]>
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 | ||
131 | #define CREATE_TRACE_POINTS | |
132 | #include <trace/events/filelock.h> | |
133 | ||
134 | #include <linux/uaccess.h> | |
135 | ||
136 | #define IS_POSIX(fl) (fl->fl_flags & FL_POSIX) | |
137 | #define IS_FLOCK(fl) (fl->fl_flags & FL_FLOCK) | |
138 | #define IS_LEASE(fl) (fl->fl_flags & (FL_LEASE|FL_DELEG|FL_LAYOUT)) | |
139 | #define IS_OFDLCK(fl) (fl->fl_flags & FL_OFDLCK) | |
140 | #define IS_REMOTELCK(fl) (fl->fl_pid <= 0) | |
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. | |
162 | * Global serialization is done using file_rwsem. | |
163 | * | |
164 | * Note that alterations to the list also require that the relevant flc_lock is | |
165 | * held. | |
166 | */ | |
167 | struct file_lock_list_struct { | |
168 | spinlock_t lock; | |
169 | struct hlist_head hlist; | |
170 | }; | |
171 | static DEFINE_PER_CPU(struct file_lock_list_struct, file_lock_list); | |
172 | DEFINE_STATIC_PERCPU_RWSEM(file_rwsem); | |
173 | ||
174 | /* | |
175 | * The blocked_hash is used to find POSIX lock loops for deadlock detection. | |
176 | * It is protected by blocked_lock_lock. | |
177 | * | |
178 | * We hash locks by lockowner in order to optimize searching for the lock a | |
179 | * particular lockowner is waiting on. | |
180 | * | |
181 | * FIXME: make this value scale via some heuristic? We generally will want more | |
182 | * buckets when we have more lockowners holding locks, but that's a little | |
183 | * difficult to determine without knowing what the workload will look like. | |
184 | */ | |
185 | #define BLOCKED_HASH_BITS 7 | |
186 | static DEFINE_HASHTABLE(blocked_hash, BLOCKED_HASH_BITS); | |
187 | ||
188 | /* | |
189 | * This lock protects the blocked_hash. Generally, if you're accessing it, you | |
190 | * want to be holding this lock. | |
191 | * | |
192 | * In addition, it also protects the fl->fl_blocked_requests list, and the | |
193 | * fl->fl_blocker pointer for file_lock structures that are acting as lock | |
194 | * requests (in contrast to those that are acting as records of acquired locks). | |
195 | * | |
196 | * Note that when we acquire this lock in order to change the above fields, | |
197 | * we often hold the flc_lock as well. In certain cases, when reading the fields | |
198 | * protected by this lock, we can skip acquiring it iff we already hold the | |
199 | * flc_lock. | |
200 | */ | |
201 | static DEFINE_SPINLOCK(blocked_lock_lock); | |
202 | ||
203 | static struct kmem_cache *flctx_cache __read_mostly; | |
204 | static struct kmem_cache *filelock_cache __read_mostly; | |
205 | ||
206 | static struct file_lock_context * | |
207 | locks_get_lock_context(struct inode *inode, int type) | |
208 | { | |
209 | struct file_lock_context *ctx; | |
210 | ||
211 | /* paired with cmpxchg() below */ | |
212 | ctx = smp_load_acquire(&inode->i_flctx); | |
213 | if (likely(ctx) || type == F_UNLCK) | |
214 | goto out; | |
215 | ||
216 | ctx = kmem_cache_alloc(flctx_cache, GFP_KERNEL); | |
217 | if (!ctx) | |
218 | goto out; | |
219 | ||
220 | spin_lock_init(&ctx->flc_lock); | |
221 | INIT_LIST_HEAD(&ctx->flc_flock); | |
222 | INIT_LIST_HEAD(&ctx->flc_posix); | |
223 | INIT_LIST_HEAD(&ctx->flc_lease); | |
224 | ||
225 | /* | |
226 | * Assign the pointer if it's not already assigned. If it is, then | |
227 | * free the context we just allocated. | |
228 | */ | |
229 | if (cmpxchg(&inode->i_flctx, NULL, ctx)) { | |
230 | kmem_cache_free(flctx_cache, ctx); | |
231 | ctx = smp_load_acquire(&inode->i_flctx); | |
232 | } | |
233 | out: | |
234 | trace_locks_get_lock_context(inode, type, ctx); | |
235 | return ctx; | |
236 | } | |
237 | ||
238 | static void | |
239 | locks_dump_ctx_list(struct list_head *list, char *list_type) | |
240 | { | |
241 | struct file_lock *fl; | |
242 | ||
243 | list_for_each_entry(fl, list, fl_list) { | |
244 | pr_warn("%s: fl_owner=%p fl_flags=0x%x fl_type=0x%x fl_pid=%u\n", list_type, fl->fl_owner, fl->fl_flags, fl->fl_type, fl->fl_pid); | |
245 | } | |
246 | } | |
247 | ||
248 | static void | |
249 | locks_check_ctx_lists(struct inode *inode) | |
250 | { | |
251 | struct file_lock_context *ctx = inode->i_flctx; | |
252 | ||
253 | if (unlikely(!list_empty(&ctx->flc_flock) || | |
254 | !list_empty(&ctx->flc_posix) || | |
255 | !list_empty(&ctx->flc_lease))) { | |
256 | pr_warn("Leaked locks on dev=0x%x:0x%x ino=0x%lx:\n", | |
257 | MAJOR(inode->i_sb->s_dev), MINOR(inode->i_sb->s_dev), | |
258 | inode->i_ino); | |
259 | locks_dump_ctx_list(&ctx->flc_flock, "FLOCK"); | |
260 | locks_dump_ctx_list(&ctx->flc_posix, "POSIX"); | |
261 | locks_dump_ctx_list(&ctx->flc_lease, "LEASE"); | |
262 | } | |
263 | } | |
264 | ||
265 | static void | |
266 | locks_check_ctx_file_list(struct file *filp, struct list_head *list, | |
267 | char *list_type) | |
268 | { | |
269 | struct file_lock *fl; | |
270 | struct inode *inode = locks_inode(filp); | |
271 | ||
272 | list_for_each_entry(fl, list, fl_list) | |
273 | if (fl->fl_file == filp) | |
274 | pr_warn("Leaked %s lock on dev=0x%x:0x%x ino=0x%lx " | |
275 | " fl_owner=%p fl_flags=0x%x fl_type=0x%x fl_pid=%u\n", | |
276 | list_type, MAJOR(inode->i_sb->s_dev), | |
277 | MINOR(inode->i_sb->s_dev), inode->i_ino, | |
278 | fl->fl_owner, fl->fl_flags, fl->fl_type, fl->fl_pid); | |
279 | } | |
280 | ||
281 | void | |
282 | locks_free_lock_context(struct inode *inode) | |
283 | { | |
284 | struct file_lock_context *ctx = inode->i_flctx; | |
285 | ||
286 | if (unlikely(ctx)) { | |
287 | locks_check_ctx_lists(inode); | |
288 | kmem_cache_free(flctx_cache, ctx); | |
289 | } | |
290 | } | |
291 | ||
292 | static void locks_init_lock_heads(struct file_lock *fl) | |
293 | { | |
294 | INIT_HLIST_NODE(&fl->fl_link); | |
295 | INIT_LIST_HEAD(&fl->fl_list); | |
296 | INIT_LIST_HEAD(&fl->fl_blocked_requests); | |
297 | INIT_LIST_HEAD(&fl->fl_blocked_member); | |
298 | init_waitqueue_head(&fl->fl_wait); | |
299 | } | |
300 | ||
301 | /* Allocate an empty lock structure. */ | |
302 | struct file_lock *locks_alloc_lock(void) | |
303 | { | |
304 | struct file_lock *fl = kmem_cache_zalloc(filelock_cache, GFP_KERNEL); | |
305 | ||
306 | if (fl) | |
307 | locks_init_lock_heads(fl); | |
308 | ||
309 | return fl; | |
310 | } | |
311 | EXPORT_SYMBOL_GPL(locks_alloc_lock); | |
312 | ||
313 | void locks_release_private(struct file_lock *fl) | |
314 | { | |
315 | if (fl->fl_ops) { | |
316 | if (fl->fl_ops->fl_release_private) | |
317 | fl->fl_ops->fl_release_private(fl); | |
318 | fl->fl_ops = NULL; | |
319 | } | |
320 | ||
321 | if (fl->fl_lmops) { | |
322 | if (fl->fl_lmops->lm_put_owner) { | |
323 | fl->fl_lmops->lm_put_owner(fl->fl_owner); | |
324 | fl->fl_owner = NULL; | |
325 | } | |
326 | fl->fl_lmops = NULL; | |
327 | } | |
328 | } | |
329 | EXPORT_SYMBOL_GPL(locks_release_private); | |
330 | ||
331 | /* Free a lock which is not in use. */ | |
332 | void locks_free_lock(struct file_lock *fl) | |
333 | { | |
334 | BUG_ON(waitqueue_active(&fl->fl_wait)); | |
335 | BUG_ON(!list_empty(&fl->fl_list)); | |
336 | BUG_ON(!list_empty(&fl->fl_blocked_requests)); | |
337 | BUG_ON(!list_empty(&fl->fl_blocked_member)); | |
338 | BUG_ON(!hlist_unhashed(&fl->fl_link)); | |
339 | ||
340 | locks_release_private(fl); | |
341 | kmem_cache_free(filelock_cache, fl); | |
342 | } | |
343 | EXPORT_SYMBOL(locks_free_lock); | |
344 | ||
345 | static void | |
346 | locks_dispose_list(struct list_head *dispose) | |
347 | { | |
348 | struct file_lock *fl; | |
349 | ||
350 | while (!list_empty(dispose)) { | |
351 | fl = list_first_entry(dispose, struct file_lock, fl_list); | |
352 | list_del_init(&fl->fl_list); | |
353 | locks_free_lock(fl); | |
354 | } | |
355 | } | |
356 | ||
357 | void locks_init_lock(struct file_lock *fl) | |
358 | { | |
359 | memset(fl, 0, sizeof(struct file_lock)); | |
360 | locks_init_lock_heads(fl); | |
361 | } | |
362 | ||
363 | EXPORT_SYMBOL(locks_init_lock); | |
364 | ||
365 | /* | |
366 | * Initialize a new lock from an existing file_lock structure. | |
367 | */ | |
368 | void locks_copy_conflock(struct file_lock *new, struct file_lock *fl) | |
369 | { | |
370 | new->fl_owner = fl->fl_owner; | |
371 | new->fl_pid = fl->fl_pid; | |
372 | new->fl_file = NULL; | |
373 | new->fl_flags = fl->fl_flags; | |
374 | new->fl_type = fl->fl_type; | |
375 | new->fl_start = fl->fl_start; | |
376 | new->fl_end = fl->fl_end; | |
377 | new->fl_lmops = fl->fl_lmops; | |
378 | new->fl_ops = NULL; | |
379 | ||
380 | if (fl->fl_lmops) { | |
381 | if (fl->fl_lmops->lm_get_owner) | |
382 | fl->fl_lmops->lm_get_owner(fl->fl_owner); | |
383 | } | |
384 | } | |
385 | EXPORT_SYMBOL(locks_copy_conflock); | |
386 | ||
387 | void locks_copy_lock(struct file_lock *new, struct file_lock *fl) | |
388 | { | |
389 | /* "new" must be a freshly-initialized lock */ | |
390 | WARN_ON_ONCE(new->fl_ops); | |
391 | ||
392 | locks_copy_conflock(new, fl); | |
393 | ||
394 | new->fl_file = fl->fl_file; | |
395 | new->fl_ops = fl->fl_ops; | |
396 | ||
397 | if (fl->fl_ops) { | |
398 | if (fl->fl_ops->fl_copy_lock) | |
399 | fl->fl_ops->fl_copy_lock(new, fl); | |
400 | } | |
401 | } | |
402 | ||
403 | EXPORT_SYMBOL(locks_copy_lock); | |
404 | ||
405 | static inline int flock_translate_cmd(int cmd) { | |
406 | if (cmd & LOCK_MAND) | |
407 | return cmd & (LOCK_MAND | LOCK_RW); | |
408 | switch (cmd) { | |
409 | case LOCK_SH: | |
410 | return F_RDLCK; | |
411 | case LOCK_EX: | |
412 | return F_WRLCK; | |
413 | case LOCK_UN: | |
414 | return F_UNLCK; | |
415 | } | |
416 | return -EINVAL; | |
417 | } | |
418 | ||
419 | /* Fill in a file_lock structure with an appropriate FLOCK lock. */ | |
420 | static struct file_lock * | |
421 | flock_make_lock(struct file *filp, unsigned int cmd) | |
422 | { | |
423 | struct file_lock *fl; | |
424 | int type = flock_translate_cmd(cmd); | |
425 | ||
426 | if (type < 0) | |
427 | return ERR_PTR(type); | |
428 | ||
429 | fl = locks_alloc_lock(); | |
430 | if (fl == NULL) | |
431 | return ERR_PTR(-ENOMEM); | |
432 | ||
433 | fl->fl_file = filp; | |
434 | fl->fl_owner = filp; | |
435 | fl->fl_pid = current->tgid; | |
436 | fl->fl_flags = FL_FLOCK; | |
437 | fl->fl_type = type; | |
438 | fl->fl_end = OFFSET_MAX; | |
439 | ||
440 | return fl; | |
441 | } | |
442 | ||
443 | static int assign_type(struct file_lock *fl, long type) | |
444 | { | |
445 | switch (type) { | |
446 | case F_RDLCK: | |
447 | case F_WRLCK: | |
448 | case F_UNLCK: | |
449 | fl->fl_type = type; | |
450 | break; | |
451 | default: | |
452 | return -EINVAL; | |
453 | } | |
454 | return 0; | |
455 | } | |
456 | ||
457 | static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl, | |
458 | struct flock64 *l) | |
459 | { | |
460 | switch (l->l_whence) { | |
461 | case SEEK_SET: | |
462 | fl->fl_start = 0; | |
463 | break; | |
464 | case SEEK_CUR: | |
465 | fl->fl_start = filp->f_pos; | |
466 | break; | |
467 | case SEEK_END: | |
468 | fl->fl_start = i_size_read(file_inode(filp)); | |
469 | break; | |
470 | default: | |
471 | return -EINVAL; | |
472 | } | |
473 | if (l->l_start > OFFSET_MAX - fl->fl_start) | |
474 | return -EOVERFLOW; | |
475 | fl->fl_start += l->l_start; | |
476 | if (fl->fl_start < 0) | |
477 | return -EINVAL; | |
478 | ||
479 | /* POSIX-1996 leaves the case l->l_len < 0 undefined; | |
480 | POSIX-2001 defines it. */ | |
481 | if (l->l_len > 0) { | |
482 | if (l->l_len - 1 > OFFSET_MAX - fl->fl_start) | |
483 | return -EOVERFLOW; | |
484 | fl->fl_end = fl->fl_start + l->l_len - 1; | |
485 | ||
486 | } else if (l->l_len < 0) { | |
487 | if (fl->fl_start + l->l_len < 0) | |
488 | return -EINVAL; | |
489 | fl->fl_end = fl->fl_start - 1; | |
490 | fl->fl_start += l->l_len; | |
491 | } else | |
492 | fl->fl_end = OFFSET_MAX; | |
493 | ||
494 | fl->fl_owner = current->files; | |
495 | fl->fl_pid = current->tgid; | |
496 | fl->fl_file = filp; | |
497 | fl->fl_flags = FL_POSIX; | |
498 | fl->fl_ops = NULL; | |
499 | fl->fl_lmops = NULL; | |
500 | ||
501 | return assign_type(fl, l->l_type); | |
502 | } | |
503 | ||
504 | /* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX | |
505 | * style lock. | |
506 | */ | |
507 | static int flock_to_posix_lock(struct file *filp, struct file_lock *fl, | |
508 | struct flock *l) | |
509 | { | |
510 | struct flock64 ll = { | |
511 | .l_type = l->l_type, | |
512 | .l_whence = l->l_whence, | |
513 | .l_start = l->l_start, | |
514 | .l_len = l->l_len, | |
515 | }; | |
516 | ||
517 | return flock64_to_posix_lock(filp, fl, &ll); | |
518 | } | |
519 | ||
520 | /* default lease lock manager operations */ | |
521 | static bool | |
522 | lease_break_callback(struct file_lock *fl) | |
523 | { | |
524 | kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG); | |
525 | return false; | |
526 | } | |
527 | ||
528 | static void | |
529 | lease_setup(struct file_lock *fl, void **priv) | |
530 | { | |
531 | struct file *filp = fl->fl_file; | |
532 | struct fasync_struct *fa = *priv; | |
533 | ||
534 | /* | |
535 | * fasync_insert_entry() returns the old entry if any. If there was no | |
536 | * old entry, then it used "priv" and inserted it into the fasync list. | |
537 | * Clear the pointer to indicate that it shouldn't be freed. | |
538 | */ | |
539 | if (!fasync_insert_entry(fa->fa_fd, filp, &fl->fl_fasync, fa)) | |
540 | *priv = NULL; | |
541 | ||
542 | __f_setown(filp, task_pid(current), PIDTYPE_TGID, 0); | |
543 | } | |
544 | ||
545 | static const struct lock_manager_operations lease_manager_ops = { | |
546 | .lm_break = lease_break_callback, | |
547 | .lm_change = lease_modify, | |
548 | .lm_setup = lease_setup, | |
549 | }; | |
550 | ||
551 | /* | |
552 | * Initialize a lease, use the default lock manager operations | |
553 | */ | |
554 | static int lease_init(struct file *filp, long type, struct file_lock *fl) | |
555 | { | |
556 | if (assign_type(fl, type) != 0) | |
557 | return -EINVAL; | |
558 | ||
559 | fl->fl_owner = filp; | |
560 | fl->fl_pid = current->tgid; | |
561 | ||
562 | fl->fl_file = filp; | |
563 | fl->fl_flags = FL_LEASE; | |
564 | fl->fl_start = 0; | |
565 | fl->fl_end = OFFSET_MAX; | |
566 | fl->fl_ops = NULL; | |
567 | fl->fl_lmops = &lease_manager_ops; | |
568 | return 0; | |
569 | } | |
570 | ||
571 | /* Allocate a file_lock initialised to this type of lease */ | |
572 | static struct file_lock *lease_alloc(struct file *filp, long type) | |
573 | { | |
574 | struct file_lock *fl = locks_alloc_lock(); | |
575 | int error = -ENOMEM; | |
576 | ||
577 | if (fl == NULL) | |
578 | return ERR_PTR(error); | |
579 | ||
580 | error = lease_init(filp, type, fl); | |
581 | if (error) { | |
582 | locks_free_lock(fl); | |
583 | return ERR_PTR(error); | |
584 | } | |
585 | return fl; | |
586 | } | |
587 | ||
588 | /* Check if two locks overlap each other. | |
589 | */ | |
590 | static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2) | |
591 | { | |
592 | return ((fl1->fl_end >= fl2->fl_start) && | |
593 | (fl2->fl_end >= fl1->fl_start)); | |
594 | } | |
595 | ||
596 | /* | |
597 | * Check whether two locks have the same owner. | |
598 | */ | |
599 | static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2) | |
600 | { | |
601 | if (fl1->fl_lmops && fl1->fl_lmops->lm_compare_owner) | |
602 | return fl2->fl_lmops == fl1->fl_lmops && | |
603 | fl1->fl_lmops->lm_compare_owner(fl1, fl2); | |
604 | return fl1->fl_owner == fl2->fl_owner; | |
605 | } | |
606 | ||
607 | /* Must be called with the flc_lock held! */ | |
608 | static void locks_insert_global_locks(struct file_lock *fl) | |
609 | { | |
610 | struct file_lock_list_struct *fll = this_cpu_ptr(&file_lock_list); | |
611 | ||
612 | percpu_rwsem_assert_held(&file_rwsem); | |
613 | ||
614 | spin_lock(&fll->lock); | |
615 | fl->fl_link_cpu = smp_processor_id(); | |
616 | hlist_add_head(&fl->fl_link, &fll->hlist); | |
617 | spin_unlock(&fll->lock); | |
618 | } | |
619 | ||
620 | /* Must be called with the flc_lock held! */ | |
621 | static void locks_delete_global_locks(struct file_lock *fl) | |
622 | { | |
623 | struct file_lock_list_struct *fll; | |
624 | ||
625 | percpu_rwsem_assert_held(&file_rwsem); | |
626 | ||
627 | /* | |
628 | * Avoid taking lock if already unhashed. This is safe since this check | |
629 | * is done while holding the flc_lock, and new insertions into the list | |
630 | * also require that it be held. | |
631 | */ | |
632 | if (hlist_unhashed(&fl->fl_link)) | |
633 | return; | |
634 | ||
635 | fll = per_cpu_ptr(&file_lock_list, fl->fl_link_cpu); | |
636 | spin_lock(&fll->lock); | |
637 | hlist_del_init(&fl->fl_link); | |
638 | spin_unlock(&fll->lock); | |
639 | } | |
640 | ||
641 | static unsigned long | |
642 | posix_owner_key(struct file_lock *fl) | |
643 | { | |
644 | if (fl->fl_lmops && fl->fl_lmops->lm_owner_key) | |
645 | return fl->fl_lmops->lm_owner_key(fl); | |
646 | return (unsigned long)fl->fl_owner; | |
647 | } | |
648 | ||
649 | static void locks_insert_global_blocked(struct file_lock *waiter) | |
650 | { | |
651 | lockdep_assert_held(&blocked_lock_lock); | |
652 | ||
653 | hash_add(blocked_hash, &waiter->fl_link, posix_owner_key(waiter)); | |
654 | } | |
655 | ||
656 | static void locks_delete_global_blocked(struct file_lock *waiter) | |
657 | { | |
658 | lockdep_assert_held(&blocked_lock_lock); | |
659 | ||
660 | hash_del(&waiter->fl_link); | |
661 | } | |
662 | ||
663 | /* Remove waiter from blocker's block list. | |
664 | * When blocker ends up pointing to itself then the list is empty. | |
665 | * | |
666 | * Must be called with blocked_lock_lock held. | |
667 | */ | |
668 | static void __locks_delete_block(struct file_lock *waiter) | |
669 | { | |
670 | locks_delete_global_blocked(waiter); | |
671 | list_del_init(&waiter->fl_blocked_member); | |
672 | waiter->fl_blocker = NULL; | |
673 | } | |
674 | ||
675 | static void locks_delete_block(struct file_lock *waiter) | |
676 | { | |
677 | spin_lock(&blocked_lock_lock); | |
678 | __locks_delete_block(waiter); | |
679 | spin_unlock(&blocked_lock_lock); | |
680 | } | |
681 | ||
682 | /* Insert waiter into blocker's block list. | |
683 | * We use a circular list so that processes can be easily woken up in | |
684 | * the order they blocked. The documentation doesn't require this but | |
685 | * it seems like the reasonable thing to do. | |
686 | * | |
687 | * Must be called with both the flc_lock and blocked_lock_lock held. The | |
688 | * fl_blocked_requests list itself is protected by the blocked_lock_lock, | |
689 | * but by ensuring that the flc_lock is also held on insertions we can avoid | |
690 | * taking the blocked_lock_lock in some cases when we see that the | |
691 | * fl_blocked_requests list is empty. | |
692 | */ | |
693 | static void __locks_insert_block(struct file_lock *blocker, | |
694 | struct file_lock *waiter) | |
695 | { | |
696 | BUG_ON(!list_empty(&waiter->fl_blocked_member)); | |
697 | waiter->fl_blocker = blocker; | |
698 | list_add_tail(&waiter->fl_blocked_member, &blocker->fl_blocked_requests); | |
699 | if (IS_POSIX(blocker) && !IS_OFDLCK(blocker)) | |
700 | locks_insert_global_blocked(waiter); | |
701 | } | |
702 | ||
703 | /* Must be called with flc_lock held. */ | |
704 | static void locks_insert_block(struct file_lock *blocker, | |
705 | struct file_lock *waiter) | |
706 | { | |
707 | spin_lock(&blocked_lock_lock); | |
708 | __locks_insert_block(blocker, waiter); | |
709 | spin_unlock(&blocked_lock_lock); | |
710 | } | |
711 | ||
712 | /* | |
713 | * Wake up processes blocked waiting for blocker. | |
714 | * | |
715 | * Must be called with the inode->flc_lock held! | |
716 | */ | |
717 | static void locks_wake_up_blocks(struct file_lock *blocker) | |
718 | { | |
719 | /* | |
720 | * Avoid taking global lock if list is empty. This is safe since new | |
721 | * blocked requests are only added to the list under the flc_lock, and | |
722 | * the flc_lock is always held here. Note that removal from the | |
723 | * fl_blocked_requests list does not require the flc_lock, so we must | |
724 | * recheck list_empty() after acquiring the blocked_lock_lock. | |
725 | */ | |
726 | if (list_empty(&blocker->fl_blocked_requests)) | |
727 | return; | |
728 | ||
729 | spin_lock(&blocked_lock_lock); | |
730 | while (!list_empty(&blocker->fl_blocked_requests)) { | |
731 | struct file_lock *waiter; | |
732 | ||
733 | waiter = list_first_entry(&blocker->fl_blocked_requests, | |
734 | struct file_lock, fl_blocked_member); | |
735 | __locks_delete_block(waiter); | |
736 | if (waiter->fl_lmops && waiter->fl_lmops->lm_notify) | |
737 | waiter->fl_lmops->lm_notify(waiter); | |
738 | else | |
739 | wake_up(&waiter->fl_wait); | |
740 | } | |
741 | spin_unlock(&blocked_lock_lock); | |
742 | } | |
743 | ||
744 | static void | |
745 | locks_insert_lock_ctx(struct file_lock *fl, struct list_head *before) | |
746 | { | |
747 | list_add_tail(&fl->fl_list, before); | |
748 | locks_insert_global_locks(fl); | |
749 | } | |
750 | ||
751 | static void | |
752 | locks_unlink_lock_ctx(struct file_lock *fl) | |
753 | { | |
754 | locks_delete_global_locks(fl); | |
755 | list_del_init(&fl->fl_list); | |
756 | locks_wake_up_blocks(fl); | |
757 | } | |
758 | ||
759 | static void | |
760 | locks_delete_lock_ctx(struct file_lock *fl, struct list_head *dispose) | |
761 | { | |
762 | locks_unlink_lock_ctx(fl); | |
763 | if (dispose) | |
764 | list_add(&fl->fl_list, dispose); | |
765 | else | |
766 | locks_free_lock(fl); | |
767 | } | |
768 | ||
769 | /* Determine if lock sys_fl blocks lock caller_fl. Common functionality | |
770 | * checks for shared/exclusive status of overlapping locks. | |
771 | */ | |
772 | static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl) | |
773 | { | |
774 | if (sys_fl->fl_type == F_WRLCK) | |
775 | return 1; | |
776 | if (caller_fl->fl_type == F_WRLCK) | |
777 | return 1; | |
778 | return 0; | |
779 | } | |
780 | ||
781 | /* Determine if lock sys_fl blocks lock caller_fl. POSIX specific | |
782 | * checking before calling the locks_conflict(). | |
783 | */ | |
784 | static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl) | |
785 | { | |
786 | /* POSIX locks owned by the same process do not conflict with | |
787 | * each other. | |
788 | */ | |
789 | if (posix_same_owner(caller_fl, sys_fl)) | |
790 | return (0); | |
791 | ||
792 | /* Check whether they overlap */ | |
793 | if (!locks_overlap(caller_fl, sys_fl)) | |
794 | return 0; | |
795 | ||
796 | return (locks_conflict(caller_fl, sys_fl)); | |
797 | } | |
798 | ||
799 | /* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific | |
800 | * checking before calling the locks_conflict(). | |
801 | */ | |
802 | static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl) | |
803 | { | |
804 | /* FLOCK locks referring to the same filp do not conflict with | |
805 | * each other. | |
806 | */ | |
807 | if (caller_fl->fl_file == sys_fl->fl_file) | |
808 | return (0); | |
809 | if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND)) | |
810 | return 0; | |
811 | ||
812 | return (locks_conflict(caller_fl, sys_fl)); | |
813 | } | |
814 | ||
815 | void | |
816 | posix_test_lock(struct file *filp, struct file_lock *fl) | |
817 | { | |
818 | struct file_lock *cfl; | |
819 | struct file_lock_context *ctx; | |
820 | struct inode *inode = locks_inode(filp); | |
821 | ||
822 | ctx = smp_load_acquire(&inode->i_flctx); | |
823 | if (!ctx || list_empty_careful(&ctx->flc_posix)) { | |
824 | fl->fl_type = F_UNLCK; | |
825 | return; | |
826 | } | |
827 | ||
828 | spin_lock(&ctx->flc_lock); | |
829 | list_for_each_entry(cfl, &ctx->flc_posix, fl_list) { | |
830 | if (posix_locks_conflict(fl, cfl)) { | |
831 | locks_copy_conflock(fl, cfl); | |
832 | goto out; | |
833 | } | |
834 | } | |
835 | fl->fl_type = F_UNLCK; | |
836 | out: | |
837 | spin_unlock(&ctx->flc_lock); | |
838 | return; | |
839 | } | |
840 | EXPORT_SYMBOL(posix_test_lock); | |
841 | ||
842 | /* | |
843 | * Deadlock detection: | |
844 | * | |
845 | * We attempt to detect deadlocks that are due purely to posix file | |
846 | * locks. | |
847 | * | |
848 | * We assume that a task can be waiting for at most one lock at a time. | |
849 | * So for any acquired lock, the process holding that lock may be | |
850 | * waiting on at most one other lock. That lock in turns may be held by | |
851 | * someone waiting for at most one other lock. Given a requested lock | |
852 | * caller_fl which is about to wait for a conflicting lock block_fl, we | |
853 | * follow this chain of waiters to ensure we are not about to create a | |
854 | * cycle. | |
855 | * | |
856 | * Since we do this before we ever put a process to sleep on a lock, we | |
857 | * are ensured that there is never a cycle; that is what guarantees that | |
858 | * the while() loop in posix_locks_deadlock() eventually completes. | |
859 | * | |
860 | * Note: the above assumption may not be true when handling lock | |
861 | * requests from a broken NFS client. It may also fail in the presence | |
862 | * of tasks (such as posix threads) sharing the same open file table. | |
863 | * To handle those cases, we just bail out after a few iterations. | |
864 | * | |
865 | * For FL_OFDLCK locks, the owner is the filp, not the files_struct. | |
866 | * Because the owner is not even nominally tied to a thread of | |
867 | * execution, the deadlock detection below can't reasonably work well. Just | |
868 | * skip it for those. | |
869 | * | |
870 | * In principle, we could do a more limited deadlock detection on FL_OFDLCK | |
871 | * locks that just checks for the case where two tasks are attempting to | |
872 | * upgrade from read to write locks on the same inode. | |
873 | */ | |
874 | ||
875 | #define MAX_DEADLK_ITERATIONS 10 | |
876 | ||
877 | /* Find a lock that the owner of the given block_fl is blocking on. */ | |
878 | static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl) | |
879 | { | |
880 | struct file_lock *fl; | |
881 | ||
882 | hash_for_each_possible(blocked_hash, fl, fl_link, posix_owner_key(block_fl)) { | |
883 | if (posix_same_owner(fl, block_fl)) | |
884 | return fl->fl_blocker; | |
885 | } | |
886 | return NULL; | |
887 | } | |
888 | ||
889 | /* Must be called with the blocked_lock_lock held! */ | |
890 | static int posix_locks_deadlock(struct file_lock *caller_fl, | |
891 | struct file_lock *block_fl) | |
892 | { | |
893 | int i = 0; | |
894 | ||
895 | lockdep_assert_held(&blocked_lock_lock); | |
896 | ||
897 | /* | |
898 | * This deadlock detector can't reasonably detect deadlocks with | |
899 | * FL_OFDLCK locks, since they aren't owned by a process, per-se. | |
900 | */ | |
901 | if (IS_OFDLCK(caller_fl)) | |
902 | return 0; | |
903 | ||
904 | while ((block_fl = what_owner_is_waiting_for(block_fl))) { | |
905 | if (i++ > MAX_DEADLK_ITERATIONS) | |
906 | return 0; | |
907 | if (posix_same_owner(caller_fl, block_fl)) | |
908 | return 1; | |
909 | } | |
910 | return 0; | |
911 | } | |
912 | ||
913 | /* Try to create a FLOCK lock on filp. We always insert new FLOCK locks | |
914 | * after any leases, but before any posix locks. | |
915 | * | |
916 | * Note that if called with an FL_EXISTS argument, the caller may determine | |
917 | * whether or not a lock was successfully freed by testing the return | |
918 | * value for -ENOENT. | |
919 | */ | |
920 | static int flock_lock_inode(struct inode *inode, struct file_lock *request) | |
921 | { | |
922 | struct file_lock *new_fl = NULL; | |
923 | struct file_lock *fl; | |
924 | struct file_lock_context *ctx; | |
925 | int error = 0; | |
926 | bool found = false; | |
927 | LIST_HEAD(dispose); | |
928 | ||
929 | ctx = locks_get_lock_context(inode, request->fl_type); | |
930 | if (!ctx) { | |
931 | if (request->fl_type != F_UNLCK) | |
932 | return -ENOMEM; | |
933 | return (request->fl_flags & FL_EXISTS) ? -ENOENT : 0; | |
934 | } | |
935 | ||
936 | if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) { | |
937 | new_fl = locks_alloc_lock(); | |
938 | if (!new_fl) | |
939 | return -ENOMEM; | |
940 | } | |
941 | ||
942 | percpu_down_read_preempt_disable(&file_rwsem); | |
943 | spin_lock(&ctx->flc_lock); | |
944 | if (request->fl_flags & FL_ACCESS) | |
945 | goto find_conflict; | |
946 | ||
947 | list_for_each_entry(fl, &ctx->flc_flock, fl_list) { | |
948 | if (request->fl_file != fl->fl_file) | |
949 | continue; | |
950 | if (request->fl_type == fl->fl_type) | |
951 | goto out; | |
952 | found = true; | |
953 | locks_delete_lock_ctx(fl, &dispose); | |
954 | break; | |
955 | } | |
956 | ||
957 | if (request->fl_type == F_UNLCK) { | |
958 | if ((request->fl_flags & FL_EXISTS) && !found) | |
959 | error = -ENOENT; | |
960 | goto out; | |
961 | } | |
962 | ||
963 | find_conflict: | |
964 | list_for_each_entry(fl, &ctx->flc_flock, fl_list) { | |
965 | if (!flock_locks_conflict(request, fl)) | |
966 | continue; | |
967 | error = -EAGAIN; | |
968 | if (!(request->fl_flags & FL_SLEEP)) | |
969 | goto out; | |
970 | error = FILE_LOCK_DEFERRED; | |
971 | locks_insert_block(fl, request); | |
972 | goto out; | |
973 | } | |
974 | if (request->fl_flags & FL_ACCESS) | |
975 | goto out; | |
976 | locks_copy_lock(new_fl, request); | |
977 | locks_insert_lock_ctx(new_fl, &ctx->flc_flock); | |
978 | new_fl = NULL; | |
979 | error = 0; | |
980 | ||
981 | out: | |
982 | spin_unlock(&ctx->flc_lock); | |
983 | percpu_up_read_preempt_enable(&file_rwsem); | |
984 | if (new_fl) | |
985 | locks_free_lock(new_fl); | |
986 | locks_dispose_list(&dispose); | |
987 | trace_flock_lock_inode(inode, request, error); | |
988 | return error; | |
989 | } | |
990 | ||
991 | static int posix_lock_inode(struct inode *inode, struct file_lock *request, | |
992 | struct file_lock *conflock) | |
993 | { | |
994 | struct file_lock *fl, *tmp; | |
995 | struct file_lock *new_fl = NULL; | |
996 | struct file_lock *new_fl2 = NULL; | |
997 | struct file_lock *left = NULL; | |
998 | struct file_lock *right = NULL; | |
999 | struct file_lock_context *ctx; | |
1000 | int error; | |
1001 | bool added = false; | |
1002 | LIST_HEAD(dispose); | |
1003 | ||
1004 | ctx = locks_get_lock_context(inode, request->fl_type); | |
1005 | if (!ctx) | |
1006 | return (request->fl_type == F_UNLCK) ? 0 : -ENOMEM; | |
1007 | ||
1008 | /* | |
1009 | * We may need two file_lock structures for this operation, | |
1010 | * so we get them in advance to avoid races. | |
1011 | * | |
1012 | * In some cases we can be sure, that no new locks will be needed | |
1013 | */ | |
1014 | if (!(request->fl_flags & FL_ACCESS) && | |
1015 | (request->fl_type != F_UNLCK || | |
1016 | request->fl_start != 0 || request->fl_end != OFFSET_MAX)) { | |
1017 | new_fl = locks_alloc_lock(); | |
1018 | new_fl2 = locks_alloc_lock(); | |
1019 | } | |
1020 | ||
1021 | percpu_down_read_preempt_disable(&file_rwsem); | |
1022 | spin_lock(&ctx->flc_lock); | |
1023 | /* | |
1024 | * New lock request. Walk all POSIX locks and look for conflicts. If | |
1025 | * there are any, either return error or put the request on the | |
1026 | * blocker's list of waiters and the global blocked_hash. | |
1027 | */ | |
1028 | if (request->fl_type != F_UNLCK) { | |
1029 | list_for_each_entry(fl, &ctx->flc_posix, fl_list) { | |
1030 | if (!posix_locks_conflict(request, fl)) | |
1031 | continue; | |
1032 | if (conflock) | |
1033 | locks_copy_conflock(conflock, fl); | |
1034 | error = -EAGAIN; | |
1035 | if (!(request->fl_flags & FL_SLEEP)) | |
1036 | goto out; | |
1037 | /* | |
1038 | * Deadlock detection and insertion into the blocked | |
1039 | * locks list must be done while holding the same lock! | |
1040 | */ | |
1041 | error = -EDEADLK; | |
1042 | spin_lock(&blocked_lock_lock); | |
1043 | if (likely(!posix_locks_deadlock(request, fl))) { | |
1044 | error = FILE_LOCK_DEFERRED; | |
1045 | __locks_insert_block(fl, request); | |
1046 | } | |
1047 | spin_unlock(&blocked_lock_lock); | |
1048 | goto out; | |
1049 | } | |
1050 | } | |
1051 | ||
1052 | /* If we're just looking for a conflict, we're done. */ | |
1053 | error = 0; | |
1054 | if (request->fl_flags & FL_ACCESS) | |
1055 | goto out; | |
1056 | ||
1057 | /* Find the first old lock with the same owner as the new lock */ | |
1058 | list_for_each_entry(fl, &ctx->flc_posix, fl_list) { | |
1059 | if (posix_same_owner(request, fl)) | |
1060 | break; | |
1061 | } | |
1062 | ||
1063 | /* Process locks with this owner. */ | |
1064 | list_for_each_entry_safe_from(fl, tmp, &ctx->flc_posix, fl_list) { | |
1065 | if (!posix_same_owner(request, fl)) | |
1066 | break; | |
1067 | ||
1068 | /* Detect adjacent or overlapping regions (if same lock type) */ | |
1069 | if (request->fl_type == fl->fl_type) { | |
1070 | /* In all comparisons of start vs end, use | |
1071 | * "start - 1" rather than "end + 1". If end | |
1072 | * is OFFSET_MAX, end + 1 will become negative. | |
1073 | */ | |
1074 | if (fl->fl_end < request->fl_start - 1) | |
1075 | continue; | |
1076 | /* If the next lock in the list has entirely bigger | |
1077 | * addresses than the new one, insert the lock here. | |
1078 | */ | |
1079 | if (fl->fl_start - 1 > request->fl_end) | |
1080 | break; | |
1081 | ||
1082 | /* If we come here, the new and old lock are of the | |
1083 | * same type and adjacent or overlapping. Make one | |
1084 | * lock yielding from the lower start address of both | |
1085 | * locks to the higher end address. | |
1086 | */ | |
1087 | if (fl->fl_start > request->fl_start) | |
1088 | fl->fl_start = request->fl_start; | |
1089 | else | |
1090 | request->fl_start = fl->fl_start; | |
1091 | if (fl->fl_end < request->fl_end) | |
1092 | fl->fl_end = request->fl_end; | |
1093 | else | |
1094 | request->fl_end = fl->fl_end; | |
1095 | if (added) { | |
1096 | locks_delete_lock_ctx(fl, &dispose); | |
1097 | continue; | |
1098 | } | |
1099 | request = fl; | |
1100 | added = true; | |
1101 | } else { | |
1102 | /* Processing for different lock types is a bit | |
1103 | * more complex. | |
1104 | */ | |
1105 | if (fl->fl_end < request->fl_start) | |
1106 | continue; | |
1107 | if (fl->fl_start > request->fl_end) | |
1108 | break; | |
1109 | if (request->fl_type == F_UNLCK) | |
1110 | added = true; | |
1111 | if (fl->fl_start < request->fl_start) | |
1112 | left = fl; | |
1113 | /* If the next lock in the list has a higher end | |
1114 | * address than the new one, insert the new one here. | |
1115 | */ | |
1116 | if (fl->fl_end > request->fl_end) { | |
1117 | right = fl; | |
1118 | break; | |
1119 | } | |
1120 | if (fl->fl_start >= request->fl_start) { | |
1121 | /* The new lock completely replaces an old | |
1122 | * one (This may happen several times). | |
1123 | */ | |
1124 | if (added) { | |
1125 | locks_delete_lock_ctx(fl, &dispose); | |
1126 | continue; | |
1127 | } | |
1128 | /* | |
1129 | * Replace the old lock with new_fl, and | |
1130 | * remove the old one. It's safe to do the | |
1131 | * insert here since we know that we won't be | |
1132 | * using new_fl later, and that the lock is | |
1133 | * just replacing an existing lock. | |
1134 | */ | |
1135 | error = -ENOLCK; | |
1136 | if (!new_fl) | |
1137 | goto out; | |
1138 | locks_copy_lock(new_fl, request); | |
1139 | request = new_fl; | |
1140 | new_fl = NULL; | |
1141 | locks_insert_lock_ctx(request, &fl->fl_list); | |
1142 | locks_delete_lock_ctx(fl, &dispose); | |
1143 | added = true; | |
1144 | } | |
1145 | } | |
1146 | } | |
1147 | ||
1148 | /* | |
1149 | * The above code only modifies existing locks in case of merging or | |
1150 | * replacing. If new lock(s) need to be inserted all modifications are | |
1151 | * done below this, so it's safe yet to bail out. | |
1152 | */ | |
1153 | error = -ENOLCK; /* "no luck" */ | |
1154 | if (right && left == right && !new_fl2) | |
1155 | goto out; | |
1156 | ||
1157 | error = 0; | |
1158 | if (!added) { | |
1159 | if (request->fl_type == F_UNLCK) { | |
1160 | if (request->fl_flags & FL_EXISTS) | |
1161 | error = -ENOENT; | |
1162 | goto out; | |
1163 | } | |
1164 | ||
1165 | if (!new_fl) { | |
1166 | error = -ENOLCK; | |
1167 | goto out; | |
1168 | } | |
1169 | locks_copy_lock(new_fl, request); | |
1170 | locks_insert_lock_ctx(new_fl, &fl->fl_list); | |
1171 | fl = new_fl; | |
1172 | new_fl = NULL; | |
1173 | } | |
1174 | if (right) { | |
1175 | if (left == right) { | |
1176 | /* The new lock breaks the old one in two pieces, | |
1177 | * so we have to use the second new lock. | |
1178 | */ | |
1179 | left = new_fl2; | |
1180 | new_fl2 = NULL; | |
1181 | locks_copy_lock(left, right); | |
1182 | locks_insert_lock_ctx(left, &fl->fl_list); | |
1183 | } | |
1184 | right->fl_start = request->fl_end + 1; | |
1185 | locks_wake_up_blocks(right); | |
1186 | } | |
1187 | if (left) { | |
1188 | left->fl_end = request->fl_start - 1; | |
1189 | locks_wake_up_blocks(left); | |
1190 | } | |
1191 | out: | |
1192 | spin_unlock(&ctx->flc_lock); | |
1193 | percpu_up_read_preempt_enable(&file_rwsem); | |
1194 | /* | |
1195 | * Free any unused locks. | |
1196 | */ | |
1197 | if (new_fl) | |
1198 | locks_free_lock(new_fl); | |
1199 | if (new_fl2) | |
1200 | locks_free_lock(new_fl2); | |
1201 | locks_dispose_list(&dispose); | |
1202 | trace_posix_lock_inode(inode, request, error); | |
1203 | ||
1204 | return error; | |
1205 | } | |
1206 | ||
1207 | /** | |
1208 | * posix_lock_file - Apply a POSIX-style lock to a file | |
1209 | * @filp: The file to apply the lock to | |
1210 | * @fl: The lock to be applied | |
1211 | * @conflock: Place to return a copy of the conflicting lock, if found. | |
1212 | * | |
1213 | * Add a POSIX style lock to a file. | |
1214 | * We merge adjacent & overlapping locks whenever possible. | |
1215 | * POSIX locks are sorted by owner task, then by starting address | |
1216 | * | |
1217 | * Note that if called with an FL_EXISTS argument, the caller may determine | |
1218 | * whether or not a lock was successfully freed by testing the return | |
1219 | * value for -ENOENT. | |
1220 | */ | |
1221 | int posix_lock_file(struct file *filp, struct file_lock *fl, | |
1222 | struct file_lock *conflock) | |
1223 | { | |
1224 | return posix_lock_inode(locks_inode(filp), fl, conflock); | |
1225 | } | |
1226 | EXPORT_SYMBOL(posix_lock_file); | |
1227 | ||
1228 | /** | |
1229 | * posix_lock_inode_wait - Apply a POSIX-style lock to a file | |
1230 | * @inode: inode of file to which lock request should be applied | |
1231 | * @fl: The lock to be applied | |
1232 | * | |
1233 | * Apply a POSIX style lock request to an inode. | |
1234 | */ | |
1235 | static int posix_lock_inode_wait(struct inode *inode, struct file_lock *fl) | |
1236 | { | |
1237 | int error; | |
1238 | might_sleep (); | |
1239 | for (;;) { | |
1240 | error = posix_lock_inode(inode, fl, NULL); | |
1241 | if (error != FILE_LOCK_DEFERRED) | |
1242 | break; | |
1243 | error = wait_event_interruptible(fl->fl_wait, !fl->fl_blocker); | |
1244 | if (!error) | |
1245 | continue; | |
1246 | ||
1247 | locks_delete_block(fl); | |
1248 | break; | |
1249 | } | |
1250 | return error; | |
1251 | } | |
1252 | ||
1253 | #ifdef CONFIG_MANDATORY_FILE_LOCKING | |
1254 | /** | |
1255 | * locks_mandatory_locked - Check for an active lock | |
1256 | * @file: the file to check | |
1257 | * | |
1258 | * Searches the inode's list of locks to find any POSIX locks which conflict. | |
1259 | * This function is called from locks_verify_locked() only. | |
1260 | */ | |
1261 | int locks_mandatory_locked(struct file *file) | |
1262 | { | |
1263 | int ret; | |
1264 | struct inode *inode = locks_inode(file); | |
1265 | struct file_lock_context *ctx; | |
1266 | struct file_lock *fl; | |
1267 | ||
1268 | ctx = smp_load_acquire(&inode->i_flctx); | |
1269 | if (!ctx || list_empty_careful(&ctx->flc_posix)) | |
1270 | return 0; | |
1271 | ||
1272 | /* | |
1273 | * Search the lock list for this inode for any POSIX locks. | |
1274 | */ | |
1275 | spin_lock(&ctx->flc_lock); | |
1276 | ret = 0; | |
1277 | list_for_each_entry(fl, &ctx->flc_posix, fl_list) { | |
1278 | if (fl->fl_owner != current->files && | |
1279 | fl->fl_owner != file) { | |
1280 | ret = -EAGAIN; | |
1281 | break; | |
1282 | } | |
1283 | } | |
1284 | spin_unlock(&ctx->flc_lock); | |
1285 | return ret; | |
1286 | } | |
1287 | ||
1288 | /** | |
1289 | * locks_mandatory_area - Check for a conflicting lock | |
1290 | * @inode: the file to check | |
1291 | * @filp: how the file was opened (if it was) | |
1292 | * @start: first byte in the file to check | |
1293 | * @end: lastbyte in the file to check | |
1294 | * @type: %F_WRLCK for a write lock, else %F_RDLCK | |
1295 | * | |
1296 | * Searches the inode's list of locks to find any POSIX locks which conflict. | |
1297 | */ | |
1298 | int locks_mandatory_area(struct inode *inode, struct file *filp, loff_t start, | |
1299 | loff_t end, unsigned char type) | |
1300 | { | |
1301 | struct file_lock fl; | |
1302 | int error; | |
1303 | bool sleep = false; | |
1304 | ||
1305 | locks_init_lock(&fl); | |
1306 | fl.fl_pid = current->tgid; | |
1307 | fl.fl_file = filp; | |
1308 | fl.fl_flags = FL_POSIX | FL_ACCESS; | |
1309 | if (filp && !(filp->f_flags & O_NONBLOCK)) | |
1310 | sleep = true; | |
1311 | fl.fl_type = type; | |
1312 | fl.fl_start = start; | |
1313 | fl.fl_end = end; | |
1314 | ||
1315 | for (;;) { | |
1316 | if (filp) { | |
1317 | fl.fl_owner = filp; | |
1318 | fl.fl_flags &= ~FL_SLEEP; | |
1319 | error = posix_lock_inode(inode, &fl, NULL); | |
1320 | if (!error) | |
1321 | break; | |
1322 | } | |
1323 | ||
1324 | if (sleep) | |
1325 | fl.fl_flags |= FL_SLEEP; | |
1326 | fl.fl_owner = current->files; | |
1327 | error = posix_lock_inode(inode, &fl, NULL); | |
1328 | if (error != FILE_LOCK_DEFERRED) | |
1329 | break; | |
1330 | error = wait_event_interruptible(fl.fl_wait, !fl.fl_blocker); | |
1331 | if (!error) { | |
1332 | /* | |
1333 | * If we've been sleeping someone might have | |
1334 | * changed the permissions behind our back. | |
1335 | */ | |
1336 | if (__mandatory_lock(inode)) | |
1337 | continue; | |
1338 | } | |
1339 | ||
1340 | locks_delete_block(&fl); | |
1341 | break; | |
1342 | } | |
1343 | ||
1344 | return error; | |
1345 | } | |
1346 | ||
1347 | EXPORT_SYMBOL(locks_mandatory_area); | |
1348 | #endif /* CONFIG_MANDATORY_FILE_LOCKING */ | |
1349 | ||
1350 | static void lease_clear_pending(struct file_lock *fl, int arg) | |
1351 | { | |
1352 | switch (arg) { | |
1353 | case F_UNLCK: | |
1354 | fl->fl_flags &= ~FL_UNLOCK_PENDING; | |
1355 | /* fall through: */ | |
1356 | case F_RDLCK: | |
1357 | fl->fl_flags &= ~FL_DOWNGRADE_PENDING; | |
1358 | } | |
1359 | } | |
1360 | ||
1361 | /* We already had a lease on this file; just change its type */ | |
1362 | int lease_modify(struct file_lock *fl, int arg, struct list_head *dispose) | |
1363 | { | |
1364 | int error = assign_type(fl, arg); | |
1365 | ||
1366 | if (error) | |
1367 | return error; | |
1368 | lease_clear_pending(fl, arg); | |
1369 | locks_wake_up_blocks(fl); | |
1370 | if (arg == F_UNLCK) { | |
1371 | struct file *filp = fl->fl_file; | |
1372 | ||
1373 | f_delown(filp); | |
1374 | filp->f_owner.signum = 0; | |
1375 | fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync); | |
1376 | if (fl->fl_fasync != NULL) { | |
1377 | printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync); | |
1378 | fl->fl_fasync = NULL; | |
1379 | } | |
1380 | locks_delete_lock_ctx(fl, dispose); | |
1381 | } | |
1382 | return 0; | |
1383 | } | |
1384 | EXPORT_SYMBOL(lease_modify); | |
1385 | ||
1386 | static bool past_time(unsigned long then) | |
1387 | { | |
1388 | if (!then) | |
1389 | /* 0 is a special value meaning "this never expires": */ | |
1390 | return false; | |
1391 | return time_after(jiffies, then); | |
1392 | } | |
1393 | ||
1394 | static void time_out_leases(struct inode *inode, struct list_head *dispose) | |
1395 | { | |
1396 | struct file_lock_context *ctx = inode->i_flctx; | |
1397 | struct file_lock *fl, *tmp; | |
1398 | ||
1399 | lockdep_assert_held(&ctx->flc_lock); | |
1400 | ||
1401 | list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list) { | |
1402 | trace_time_out_leases(inode, fl); | |
1403 | if (past_time(fl->fl_downgrade_time)) | |
1404 | lease_modify(fl, F_RDLCK, dispose); | |
1405 | if (past_time(fl->fl_break_time)) | |
1406 | lease_modify(fl, F_UNLCK, dispose); | |
1407 | } | |
1408 | } | |
1409 | ||
1410 | static bool leases_conflict(struct file_lock *lease, struct file_lock *breaker) | |
1411 | { | |
1412 | if ((breaker->fl_flags & FL_LAYOUT) != (lease->fl_flags & FL_LAYOUT)) | |
1413 | return false; | |
1414 | if ((breaker->fl_flags & FL_DELEG) && (lease->fl_flags & FL_LEASE)) | |
1415 | return false; | |
1416 | return locks_conflict(breaker, lease); | |
1417 | } | |
1418 | ||
1419 | static bool | |
1420 | any_leases_conflict(struct inode *inode, struct file_lock *breaker) | |
1421 | { | |
1422 | struct file_lock_context *ctx = inode->i_flctx; | |
1423 | struct file_lock *fl; | |
1424 | ||
1425 | lockdep_assert_held(&ctx->flc_lock); | |
1426 | ||
1427 | list_for_each_entry(fl, &ctx->flc_lease, fl_list) { | |
1428 | if (leases_conflict(fl, breaker)) | |
1429 | return true; | |
1430 | } | |
1431 | return false; | |
1432 | } | |
1433 | ||
1434 | /** | |
1435 | * __break_lease - revoke all outstanding leases on file | |
1436 | * @inode: the inode of the file to return | |
1437 | * @mode: O_RDONLY: break only write leases; O_WRONLY or O_RDWR: | |
1438 | * break all leases | |
1439 | * @type: FL_LEASE: break leases and delegations; FL_DELEG: break | |
1440 | * only delegations | |
1441 | * | |
1442 | * break_lease (inlined for speed) has checked there already is at least | |
1443 | * some kind of lock (maybe a lease) on this file. Leases are broken on | |
1444 | * a call to open() or truncate(). This function can sleep unless you | |
1445 | * specified %O_NONBLOCK to your open(). | |
1446 | */ | |
1447 | int __break_lease(struct inode *inode, unsigned int mode, unsigned int type) | |
1448 | { | |
1449 | int error = 0; | |
1450 | struct file_lock_context *ctx; | |
1451 | struct file_lock *new_fl, *fl, *tmp; | |
1452 | unsigned long break_time; | |
1453 | int want_write = (mode & O_ACCMODE) != O_RDONLY; | |
1454 | LIST_HEAD(dispose); | |
1455 | ||
1456 | new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK); | |
1457 | if (IS_ERR(new_fl)) | |
1458 | return PTR_ERR(new_fl); | |
1459 | new_fl->fl_flags = type; | |
1460 | ||
1461 | /* typically we will check that ctx is non-NULL before calling */ | |
1462 | ctx = smp_load_acquire(&inode->i_flctx); | |
1463 | if (!ctx) { | |
1464 | WARN_ON_ONCE(1); | |
1465 | return error; | |
1466 | } | |
1467 | ||
1468 | percpu_down_read_preempt_disable(&file_rwsem); | |
1469 | spin_lock(&ctx->flc_lock); | |
1470 | ||
1471 | time_out_leases(inode, &dispose); | |
1472 | ||
1473 | if (!any_leases_conflict(inode, new_fl)) | |
1474 | goto out; | |
1475 | ||
1476 | break_time = 0; | |
1477 | if (lease_break_time > 0) { | |
1478 | break_time = jiffies + lease_break_time * HZ; | |
1479 | if (break_time == 0) | |
1480 | break_time++; /* so that 0 means no break time */ | |
1481 | } | |
1482 | ||
1483 | list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list) { | |
1484 | if (!leases_conflict(fl, new_fl)) | |
1485 | continue; | |
1486 | if (want_write) { | |
1487 | if (fl->fl_flags & FL_UNLOCK_PENDING) | |
1488 | continue; | |
1489 | fl->fl_flags |= FL_UNLOCK_PENDING; | |
1490 | fl->fl_break_time = break_time; | |
1491 | } else { | |
1492 | if (lease_breaking(fl)) | |
1493 | continue; | |
1494 | fl->fl_flags |= FL_DOWNGRADE_PENDING; | |
1495 | fl->fl_downgrade_time = break_time; | |
1496 | } | |
1497 | if (fl->fl_lmops->lm_break(fl)) | |
1498 | locks_delete_lock_ctx(fl, &dispose); | |
1499 | } | |
1500 | ||
1501 | if (list_empty(&ctx->flc_lease)) | |
1502 | goto out; | |
1503 | ||
1504 | if (mode & O_NONBLOCK) { | |
1505 | trace_break_lease_noblock(inode, new_fl); | |
1506 | error = -EWOULDBLOCK; | |
1507 | goto out; | |
1508 | } | |
1509 | ||
1510 | restart: | |
1511 | fl = list_first_entry(&ctx->flc_lease, struct file_lock, fl_list); | |
1512 | break_time = fl->fl_break_time; | |
1513 | if (break_time != 0) | |
1514 | break_time -= jiffies; | |
1515 | if (break_time == 0) | |
1516 | break_time++; | |
1517 | locks_insert_block(fl, new_fl); | |
1518 | trace_break_lease_block(inode, new_fl); | |
1519 | spin_unlock(&ctx->flc_lock); | |
1520 | percpu_up_read_preempt_enable(&file_rwsem); | |
1521 | ||
1522 | locks_dispose_list(&dispose); | |
1523 | error = wait_event_interruptible_timeout(new_fl->fl_wait, | |
1524 | !new_fl->fl_blocker, break_time); | |
1525 | ||
1526 | percpu_down_read_preempt_disable(&file_rwsem); | |
1527 | spin_lock(&ctx->flc_lock); | |
1528 | trace_break_lease_unblock(inode, new_fl); | |
1529 | locks_delete_block(new_fl); | |
1530 | if (error >= 0) { | |
1531 | /* | |
1532 | * Wait for the next conflicting lease that has not been | |
1533 | * broken yet | |
1534 | */ | |
1535 | if (error == 0) | |
1536 | time_out_leases(inode, &dispose); | |
1537 | if (any_leases_conflict(inode, new_fl)) | |
1538 | goto restart; | |
1539 | error = 0; | |
1540 | } | |
1541 | out: | |
1542 | spin_unlock(&ctx->flc_lock); | |
1543 | percpu_up_read_preempt_enable(&file_rwsem); | |
1544 | locks_dispose_list(&dispose); | |
1545 | locks_free_lock(new_fl); | |
1546 | return error; | |
1547 | } | |
1548 | ||
1549 | EXPORT_SYMBOL(__break_lease); | |
1550 | ||
1551 | /** | |
1552 | * lease_get_mtime - update modified time of an inode with exclusive lease | |
1553 | * @inode: the inode | |
1554 | * @time: pointer to a timespec which contains the last modified time | |
1555 | * | |
1556 | * This is to force NFS clients to flush their caches for files with | |
1557 | * exclusive leases. The justification is that if someone has an | |
1558 | * exclusive lease, then they could be modifying it. | |
1559 | */ | |
1560 | void lease_get_mtime(struct inode *inode, struct timespec64 *time) | |
1561 | { | |
1562 | bool has_lease = false; | |
1563 | struct file_lock_context *ctx; | |
1564 | struct file_lock *fl; | |
1565 | ||
1566 | ctx = smp_load_acquire(&inode->i_flctx); | |
1567 | if (ctx && !list_empty_careful(&ctx->flc_lease)) { | |
1568 | spin_lock(&ctx->flc_lock); | |
1569 | fl = list_first_entry_or_null(&ctx->flc_lease, | |
1570 | struct file_lock, fl_list); | |
1571 | if (fl && (fl->fl_type == F_WRLCK)) | |
1572 | has_lease = true; | |
1573 | spin_unlock(&ctx->flc_lock); | |
1574 | } | |
1575 | ||
1576 | if (has_lease) | |
1577 | *time = current_time(inode); | |
1578 | } | |
1579 | ||
1580 | EXPORT_SYMBOL(lease_get_mtime); | |
1581 | ||
1582 | /** | |
1583 | * fcntl_getlease - Enquire what lease is currently active | |
1584 | * @filp: the file | |
1585 | * | |
1586 | * The value returned by this function will be one of | |
1587 | * (if no lease break is pending): | |
1588 | * | |
1589 | * %F_RDLCK to indicate a shared lease is held. | |
1590 | * | |
1591 | * %F_WRLCK to indicate an exclusive lease is held. | |
1592 | * | |
1593 | * %F_UNLCK to indicate no lease is held. | |
1594 | * | |
1595 | * (if a lease break is pending): | |
1596 | * | |
1597 | * %F_RDLCK to indicate an exclusive lease needs to be | |
1598 | * changed to a shared lease (or removed). | |
1599 | * | |
1600 | * %F_UNLCK to indicate the lease needs to be removed. | |
1601 | * | |
1602 | * XXX: sfr & willy disagree over whether F_INPROGRESS | |
1603 | * should be returned to userspace. | |
1604 | */ | |
1605 | int fcntl_getlease(struct file *filp) | |
1606 | { | |
1607 | struct file_lock *fl; | |
1608 | struct inode *inode = locks_inode(filp); | |
1609 | struct file_lock_context *ctx; | |
1610 | int type = F_UNLCK; | |
1611 | LIST_HEAD(dispose); | |
1612 | ||
1613 | ctx = smp_load_acquire(&inode->i_flctx); | |
1614 | if (ctx && !list_empty_careful(&ctx->flc_lease)) { | |
1615 | percpu_down_read_preempt_disable(&file_rwsem); | |
1616 | spin_lock(&ctx->flc_lock); | |
1617 | time_out_leases(inode, &dispose); | |
1618 | list_for_each_entry(fl, &ctx->flc_lease, fl_list) { | |
1619 | if (fl->fl_file != filp) | |
1620 | continue; | |
1621 | type = target_leasetype(fl); | |
1622 | break; | |
1623 | } | |
1624 | spin_unlock(&ctx->flc_lock); | |
1625 | percpu_up_read_preempt_enable(&file_rwsem); | |
1626 | ||
1627 | locks_dispose_list(&dispose); | |
1628 | } | |
1629 | return type; | |
1630 | } | |
1631 | ||
1632 | /** | |
1633 | * check_conflicting_open - see if the given dentry points to a file that has | |
1634 | * an existing open that would conflict with the | |
1635 | * desired lease. | |
1636 | * @dentry: dentry to check | |
1637 | * @arg: type of lease that we're trying to acquire | |
1638 | * @flags: current lock flags | |
1639 | * | |
1640 | * Check to see if there's an existing open fd on this file that would | |
1641 | * conflict with the lease we're trying to set. | |
1642 | */ | |
1643 | static int | |
1644 | check_conflicting_open(const struct dentry *dentry, const long arg, int flags) | |
1645 | { | |
1646 | int ret = 0; | |
1647 | struct inode *inode = dentry->d_inode; | |
1648 | ||
1649 | if (flags & FL_LAYOUT) | |
1650 | return 0; | |
1651 | ||
1652 | if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0)) | |
1653 | return -EAGAIN; | |
1654 | ||
1655 | if ((arg == F_WRLCK) && ((d_count(dentry) > 1) || | |
1656 | (atomic_read(&inode->i_count) > 1))) | |
1657 | ret = -EAGAIN; | |
1658 | ||
1659 | return ret; | |
1660 | } | |
1661 | ||
1662 | static int | |
1663 | generic_add_lease(struct file *filp, long arg, struct file_lock **flp, void **priv) | |
1664 | { | |
1665 | struct file_lock *fl, *my_fl = NULL, *lease; | |
1666 | struct dentry *dentry = filp->f_path.dentry; | |
1667 | struct inode *inode = dentry->d_inode; | |
1668 | struct file_lock_context *ctx; | |
1669 | bool is_deleg = (*flp)->fl_flags & FL_DELEG; | |
1670 | int error; | |
1671 | LIST_HEAD(dispose); | |
1672 | ||
1673 | lease = *flp; | |
1674 | trace_generic_add_lease(inode, lease); | |
1675 | ||
1676 | /* Note that arg is never F_UNLCK here */ | |
1677 | ctx = locks_get_lock_context(inode, arg); | |
1678 | if (!ctx) | |
1679 | return -ENOMEM; | |
1680 | ||
1681 | /* | |
1682 | * In the delegation case we need mutual exclusion with | |
1683 | * a number of operations that take the i_mutex. We trylock | |
1684 | * because delegations are an optional optimization, and if | |
1685 | * there's some chance of a conflict--we'd rather not | |
1686 | * bother, maybe that's a sign this just isn't a good file to | |
1687 | * hand out a delegation on. | |
1688 | */ | |
1689 | if (is_deleg && !inode_trylock(inode)) | |
1690 | return -EAGAIN; | |
1691 | ||
1692 | if (is_deleg && arg == F_WRLCK) { | |
1693 | /* Write delegations are not currently supported: */ | |
1694 | inode_unlock(inode); | |
1695 | WARN_ON_ONCE(1); | |
1696 | return -EINVAL; | |
1697 | } | |
1698 | ||
1699 | percpu_down_read_preempt_disable(&file_rwsem); | |
1700 | spin_lock(&ctx->flc_lock); | |
1701 | time_out_leases(inode, &dispose); | |
1702 | error = check_conflicting_open(dentry, arg, lease->fl_flags); | |
1703 | if (error) | |
1704 | goto out; | |
1705 | ||
1706 | /* | |
1707 | * At this point, we know that if there is an exclusive | |
1708 | * lease on this file, then we hold it on this filp | |
1709 | * (otherwise our open of this file would have blocked). | |
1710 | * And if we are trying to acquire an exclusive lease, | |
1711 | * then the file is not open by anyone (including us) | |
1712 | * except for this filp. | |
1713 | */ | |
1714 | error = -EAGAIN; | |
1715 | list_for_each_entry(fl, &ctx->flc_lease, fl_list) { | |
1716 | if (fl->fl_file == filp && | |
1717 | fl->fl_owner == lease->fl_owner) { | |
1718 | my_fl = fl; | |
1719 | continue; | |
1720 | } | |
1721 | ||
1722 | /* | |
1723 | * No exclusive leases if someone else has a lease on | |
1724 | * this file: | |
1725 | */ | |
1726 | if (arg == F_WRLCK) | |
1727 | goto out; | |
1728 | /* | |
1729 | * Modifying our existing lease is OK, but no getting a | |
1730 | * new lease if someone else is opening for write: | |
1731 | */ | |
1732 | if (fl->fl_flags & FL_UNLOCK_PENDING) | |
1733 | goto out; | |
1734 | } | |
1735 | ||
1736 | if (my_fl != NULL) { | |
1737 | lease = my_fl; | |
1738 | error = lease->fl_lmops->lm_change(lease, arg, &dispose); | |
1739 | if (error) | |
1740 | goto out; | |
1741 | goto out_setup; | |
1742 | } | |
1743 | ||
1744 | error = -EINVAL; | |
1745 | if (!leases_enable) | |
1746 | goto out; | |
1747 | ||
1748 | locks_insert_lock_ctx(lease, &ctx->flc_lease); | |
1749 | /* | |
1750 | * The check in break_lease() is lockless. It's possible for another | |
1751 | * open to race in after we did the earlier check for a conflicting | |
1752 | * open but before the lease was inserted. Check again for a | |
1753 | * conflicting open and cancel the lease if there is one. | |
1754 | * | |
1755 | * We also add a barrier here to ensure that the insertion of the lock | |
1756 | * precedes these checks. | |
1757 | */ | |
1758 | smp_mb(); | |
1759 | error = check_conflicting_open(dentry, arg, lease->fl_flags); | |
1760 | if (error) { | |
1761 | locks_unlink_lock_ctx(lease); | |
1762 | goto out; | |
1763 | } | |
1764 | ||
1765 | out_setup: | |
1766 | if (lease->fl_lmops->lm_setup) | |
1767 | lease->fl_lmops->lm_setup(lease, priv); | |
1768 | out: | |
1769 | spin_unlock(&ctx->flc_lock); | |
1770 | percpu_up_read_preempt_enable(&file_rwsem); | |
1771 | locks_dispose_list(&dispose); | |
1772 | if (is_deleg) | |
1773 | inode_unlock(inode); | |
1774 | if (!error && !my_fl) | |
1775 | *flp = NULL; | |
1776 | return error; | |
1777 | } | |
1778 | ||
1779 | static int generic_delete_lease(struct file *filp, void *owner) | |
1780 | { | |
1781 | int error = -EAGAIN; | |
1782 | struct file_lock *fl, *victim = NULL; | |
1783 | struct inode *inode = locks_inode(filp); | |
1784 | struct file_lock_context *ctx; | |
1785 | LIST_HEAD(dispose); | |
1786 | ||
1787 | ctx = smp_load_acquire(&inode->i_flctx); | |
1788 | if (!ctx) { | |
1789 | trace_generic_delete_lease(inode, NULL); | |
1790 | return error; | |
1791 | } | |
1792 | ||
1793 | percpu_down_read_preempt_disable(&file_rwsem); | |
1794 | spin_lock(&ctx->flc_lock); | |
1795 | list_for_each_entry(fl, &ctx->flc_lease, fl_list) { | |
1796 | if (fl->fl_file == filp && | |
1797 | fl->fl_owner == owner) { | |
1798 | victim = fl; | |
1799 | break; | |
1800 | } | |
1801 | } | |
1802 | trace_generic_delete_lease(inode, victim); | |
1803 | if (victim) | |
1804 | error = fl->fl_lmops->lm_change(victim, F_UNLCK, &dispose); | |
1805 | spin_unlock(&ctx->flc_lock); | |
1806 | percpu_up_read_preempt_enable(&file_rwsem); | |
1807 | locks_dispose_list(&dispose); | |
1808 | return error; | |
1809 | } | |
1810 | ||
1811 | /** | |
1812 | * generic_setlease - sets a lease on an open file | |
1813 | * @filp: file pointer | |
1814 | * @arg: type of lease to obtain | |
1815 | * @flp: input - file_lock to use, output - file_lock inserted | |
1816 | * @priv: private data for lm_setup (may be NULL if lm_setup | |
1817 | * doesn't require it) | |
1818 | * | |
1819 | * The (input) flp->fl_lmops->lm_break function is required | |
1820 | * by break_lease(). | |
1821 | */ | |
1822 | int generic_setlease(struct file *filp, long arg, struct file_lock **flp, | |
1823 | void **priv) | |
1824 | { | |
1825 | struct inode *inode = locks_inode(filp); | |
1826 | int error; | |
1827 | ||
1828 | if ((!uid_eq(current_fsuid(), inode->i_uid)) && !capable(CAP_LEASE)) | |
1829 | return -EACCES; | |
1830 | if (!S_ISREG(inode->i_mode)) | |
1831 | return -EINVAL; | |
1832 | error = security_file_lock(filp, arg); | |
1833 | if (error) | |
1834 | return error; | |
1835 | ||
1836 | switch (arg) { | |
1837 | case F_UNLCK: | |
1838 | return generic_delete_lease(filp, *priv); | |
1839 | case F_RDLCK: | |
1840 | case F_WRLCK: | |
1841 | if (!(*flp)->fl_lmops->lm_break) { | |
1842 | WARN_ON_ONCE(1); | |
1843 | return -ENOLCK; | |
1844 | } | |
1845 | ||
1846 | return generic_add_lease(filp, arg, flp, priv); | |
1847 | default: | |
1848 | return -EINVAL; | |
1849 | } | |
1850 | } | |
1851 | EXPORT_SYMBOL(generic_setlease); | |
1852 | ||
1853 | /** | |
1854 | * vfs_setlease - sets a lease on an open file | |
1855 | * @filp: file pointer | |
1856 | * @arg: type of lease to obtain | |
1857 | * @lease: file_lock to use when adding a lease | |
1858 | * @priv: private info for lm_setup when adding a lease (may be | |
1859 | * NULL if lm_setup doesn't require it) | |
1860 | * | |
1861 | * Call this to establish a lease on the file. The "lease" argument is not | |
1862 | * used for F_UNLCK requests and may be NULL. For commands that set or alter | |
1863 | * an existing lease, the ``(*lease)->fl_lmops->lm_break`` operation must be | |
1864 | * set; if not, this function will return -ENOLCK (and generate a scary-looking | |
1865 | * stack trace). | |
1866 | * | |
1867 | * The "priv" pointer is passed directly to the lm_setup function as-is. It | |
1868 | * may be NULL if the lm_setup operation doesn't require it. | |
1869 | */ | |
1870 | int | |
1871 | vfs_setlease(struct file *filp, long arg, struct file_lock **lease, void **priv) | |
1872 | { | |
1873 | if (filp->f_op->setlease) | |
1874 | return filp->f_op->setlease(filp, arg, lease, priv); | |
1875 | else | |
1876 | return generic_setlease(filp, arg, lease, priv); | |
1877 | } | |
1878 | EXPORT_SYMBOL_GPL(vfs_setlease); | |
1879 | ||
1880 | static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg) | |
1881 | { | |
1882 | struct file_lock *fl; | |
1883 | struct fasync_struct *new; | |
1884 | int error; | |
1885 | ||
1886 | fl = lease_alloc(filp, arg); | |
1887 | if (IS_ERR(fl)) | |
1888 | return PTR_ERR(fl); | |
1889 | ||
1890 | new = fasync_alloc(); | |
1891 | if (!new) { | |
1892 | locks_free_lock(fl); | |
1893 | return -ENOMEM; | |
1894 | } | |
1895 | new->fa_fd = fd; | |
1896 | ||
1897 | error = vfs_setlease(filp, arg, &fl, (void **)&new); | |
1898 | if (fl) | |
1899 | locks_free_lock(fl); | |
1900 | if (new) | |
1901 | fasync_free(new); | |
1902 | return error; | |
1903 | } | |
1904 | ||
1905 | /** | |
1906 | * fcntl_setlease - sets a lease on an open file | |
1907 | * @fd: open file descriptor | |
1908 | * @filp: file pointer | |
1909 | * @arg: type of lease to obtain | |
1910 | * | |
1911 | * Call this fcntl to establish a lease on the file. | |
1912 | * Note that you also need to call %F_SETSIG to | |
1913 | * receive a signal when the lease is broken. | |
1914 | */ | |
1915 | int fcntl_setlease(unsigned int fd, struct file *filp, long arg) | |
1916 | { | |
1917 | if (arg == F_UNLCK) | |
1918 | return vfs_setlease(filp, F_UNLCK, NULL, (void **)&filp); | |
1919 | return do_fcntl_add_lease(fd, filp, arg); | |
1920 | } | |
1921 | ||
1922 | /** | |
1923 | * flock_lock_inode_wait - Apply a FLOCK-style lock to a file | |
1924 | * @inode: inode of the file to apply to | |
1925 | * @fl: The lock to be applied | |
1926 | * | |
1927 | * Apply a FLOCK style lock request to an inode. | |
1928 | */ | |
1929 | static int flock_lock_inode_wait(struct inode *inode, struct file_lock *fl) | |
1930 | { | |
1931 | int error; | |
1932 | might_sleep(); | |
1933 | for (;;) { | |
1934 | error = flock_lock_inode(inode, fl); | |
1935 | if (error != FILE_LOCK_DEFERRED) | |
1936 | break; | |
1937 | error = wait_event_interruptible(fl->fl_wait, !fl->fl_blocker); | |
1938 | if (!error) | |
1939 | continue; | |
1940 | ||
1941 | locks_delete_block(fl); | |
1942 | break; | |
1943 | } | |
1944 | return error; | |
1945 | } | |
1946 | ||
1947 | /** | |
1948 | * locks_lock_inode_wait - Apply a lock to an inode | |
1949 | * @inode: inode of the file to apply to | |
1950 | * @fl: The lock to be applied | |
1951 | * | |
1952 | * Apply a POSIX or FLOCK style lock request to an inode. | |
1953 | */ | |
1954 | int locks_lock_inode_wait(struct inode *inode, struct file_lock *fl) | |
1955 | { | |
1956 | int res = 0; | |
1957 | switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) { | |
1958 | case FL_POSIX: | |
1959 | res = posix_lock_inode_wait(inode, fl); | |
1960 | break; | |
1961 | case FL_FLOCK: | |
1962 | res = flock_lock_inode_wait(inode, fl); | |
1963 | break; | |
1964 | default: | |
1965 | BUG(); | |
1966 | } | |
1967 | return res; | |
1968 | } | |
1969 | EXPORT_SYMBOL(locks_lock_inode_wait); | |
1970 | ||
1971 | /** | |
1972 | * sys_flock: - flock() system call. | |
1973 | * @fd: the file descriptor to lock. | |
1974 | * @cmd: the type of lock to apply. | |
1975 | * | |
1976 | * Apply a %FL_FLOCK style lock to an open file descriptor. | |
1977 | * The @cmd can be one of: | |
1978 | * | |
1979 | * - %LOCK_SH -- a shared lock. | |
1980 | * - %LOCK_EX -- an exclusive lock. | |
1981 | * - %LOCK_UN -- remove an existing lock. | |
1982 | * - %LOCK_MAND -- a 'mandatory' flock. | |
1983 | * This exists to emulate Windows Share Modes. | |
1984 | * | |
1985 | * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other | |
1986 | * processes read and write access respectively. | |
1987 | */ | |
1988 | SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd) | |
1989 | { | |
1990 | struct fd f = fdget(fd); | |
1991 | struct file_lock *lock; | |
1992 | int can_sleep, unlock; | |
1993 | int error; | |
1994 | ||
1995 | error = -EBADF; | |
1996 | if (!f.file) | |
1997 | goto out; | |
1998 | ||
1999 | can_sleep = !(cmd & LOCK_NB); | |
2000 | cmd &= ~LOCK_NB; | |
2001 | unlock = (cmd == LOCK_UN); | |
2002 | ||
2003 | if (!unlock && !(cmd & LOCK_MAND) && | |
2004 | !(f.file->f_mode & (FMODE_READ|FMODE_WRITE))) | |
2005 | goto out_putf; | |
2006 | ||
2007 | lock = flock_make_lock(f.file, cmd); | |
2008 | if (IS_ERR(lock)) { | |
2009 | error = PTR_ERR(lock); | |
2010 | goto out_putf; | |
2011 | } | |
2012 | ||
2013 | if (can_sleep) | |
2014 | lock->fl_flags |= FL_SLEEP; | |
2015 | ||
2016 | error = security_file_lock(f.file, lock->fl_type); | |
2017 | if (error) | |
2018 | goto out_free; | |
2019 | ||
2020 | if (f.file->f_op->flock) | |
2021 | error = f.file->f_op->flock(f.file, | |
2022 | (can_sleep) ? F_SETLKW : F_SETLK, | |
2023 | lock); | |
2024 | else | |
2025 | error = locks_lock_file_wait(f.file, lock); | |
2026 | ||
2027 | out_free: | |
2028 | locks_free_lock(lock); | |
2029 | ||
2030 | out_putf: | |
2031 | fdput(f); | |
2032 | out: | |
2033 | return error; | |
2034 | } | |
2035 | ||
2036 | /** | |
2037 | * vfs_test_lock - test file byte range lock | |
2038 | * @filp: The file to test lock for | |
2039 | * @fl: The lock to test; also used to hold result | |
2040 | * | |
2041 | * Returns -ERRNO on failure. Indicates presence of conflicting lock by | |
2042 | * setting conf->fl_type to something other than F_UNLCK. | |
2043 | */ | |
2044 | int vfs_test_lock(struct file *filp, struct file_lock *fl) | |
2045 | { | |
2046 | if (filp->f_op->lock) | |
2047 | return filp->f_op->lock(filp, F_GETLK, fl); | |
2048 | posix_test_lock(filp, fl); | |
2049 | return 0; | |
2050 | } | |
2051 | EXPORT_SYMBOL_GPL(vfs_test_lock); | |
2052 | ||
2053 | /** | |
2054 | * locks_translate_pid - translate a file_lock's fl_pid number into a namespace | |
2055 | * @fl: The file_lock who's fl_pid should be translated | |
2056 | * @ns: The namespace into which the pid should be translated | |
2057 | * | |
2058 | * Used to tranlate a fl_pid into a namespace virtual pid number | |
2059 | */ | |
2060 | static pid_t locks_translate_pid(struct file_lock *fl, struct pid_namespace *ns) | |
2061 | { | |
2062 | pid_t vnr; | |
2063 | struct pid *pid; | |
2064 | ||
2065 | if (IS_OFDLCK(fl)) | |
2066 | return -1; | |
2067 | if (IS_REMOTELCK(fl)) | |
2068 | return fl->fl_pid; | |
2069 | /* | |
2070 | * If the flock owner process is dead and its pid has been already | |
2071 | * freed, the translation below won't work, but we still want to show | |
2072 | * flock owner pid number in init pidns. | |
2073 | */ | |
2074 | if (ns == &init_pid_ns) | |
2075 | return (pid_t)fl->fl_pid; | |
2076 | ||
2077 | rcu_read_lock(); | |
2078 | pid = find_pid_ns(fl->fl_pid, &init_pid_ns); | |
2079 | vnr = pid_nr_ns(pid, ns); | |
2080 | rcu_read_unlock(); | |
2081 | return vnr; | |
2082 | } | |
2083 | ||
2084 | static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl) | |
2085 | { | |
2086 | flock->l_pid = locks_translate_pid(fl, task_active_pid_ns(current)); | |
2087 | #if BITS_PER_LONG == 32 | |
2088 | /* | |
2089 | * Make sure we can represent the posix lock via | |
2090 | * legacy 32bit flock. | |
2091 | */ | |
2092 | if (fl->fl_start > OFFT_OFFSET_MAX) | |
2093 | return -EOVERFLOW; | |
2094 | if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX) | |
2095 | return -EOVERFLOW; | |
2096 | #endif | |
2097 | flock->l_start = fl->fl_start; | |
2098 | flock->l_len = fl->fl_end == OFFSET_MAX ? 0 : | |
2099 | fl->fl_end - fl->fl_start + 1; | |
2100 | flock->l_whence = 0; | |
2101 | flock->l_type = fl->fl_type; | |
2102 | return 0; | |
2103 | } | |
2104 | ||
2105 | #if BITS_PER_LONG == 32 | |
2106 | static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl) | |
2107 | { | |
2108 | flock->l_pid = locks_translate_pid(fl, task_active_pid_ns(current)); | |
2109 | flock->l_start = fl->fl_start; | |
2110 | flock->l_len = fl->fl_end == OFFSET_MAX ? 0 : | |
2111 | fl->fl_end - fl->fl_start + 1; | |
2112 | flock->l_whence = 0; | |
2113 | flock->l_type = fl->fl_type; | |
2114 | } | |
2115 | #endif | |
2116 | ||
2117 | /* Report the first existing lock that would conflict with l. | |
2118 | * This implements the F_GETLK command of fcntl(). | |
2119 | */ | |
2120 | int fcntl_getlk(struct file *filp, unsigned int cmd, struct flock *flock) | |
2121 | { | |
2122 | struct file_lock *fl; | |
2123 | int error; | |
2124 | ||
2125 | fl = locks_alloc_lock(); | |
2126 | if (fl == NULL) | |
2127 | return -ENOMEM; | |
2128 | error = -EINVAL; | |
2129 | if (flock->l_type != F_RDLCK && flock->l_type != F_WRLCK) | |
2130 | goto out; | |
2131 | ||
2132 | error = flock_to_posix_lock(filp, fl, flock); | |
2133 | if (error) | |
2134 | goto out; | |
2135 | ||
2136 | if (cmd == F_OFD_GETLK) { | |
2137 | error = -EINVAL; | |
2138 | if (flock->l_pid != 0) | |
2139 | goto out; | |
2140 | ||
2141 | cmd = F_GETLK; | |
2142 | fl->fl_flags |= FL_OFDLCK; | |
2143 | fl->fl_owner = filp; | |
2144 | } | |
2145 | ||
2146 | error = vfs_test_lock(filp, fl); | |
2147 | if (error) | |
2148 | goto out; | |
2149 | ||
2150 | flock->l_type = fl->fl_type; | |
2151 | if (fl->fl_type != F_UNLCK) { | |
2152 | error = posix_lock_to_flock(flock, fl); | |
2153 | if (error) | |
2154 | goto out; | |
2155 | } | |
2156 | out: | |
2157 | locks_free_lock(fl); | |
2158 | return error; | |
2159 | } | |
2160 | ||
2161 | /** | |
2162 | * vfs_lock_file - file byte range lock | |
2163 | * @filp: The file to apply the lock to | |
2164 | * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.) | |
2165 | * @fl: The lock to be applied | |
2166 | * @conf: Place to return a copy of the conflicting lock, if found. | |
2167 | * | |
2168 | * A caller that doesn't care about the conflicting lock may pass NULL | |
2169 | * as the final argument. | |
2170 | * | |
2171 | * If the filesystem defines a private ->lock() method, then @conf will | |
2172 | * be left unchanged; so a caller that cares should initialize it to | |
2173 | * some acceptable default. | |
2174 | * | |
2175 | * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX | |
2176 | * locks, the ->lock() interface may return asynchronously, before the lock has | |
2177 | * been granted or denied by the underlying filesystem, if (and only if) | |
2178 | * lm_grant is set. Callers expecting ->lock() to return asynchronously | |
2179 | * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if) | |
2180 | * the request is for a blocking lock. When ->lock() does return asynchronously, | |
2181 | * it must return FILE_LOCK_DEFERRED, and call ->lm_grant() when the lock | |
2182 | * request completes. | |
2183 | * If the request is for non-blocking lock the file system should return | |
2184 | * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine | |
2185 | * with the result. If the request timed out the callback routine will return a | |
2186 | * nonzero return code and the file system should release the lock. The file | |
2187 | * system is also responsible to keep a corresponding posix lock when it | |
2188 | * grants a lock so the VFS can find out which locks are locally held and do | |
2189 | * the correct lock cleanup when required. | |
2190 | * The underlying filesystem must not drop the kernel lock or call | |
2191 | * ->lm_grant() before returning to the caller with a FILE_LOCK_DEFERRED | |
2192 | * return code. | |
2193 | */ | |
2194 | int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf) | |
2195 | { | |
2196 | if (filp->f_op->lock) | |
2197 | return filp->f_op->lock(filp, cmd, fl); | |
2198 | else | |
2199 | return posix_lock_file(filp, fl, conf); | |
2200 | } | |
2201 | EXPORT_SYMBOL_GPL(vfs_lock_file); | |
2202 | ||
2203 | static int do_lock_file_wait(struct file *filp, unsigned int cmd, | |
2204 | struct file_lock *fl) | |
2205 | { | |
2206 | int error; | |
2207 | ||
2208 | error = security_file_lock(filp, fl->fl_type); | |
2209 | if (error) | |
2210 | return error; | |
2211 | ||
2212 | for (;;) { | |
2213 | error = vfs_lock_file(filp, cmd, fl, NULL); | |
2214 | if (error != FILE_LOCK_DEFERRED) | |
2215 | break; | |
2216 | error = wait_event_interruptible(fl->fl_wait, !fl->fl_blocker); | |
2217 | if (!error) | |
2218 | continue; | |
2219 | ||
2220 | locks_delete_block(fl); | |
2221 | break; | |
2222 | } | |
2223 | ||
2224 | return error; | |
2225 | } | |
2226 | ||
2227 | /* Ensure that fl->fl_file has compatible f_mode for F_SETLK calls */ | |
2228 | static int | |
2229 | check_fmode_for_setlk(struct file_lock *fl) | |
2230 | { | |
2231 | switch (fl->fl_type) { | |
2232 | case F_RDLCK: | |
2233 | if (!(fl->fl_file->f_mode & FMODE_READ)) | |
2234 | return -EBADF; | |
2235 | break; | |
2236 | case F_WRLCK: | |
2237 | if (!(fl->fl_file->f_mode & FMODE_WRITE)) | |
2238 | return -EBADF; | |
2239 | } | |
2240 | return 0; | |
2241 | } | |
2242 | ||
2243 | /* Apply the lock described by l to an open file descriptor. | |
2244 | * This implements both the F_SETLK and F_SETLKW commands of fcntl(). | |
2245 | */ | |
2246 | int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd, | |
2247 | struct flock *flock) | |
2248 | { | |
2249 | struct file_lock *file_lock = locks_alloc_lock(); | |
2250 | struct inode *inode = locks_inode(filp); | |
2251 | struct file *f; | |
2252 | int error; | |
2253 | ||
2254 | if (file_lock == NULL) | |
2255 | return -ENOLCK; | |
2256 | ||
2257 | /* Don't allow mandatory locks on files that may be memory mapped | |
2258 | * and shared. | |
2259 | */ | |
2260 | if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) { | |
2261 | error = -EAGAIN; | |
2262 | goto out; | |
2263 | } | |
2264 | ||
2265 | error = flock_to_posix_lock(filp, file_lock, flock); | |
2266 | if (error) | |
2267 | goto out; | |
2268 | ||
2269 | error = check_fmode_for_setlk(file_lock); | |
2270 | if (error) | |
2271 | goto out; | |
2272 | ||
2273 | /* | |
2274 | * If the cmd is requesting file-private locks, then set the | |
2275 | * FL_OFDLCK flag and override the owner. | |
2276 | */ | |
2277 | switch (cmd) { | |
2278 | case F_OFD_SETLK: | |
2279 | error = -EINVAL; | |
2280 | if (flock->l_pid != 0) | |
2281 | goto out; | |
2282 | ||
2283 | cmd = F_SETLK; | |
2284 | file_lock->fl_flags |= FL_OFDLCK; | |
2285 | file_lock->fl_owner = filp; | |
2286 | break; | |
2287 | case F_OFD_SETLKW: | |
2288 | error = -EINVAL; | |
2289 | if (flock->l_pid != 0) | |
2290 | goto out; | |
2291 | ||
2292 | cmd = F_SETLKW; | |
2293 | file_lock->fl_flags |= FL_OFDLCK; | |
2294 | file_lock->fl_owner = filp; | |
2295 | /* Fallthrough */ | |
2296 | case F_SETLKW: | |
2297 | file_lock->fl_flags |= FL_SLEEP; | |
2298 | } | |
2299 | ||
2300 | error = do_lock_file_wait(filp, cmd, file_lock); | |
2301 | ||
2302 | /* | |
2303 | * Attempt to detect a close/fcntl race and recover by releasing the | |
2304 | * lock that was just acquired. There is no need to do that when we're | |
2305 | * unlocking though, or for OFD locks. | |
2306 | */ | |
2307 | if (!error && file_lock->fl_type != F_UNLCK && | |
2308 | !(file_lock->fl_flags & FL_OFDLCK)) { | |
2309 | /* | |
2310 | * We need that spin_lock here - it prevents reordering between | |
2311 | * update of i_flctx->flc_posix and check for it done in | |
2312 | * close(). rcu_read_lock() wouldn't do. | |
2313 | */ | |
2314 | spin_lock(¤t->files->file_lock); | |
2315 | f = fcheck(fd); | |
2316 | spin_unlock(¤t->files->file_lock); | |
2317 | if (f != filp) { | |
2318 | file_lock->fl_type = F_UNLCK; | |
2319 | error = do_lock_file_wait(filp, cmd, file_lock); | |
2320 | WARN_ON_ONCE(error); | |
2321 | error = -EBADF; | |
2322 | } | |
2323 | } | |
2324 | out: | |
2325 | trace_fcntl_setlk(inode, file_lock, error); | |
2326 | locks_free_lock(file_lock); | |
2327 | return error; | |
2328 | } | |
2329 | ||
2330 | #if BITS_PER_LONG == 32 | |
2331 | /* Report the first existing lock that would conflict with l. | |
2332 | * This implements the F_GETLK command of fcntl(). | |
2333 | */ | |
2334 | int fcntl_getlk64(struct file *filp, unsigned int cmd, struct flock64 *flock) | |
2335 | { | |
2336 | struct file_lock *fl; | |
2337 | int error; | |
2338 | ||
2339 | fl = locks_alloc_lock(); | |
2340 | if (fl == NULL) | |
2341 | return -ENOMEM; | |
2342 | ||
2343 | error = -EINVAL; | |
2344 | if (flock->l_type != F_RDLCK && flock->l_type != F_WRLCK) | |
2345 | goto out; | |
2346 | ||
2347 | error = flock64_to_posix_lock(filp, fl, flock); | |
2348 | if (error) | |
2349 | goto out; | |
2350 | ||
2351 | if (cmd == F_OFD_GETLK) { | |
2352 | error = -EINVAL; | |
2353 | if (flock->l_pid != 0) | |
2354 | goto out; | |
2355 | ||
2356 | cmd = F_GETLK64; | |
2357 | fl->fl_flags |= FL_OFDLCK; | |
2358 | fl->fl_owner = filp; | |
2359 | } | |
2360 | ||
2361 | error = vfs_test_lock(filp, fl); | |
2362 | if (error) | |
2363 | goto out; | |
2364 | ||
2365 | flock->l_type = fl->fl_type; | |
2366 | if (fl->fl_type != F_UNLCK) | |
2367 | posix_lock_to_flock64(flock, fl); | |
2368 | ||
2369 | out: | |
2370 | locks_free_lock(fl); | |
2371 | return error; | |
2372 | } | |
2373 | ||
2374 | /* Apply the lock described by l to an open file descriptor. | |
2375 | * This implements both the F_SETLK and F_SETLKW commands of fcntl(). | |
2376 | */ | |
2377 | int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd, | |
2378 | struct flock64 *flock) | |
2379 | { | |
2380 | struct file_lock *file_lock = locks_alloc_lock(); | |
2381 | struct inode *inode = locks_inode(filp); | |
2382 | struct file *f; | |
2383 | int error; | |
2384 | ||
2385 | if (file_lock == NULL) | |
2386 | return -ENOLCK; | |
2387 | ||
2388 | /* Don't allow mandatory locks on files that may be memory mapped | |
2389 | * and shared. | |
2390 | */ | |
2391 | if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) { | |
2392 | error = -EAGAIN; | |
2393 | goto out; | |
2394 | } | |
2395 | ||
2396 | error = flock64_to_posix_lock(filp, file_lock, flock); | |
2397 | if (error) | |
2398 | goto out; | |
2399 | ||
2400 | error = check_fmode_for_setlk(file_lock); | |
2401 | if (error) | |
2402 | goto out; | |
2403 | ||
2404 | /* | |
2405 | * If the cmd is requesting file-private locks, then set the | |
2406 | * FL_OFDLCK flag and override the owner. | |
2407 | */ | |
2408 | switch (cmd) { | |
2409 | case F_OFD_SETLK: | |
2410 | error = -EINVAL; | |
2411 | if (flock->l_pid != 0) | |
2412 | goto out; | |
2413 | ||
2414 | cmd = F_SETLK64; | |
2415 | file_lock->fl_flags |= FL_OFDLCK; | |
2416 | file_lock->fl_owner = filp; | |
2417 | break; | |
2418 | case F_OFD_SETLKW: | |
2419 | error = -EINVAL; | |
2420 | if (flock->l_pid != 0) | |
2421 | goto out; | |
2422 | ||
2423 | cmd = F_SETLKW64; | |
2424 | file_lock->fl_flags |= FL_OFDLCK; | |
2425 | file_lock->fl_owner = filp; | |
2426 | /* Fallthrough */ | |
2427 | case F_SETLKW64: | |
2428 | file_lock->fl_flags |= FL_SLEEP; | |
2429 | } | |
2430 | ||
2431 | error = do_lock_file_wait(filp, cmd, file_lock); | |
2432 | ||
2433 | /* | |
2434 | * Attempt to detect a close/fcntl race and recover by releasing the | |
2435 | * lock that was just acquired. There is no need to do that when we're | |
2436 | * unlocking though, or for OFD locks. | |
2437 | */ | |
2438 | if (!error && file_lock->fl_type != F_UNLCK && | |
2439 | !(file_lock->fl_flags & FL_OFDLCK)) { | |
2440 | /* | |
2441 | * We need that spin_lock here - it prevents reordering between | |
2442 | * update of i_flctx->flc_posix and check for it done in | |
2443 | * close(). rcu_read_lock() wouldn't do. | |
2444 | */ | |
2445 | spin_lock(¤t->files->file_lock); | |
2446 | f = fcheck(fd); | |
2447 | spin_unlock(¤t->files->file_lock); | |
2448 | if (f != filp) { | |
2449 | file_lock->fl_type = F_UNLCK; | |
2450 | error = do_lock_file_wait(filp, cmd, file_lock); | |
2451 | WARN_ON_ONCE(error); | |
2452 | error = -EBADF; | |
2453 | } | |
2454 | } | |
2455 | out: | |
2456 | locks_free_lock(file_lock); | |
2457 | return error; | |
2458 | } | |
2459 | #endif /* BITS_PER_LONG == 32 */ | |
2460 | ||
2461 | /* | |
2462 | * This function is called when the file is being removed | |
2463 | * from the task's fd array. POSIX locks belonging to this task | |
2464 | * are deleted at this time. | |
2465 | */ | |
2466 | void locks_remove_posix(struct file *filp, fl_owner_t owner) | |
2467 | { | |
2468 | int error; | |
2469 | struct inode *inode = locks_inode(filp); | |
2470 | struct file_lock lock; | |
2471 | struct file_lock_context *ctx; | |
2472 | ||
2473 | /* | |
2474 | * If there are no locks held on this file, we don't need to call | |
2475 | * posix_lock_file(). Another process could be setting a lock on this | |
2476 | * file at the same time, but we wouldn't remove that lock anyway. | |
2477 | */ | |
2478 | ctx = smp_load_acquire(&inode->i_flctx); | |
2479 | if (!ctx || list_empty(&ctx->flc_posix)) | |
2480 | return; | |
2481 | ||
2482 | lock.fl_type = F_UNLCK; | |
2483 | lock.fl_flags = FL_POSIX | FL_CLOSE; | |
2484 | lock.fl_start = 0; | |
2485 | lock.fl_end = OFFSET_MAX; | |
2486 | lock.fl_owner = owner; | |
2487 | lock.fl_pid = current->tgid; | |
2488 | lock.fl_file = filp; | |
2489 | lock.fl_ops = NULL; | |
2490 | lock.fl_lmops = NULL; | |
2491 | ||
2492 | error = vfs_lock_file(filp, F_SETLK, &lock, NULL); | |
2493 | ||
2494 | if (lock.fl_ops && lock.fl_ops->fl_release_private) | |
2495 | lock.fl_ops->fl_release_private(&lock); | |
2496 | trace_locks_remove_posix(inode, &lock, error); | |
2497 | } | |
2498 | ||
2499 | EXPORT_SYMBOL(locks_remove_posix); | |
2500 | ||
2501 | /* The i_flctx must be valid when calling into here */ | |
2502 | static void | |
2503 | locks_remove_flock(struct file *filp, struct file_lock_context *flctx) | |
2504 | { | |
2505 | struct file_lock fl = { | |
2506 | .fl_owner = filp, | |
2507 | .fl_pid = current->tgid, | |
2508 | .fl_file = filp, | |
2509 | .fl_flags = FL_FLOCK | FL_CLOSE, | |
2510 | .fl_type = F_UNLCK, | |
2511 | .fl_end = OFFSET_MAX, | |
2512 | }; | |
2513 | struct inode *inode = locks_inode(filp); | |
2514 | ||
2515 | if (list_empty(&flctx->flc_flock)) | |
2516 | return; | |
2517 | ||
2518 | if (filp->f_op->flock) | |
2519 | filp->f_op->flock(filp, F_SETLKW, &fl); | |
2520 | else | |
2521 | flock_lock_inode(inode, &fl); | |
2522 | ||
2523 | if (fl.fl_ops && fl.fl_ops->fl_release_private) | |
2524 | fl.fl_ops->fl_release_private(&fl); | |
2525 | } | |
2526 | ||
2527 | /* The i_flctx must be valid when calling into here */ | |
2528 | static void | |
2529 | locks_remove_lease(struct file *filp, struct file_lock_context *ctx) | |
2530 | { | |
2531 | struct file_lock *fl, *tmp; | |
2532 | LIST_HEAD(dispose); | |
2533 | ||
2534 | if (list_empty(&ctx->flc_lease)) | |
2535 | return; | |
2536 | ||
2537 | percpu_down_read_preempt_disable(&file_rwsem); | |
2538 | spin_lock(&ctx->flc_lock); | |
2539 | list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list) | |
2540 | if (filp == fl->fl_file) | |
2541 | lease_modify(fl, F_UNLCK, &dispose); | |
2542 | spin_unlock(&ctx->flc_lock); | |
2543 | percpu_up_read_preempt_enable(&file_rwsem); | |
2544 | ||
2545 | locks_dispose_list(&dispose); | |
2546 | } | |
2547 | ||
2548 | /* | |
2549 | * This function is called on the last close of an open file. | |
2550 | */ | |
2551 | void locks_remove_file(struct file *filp) | |
2552 | { | |
2553 | struct file_lock_context *ctx; | |
2554 | ||
2555 | ctx = smp_load_acquire(&locks_inode(filp)->i_flctx); | |
2556 | if (!ctx) | |
2557 | return; | |
2558 | ||
2559 | /* remove any OFD locks */ | |
2560 | locks_remove_posix(filp, filp); | |
2561 | ||
2562 | /* remove flock locks */ | |
2563 | locks_remove_flock(filp, ctx); | |
2564 | ||
2565 | /* remove any leases */ | |
2566 | locks_remove_lease(filp, ctx); | |
2567 | ||
2568 | spin_lock(&ctx->flc_lock); | |
2569 | locks_check_ctx_file_list(filp, &ctx->flc_posix, "POSIX"); | |
2570 | locks_check_ctx_file_list(filp, &ctx->flc_flock, "FLOCK"); | |
2571 | locks_check_ctx_file_list(filp, &ctx->flc_lease, "LEASE"); | |
2572 | spin_unlock(&ctx->flc_lock); | |
2573 | } | |
2574 | ||
2575 | /** | |
2576 | * posix_unblock_lock - stop waiting for a file lock | |
2577 | * @waiter: the lock which was waiting | |
2578 | * | |
2579 | * lockd needs to block waiting for locks. | |
2580 | */ | |
2581 | int | |
2582 | posix_unblock_lock(struct file_lock *waiter) | |
2583 | { | |
2584 | int status = 0; | |
2585 | ||
2586 | spin_lock(&blocked_lock_lock); | |
2587 | if (waiter->fl_blocker) | |
2588 | __locks_delete_block(waiter); | |
2589 | else | |
2590 | status = -ENOENT; | |
2591 | spin_unlock(&blocked_lock_lock); | |
2592 | return status; | |
2593 | } | |
2594 | EXPORT_SYMBOL(posix_unblock_lock); | |
2595 | ||
2596 | /** | |
2597 | * vfs_cancel_lock - file byte range unblock lock | |
2598 | * @filp: The file to apply the unblock to | |
2599 | * @fl: The lock to be unblocked | |
2600 | * | |
2601 | * Used by lock managers to cancel blocked requests | |
2602 | */ | |
2603 | int vfs_cancel_lock(struct file *filp, struct file_lock *fl) | |
2604 | { | |
2605 | if (filp->f_op->lock) | |
2606 | return filp->f_op->lock(filp, F_CANCELLK, fl); | |
2607 | return 0; | |
2608 | } | |
2609 | ||
2610 | EXPORT_SYMBOL_GPL(vfs_cancel_lock); | |
2611 | ||
2612 | #ifdef CONFIG_PROC_FS | |
2613 | #include <linux/proc_fs.h> | |
2614 | #include <linux/seq_file.h> | |
2615 | ||
2616 | struct locks_iterator { | |
2617 | int li_cpu; | |
2618 | loff_t li_pos; | |
2619 | }; | |
2620 | ||
2621 | static void lock_get_status(struct seq_file *f, struct file_lock *fl, | |
2622 | loff_t id, char *pfx) | |
2623 | { | |
2624 | struct inode *inode = NULL; | |
2625 | unsigned int fl_pid; | |
2626 | struct pid_namespace *proc_pidns = file_inode(f->file)->i_sb->s_fs_info; | |
2627 | ||
2628 | fl_pid = locks_translate_pid(fl, proc_pidns); | |
2629 | /* | |
2630 | * If lock owner is dead (and pid is freed) or not visible in current | |
2631 | * pidns, zero is shown as a pid value. Check lock info from | |
2632 | * init_pid_ns to get saved lock pid value. | |
2633 | */ | |
2634 | ||
2635 | if (fl->fl_file != NULL) | |
2636 | inode = locks_inode(fl->fl_file); | |
2637 | ||
2638 | seq_printf(f, "%lld:%s ", id, pfx); | |
2639 | if (IS_POSIX(fl)) { | |
2640 | if (fl->fl_flags & FL_ACCESS) | |
2641 | seq_puts(f, "ACCESS"); | |
2642 | else if (IS_OFDLCK(fl)) | |
2643 | seq_puts(f, "OFDLCK"); | |
2644 | else | |
2645 | seq_puts(f, "POSIX "); | |
2646 | ||
2647 | seq_printf(f, " %s ", | |
2648 | (inode == NULL) ? "*NOINODE*" : | |
2649 | mandatory_lock(inode) ? "MANDATORY" : "ADVISORY "); | |
2650 | } else if (IS_FLOCK(fl)) { | |
2651 | if (fl->fl_type & LOCK_MAND) { | |
2652 | seq_puts(f, "FLOCK MSNFS "); | |
2653 | } else { | |
2654 | seq_puts(f, "FLOCK ADVISORY "); | |
2655 | } | |
2656 | } else if (IS_LEASE(fl)) { | |
2657 | if (fl->fl_flags & FL_DELEG) | |
2658 | seq_puts(f, "DELEG "); | |
2659 | else | |
2660 | seq_puts(f, "LEASE "); | |
2661 | ||
2662 | if (lease_breaking(fl)) | |
2663 | seq_puts(f, "BREAKING "); | |
2664 | else if (fl->fl_file) | |
2665 | seq_puts(f, "ACTIVE "); | |
2666 | else | |
2667 | seq_puts(f, "BREAKER "); | |
2668 | } else { | |
2669 | seq_puts(f, "UNKNOWN UNKNOWN "); | |
2670 | } | |
2671 | if (fl->fl_type & LOCK_MAND) { | |
2672 | seq_printf(f, "%s ", | |
2673 | (fl->fl_type & LOCK_READ) | |
2674 | ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ " | |
2675 | : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE "); | |
2676 | } else { | |
2677 | seq_printf(f, "%s ", | |
2678 | (lease_breaking(fl)) | |
2679 | ? (fl->fl_type == F_UNLCK) ? "UNLCK" : "READ " | |
2680 | : (fl->fl_type == F_WRLCK) ? "WRITE" : "READ "); | |
2681 | } | |
2682 | if (inode) { | |
2683 | /* userspace relies on this representation of dev_t */ | |
2684 | seq_printf(f, "%d %02x:%02x:%ld ", fl_pid, | |
2685 | MAJOR(inode->i_sb->s_dev), | |
2686 | MINOR(inode->i_sb->s_dev), inode->i_ino); | |
2687 | } else { | |
2688 | seq_printf(f, "%d <none>:0 ", fl_pid); | |
2689 | } | |
2690 | if (IS_POSIX(fl)) { | |
2691 | if (fl->fl_end == OFFSET_MAX) | |
2692 | seq_printf(f, "%Ld EOF\n", fl->fl_start); | |
2693 | else | |
2694 | seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end); | |
2695 | } else { | |
2696 | seq_puts(f, "0 EOF\n"); | |
2697 | } | |
2698 | } | |
2699 | ||
2700 | static int locks_show(struct seq_file *f, void *v) | |
2701 | { | |
2702 | struct locks_iterator *iter = f->private; | |
2703 | struct file_lock *fl, *bfl; | |
2704 | struct pid_namespace *proc_pidns = file_inode(f->file)->i_sb->s_fs_info; | |
2705 | ||
2706 | fl = hlist_entry(v, struct file_lock, fl_link); | |
2707 | ||
2708 | if (locks_translate_pid(fl, proc_pidns) == 0) | |
2709 | return 0; | |
2710 | ||
2711 | lock_get_status(f, fl, iter->li_pos, ""); | |
2712 | ||
2713 | list_for_each_entry(bfl, &fl->fl_blocked_requests, fl_blocked_member) | |
2714 | lock_get_status(f, bfl, iter->li_pos, " ->"); | |
2715 | ||
2716 | return 0; | |
2717 | } | |
2718 | ||
2719 | static void __show_fd_locks(struct seq_file *f, | |
2720 | struct list_head *head, int *id, | |
2721 | struct file *filp, struct files_struct *files) | |
2722 | { | |
2723 | struct file_lock *fl; | |
2724 | ||
2725 | list_for_each_entry(fl, head, fl_list) { | |
2726 | ||
2727 | if (filp != fl->fl_file) | |
2728 | continue; | |
2729 | if (fl->fl_owner != files && | |
2730 | fl->fl_owner != filp) | |
2731 | continue; | |
2732 | ||
2733 | (*id)++; | |
2734 | seq_puts(f, "lock:\t"); | |
2735 | lock_get_status(f, fl, *id, ""); | |
2736 | } | |
2737 | } | |
2738 | ||
2739 | void show_fd_locks(struct seq_file *f, | |
2740 | struct file *filp, struct files_struct *files) | |
2741 | { | |
2742 | struct inode *inode = locks_inode(filp); | |
2743 | struct file_lock_context *ctx; | |
2744 | int id = 0; | |
2745 | ||
2746 | ctx = smp_load_acquire(&inode->i_flctx); | |
2747 | if (!ctx) | |
2748 | return; | |
2749 | ||
2750 | spin_lock(&ctx->flc_lock); | |
2751 | __show_fd_locks(f, &ctx->flc_flock, &id, filp, files); | |
2752 | __show_fd_locks(f, &ctx->flc_posix, &id, filp, files); | |
2753 | __show_fd_locks(f, &ctx->flc_lease, &id, filp, files); | |
2754 | spin_unlock(&ctx->flc_lock); | |
2755 | } | |
2756 | ||
2757 | static void *locks_start(struct seq_file *f, loff_t *pos) | |
2758 | __acquires(&blocked_lock_lock) | |
2759 | { | |
2760 | struct locks_iterator *iter = f->private; | |
2761 | ||
2762 | iter->li_pos = *pos + 1; | |
2763 | percpu_down_write(&file_rwsem); | |
2764 | spin_lock(&blocked_lock_lock); | |
2765 | return seq_hlist_start_percpu(&file_lock_list.hlist, &iter->li_cpu, *pos); | |
2766 | } | |
2767 | ||
2768 | static void *locks_next(struct seq_file *f, void *v, loff_t *pos) | |
2769 | { | |
2770 | struct locks_iterator *iter = f->private; | |
2771 | ||
2772 | ++iter->li_pos; | |
2773 | return seq_hlist_next_percpu(v, &file_lock_list.hlist, &iter->li_cpu, pos); | |
2774 | } | |
2775 | ||
2776 | static void locks_stop(struct seq_file *f, void *v) | |
2777 | __releases(&blocked_lock_lock) | |
2778 | { | |
2779 | spin_unlock(&blocked_lock_lock); | |
2780 | percpu_up_write(&file_rwsem); | |
2781 | } | |
2782 | ||
2783 | static const struct seq_operations locks_seq_operations = { | |
2784 | .start = locks_start, | |
2785 | .next = locks_next, | |
2786 | .stop = locks_stop, | |
2787 | .show = locks_show, | |
2788 | }; | |
2789 | ||
2790 | static int __init proc_locks_init(void) | |
2791 | { | |
2792 | proc_create_seq_private("locks", 0, NULL, &locks_seq_operations, | |
2793 | sizeof(struct locks_iterator), NULL); | |
2794 | return 0; | |
2795 | } | |
2796 | fs_initcall(proc_locks_init); | |
2797 | #endif | |
2798 | ||
2799 | static int __init filelock_init(void) | |
2800 | { | |
2801 | int i; | |
2802 | ||
2803 | flctx_cache = kmem_cache_create("file_lock_ctx", | |
2804 | sizeof(struct file_lock_context), 0, SLAB_PANIC, NULL); | |
2805 | ||
2806 | filelock_cache = kmem_cache_create("file_lock_cache", | |
2807 | sizeof(struct file_lock), 0, SLAB_PANIC, NULL); | |
2808 | ||
2809 | ||
2810 | for_each_possible_cpu(i) { | |
2811 | struct file_lock_list_struct *fll = per_cpu_ptr(&file_lock_list, i); | |
2812 | ||
2813 | spin_lock_init(&fll->lock); | |
2814 | INIT_HLIST_HEAD(&fll->hlist); | |
2815 | } | |
2816 | ||
2817 | return 0; | |
2818 | } | |
2819 | ||
2820 | core_initcall(filelock_init); |