks_remove_posix(struct file *filp, fl_owner_t owner) | |
2403 | { | |
2404 | int error; | |
2405 | struct file_lock lock; | |
2406 | struct file_lock_context *ctx; | |
2407 | ||
2408 | /* | |
2409 | * If there are no locks held on this file, we don't need to call | |
2410 | * posix_lock_file(). Another process could be setting a lock on this | |
2411 | * file at the same time, but we wouldn't remove that lock anyway. | |
2412 | */ | |
2413 | ctx = smp_load_acquire(&file_inode(filp)->i_flctx); | |
2414 | if (!ctx || list_empty(&ctx->flc_posix)) | |
2415 | return; | |
2416 | ||
2417 | lock.fl_type = F_UNLCK; | |
2418 | lock.fl_flags = FL_POSIX | FL_CLOSE; | |
2419 | lock.fl_start = 0; | |
2420 | lock.fl_end = OFFSET_MAX; | |
2421 | lock.fl_owner = owner; | |
2422 | lock.fl_pid = current->tgid; | |
2423 | lock.fl_file = filp; | |
2424 | lock.fl_ops = NULL; | |
2425 | lock.fl_lmops = NULL; | |
2426 | ||
2427 | error = vfs_lock_file(filp, F_SETLK, &lock, NULL); | |
2428 | ||
2429 | if (lock.fl_ops && lock.fl_ops->fl_release_private) | |
2430 | lock.fl_ops->fl_release_private(&lock); | |
2431 | trace_locks_remove_posix(file_inode(filp), &lock, error); | |
2432 | } | |
2433 | ||
2434 | EXPORT_SYMBOL(locks_remove_posix); | |
2435 | ||
2436 | /* The i_flctx must be valid when calling into here */ | |
2437 | static void | |
2438 | locks_remove_flock(struct file *filp, struct file_lock_context *flctx) | |
2439 | { | |
2440 | struct file_lock fl = { | |
2441 | .fl_owner = filp, | |
2442 | .fl_pid = current->tgid, | |
2443 | .fl_file = filp, | |
2444 | .fl_flags = FL_FLOCK, | |
2445 | .fl_type = F_UNLCK, | |
2446 | .fl_end = OFFSET_MAX, | |
2447 | }; | |
2448 | struct inode *inode = file_inode(filp); | |
2449 | ||
2450 | if (list_empty(&flctx->flc_flock)) | |
2451 | return; | |
2452 | ||
2453 | if (filp->f_op->flock) | |
2454 | filp->f_op->flock(filp, F_SETLKW, &fl); | |
2455 | else | |
2456 | flock_lock_inode(inode, &fl); | |
2457 | ||
2458 | if (fl.fl_ops && fl.fl_ops->fl_release_private) | |
2459 | fl.fl_ops->fl_release_private(&fl); | |
2460 | } | |
2461 | ||
2462 | /* The i_flctx must be valid when calling into here */ | |
2463 | static void | |
2464 | locks_remove_lease(struct file *filp, struct file_lock_context *ctx) | |
2465 | { | |
2466 | struct file_lock *fl, *tmp; | |
2467 | LIST_HEAD(dispose); | |
2468 | ||
2469 | if (list_empty(&ctx->flc_lease)) | |
2470 | return; | |
2471 | ||
2472 | spin_lock(&ctx->flc_lock); | |
2473 | list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list) | |
2474 | if (filp == fl->fl_file) | |
2475 | lease_modify(fl, F_UNLCK, &dispose); | |
2476 | spin_unlock(&ctx->flc_lock); | |
2477 | locks_dispose_list(&dispose); | |
2478 | } | |
2479 | ||
2480 | /* | |
2481 | * This function is called on the last close of an open file. | |
2482 | */ | |
2483 | void locks_remove_file(struct file *filp) | |
2484 | { | |
2485 | struct file_lock_context *ctx; | |
2486 | ||
2487 | ctx = smp_load_acquire(&file_inode(filp)->i_flctx); | |
2488 | if (!ctx) | |
2489 | return; | |
2490 | ||
2491 | /* remove any OFD locks */ | |
2492 | locks_remove_posix(filp, filp); | |
2493 | ||
2494 | /* remove flock locks */ | |
2495 | locks_remove_flock(filp, ctx); | |
2496 | ||
2497 | /* remove any leases */ | |
2498 | locks_remove_lease(filp, ctx); | |
2499 | } | |
2500 | ||
2501 | /** | |
2502 | * posix_unblock_lock - stop waiting for a file lock | |
2503 | * @waiter: the lock which was waiting | |
2504 | * | |
2505 | * lockd needs to block waiting for locks. | |
2506 | */ | |
2507 | int | |
2508 | posix_unblock_lock(struct file_lock *waiter) | |
2509 | { | |
2510 | int status = 0; | |
2511 | ||
2512 | spin_lock(&blocked_lock_lock); | |
2513 | if (waiter->fl_next) | |
2514 | __locks_delete_block(waiter); | |
2515 | else | |
2516 | status = -ENOENT; | |
2517 | spin_unlock(&blocked_lock_lock); | |
2518 | return status; | |
2519 | } | |
2520 | EXPORT_SYMBOL(posix_unblock_lock); | |
2521 | ||
2522 | /** | |
2523 | * vfs_cancel_lock - file byte range unblock lock | |
2524 | * @filp: The file to apply the unblock to | |
2525 | * @fl: The lock to be unblocked | |
2526 | * | |
2527 | * Used by lock managers to cancel blocked requests | |
2528 | */ | |
2529 | int vfs_cancel_lock(struct file *filp, struct file_lock *fl) | |
2530 | { | |
2531 | if (filp->f_op->lock) | |
2532 | return filp->f_op->lock(filp, F_CANCELLK, fl); | |
2533 | return 0; | |
2534 | } | |
2535 | ||
2536 | EXPORT_SYMBOL_GPL(vfs_cancel_lock); | |
2537 | ||
2538 | #ifdef CONFIG_PROC_FS | |
2539 | #include <linux/proc_fs.h> | |
2540 | #include <linux/seq_file.h> | |
2541 | ||
2542 | struct locks_iterator { | |
2543 | int li_cpu; | |
2544 | loff_t li_pos; | |
2545 | }; | |
2546 | ||
2547 | static void lock_get_status(struct seq_file *f, struct file_lock *fl, | |
2548 | loff_t id, char *pfx) | |
2549 | { | |
2550 | struct inode *inode = NULL; | |
2551 | unsigned int fl_pid; | |
2552 | ||
2553 | if (fl->fl_nspid) | |
2554 | fl_pid = pid_vnr(fl->fl_nspid); | |
2555 | else | |
2556 | fl_pid = fl->fl_pid; | |
2557 | ||
2558 | if (fl->fl_file != NULL) | |
2559 | inode = file_inode(fl->fl_file); | |
2560 | ||
2561 | seq_printf(f, "%lld:%s ", id, pfx); | |
2562 | if (IS_POSIX(fl)) { | |
2563 | if (fl->fl_flags & FL_ACCESS) | |
2564 | seq_puts(f, "ACCESS"); | |
2565 | else if (IS_OFDLCK(fl)) | |
2566 | seq_puts(f, "OFDLCK"); | |
2567 | else | |
2568 | seq_puts(f, "POSIX "); | |
2569 | ||
2570 | seq_printf(f, " %s ", | |
2571 | (inode == NULL) ? "*NOINODE*" : | |
2572 | mandatory_lock(inode) ? "MANDATORY" : "ADVISORY "); | |
2573 | } else if (IS_FLOCK(fl)) { | |
2574 | if (fl->fl_type & LOCK_MAND) { | |
2575 | seq_puts(f, "FLOCK MSNFS "); | |
2576 | } else { | |
2577 | seq_puts(f, "FLOCK ADVISORY "); | |
2578 | } | |
2579 | } else if (IS_LEASE(fl)) { | |
2580 | if (fl->fl_flags & FL_DELEG) | |
2581 | seq_puts(f, "DELEG "); | |
2582 | else | |
2583 | seq_puts(f, "LEASE "); | |
2584 | ||
2585 | if (lease_breaking(fl)) | |
2586 | seq_puts(f, "BREAKING "); | |
2587 | else if (fl->fl_file) | |
2588 | seq_puts(f, "ACTIVE "); | |
2589 | else | |
2590 | seq_puts(f, "BREAKER "); | |
2591 | } else { | |
2592 | seq_puts(f, "UNKNOWN UNKNOWN "); | |
2593 | } | |
2594 | if (fl->fl_type & LOCK_MAND) { | |
2595 | seq_printf(f, "%s ", | |
2596 | (fl->fl_type & LOCK_READ) | |
2597 | ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ " | |
2598 | : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE "); | |
2599 | } else { | |
2600 | seq_printf(f, "%s ", | |
2601 | (lease_breaking(fl)) | |
2602 | ? (fl->fl_type == F_UNLCK) ? "UNLCK" : "READ " | |
2603 | : (fl->fl_type == F_WRLCK) ? "WRITE" : "READ "); | |
2604 | } | |
2605 | if (inode) { | |
2606 | /* userspace relies on this representation of dev_t */ | |
2607 | seq_printf(f, "%d %02x:%02x:%ld ", fl_pid, | |
2608 | MAJOR(inode->i_sb->s_dev), | |
2609 | MINOR(inode->i_sb->s_dev), inode->i_ino); | |
2610 | } else { | |
2611 | seq_printf(f, "%d <none>:0 ", fl_pid); | |
2612 | } | |
2613 | if (IS_POSIX(fl)) { | |
2614 | if (fl->fl_end == OFFSET_MAX) | |
2615 | seq_printf(f, "%Ld EOF\n", fl->fl_start); | |
2616 | else | |
2617 | seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end); | |
2618 | } else { | |
2619 | seq_puts(f, "0 EOF\n"); | |
2620 | } | |
2621 | } | |
2622 | ||
2623 | static int locks_show(struct seq_file *f, void *v) | |
2624 | { | |
2625 | struct locks_iterator *iter = f->private; | |
2626 | struct file_lock *fl, *bfl; | |
2627 | ||
2628 | fl = hlist_entry(v, struct file_lock, fl_link); | |
2629 | ||
2630 | lock_get_status(f, fl, iter->li_pos, ""); | |
2631 | ||
2632 | list_for_each_entry(bfl, &fl->fl_block, fl_block) | |
2633 | lock_get_status(f, bfl, iter->li_pos, " ->"); | |
2634 | ||
2635 | return 0; | |
2636 | } | |
2637 | ||
2638 | static void __show_fd_locks(struct seq_file *f, | |
2639 | struct list_head *head, int *id, | |
2640 | struct file *filp, struct files_struct *files) | |
2641 | { | |
2642 | struct file_lock *fl; | |
2643 | ||
2644 | list_for_each_entry(fl, head, fl_list) { | |
2645 | ||
2646 | if (filp != fl->fl_file) | |
2647 | continue; | |
2648 | if (fl->fl_owner != files && | |
2649 | fl->fl_owner != filp) | |
2650 | continue; | |
2651 | ||
2652 | (*id)++; | |
2653 | seq_puts(f, "lock:\t"); | |
2654 | lock_get_status(f, fl, *id, ""); | |
2655 | } | |
2656 | } | |
2657 | ||
2658 | void show_fd_locks(struct seq_file *f, | |
2659 | struct file *filp, struct files_struct *files) | |
2660 | { | |
2661 | struct inode *inode = file_inode(filp); | |
2662 | struct file_lock_context *ctx; | |
2663 | int id = 0; | |
2664 | ||
2665 | ctx = smp_load_acquire(&inode->i_flctx); | |
2666 | if (!ctx) | |
2667 | return; | |
2668 | ||
2669 | spin_lock(&ctx->flc_lock); | |
2670 | __show_fd_locks(f, &ctx->flc_flock, &id, filp, files); | |
2671 | __show_fd_locks(f, &ctx->flc_posix, &id, filp, files); | |
2672 | __show_fd_locks(f, &ctx->flc_lease, &id, filp, files); | |
2673 | spin_unlock(&ctx->flc_lock); | |
2674 | } | |
2675 | ||
2676 | static void *locks_start(struct seq_file *f, loff_t *pos) | |
2677 | __acquires(&blocked_lock_lock) | |
2678 | { | |
2679 | struct locks_iterator *iter = f->private; | |
2680 | ||
2681 | iter->li_pos = *pos + 1; | |
2682 | lg_global_lock(&file_lock_lglock); | |
2683 | spin_lock(&blocked_lock_lock); | |
2684 | return seq_hlist_start_percpu(&file_lock_list, &iter->li_cpu, *pos); | |
2685 | } | |
2686 | ||
2687 | static void *locks_next(struct seq_file *f, void *v, loff_t *pos) | |
2688 | { | |
2689 | struct locks_iterator *iter = f->private; | |
2690 | ||
2691 | ++iter->li_pos; | |
2692 | return seq_hlist_next_percpu(v, &file_lock_list, &iter->li_cpu, pos); | |
2693 | } | |
2694 | ||
2695 | static void locks_stop(struct seq_file *f, void *v) | |
2696 | __releases(&blocked_lock_lock) | |
2697 | { | |
2698 | spin_unlock(&blocked_lock_lock); | |
2699 | lg_global_unlock(&file_lock_lglock); | |
2700 | } | |
2701 | ||
2702 | static const struct seq_operations locks_seq_operations = { | |
2703 | .start = locks_start, | |
2704 | .next = locks_next, | |
2705 | .stop = locks_stop, | |
2706 | .show = locks_show, | |
2707 | }; | |
2708 | ||
2709 | static int locks_open(struct inode *inode, struct file *filp) | |
2710 | { | |
2711 | return seq_open_private(filp, &locks_seq_operations, | |
2712 | sizeof(struct locks_iterator)); | |
2713 | } | |
2714 | ||
2715 | static const struct file_operations proc_locks_operations = { | |
2716 | .open = locks_open, | |
2717 | .read = seq_read, | |
2718 | .llseek = seq_lseek, | |
2719 | .release = seq_release_private, | |
2720 | }; | |
2721 | ||
2722 | static int __init proc_locks_init(void) | |
2723 | { | |
2724 | proc_create("locks", 0, NULL, &proc_locks_operations); | |
2725 | return 0; | |
2726 | } | |
2727 | fs_initcall(proc_locks_init); | |
2728 | #endif | |
2729 | ||
2730 | static int __init filelock_init(void) | |
2731 | { | |
2732 | int i; | |
2733 | ||
2734 | flctx_cache = kmem_cache_create("file_lock_ctx", | |
2735 | sizeof(struct file_lock_context), 0, SLAB_PANIC, NULL); | |
2736 | ||
2737 | filelock_cache = kmem_cache_create("file_lock_cache", | |
2738 | sizeof(struct file_lock), 0, SLAB_PANIC, NULL); | |
2739 | ||
2740 | lg_lock_init(&file_lock_lglock, "file_lock_lglock"); | |
2741 | ||
2742 | for_each_possible_cpu(i) | |
2743 | INIT_HLIST_HEAD(per_cpu_ptr(&file_lock_list, i)); | |
2744 | ||
2745 | return 0; | |
2746 | } | |
2747 | ||
2748 | core_initcall(filelock_init); |