1 | // SPDX-License-Identifier: GPL-2.0-only |
2 | /* |
3 | * linux/fs/locks.c |
4 | * |
5 | * We implement four types of file locks: BSD locks, posix locks, open |
6 | * file description locks, and leases. For details about BSD locks, |
7 | * see the flock(2) man page; for details about the other three, see |
8 | * fcntl(2). |
9 | * |
10 | * |
11 | * Locking conflicts and dependencies: |
12 | * If multiple threads attempt to lock the same byte (or flock the same file) |
13 | * only one can be granted the lock, and other must wait their turn. |
14 | * The first lock has been "applied" or "granted", the others are "waiting" |
15 | * and are "blocked" by the "applied" lock.. |
16 | * |
17 | * Waiting and applied locks are all kept in trees whose properties are: |
18 | * |
19 | * - the root of a tree may be an applied or waiting lock. |
20 | * - every other node in the tree is a waiting lock that |
21 | * conflicts with every ancestor of that node. |
22 | * |
23 | * Every such tree begins life as a waiting singleton which obviously |
24 | * satisfies the above properties. |
25 | * |
26 | * The only ways we modify trees preserve these properties: |
27 | * |
28 | * 1. We may add a new leaf node, but only after first verifying that it |
29 | * conflicts with all of its ancestors. |
30 | * 2. We may remove the root of a tree, creating a new singleton |
31 | * tree from the root and N new trees rooted in the immediate |
32 | * children. |
33 | * 3. If the root of a tree is not currently an applied lock, we may |
34 | * apply it (if possible). |
35 | * 4. We may upgrade the root of the tree (either extend its range, |
36 | * or upgrade its entire range from read to write). |
37 | * |
38 | * When an applied lock is modified in a way that reduces or downgrades any |
39 | * part of its range, we remove all its children (2 above). This particularly |
40 | * happens when a lock is unlocked. |
41 | * |
42 | * For each of those child trees we "wake up" the thread which is |
43 | * waiting for the lock so it can continue handling as follows: if the |
44 | * root of the tree applies, we do so (3). If it doesn't, it must |
45 | * conflict with some applied lock. We remove (wake up) all of its children |
46 | * (2), and add it is a new leaf to the tree rooted in the applied |
47 | * lock (1). We then repeat the process recursively with those |
48 | * children. |
49 | * |
50 | */ |
51 | |
52 | #include <linux/capability.h> |
53 | #include <linux/file.h> |
54 | #include <linux/fdtable.h> |
55 | #include <linux/filelock.h> |
56 | #include <linux/fs.h> |
57 | #include <linux/init.h> |
58 | #include <linux/security.h> |
59 | #include <linux/slab.h> |
60 | #include <linux/syscalls.h> |
61 | #include <linux/time.h> |
62 | #include <linux/rcupdate.h> |
63 | #include <linux/pid_namespace.h> |
64 | #include <linux/hashtable.h> |
65 | #include <linux/percpu.h> |
66 | #include <linux/sysctl.h> |
67 | |
68 | #define CREATE_TRACE_POINTS |
69 | #include <trace/events/filelock.h> |
70 | |
71 | #include <linux/uaccess.h> |
72 | |
73 | #define IS_POSIX(fl) (fl->fl_flags & FL_POSIX) |
74 | #define IS_FLOCK(fl) (fl->fl_flags & FL_FLOCK) |
75 | #define IS_LEASE(fl) (fl->fl_flags & (FL_LEASE|FL_DELEG|FL_LAYOUT)) |
76 | #define IS_OFDLCK(fl) (fl->fl_flags & FL_OFDLCK) |
77 | #define IS_REMOTELCK(fl) (fl->fl_pid <= 0) |
78 | |
79 | static bool lease_breaking(struct file_lock *fl) |
80 | { |
81 | return fl->fl_flags & (FL_UNLOCK_PENDING | FL_DOWNGRADE_PENDING); |
82 | } |
83 | |
84 | static int target_leasetype(struct file_lock *fl) |
85 | { |
86 | if (fl->fl_flags & FL_UNLOCK_PENDING) |
87 | return F_UNLCK; |
88 | if (fl->fl_flags & FL_DOWNGRADE_PENDING) |
89 | return F_RDLCK; |
90 | return fl->fl_type; |
91 | } |
92 | |
93 | static int leases_enable = 1; |
94 | static int lease_break_time = 45; |
95 | |
96 | #ifdef CONFIG_SYSCTL |
97 | static struct ctl_table locks_sysctls[] = { |
98 | { |
99 | .procname = "leases-enable" , |
100 | .data = &leases_enable, |
101 | .maxlen = sizeof(int), |
102 | .mode = 0644, |
103 | .proc_handler = proc_dointvec, |
104 | }, |
105 | #ifdef CONFIG_MMU |
106 | { |
107 | .procname = "lease-break-time" , |
108 | .data = &lease_break_time, |
109 | .maxlen = sizeof(int), |
110 | .mode = 0644, |
111 | .proc_handler = proc_dointvec, |
112 | }, |
113 | #endif /* CONFIG_MMU */ |
114 | {} |
115 | }; |
116 | |
117 | static int __init init_fs_locks_sysctls(void) |
118 | { |
119 | register_sysctl_init("fs" , locks_sysctls); |
120 | return 0; |
121 | } |
122 | early_initcall(init_fs_locks_sysctls); |
123 | #endif /* CONFIG_SYSCTL */ |
124 | |
125 | /* |
126 | * The global file_lock_list is only used for displaying /proc/locks, so we |
127 | * keep a list on each CPU, with each list protected by its own spinlock. |
128 | * Global serialization is done using file_rwsem. |
129 | * |
130 | * Note that alterations to the list also require that the relevant flc_lock is |
131 | * held. |
132 | */ |
133 | struct file_lock_list_struct { |
134 | spinlock_t lock; |
135 | struct hlist_head hlist; |
136 | }; |
137 | static DEFINE_PER_CPU(struct file_lock_list_struct, file_lock_list); |
138 | DEFINE_STATIC_PERCPU_RWSEM(file_rwsem); |
139 | |
140 | |
141 | /* |
142 | * The blocked_hash is used to find POSIX lock loops for deadlock detection. |
143 | * It is protected by blocked_lock_lock. |
144 | * |
145 | * We hash locks by lockowner in order to optimize searching for the lock a |
146 | * particular lockowner is waiting on. |
147 | * |
148 | * FIXME: make this value scale via some heuristic? We generally will want more |
149 | * buckets when we have more lockowners holding locks, but that's a little |
150 | * difficult to determine without knowing what the workload will look like. |
151 | */ |
152 | #define BLOCKED_HASH_BITS 7 |
153 | static DEFINE_HASHTABLE(blocked_hash, BLOCKED_HASH_BITS); |
154 | |
155 | /* |
156 | * This lock protects the blocked_hash. Generally, if you're accessing it, you |
157 | * want to be holding this lock. |
158 | * |
159 | * In addition, it also protects the fl->fl_blocked_requests list, and the |
160 | * fl->fl_blocker pointer for file_lock structures that are acting as lock |
161 | * requests (in contrast to those that are acting as records of acquired locks). |
162 | * |
163 | * Note that when we acquire this lock in order to change the above fields, |
164 | * we often hold the flc_lock as well. In certain cases, when reading the fields |
165 | * protected by this lock, we can skip acquiring it iff we already hold the |
166 | * flc_lock. |
167 | */ |
168 | static DEFINE_SPINLOCK(blocked_lock_lock); |
169 | |
170 | static struct kmem_cache *flctx_cache __ro_after_init; |
171 | static struct kmem_cache *filelock_cache __ro_after_init; |
172 | |
173 | static struct file_lock_context * |
174 | locks_get_lock_context(struct inode *inode, int type) |
175 | { |
176 | struct file_lock_context *ctx; |
177 | |
178 | /* paired with cmpxchg() below */ |
179 | ctx = locks_inode_context(inode); |
180 | if (likely(ctx) || type == F_UNLCK) |
181 | goto out; |
182 | |
183 | ctx = kmem_cache_alloc(cachep: flctx_cache, GFP_KERNEL); |
184 | if (!ctx) |
185 | goto out; |
186 | |
187 | spin_lock_init(&ctx->flc_lock); |
188 | INIT_LIST_HEAD(list: &ctx->flc_flock); |
189 | INIT_LIST_HEAD(list: &ctx->flc_posix); |
190 | INIT_LIST_HEAD(list: &ctx->flc_lease); |
191 | |
192 | /* |
193 | * Assign the pointer if it's not already assigned. If it is, then |
194 | * free the context we just allocated. |
195 | */ |
196 | if (cmpxchg(&inode->i_flctx, NULL, ctx)) { |
197 | kmem_cache_free(s: flctx_cache, objp: ctx); |
198 | ctx = locks_inode_context(inode); |
199 | } |
200 | out: |
201 | trace_locks_get_lock_context(inode, type, ctx); |
202 | return ctx; |
203 | } |
204 | |
205 | static void |
206 | locks_dump_ctx_list(struct list_head *list, char *list_type) |
207 | { |
208 | struct file_lock *fl; |
209 | |
210 | list_for_each_entry(fl, list, fl_list) { |
211 | 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); |
212 | } |
213 | } |
214 | |
215 | static void |
216 | locks_check_ctx_lists(struct inode *inode) |
217 | { |
218 | struct file_lock_context *ctx = inode->i_flctx; |
219 | |
220 | if (unlikely(!list_empty(&ctx->flc_flock) || |
221 | !list_empty(&ctx->flc_posix) || |
222 | !list_empty(&ctx->flc_lease))) { |
223 | pr_warn("Leaked locks on dev=0x%x:0x%x ino=0x%lx:\n" , |
224 | MAJOR(inode->i_sb->s_dev), MINOR(inode->i_sb->s_dev), |
225 | inode->i_ino); |
226 | locks_dump_ctx_list(list: &ctx->flc_flock, list_type: "FLOCK" ); |
227 | locks_dump_ctx_list(list: &ctx->flc_posix, list_type: "POSIX" ); |
228 | locks_dump_ctx_list(list: &ctx->flc_lease, list_type: "LEASE" ); |
229 | } |
230 | } |
231 | |
232 | static void |
233 | locks_check_ctx_file_list(struct file *filp, struct list_head *list, |
234 | char *list_type) |
235 | { |
236 | struct file_lock *fl; |
237 | struct inode *inode = file_inode(f: filp); |
238 | |
239 | list_for_each_entry(fl, list, fl_list) |
240 | if (fl->fl_file == filp) |
241 | pr_warn("Leaked %s lock on dev=0x%x:0x%x ino=0x%lx " |
242 | " fl_owner=%p fl_flags=0x%x fl_type=0x%x fl_pid=%u\n" , |
243 | list_type, MAJOR(inode->i_sb->s_dev), |
244 | MINOR(inode->i_sb->s_dev), inode->i_ino, |
245 | fl->fl_owner, fl->fl_flags, fl->fl_type, fl->fl_pid); |
246 | } |
247 | |
248 | void |
249 | locks_free_lock_context(struct inode *inode) |
250 | { |
251 | struct file_lock_context *ctx = locks_inode_context(inode); |
252 | |
253 | if (unlikely(ctx)) { |
254 | locks_check_ctx_lists(inode); |
255 | kmem_cache_free(s: flctx_cache, objp: ctx); |
256 | } |
257 | } |
258 | |
259 | static void locks_init_lock_heads(struct file_lock *fl) |
260 | { |
261 | INIT_HLIST_NODE(h: &fl->fl_link); |
262 | INIT_LIST_HEAD(list: &fl->fl_list); |
263 | INIT_LIST_HEAD(list: &fl->fl_blocked_requests); |
264 | INIT_LIST_HEAD(list: &fl->fl_blocked_member); |
265 | init_waitqueue_head(&fl->fl_wait); |
266 | } |
267 | |
268 | /* Allocate an empty lock structure. */ |
269 | struct file_lock *locks_alloc_lock(void) |
270 | { |
271 | struct file_lock *fl = kmem_cache_zalloc(k: filelock_cache, GFP_KERNEL); |
272 | |
273 | if (fl) |
274 | locks_init_lock_heads(fl); |
275 | |
276 | return fl; |
277 | } |
278 | EXPORT_SYMBOL_GPL(locks_alloc_lock); |
279 | |
280 | void locks_release_private(struct file_lock *fl) |
281 | { |
282 | BUG_ON(waitqueue_active(&fl->fl_wait)); |
283 | BUG_ON(!list_empty(&fl->fl_list)); |
284 | BUG_ON(!list_empty(&fl->fl_blocked_requests)); |
285 | BUG_ON(!list_empty(&fl->fl_blocked_member)); |
286 | BUG_ON(!hlist_unhashed(&fl->fl_link)); |
287 | |
288 | if (fl->fl_ops) { |
289 | if (fl->fl_ops->fl_release_private) |
290 | fl->fl_ops->fl_release_private(fl); |
291 | fl->fl_ops = NULL; |
292 | } |
293 | |
294 | if (fl->fl_lmops) { |
295 | if (fl->fl_lmops->lm_put_owner) { |
296 | fl->fl_lmops->lm_put_owner(fl->fl_owner); |
297 | fl->fl_owner = NULL; |
298 | } |
299 | fl->fl_lmops = NULL; |
300 | } |
301 | } |
302 | EXPORT_SYMBOL_GPL(locks_release_private); |
303 | |
304 | /** |
305 | * locks_owner_has_blockers - Check for blocking lock requests |
306 | * @flctx: file lock context |
307 | * @owner: lock owner |
308 | * |
309 | * Return values: |
310 | * %true: @owner has at least one blocker |
311 | * %false: @owner has no blockers |
312 | */ |
313 | bool locks_owner_has_blockers(struct file_lock_context *flctx, |
314 | fl_owner_t owner) |
315 | { |
316 | struct file_lock *fl; |
317 | |
318 | spin_lock(lock: &flctx->flc_lock); |
319 | list_for_each_entry(fl, &flctx->flc_posix, fl_list) { |
320 | if (fl->fl_owner != owner) |
321 | continue; |
322 | if (!list_empty(head: &fl->fl_blocked_requests)) { |
323 | spin_unlock(lock: &flctx->flc_lock); |
324 | return true; |
325 | } |
326 | } |
327 | spin_unlock(lock: &flctx->flc_lock); |
328 | return false; |
329 | } |
330 | EXPORT_SYMBOL_GPL(locks_owner_has_blockers); |
331 | |
332 | /* Free a lock which is not in use. */ |
333 | void locks_free_lock(struct file_lock *fl) |
334 | { |
335 | locks_release_private(fl); |
336 | kmem_cache_free(s: filelock_cache, objp: fl); |
337 | } |
338 | EXPORT_SYMBOL(locks_free_lock); |
339 | |
340 | static void |
341 | locks_dispose_list(struct list_head *dispose) |
342 | { |
343 | struct file_lock *fl; |
344 | |
345 | while (!list_empty(head: dispose)) { |
346 | fl = list_first_entry(dispose, struct file_lock, fl_list); |
347 | list_del_init(entry: &fl->fl_list); |
348 | locks_free_lock(fl); |
349 | } |
350 | } |
351 | |
352 | void locks_init_lock(struct file_lock *fl) |
353 | { |
354 | memset(fl, 0, sizeof(struct file_lock)); |
355 | locks_init_lock_heads(fl); |
356 | } |
357 | EXPORT_SYMBOL(locks_init_lock); |
358 | |
359 | /* |
360 | * Initialize a new lock from an existing file_lock structure. |
361 | */ |
362 | void locks_copy_conflock(struct file_lock *new, struct file_lock *fl) |
363 | { |
364 | new->fl_owner = fl->fl_owner; |
365 | new->fl_pid = fl->fl_pid; |
366 | new->fl_file = NULL; |
367 | new->fl_flags = fl->fl_flags; |
368 | new->fl_type = fl->fl_type; |
369 | new->fl_start = fl->fl_start; |
370 | new->fl_end = fl->fl_end; |
371 | new->fl_lmops = fl->fl_lmops; |
372 | new->fl_ops = NULL; |
373 | |
374 | if (fl->fl_lmops) { |
375 | if (fl->fl_lmops->lm_get_owner) |
376 | fl->fl_lmops->lm_get_owner(fl->fl_owner); |
377 | } |
378 | } |
379 | EXPORT_SYMBOL(locks_copy_conflock); |
380 | |
381 | void locks_copy_lock(struct file_lock *new, struct file_lock *fl) |
382 | { |
383 | /* "new" must be a freshly-initialized lock */ |
384 | WARN_ON_ONCE(new->fl_ops); |
385 | |
386 | locks_copy_conflock(new, fl); |
387 | |
388 | new->fl_file = fl->fl_file; |
389 | new->fl_ops = fl->fl_ops; |
390 | |
391 | if (fl->fl_ops) { |
392 | if (fl->fl_ops->fl_copy_lock) |
393 | fl->fl_ops->fl_copy_lock(new, fl); |
394 | } |
395 | } |
396 | EXPORT_SYMBOL(locks_copy_lock); |
397 | |
398 | static void locks_move_blocks(struct file_lock *new, struct file_lock *fl) |
399 | { |
400 | struct file_lock *f; |
401 | |
402 | /* |
403 | * As ctx->flc_lock is held, new requests cannot be added to |
404 | * ->fl_blocked_requests, so we don't need a lock to check if it |
405 | * is empty. |
406 | */ |
407 | if (list_empty(head: &fl->fl_blocked_requests)) |
408 | return; |
409 | spin_lock(lock: &blocked_lock_lock); |
410 | list_splice_init(list: &fl->fl_blocked_requests, head: &new->fl_blocked_requests); |
411 | list_for_each_entry(f, &new->fl_blocked_requests, fl_blocked_member) |
412 | f->fl_blocker = new; |
413 | spin_unlock(lock: &blocked_lock_lock); |
414 | } |
415 | |
416 | static inline int flock_translate_cmd(int cmd) { |
417 | switch (cmd) { |
418 | case LOCK_SH: |
419 | return F_RDLCK; |
420 | case LOCK_EX: |
421 | return F_WRLCK; |
422 | case LOCK_UN: |
423 | return F_UNLCK; |
424 | } |
425 | return -EINVAL; |
426 | } |
427 | |
428 | /* Fill in a file_lock structure with an appropriate FLOCK lock. */ |
429 | static void flock_make_lock(struct file *filp, struct file_lock *fl, int type) |
430 | { |
431 | locks_init_lock(fl); |
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 | |
441 | static int assign_type(struct file_lock *fl, int type) |
442 | { |
443 | switch (type) { |
444 | case F_RDLCK: |
445 | case F_WRLCK: |
446 | case F_UNLCK: |
447 | fl->fl_type = type; |
448 | break; |
449 | default: |
450 | return -EINVAL; |
451 | } |
452 | return 0; |
453 | } |
454 | |
455 | static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl, |
456 | struct flock64 *l) |
457 | { |
458 | switch (l->l_whence) { |
459 | case SEEK_SET: |
460 | fl->fl_start = 0; |
461 | break; |
462 | case SEEK_CUR: |
463 | fl->fl_start = filp->f_pos; |
464 | break; |
465 | case SEEK_END: |
466 | fl->fl_start = i_size_read(inode: file_inode(f: filp)); |
467 | break; |
468 | default: |
469 | return -EINVAL; |
470 | } |
471 | if (l->l_start > OFFSET_MAX - fl->fl_start) |
472 | return -EOVERFLOW; |
473 | fl->fl_start += l->l_start; |
474 | if (fl->fl_start < 0) |
475 | return -EINVAL; |
476 | |
477 | /* POSIX-1996 leaves the case l->l_len < 0 undefined; |
478 | POSIX-2001 defines it. */ |
479 | if (l->l_len > 0) { |
480 | if (l->l_len - 1 > OFFSET_MAX - fl->fl_start) |
481 | return -EOVERFLOW; |
482 | fl->fl_end = fl->fl_start + (l->l_len - 1); |
483 | |
484 | } else if (l->l_len < 0) { |
485 | if (fl->fl_start + l->l_len < 0) |
486 | return -EINVAL; |
487 | fl->fl_end = fl->fl_start - 1; |
488 | fl->fl_start += l->l_len; |
489 | } else |
490 | fl->fl_end = OFFSET_MAX; |
491 | |
492 | fl->fl_owner = current->files; |
493 | fl->fl_pid = current->tgid; |
494 | fl->fl_file = filp; |
495 | fl->fl_flags = FL_POSIX; |
496 | fl->fl_ops = NULL; |
497 | fl->fl_lmops = NULL; |
498 | |
499 | return assign_type(fl, type: l->l_type); |
500 | } |
501 | |
502 | /* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX |
503 | * style lock. |
504 | */ |
505 | static int flock_to_posix_lock(struct file *filp, struct file_lock *fl, |
506 | struct flock *l) |
507 | { |
508 | struct flock64 ll = { |
509 | .l_type = l->l_type, |
510 | .l_whence = l->l_whence, |
511 | .l_start = l->l_start, |
512 | .l_len = l->l_len, |
513 | }; |
514 | |
515 | return flock64_to_posix_lock(filp, fl, l: &ll); |
516 | } |
517 | |
518 | /* default lease lock manager operations */ |
519 | static bool |
520 | lease_break_callback(struct file_lock *fl) |
521 | { |
522 | kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG); |
523 | return false; |
524 | } |
525 | |
526 | static void |
527 | lease_setup(struct file_lock *fl, void **priv) |
528 | { |
529 | struct file *filp = fl->fl_file; |
530 | struct fasync_struct *fa = *priv; |
531 | |
532 | /* |
533 | * fasync_insert_entry() returns the old entry if any. If there was no |
534 | * old entry, then it used "priv" and inserted it into the fasync list. |
535 | * Clear the pointer to indicate that it shouldn't be freed. |
536 | */ |
537 | if (!fasync_insert_entry(fa->fa_fd, filp, &fl->fl_fasync, fa)) |
538 | *priv = NULL; |
539 | |
540 | __f_setown(filp, task_pid(current), PIDTYPE_TGID, force: 0); |
541 | } |
542 | |
543 | static const struct lock_manager_operations lease_manager_ops = { |
544 | .lm_break = lease_break_callback, |
545 | .lm_change = lease_modify, |
546 | .lm_setup = lease_setup, |
547 | }; |
548 | |
549 | /* |
550 | * Initialize a lease, use the default lock manager operations |
551 | */ |
552 | static int lease_init(struct file *filp, int type, struct file_lock *fl) |
553 | { |
554 | if (assign_type(fl, type) != 0) |
555 | return -EINVAL; |
556 | |
557 | fl->fl_owner = filp; |
558 | fl->fl_pid = current->tgid; |
559 | |
560 | fl->fl_file = filp; |
561 | fl->fl_flags = FL_LEASE; |
562 | fl->fl_start = 0; |
563 | fl->fl_end = OFFSET_MAX; |
564 | fl->fl_ops = NULL; |
565 | fl->fl_lmops = &lease_manager_ops; |
566 | return 0; |
567 | } |
568 | |
569 | /* Allocate a file_lock initialised to this type of lease */ |
570 | static struct file_lock *lease_alloc(struct file *filp, int type) |
571 | { |
572 | struct file_lock *fl = locks_alloc_lock(); |
573 | int error = -ENOMEM; |
574 | |
575 | if (fl == NULL) |
576 | return ERR_PTR(error); |
577 | |
578 | error = lease_init(filp, type, fl); |
579 | if (error) { |
580 | locks_free_lock(fl); |
581 | return ERR_PTR(error); |
582 | } |
583 | return fl; |
584 | } |
585 | |
586 | /* Check if two locks overlap each other. |
587 | */ |
588 | static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2) |
589 | { |
590 | return ((fl1->fl_end >= fl2->fl_start) && |
591 | (fl2->fl_end >= fl1->fl_start)); |
592 | } |
593 | |
594 | /* |
595 | * Check whether two locks have the same owner. |
596 | */ |
597 | static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2) |
598 | { |
599 | return fl1->fl_owner == fl2->fl_owner; |
600 | } |
601 | |
602 | /* Must be called with the flc_lock held! */ |
603 | static void locks_insert_global_locks(struct file_lock *fl) |
604 | { |
605 | struct file_lock_list_struct *fll = this_cpu_ptr(&file_lock_list); |
606 | |
607 | percpu_rwsem_assert_held(&file_rwsem); |
608 | |
609 | spin_lock(lock: &fll->lock); |
610 | fl->fl_link_cpu = smp_processor_id(); |
611 | hlist_add_head(n: &fl->fl_link, h: &fll->hlist); |
612 | spin_unlock(lock: &fll->lock); |
613 | } |
614 | |
615 | /* Must be called with the flc_lock held! */ |
616 | static void locks_delete_global_locks(struct file_lock *fl) |
617 | { |
618 | struct file_lock_list_struct *fll; |
619 | |
620 | percpu_rwsem_assert_held(&file_rwsem); |
621 | |
622 | /* |
623 | * Avoid taking lock if already unhashed. This is safe since this check |
624 | * is done while holding the flc_lock, and new insertions into the list |
625 | * also require that it be held. |
626 | */ |
627 | if (hlist_unhashed(h: &fl->fl_link)) |
628 | return; |
629 | |
630 | fll = per_cpu_ptr(&file_lock_list, fl->fl_link_cpu); |
631 | spin_lock(lock: &fll->lock); |
632 | hlist_del_init(n: &fl->fl_link); |
633 | spin_unlock(lock: &fll->lock); |
634 | } |
635 | |
636 | static unsigned long |
637 | posix_owner_key(struct file_lock *fl) |
638 | { |
639 | return (unsigned long)fl->fl_owner; |
640 | } |
641 | |
642 | static void locks_insert_global_blocked(struct file_lock *waiter) |
643 | { |
644 | lockdep_assert_held(&blocked_lock_lock); |
645 | |
646 | hash_add(blocked_hash, &waiter->fl_link, posix_owner_key(waiter)); |
647 | } |
648 | |
649 | static void locks_delete_global_blocked(struct file_lock *waiter) |
650 | { |
651 | lockdep_assert_held(&blocked_lock_lock); |
652 | |
653 | hash_del(node: &waiter->fl_link); |
654 | } |
655 | |
656 | /* Remove waiter from blocker's block list. |
657 | * When blocker ends up pointing to itself then the list is empty. |
658 | * |
659 | * Must be called with blocked_lock_lock held. |
660 | */ |
661 | static void __locks_delete_block(struct file_lock *waiter) |
662 | { |
663 | locks_delete_global_blocked(waiter); |
664 | list_del_init(entry: &waiter->fl_blocked_member); |
665 | } |
666 | |
667 | static void __locks_wake_up_blocks(struct file_lock *blocker) |
668 | { |
669 | while (!list_empty(head: &blocker->fl_blocked_requests)) { |
670 | struct file_lock *waiter; |
671 | |
672 | waiter = list_first_entry(&blocker->fl_blocked_requests, |
673 | struct file_lock, fl_blocked_member); |
674 | __locks_delete_block(waiter); |
675 | if (waiter->fl_lmops && waiter->fl_lmops->lm_notify) |
676 | waiter->fl_lmops->lm_notify(waiter); |
677 | else |
678 | wake_up(&waiter->fl_wait); |
679 | |
680 | /* |
681 | * The setting of fl_blocker to NULL marks the "done" |
682 | * point in deleting a block. Paired with acquire at the top |
683 | * of locks_delete_block(). |
684 | */ |
685 | smp_store_release(&waiter->fl_blocker, NULL); |
686 | } |
687 | } |
688 | |
689 | /** |
690 | * locks_delete_block - stop waiting for a file lock |
691 | * @waiter: the lock which was waiting |
692 | * |
693 | * lockd/nfsd need to disconnect the lock while working on it. |
694 | */ |
695 | int locks_delete_block(struct file_lock *waiter) |
696 | { |
697 | int status = -ENOENT; |
698 | |
699 | /* |
700 | * If fl_blocker is NULL, it won't be set again as this thread "owns" |
701 | * the lock and is the only one that might try to claim the lock. |
702 | * |
703 | * We use acquire/release to manage fl_blocker so that we can |
704 | * optimize away taking the blocked_lock_lock in many cases. |
705 | * |
706 | * The smp_load_acquire guarantees two things: |
707 | * |
708 | * 1/ that fl_blocked_requests can be tested locklessly. If something |
709 | * was recently added to that list it must have been in a locked region |
710 | * *before* the locked region when fl_blocker was set to NULL. |
711 | * |
712 | * 2/ that no other thread is accessing 'waiter', so it is safe to free |
713 | * it. __locks_wake_up_blocks is careful not to touch waiter after |
714 | * fl_blocker is released. |
715 | * |
716 | * If a lockless check of fl_blocker shows it to be NULL, we know that |
717 | * no new locks can be inserted into its fl_blocked_requests list, and |
718 | * can avoid doing anything further if the list is empty. |
719 | */ |
720 | if (!smp_load_acquire(&waiter->fl_blocker) && |
721 | list_empty(head: &waiter->fl_blocked_requests)) |
722 | return status; |
723 | |
724 | spin_lock(lock: &blocked_lock_lock); |
725 | if (waiter->fl_blocker) |
726 | status = 0; |
727 | __locks_wake_up_blocks(blocker: waiter); |
728 | __locks_delete_block(waiter); |
729 | |
730 | /* |
731 | * The setting of fl_blocker to NULL marks the "done" point in deleting |
732 | * a block. Paired with acquire at the top of this function. |
733 | */ |
734 | smp_store_release(&waiter->fl_blocker, NULL); |
735 | spin_unlock(lock: &blocked_lock_lock); |
736 | return status; |
737 | } |
738 | EXPORT_SYMBOL(locks_delete_block); |
739 | |
740 | /* Insert waiter into blocker's block list. |
741 | * We use a circular list so that processes can be easily woken up in |
742 | * the order they blocked. The documentation doesn't require this but |
743 | * it seems like the reasonable thing to do. |
744 | * |
745 | * Must be called with both the flc_lock and blocked_lock_lock held. The |
746 | * fl_blocked_requests list itself is protected by the blocked_lock_lock, |
747 | * but by ensuring that the flc_lock is also held on insertions we can avoid |
748 | * taking the blocked_lock_lock in some cases when we see that the |
749 | * fl_blocked_requests list is empty. |
750 | * |
751 | * Rather than just adding to the list, we check for conflicts with any existing |
752 | * waiters, and add beneath any waiter that blocks the new waiter. |
753 | * Thus wakeups don't happen until needed. |
754 | */ |
755 | static void __locks_insert_block(struct file_lock *blocker, |
756 | struct file_lock *waiter, |
757 | bool conflict(struct file_lock *, |
758 | struct file_lock *)) |
759 | { |
760 | struct file_lock *fl; |
761 | BUG_ON(!list_empty(&waiter->fl_blocked_member)); |
762 | |
763 | new_blocker: |
764 | list_for_each_entry(fl, &blocker->fl_blocked_requests, fl_blocked_member) |
765 | if (conflict(fl, waiter)) { |
766 | blocker = fl; |
767 | goto new_blocker; |
768 | } |
769 | waiter->fl_blocker = blocker; |
770 | list_add_tail(new: &waiter->fl_blocked_member, head: &blocker->fl_blocked_requests); |
771 | if (IS_POSIX(blocker) && !IS_OFDLCK(blocker)) |
772 | locks_insert_global_blocked(waiter); |
773 | |
774 | /* The requests in waiter->fl_blocked are known to conflict with |
775 | * waiter, but might not conflict with blocker, or the requests |
776 | * and lock which block it. So they all need to be woken. |
777 | */ |
778 | __locks_wake_up_blocks(blocker: waiter); |
779 | } |
780 | |
781 | /* Must be called with flc_lock held. */ |
782 | static void locks_insert_block(struct file_lock *blocker, |
783 | struct file_lock *waiter, |
784 | bool conflict(struct file_lock *, |
785 | struct file_lock *)) |
786 | { |
787 | spin_lock(lock: &blocked_lock_lock); |
788 | __locks_insert_block(blocker, waiter, conflict); |
789 | spin_unlock(lock: &blocked_lock_lock); |
790 | } |
791 | |
792 | /* |
793 | * Wake up processes blocked waiting for blocker. |
794 | * |
795 | * Must be called with the inode->flc_lock held! |
796 | */ |
797 | static void locks_wake_up_blocks(struct file_lock *blocker) |
798 | { |
799 | /* |
800 | * Avoid taking global lock if list is empty. This is safe since new |
801 | * blocked requests are only added to the list under the flc_lock, and |
802 | * the flc_lock is always held here. Note that removal from the |
803 | * fl_blocked_requests list does not require the flc_lock, so we must |
804 | * recheck list_empty() after acquiring the blocked_lock_lock. |
805 | */ |
806 | if (list_empty(head: &blocker->fl_blocked_requests)) |
807 | return; |
808 | |
809 | spin_lock(lock: &blocked_lock_lock); |
810 | __locks_wake_up_blocks(blocker); |
811 | spin_unlock(lock: &blocked_lock_lock); |
812 | } |
813 | |
814 | static void |
815 | locks_insert_lock_ctx(struct file_lock *fl, struct list_head *before) |
816 | { |
817 | list_add_tail(new: &fl->fl_list, head: before); |
818 | locks_insert_global_locks(fl); |
819 | } |
820 | |
821 | static void |
822 | locks_unlink_lock_ctx(struct file_lock *fl) |
823 | { |
824 | locks_delete_global_locks(fl); |
825 | list_del_init(entry: &fl->fl_list); |
826 | locks_wake_up_blocks(blocker: fl); |
827 | } |
828 | |
829 | static void |
830 | locks_delete_lock_ctx(struct file_lock *fl, struct list_head *dispose) |
831 | { |
832 | locks_unlink_lock_ctx(fl); |
833 | if (dispose) |
834 | list_add(new: &fl->fl_list, head: dispose); |
835 | else |
836 | locks_free_lock(fl); |
837 | } |
838 | |
839 | /* Determine if lock sys_fl blocks lock caller_fl. Common functionality |
840 | * checks for shared/exclusive status of overlapping locks. |
841 | */ |
842 | static bool locks_conflict(struct file_lock *caller_fl, |
843 | struct file_lock *sys_fl) |
844 | { |
845 | if (sys_fl->fl_type == F_WRLCK) |
846 | return true; |
847 | if (caller_fl->fl_type == F_WRLCK) |
848 | return true; |
849 | return false; |
850 | } |
851 | |
852 | /* Determine if lock sys_fl blocks lock caller_fl. POSIX specific |
853 | * checking before calling the locks_conflict(). |
854 | */ |
855 | static bool posix_locks_conflict(struct file_lock *caller_fl, |
856 | struct file_lock *sys_fl) |
857 | { |
858 | /* POSIX locks owned by the same process do not conflict with |
859 | * each other. |
860 | */ |
861 | if (posix_same_owner(fl1: caller_fl, fl2: sys_fl)) |
862 | return false; |
863 | |
864 | /* Check whether they overlap */ |
865 | if (!locks_overlap(fl1: caller_fl, fl2: sys_fl)) |
866 | return false; |
867 | |
868 | return locks_conflict(caller_fl, sys_fl); |
869 | } |
870 | |
871 | /* Determine if lock sys_fl blocks lock caller_fl. Used on xx_GETLK |
872 | * path so checks for additional GETLK-specific things like F_UNLCK. |
873 | */ |
874 | static bool posix_test_locks_conflict(struct file_lock *caller_fl, |
875 | struct file_lock *sys_fl) |
876 | { |
877 | /* F_UNLCK checks any locks on the same fd. */ |
878 | if (caller_fl->fl_type == F_UNLCK) { |
879 | if (!posix_same_owner(fl1: caller_fl, fl2: sys_fl)) |
880 | return false; |
881 | return locks_overlap(fl1: caller_fl, fl2: sys_fl); |
882 | } |
883 | return posix_locks_conflict(caller_fl, sys_fl); |
884 | } |
885 | |
886 | /* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific |
887 | * checking before calling the locks_conflict(). |
888 | */ |
889 | static bool flock_locks_conflict(struct file_lock *caller_fl, |
890 | struct file_lock *sys_fl) |
891 | { |
892 | /* FLOCK locks referring to the same filp do not conflict with |
893 | * each other. |
894 | */ |
895 | if (caller_fl->fl_file == sys_fl->fl_file) |
896 | return false; |
897 | |
898 | return locks_conflict(caller_fl, sys_fl); |
899 | } |
900 | |
901 | void |
902 | posix_test_lock(struct file *filp, struct file_lock *fl) |
903 | { |
904 | struct file_lock *cfl; |
905 | struct file_lock_context *ctx; |
906 | struct inode *inode = file_inode(f: filp); |
907 | void *owner; |
908 | void (*func)(void); |
909 | |
910 | ctx = locks_inode_context(inode); |
911 | if (!ctx || list_empty_careful(head: &ctx->flc_posix)) { |
912 | fl->fl_type = F_UNLCK; |
913 | return; |
914 | } |
915 | |
916 | retry: |
917 | spin_lock(lock: &ctx->flc_lock); |
918 | list_for_each_entry(cfl, &ctx->flc_posix, fl_list) { |
919 | if (!posix_test_locks_conflict(caller_fl: fl, sys_fl: cfl)) |
920 | continue; |
921 | if (cfl->fl_lmops && cfl->fl_lmops->lm_lock_expirable |
922 | && (*cfl->fl_lmops->lm_lock_expirable)(cfl)) { |
923 | owner = cfl->fl_lmops->lm_mod_owner; |
924 | func = cfl->fl_lmops->lm_expire_lock; |
925 | __module_get(module: owner); |
926 | spin_unlock(lock: &ctx->flc_lock); |
927 | (*func)(); |
928 | module_put(module: owner); |
929 | goto retry; |
930 | } |
931 | locks_copy_conflock(fl, cfl); |
932 | goto out; |
933 | } |
934 | fl->fl_type = F_UNLCK; |
935 | out: |
936 | spin_unlock(lock: &ctx->flc_lock); |
937 | return; |
938 | } |
939 | EXPORT_SYMBOL(posix_test_lock); |
940 | |
941 | /* |
942 | * Deadlock detection: |
943 | * |
944 | * We attempt to detect deadlocks that are due purely to posix file |
945 | * locks. |
946 | * |
947 | * We assume that a task can be waiting for at most one lock at a time. |
948 | * So for any acquired lock, the process holding that lock may be |
949 | * waiting on at most one other lock. That lock in turns may be held by |
950 | * someone waiting for at most one other lock. Given a requested lock |
951 | * caller_fl which is about to wait for a conflicting lock block_fl, we |
952 | * follow this chain of waiters to ensure we are not about to create a |
953 | * cycle. |
954 | * |
955 | * Since we do this before we ever put a process to sleep on a lock, we |
956 | * are ensured that there is never a cycle; that is what guarantees that |
957 | * the while() loop in posix_locks_deadlock() eventually completes. |
958 | * |
959 | * Note: the above assumption may not be true when handling lock |
960 | * requests from a broken NFS client. It may also fail in the presence |
961 | * of tasks (such as posix threads) sharing the same open file table. |
962 | * To handle those cases, we just bail out after a few iterations. |
963 | * |
964 | * For FL_OFDLCK locks, the owner is the filp, not the files_struct. |
965 | * Because the owner is not even nominally tied to a thread of |
966 | * execution, the deadlock detection below can't reasonably work well. Just |
967 | * skip it for those. |
968 | * |
969 | * In principle, we could do a more limited deadlock detection on FL_OFDLCK |
970 | * locks that just checks for the case where two tasks are attempting to |
971 | * upgrade from read to write locks on the same inode. |
972 | */ |
973 | |
974 | #define MAX_DEADLK_ITERATIONS 10 |
975 | |
976 | /* Find a lock that the owner of the given block_fl is blocking on. */ |
977 | static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl) |
978 | { |
979 | struct file_lock *fl; |
980 | |
981 | hash_for_each_possible(blocked_hash, fl, fl_link, posix_owner_key(block_fl)) { |
982 | if (posix_same_owner(fl1: fl, fl2: block_fl)) { |
983 | while (fl->fl_blocker) |
984 | fl = fl->fl_blocker; |
985 | return fl; |
986 | } |
987 | } |
988 | return NULL; |
989 | } |
990 | |
991 | /* Must be called with the blocked_lock_lock held! */ |
992 | static int posix_locks_deadlock(struct file_lock *caller_fl, |
993 | struct file_lock *block_fl) |
994 | { |
995 | int i = 0; |
996 | |
997 | lockdep_assert_held(&blocked_lock_lock); |
998 | |
999 | /* |
1000 | * This deadlock detector can't reasonably detect deadlocks with |
1001 | * FL_OFDLCK locks, since they aren't owned by a process, per-se. |
1002 | */ |
1003 | if (IS_OFDLCK(caller_fl)) |
1004 | return 0; |
1005 | |
1006 | while ((block_fl = what_owner_is_waiting_for(block_fl))) { |
1007 | if (i++ > MAX_DEADLK_ITERATIONS) |
1008 | return 0; |
1009 | if (posix_same_owner(fl1: caller_fl, fl2: block_fl)) |
1010 | return 1; |
1011 | } |
1012 | return 0; |
1013 | } |
1014 | |
1015 | /* Try to create a FLOCK lock on filp. We always insert new FLOCK locks |
1016 | * after any leases, but before any posix locks. |
1017 | * |
1018 | * Note that if called with an FL_EXISTS argument, the caller may determine |
1019 | * whether or not a lock was successfully freed by testing the return |
1020 | * value for -ENOENT. |
1021 | */ |
1022 | static int flock_lock_inode(struct inode *inode, struct file_lock *request) |
1023 | { |
1024 | struct file_lock *new_fl = NULL; |
1025 | struct file_lock *fl; |
1026 | struct file_lock_context *ctx; |
1027 | int error = 0; |
1028 | bool found = false; |
1029 | LIST_HEAD(dispose); |
1030 | |
1031 | ctx = locks_get_lock_context(inode, type: request->fl_type); |
1032 | if (!ctx) { |
1033 | if (request->fl_type != F_UNLCK) |
1034 | return -ENOMEM; |
1035 | return (request->fl_flags & FL_EXISTS) ? -ENOENT : 0; |
1036 | } |
1037 | |
1038 | if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) { |
1039 | new_fl = locks_alloc_lock(); |
1040 | if (!new_fl) |
1041 | return -ENOMEM; |
1042 | } |
1043 | |
1044 | percpu_down_read(sem: &file_rwsem); |
1045 | spin_lock(lock: &ctx->flc_lock); |
1046 | if (request->fl_flags & FL_ACCESS) |
1047 | goto find_conflict; |
1048 | |
1049 | list_for_each_entry(fl, &ctx->flc_flock, fl_list) { |
1050 | if (request->fl_file != fl->fl_file) |
1051 | continue; |
1052 | if (request->fl_type == fl->fl_type) |
1053 | goto out; |
1054 | found = true; |
1055 | locks_delete_lock_ctx(fl, dispose: &dispose); |
1056 | break; |
1057 | } |
1058 | |
1059 | if (request->fl_type == F_UNLCK) { |
1060 | if ((request->fl_flags & FL_EXISTS) && !found) |
1061 | error = -ENOENT; |
1062 | goto out; |
1063 | } |
1064 | |
1065 | find_conflict: |
1066 | list_for_each_entry(fl, &ctx->flc_flock, fl_list) { |
1067 | if (!flock_locks_conflict(caller_fl: request, sys_fl: fl)) |
1068 | continue; |
1069 | error = -EAGAIN; |
1070 | if (!(request->fl_flags & FL_SLEEP)) |
1071 | goto out; |
1072 | error = FILE_LOCK_DEFERRED; |
1073 | locks_insert_block(blocker: fl, waiter: request, conflict: flock_locks_conflict); |
1074 | goto out; |
1075 | } |
1076 | if (request->fl_flags & FL_ACCESS) |
1077 | goto out; |
1078 | locks_copy_lock(new_fl, request); |
1079 | locks_move_blocks(new: new_fl, fl: request); |
1080 | locks_insert_lock_ctx(fl: new_fl, before: &ctx->flc_flock); |
1081 | new_fl = NULL; |
1082 | error = 0; |
1083 | |
1084 | out: |
1085 | spin_unlock(lock: &ctx->flc_lock); |
1086 | percpu_up_read(sem: &file_rwsem); |
1087 | if (new_fl) |
1088 | locks_free_lock(new_fl); |
1089 | locks_dispose_list(dispose: &dispose); |
1090 | trace_flock_lock_inode(inode, fl: request, ret: error); |
1091 | return error; |
1092 | } |
1093 | |
1094 | static int posix_lock_inode(struct inode *inode, struct file_lock *request, |
1095 | struct file_lock *conflock) |
1096 | { |
1097 | struct file_lock *fl, *tmp; |
1098 | struct file_lock *new_fl = NULL; |
1099 | struct file_lock *new_fl2 = NULL; |
1100 | struct file_lock *left = NULL; |
1101 | struct file_lock *right = NULL; |
1102 | struct file_lock_context *ctx; |
1103 | int error; |
1104 | bool added = false; |
1105 | LIST_HEAD(dispose); |
1106 | void *owner; |
1107 | void (*func)(void); |
1108 | |
1109 | ctx = locks_get_lock_context(inode, type: request->fl_type); |
1110 | if (!ctx) |
1111 | return (request->fl_type == F_UNLCK) ? 0 : -ENOMEM; |
1112 | |
1113 | /* |
1114 | * We may need two file_lock structures for this operation, |
1115 | * so we get them in advance to avoid races. |
1116 | * |
1117 | * In some cases we can be sure, that no new locks will be needed |
1118 | */ |
1119 | if (!(request->fl_flags & FL_ACCESS) && |
1120 | (request->fl_type != F_UNLCK || |
1121 | request->fl_start != 0 || request->fl_end != OFFSET_MAX)) { |
1122 | new_fl = locks_alloc_lock(); |
1123 | new_fl2 = locks_alloc_lock(); |
1124 | } |
1125 | |
1126 | retry: |
1127 | percpu_down_read(sem: &file_rwsem); |
1128 | spin_lock(lock: &ctx->flc_lock); |
1129 | /* |
1130 | * New lock request. Walk all POSIX locks and look for conflicts. If |
1131 | * there are any, either return error or put the request on the |
1132 | * blocker's list of waiters and the global blocked_hash. |
1133 | */ |
1134 | if (request->fl_type != F_UNLCK) { |
1135 | list_for_each_entry(fl, &ctx->flc_posix, fl_list) { |
1136 | if (!posix_locks_conflict(caller_fl: request, sys_fl: fl)) |
1137 | continue; |
1138 | if (fl->fl_lmops && fl->fl_lmops->lm_lock_expirable |
1139 | && (*fl->fl_lmops->lm_lock_expirable)(fl)) { |
1140 | owner = fl->fl_lmops->lm_mod_owner; |
1141 | func = fl->fl_lmops->lm_expire_lock; |
1142 | __module_get(module: owner); |
1143 | spin_unlock(lock: &ctx->flc_lock); |
1144 | percpu_up_read(sem: &file_rwsem); |
1145 | (*func)(); |
1146 | module_put(module: owner); |
1147 | goto retry; |
1148 | } |
1149 | if (conflock) |
1150 | locks_copy_conflock(conflock, fl); |
1151 | error = -EAGAIN; |
1152 | if (!(request->fl_flags & FL_SLEEP)) |
1153 | goto out; |
1154 | /* |
1155 | * Deadlock detection and insertion into the blocked |
1156 | * locks list must be done while holding the same lock! |
1157 | */ |
1158 | error = -EDEADLK; |
1159 | spin_lock(lock: &blocked_lock_lock); |
1160 | /* |
1161 | * Ensure that we don't find any locks blocked on this |
1162 | * request during deadlock detection. |
1163 | */ |
1164 | __locks_wake_up_blocks(blocker: request); |
1165 | if (likely(!posix_locks_deadlock(request, fl))) { |
1166 | error = FILE_LOCK_DEFERRED; |
1167 | __locks_insert_block(blocker: fl, waiter: request, |
1168 | conflict: posix_locks_conflict); |
1169 | } |
1170 | spin_unlock(lock: &blocked_lock_lock); |
1171 | goto out; |
1172 | } |
1173 | } |
1174 | |
1175 | /* If we're just looking for a conflict, we're done. */ |
1176 | error = 0; |
1177 | if (request->fl_flags & FL_ACCESS) |
1178 | goto out; |
1179 | |
1180 | /* Find the first old lock with the same owner as the new lock */ |
1181 | list_for_each_entry(fl, &ctx->flc_posix, fl_list) { |
1182 | if (posix_same_owner(fl1: request, fl2: fl)) |
1183 | break; |
1184 | } |
1185 | |
1186 | /* Process locks with this owner. */ |
1187 | list_for_each_entry_safe_from(fl, tmp, &ctx->flc_posix, fl_list) { |
1188 | if (!posix_same_owner(fl1: request, fl2: fl)) |
1189 | break; |
1190 | |
1191 | /* Detect adjacent or overlapping regions (if same lock type) */ |
1192 | if (request->fl_type == fl->fl_type) { |
1193 | /* In all comparisons of start vs end, use |
1194 | * "start - 1" rather than "end + 1". If end |
1195 | * is OFFSET_MAX, end + 1 will become negative. |
1196 | */ |
1197 | if (fl->fl_end < request->fl_start - 1) |
1198 | continue; |
1199 | /* If the next lock in the list has entirely bigger |
1200 | * addresses than the new one, insert the lock here. |
1201 | */ |
1202 | if (fl->fl_start - 1 > request->fl_end) |
1203 | break; |
1204 | |
1205 | /* If we come here, the new and old lock are of the |
1206 | * same type and adjacent or overlapping. Make one |
1207 | * lock yielding from the lower start address of both |
1208 | * locks to the higher end address. |
1209 | */ |
1210 | if (fl->fl_start > request->fl_start) |
1211 | fl->fl_start = request->fl_start; |
1212 | else |
1213 | request->fl_start = fl->fl_start; |
1214 | if (fl->fl_end < request->fl_end) |
1215 | fl->fl_end = request->fl_end; |
1216 | else |
1217 | request->fl_end = fl->fl_end; |
1218 | if (added) { |
1219 | locks_delete_lock_ctx(fl, dispose: &dispose); |
1220 | continue; |
1221 | } |
1222 | request = fl; |
1223 | added = true; |
1224 | } else { |
1225 | /* Processing for different lock types is a bit |
1226 | * more complex. |
1227 | */ |
1228 | if (fl->fl_end < request->fl_start) |
1229 | continue; |
1230 | if (fl->fl_start > request->fl_end) |
1231 | break; |
1232 | if (request->fl_type == F_UNLCK) |
1233 | added = true; |
1234 | if (fl->fl_start < request->fl_start) |
1235 | left = fl; |
1236 | /* If the next lock in the list has a higher end |
1237 | * address than the new one, insert the new one here. |
1238 | */ |
1239 | if (fl->fl_end > request->fl_end) { |
1240 | right = fl; |
1241 | break; |
1242 | } |
1243 | if (fl->fl_start >= request->fl_start) { |
1244 | /* The new lock completely replaces an old |
1245 | * one (This may happen several times). |
1246 | */ |
1247 | if (added) { |
1248 | locks_delete_lock_ctx(fl, dispose: &dispose); |
1249 | continue; |
1250 | } |
1251 | /* |
1252 | * Replace the old lock with new_fl, and |
1253 | * remove the old one. It's safe to do the |
1254 | * insert here since we know that we won't be |
1255 | * using new_fl later, and that the lock is |
1256 | * just replacing an existing lock. |
1257 | */ |
1258 | error = -ENOLCK; |
1259 | if (!new_fl) |
1260 | goto out; |
1261 | locks_copy_lock(new_fl, request); |
1262 | locks_move_blocks(new: new_fl, fl: request); |
1263 | request = new_fl; |
1264 | new_fl = NULL; |
1265 | locks_insert_lock_ctx(fl: request, before: &fl->fl_list); |
1266 | locks_delete_lock_ctx(fl, dispose: &dispose); |
1267 | added = true; |
1268 | } |
1269 | } |
1270 | } |
1271 | |
1272 | /* |
1273 | * The above code only modifies existing locks in case of merging or |
1274 | * replacing. If new lock(s) need to be inserted all modifications are |
1275 | * done below this, so it's safe yet to bail out. |
1276 | */ |
1277 | error = -ENOLCK; /* "no luck" */ |
1278 | if (right && left == right && !new_fl2) |
1279 | goto out; |
1280 | |
1281 | error = 0; |
1282 | if (!added) { |
1283 | if (request->fl_type == F_UNLCK) { |
1284 | if (request->fl_flags & FL_EXISTS) |
1285 | error = -ENOENT; |
1286 | goto out; |
1287 | } |
1288 | |
1289 | if (!new_fl) { |
1290 | error = -ENOLCK; |
1291 | goto out; |
1292 | } |
1293 | locks_copy_lock(new_fl, request); |
1294 | locks_move_blocks(new: new_fl, fl: request); |
1295 | locks_insert_lock_ctx(fl: new_fl, before: &fl->fl_list); |
1296 | fl = new_fl; |
1297 | new_fl = NULL; |
1298 | } |
1299 | if (right) { |
1300 | if (left == right) { |
1301 | /* The new lock breaks the old one in two pieces, |
1302 | * so we have to use the second new lock. |
1303 | */ |
1304 | left = new_fl2; |
1305 | new_fl2 = NULL; |
1306 | locks_copy_lock(left, right); |
1307 | locks_insert_lock_ctx(fl: left, before: &fl->fl_list); |
1308 | } |
1309 | right->fl_start = request->fl_end + 1; |
1310 | locks_wake_up_blocks(blocker: right); |
1311 | } |
1312 | if (left) { |
1313 | left->fl_end = request->fl_start - 1; |
1314 | locks_wake_up_blocks(blocker: left); |
1315 | } |
1316 | out: |
1317 | spin_unlock(lock: &ctx->flc_lock); |
1318 | percpu_up_read(sem: &file_rwsem); |
1319 | trace_posix_lock_inode(inode, fl: request, ret: error); |
1320 | /* |
1321 | * Free any unused locks. |
1322 | */ |
1323 | if (new_fl) |
1324 | locks_free_lock(new_fl); |
1325 | if (new_fl2) |
1326 | locks_free_lock(new_fl2); |
1327 | locks_dispose_list(dispose: &dispose); |
1328 | |
1329 | return error; |
1330 | } |
1331 | |
1332 | /** |
1333 | * posix_lock_file - Apply a POSIX-style lock to a file |
1334 | * @filp: The file to apply the lock to |
1335 | * @fl: The lock to be applied |
1336 | * @conflock: Place to return a copy of the conflicting lock, if found. |
1337 | * |
1338 | * Add a POSIX style lock to a file. |
1339 | * We merge adjacent & overlapping locks whenever possible. |
1340 | * POSIX locks are sorted by owner task, then by starting address |
1341 | * |
1342 | * Note that if called with an FL_EXISTS argument, the caller may determine |
1343 | * whether or not a lock was successfully freed by testing the return |
1344 | * value for -ENOENT. |
1345 | */ |
1346 | int posix_lock_file(struct file *filp, struct file_lock *fl, |
1347 | struct file_lock *conflock) |
1348 | { |
1349 | return posix_lock_inode(inode: file_inode(f: filp), request: fl, conflock); |
1350 | } |
1351 | EXPORT_SYMBOL(posix_lock_file); |
1352 | |
1353 | /** |
1354 | * posix_lock_inode_wait - Apply a POSIX-style lock to a file |
1355 | * @inode: inode of file to which lock request should be applied |
1356 | * @fl: The lock to be applied |
1357 | * |
1358 | * Apply a POSIX style lock request to an inode. |
1359 | */ |
1360 | static int posix_lock_inode_wait(struct inode *inode, struct file_lock *fl) |
1361 | { |
1362 | int error; |
1363 | might_sleep (); |
1364 | for (;;) { |
1365 | error = posix_lock_inode(inode, request: fl, NULL); |
1366 | if (error != FILE_LOCK_DEFERRED) |
1367 | break; |
1368 | error = wait_event_interruptible(fl->fl_wait, |
1369 | list_empty(&fl->fl_blocked_member)); |
1370 | if (error) |
1371 | break; |
1372 | } |
1373 | locks_delete_block(fl); |
1374 | return error; |
1375 | } |
1376 | |
1377 | static void lease_clear_pending(struct file_lock *fl, int arg) |
1378 | { |
1379 | switch (arg) { |
1380 | case F_UNLCK: |
1381 | fl->fl_flags &= ~FL_UNLOCK_PENDING; |
1382 | fallthrough; |
1383 | case F_RDLCK: |
1384 | fl->fl_flags &= ~FL_DOWNGRADE_PENDING; |
1385 | } |
1386 | } |
1387 | |
1388 | /* We already had a lease on this file; just change its type */ |
1389 | int lease_modify(struct file_lock *fl, int arg, struct list_head *dispose) |
1390 | { |
1391 | int error = assign_type(fl, type: arg); |
1392 | |
1393 | if (error) |
1394 | return error; |
1395 | lease_clear_pending(fl, arg); |
1396 | locks_wake_up_blocks(blocker: fl); |
1397 | if (arg == F_UNLCK) { |
1398 | struct file *filp = fl->fl_file; |
1399 | |
1400 | f_delown(filp); |
1401 | filp->f_owner.signum = 0; |
1402 | fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync); |
1403 | if (fl->fl_fasync != NULL) { |
1404 | printk(KERN_ERR "locks_delete_lock: fasync == %p\n" , fl->fl_fasync); |
1405 | fl->fl_fasync = NULL; |
1406 | } |
1407 | locks_delete_lock_ctx(fl, dispose); |
1408 | } |
1409 | return 0; |
1410 | } |
1411 | EXPORT_SYMBOL(lease_modify); |
1412 | |
1413 | static bool past_time(unsigned long then) |
1414 | { |
1415 | if (!then) |
1416 | /* 0 is a special value meaning "this never expires": */ |
1417 | return false; |
1418 | return time_after(jiffies, then); |
1419 | } |
1420 | |
1421 | static void time_out_leases(struct inode *inode, struct list_head *dispose) |
1422 | { |
1423 | struct file_lock_context *ctx = inode->i_flctx; |
1424 | struct file_lock *fl, *tmp; |
1425 | |
1426 | lockdep_assert_held(&ctx->flc_lock); |
1427 | |
1428 | list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list) { |
1429 | trace_time_out_leases(inode, fl); |
1430 | if (past_time(then: fl->fl_downgrade_time)) |
1431 | lease_modify(fl, F_RDLCK, dispose); |
1432 | if (past_time(then: fl->fl_break_time)) |
1433 | lease_modify(fl, F_UNLCK, dispose); |
1434 | } |
1435 | } |
1436 | |
1437 | static bool leases_conflict(struct file_lock *lease, struct file_lock *breaker) |
1438 | { |
1439 | bool rc; |
1440 | |
1441 | if (lease->fl_lmops->lm_breaker_owns_lease |
1442 | && lease->fl_lmops->lm_breaker_owns_lease(lease)) |
1443 | return false; |
1444 | if ((breaker->fl_flags & FL_LAYOUT) != (lease->fl_flags & FL_LAYOUT)) { |
1445 | rc = false; |
1446 | goto trace; |
1447 | } |
1448 | if ((breaker->fl_flags & FL_DELEG) && (lease->fl_flags & FL_LEASE)) { |
1449 | rc = false; |
1450 | goto trace; |
1451 | } |
1452 | |
1453 | rc = locks_conflict(caller_fl: breaker, sys_fl: lease); |
1454 | trace: |
1455 | trace_leases_conflict(conflict: rc, lease, breaker); |
1456 | return rc; |
1457 | } |
1458 | |
1459 | static bool |
1460 | any_leases_conflict(struct inode *inode, struct file_lock *breaker) |
1461 | { |
1462 | struct file_lock_context *ctx = inode->i_flctx; |
1463 | struct file_lock *fl; |
1464 | |
1465 | lockdep_assert_held(&ctx->flc_lock); |
1466 | |
1467 | list_for_each_entry(fl, &ctx->flc_lease, fl_list) { |
1468 | if (leases_conflict(lease: fl, breaker)) |
1469 | return true; |
1470 | } |
1471 | return false; |
1472 | } |
1473 | |
1474 | /** |
1475 | * __break_lease - revoke all outstanding leases on file |
1476 | * @inode: the inode of the file to return |
1477 | * @mode: O_RDONLY: break only write leases; O_WRONLY or O_RDWR: |
1478 | * break all leases |
1479 | * @type: FL_LEASE: break leases and delegations; FL_DELEG: break |
1480 | * only delegations |
1481 | * |
1482 | * break_lease (inlined for speed) has checked there already is at least |
1483 | * some kind of lock (maybe a lease) on this file. Leases are broken on |
1484 | * a call to open() or truncate(). This function can sleep unless you |
1485 | * specified %O_NONBLOCK to your open(). |
1486 | */ |
1487 | int __break_lease(struct inode *inode, unsigned int mode, unsigned int type) |
1488 | { |
1489 | int error = 0; |
1490 | struct file_lock_context *ctx; |
1491 | struct file_lock *new_fl, *fl, *tmp; |
1492 | unsigned long break_time; |
1493 | int want_write = (mode & O_ACCMODE) != O_RDONLY; |
1494 | LIST_HEAD(dispose); |
1495 | |
1496 | new_fl = lease_alloc(NULL, type: want_write ? F_WRLCK : F_RDLCK); |
1497 | if (IS_ERR(ptr: new_fl)) |
1498 | return PTR_ERR(ptr: new_fl); |
1499 | new_fl->fl_flags = type; |
1500 | |
1501 | /* typically we will check that ctx is non-NULL before calling */ |
1502 | ctx = locks_inode_context(inode); |
1503 | if (!ctx) { |
1504 | WARN_ON_ONCE(1); |
1505 | goto free_lock; |
1506 | } |
1507 | |
1508 | percpu_down_read(sem: &file_rwsem); |
1509 | spin_lock(lock: &ctx->flc_lock); |
1510 | |
1511 | time_out_leases(inode, dispose: &dispose); |
1512 | |
1513 | if (!any_leases_conflict(inode, breaker: new_fl)) |
1514 | goto out; |
1515 | |
1516 | break_time = 0; |
1517 | if (lease_break_time > 0) { |
1518 | break_time = jiffies + lease_break_time * HZ; |
1519 | if (break_time == 0) |
1520 | break_time++; /* so that 0 means no break time */ |
1521 | } |
1522 | |
1523 | list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list) { |
1524 | if (!leases_conflict(lease: fl, breaker: new_fl)) |
1525 | continue; |
1526 | if (want_write) { |
1527 | if (fl->fl_flags & FL_UNLOCK_PENDING) |
1528 | continue; |
1529 | fl->fl_flags |= FL_UNLOCK_PENDING; |
1530 | fl->fl_break_time = break_time; |
1531 | } else { |
1532 | if (lease_breaking(fl)) |
1533 | continue; |
1534 | fl->fl_flags |= FL_DOWNGRADE_PENDING; |
1535 | fl->fl_downgrade_time = break_time; |
1536 | } |
1537 | if (fl->fl_lmops->lm_break(fl)) |
1538 | locks_delete_lock_ctx(fl, dispose: &dispose); |
1539 | } |
1540 | |
1541 | if (list_empty(head: &ctx->flc_lease)) |
1542 | goto out; |
1543 | |
1544 | if (mode & O_NONBLOCK) { |
1545 | trace_break_lease_noblock(inode, fl: new_fl); |
1546 | error = -EWOULDBLOCK; |
1547 | goto out; |
1548 | } |
1549 | |
1550 | restart: |
1551 | fl = list_first_entry(&ctx->flc_lease, struct file_lock, fl_list); |
1552 | break_time = fl->fl_break_time; |
1553 | if (break_time != 0) |
1554 | break_time -= jiffies; |
1555 | if (break_time == 0) |
1556 | break_time++; |
1557 | locks_insert_block(blocker: fl, waiter: new_fl, conflict: leases_conflict); |
1558 | trace_break_lease_block(inode, fl: new_fl); |
1559 | spin_unlock(lock: &ctx->flc_lock); |
1560 | percpu_up_read(sem: &file_rwsem); |
1561 | |
1562 | locks_dispose_list(dispose: &dispose); |
1563 | error = wait_event_interruptible_timeout(new_fl->fl_wait, |
1564 | list_empty(&new_fl->fl_blocked_member), |
1565 | break_time); |
1566 | |
1567 | percpu_down_read(sem: &file_rwsem); |
1568 | spin_lock(lock: &ctx->flc_lock); |
1569 | trace_break_lease_unblock(inode, fl: new_fl); |
1570 | locks_delete_block(new_fl); |
1571 | if (error >= 0) { |
1572 | /* |
1573 | * Wait for the next conflicting lease that has not been |
1574 | * broken yet |
1575 | */ |
1576 | if (error == 0) |
1577 | time_out_leases(inode, dispose: &dispose); |
1578 | if (any_leases_conflict(inode, breaker: new_fl)) |
1579 | goto restart; |
1580 | error = 0; |
1581 | } |
1582 | out: |
1583 | spin_unlock(lock: &ctx->flc_lock); |
1584 | percpu_up_read(sem: &file_rwsem); |
1585 | locks_dispose_list(dispose: &dispose); |
1586 | free_lock: |
1587 | locks_free_lock(new_fl); |
1588 | return error; |
1589 | } |
1590 | EXPORT_SYMBOL(__break_lease); |
1591 | |
1592 | /** |
1593 | * lease_get_mtime - update modified time of an inode with exclusive lease |
1594 | * @inode: the inode |
1595 | * @time: pointer to a timespec which contains the last modified time |
1596 | * |
1597 | * This is to force NFS clients to flush their caches for files with |
1598 | * exclusive leases. The justification is that if someone has an |
1599 | * exclusive lease, then they could be modifying it. |
1600 | */ |
1601 | void lease_get_mtime(struct inode *inode, struct timespec64 *time) |
1602 | { |
1603 | bool has_lease = false; |
1604 | struct file_lock_context *ctx; |
1605 | struct file_lock *fl; |
1606 | |
1607 | ctx = locks_inode_context(inode); |
1608 | if (ctx && !list_empty_careful(head: &ctx->flc_lease)) { |
1609 | spin_lock(lock: &ctx->flc_lock); |
1610 | fl = list_first_entry_or_null(&ctx->flc_lease, |
1611 | struct file_lock, fl_list); |
1612 | if (fl && (fl->fl_type == F_WRLCK)) |
1613 | has_lease = true; |
1614 | spin_unlock(lock: &ctx->flc_lock); |
1615 | } |
1616 | |
1617 | if (has_lease) |
1618 | *time = current_time(inode); |
1619 | } |
1620 | EXPORT_SYMBOL(lease_get_mtime); |
1621 | |
1622 | /** |
1623 | * fcntl_getlease - Enquire what lease is currently active |
1624 | * @filp: the file |
1625 | * |
1626 | * The value returned by this function will be one of |
1627 | * (if no lease break is pending): |
1628 | * |
1629 | * %F_RDLCK to indicate a shared lease is held. |
1630 | * |
1631 | * %F_WRLCK to indicate an exclusive lease is held. |
1632 | * |
1633 | * %F_UNLCK to indicate no lease is held. |
1634 | * |
1635 | * (if a lease break is pending): |
1636 | * |
1637 | * %F_RDLCK to indicate an exclusive lease needs to be |
1638 | * changed to a shared lease (or removed). |
1639 | * |
1640 | * %F_UNLCK to indicate the lease needs to be removed. |
1641 | * |
1642 | * XXX: sfr & willy disagree over whether F_INPROGRESS |
1643 | * should be returned to userspace. |
1644 | */ |
1645 | int fcntl_getlease(struct file *filp) |
1646 | { |
1647 | struct file_lock *fl; |
1648 | struct inode *inode = file_inode(f: filp); |
1649 | struct file_lock_context *ctx; |
1650 | int type = F_UNLCK; |
1651 | LIST_HEAD(dispose); |
1652 | |
1653 | ctx = locks_inode_context(inode); |
1654 | if (ctx && !list_empty_careful(head: &ctx->flc_lease)) { |
1655 | percpu_down_read(sem: &file_rwsem); |
1656 | spin_lock(lock: &ctx->flc_lock); |
1657 | time_out_leases(inode, dispose: &dispose); |
1658 | list_for_each_entry(fl, &ctx->flc_lease, fl_list) { |
1659 | if (fl->fl_file != filp) |
1660 | continue; |
1661 | type = target_leasetype(fl); |
1662 | break; |
1663 | } |
1664 | spin_unlock(lock: &ctx->flc_lock); |
1665 | percpu_up_read(sem: &file_rwsem); |
1666 | |
1667 | locks_dispose_list(dispose: &dispose); |
1668 | } |
1669 | return type; |
1670 | } |
1671 | |
1672 | /** |
1673 | * check_conflicting_open - see if the given file points to an inode that has |
1674 | * an existing open that would conflict with the |
1675 | * desired lease. |
1676 | * @filp: file to check |
1677 | * @arg: type of lease that we're trying to acquire |
1678 | * @flags: current lock flags |
1679 | * |
1680 | * Check to see if there's an existing open fd on this file that would |
1681 | * conflict with the lease we're trying to set. |
1682 | */ |
1683 | static int |
1684 | check_conflicting_open(struct file *filp, const int arg, int flags) |
1685 | { |
1686 | struct inode *inode = file_inode(f: filp); |
1687 | int self_wcount = 0, self_rcount = 0; |
1688 | |
1689 | if (flags & FL_LAYOUT) |
1690 | return 0; |
1691 | if (flags & FL_DELEG) |
1692 | /* We leave these checks to the caller */ |
1693 | return 0; |
1694 | |
1695 | if (arg == F_RDLCK) |
1696 | return inode_is_open_for_write(inode) ? -EAGAIN : 0; |
1697 | else if (arg != F_WRLCK) |
1698 | return 0; |
1699 | |
1700 | /* |
1701 | * Make sure that only read/write count is from lease requestor. |
1702 | * Note that this will result in denying write leases when i_writecount |
1703 | * is negative, which is what we want. (We shouldn't grant write leases |
1704 | * on files open for execution.) |
1705 | */ |
1706 | if (filp->f_mode & FMODE_WRITE) |
1707 | self_wcount = 1; |
1708 | else if (filp->f_mode & FMODE_READ) |
1709 | self_rcount = 1; |
1710 | |
1711 | if (atomic_read(v: &inode->i_writecount) != self_wcount || |
1712 | atomic_read(v: &inode->i_readcount) != self_rcount) |
1713 | return -EAGAIN; |
1714 | |
1715 | return 0; |
1716 | } |
1717 | |
1718 | static int |
1719 | generic_add_lease(struct file *filp, int arg, struct file_lock **flp, void **priv) |
1720 | { |
1721 | struct file_lock *fl, *my_fl = NULL, *lease; |
1722 | struct inode *inode = file_inode(f: filp); |
1723 | struct file_lock_context *ctx; |
1724 | bool is_deleg = (*flp)->fl_flags & FL_DELEG; |
1725 | int error; |
1726 | LIST_HEAD(dispose); |
1727 | |
1728 | lease = *flp; |
1729 | trace_generic_add_lease(inode, fl: lease); |
1730 | |
1731 | /* Note that arg is never F_UNLCK here */ |
1732 | ctx = locks_get_lock_context(inode, type: arg); |
1733 | if (!ctx) |
1734 | return -ENOMEM; |
1735 | |
1736 | /* |
1737 | * In the delegation case we need mutual exclusion with |
1738 | * a number of operations that take the i_mutex. We trylock |
1739 | * because delegations are an optional optimization, and if |
1740 | * there's some chance of a conflict--we'd rather not |
1741 | * bother, maybe that's a sign this just isn't a good file to |
1742 | * hand out a delegation on. |
1743 | */ |
1744 | if (is_deleg && !inode_trylock(inode)) |
1745 | return -EAGAIN; |
1746 | |
1747 | percpu_down_read(sem: &file_rwsem); |
1748 | spin_lock(lock: &ctx->flc_lock); |
1749 | time_out_leases(inode, dispose: &dispose); |
1750 | error = check_conflicting_open(filp, arg, flags: lease->fl_flags); |
1751 | if (error) |
1752 | goto out; |
1753 | |
1754 | /* |
1755 | * At this point, we know that if there is an exclusive |
1756 | * lease on this file, then we hold it on this filp |
1757 | * (otherwise our open of this file would have blocked). |
1758 | * And if we are trying to acquire an exclusive lease, |
1759 | * then the file is not open by anyone (including us) |
1760 | * except for this filp. |
1761 | */ |
1762 | error = -EAGAIN; |
1763 | list_for_each_entry(fl, &ctx->flc_lease, fl_list) { |
1764 | if (fl->fl_file == filp && |
1765 | fl->fl_owner == lease->fl_owner) { |
1766 | my_fl = fl; |
1767 | continue; |
1768 | } |
1769 | |
1770 | /* |
1771 | * No exclusive leases if someone else has a lease on |
1772 | * this file: |
1773 | */ |
1774 | if (arg == F_WRLCK) |
1775 | goto out; |
1776 | /* |
1777 | * Modifying our existing lease is OK, but no getting a |
1778 | * new lease if someone else is opening for write: |
1779 | */ |
1780 | if (fl->fl_flags & FL_UNLOCK_PENDING) |
1781 | goto out; |
1782 | } |
1783 | |
1784 | if (my_fl != NULL) { |
1785 | lease = my_fl; |
1786 | error = lease->fl_lmops->lm_change(lease, arg, &dispose); |
1787 | if (error) |
1788 | goto out; |
1789 | goto out_setup; |
1790 | } |
1791 | |
1792 | error = -EINVAL; |
1793 | if (!leases_enable) |
1794 | goto out; |
1795 | |
1796 | locks_insert_lock_ctx(fl: lease, before: &ctx->flc_lease); |
1797 | /* |
1798 | * The check in break_lease() is lockless. It's possible for another |
1799 | * open to race in after we did the earlier check for a conflicting |
1800 | * open but before the lease was inserted. Check again for a |
1801 | * conflicting open and cancel the lease if there is one. |
1802 | * |
1803 | * We also add a barrier here to ensure that the insertion of the lock |
1804 | * precedes these checks. |
1805 | */ |
1806 | smp_mb(); |
1807 | error = check_conflicting_open(filp, arg, flags: lease->fl_flags); |
1808 | if (error) { |
1809 | locks_unlink_lock_ctx(fl: lease); |
1810 | goto out; |
1811 | } |
1812 | |
1813 | out_setup: |
1814 | if (lease->fl_lmops->lm_setup) |
1815 | lease->fl_lmops->lm_setup(lease, priv); |
1816 | out: |
1817 | spin_unlock(lock: &ctx->flc_lock); |
1818 | percpu_up_read(sem: &file_rwsem); |
1819 | locks_dispose_list(dispose: &dispose); |
1820 | if (is_deleg) |
1821 | inode_unlock(inode); |
1822 | if (!error && !my_fl) |
1823 | *flp = NULL; |
1824 | return error; |
1825 | } |
1826 | |
1827 | static int generic_delete_lease(struct file *filp, void *owner) |
1828 | { |
1829 | int error = -EAGAIN; |
1830 | struct file_lock *fl, *victim = NULL; |
1831 | struct inode *inode = file_inode(f: filp); |
1832 | struct file_lock_context *ctx; |
1833 | LIST_HEAD(dispose); |
1834 | |
1835 | ctx = locks_inode_context(inode); |
1836 | if (!ctx) { |
1837 | trace_generic_delete_lease(inode, NULL); |
1838 | return error; |
1839 | } |
1840 | |
1841 | percpu_down_read(sem: &file_rwsem); |
1842 | spin_lock(lock: &ctx->flc_lock); |
1843 | list_for_each_entry(fl, &ctx->flc_lease, fl_list) { |
1844 | if (fl->fl_file == filp && |
1845 | fl->fl_owner == owner) { |
1846 | victim = fl; |
1847 | break; |
1848 | } |
1849 | } |
1850 | trace_generic_delete_lease(inode, fl: victim); |
1851 | if (victim) |
1852 | error = fl->fl_lmops->lm_change(victim, F_UNLCK, &dispose); |
1853 | spin_unlock(lock: &ctx->flc_lock); |
1854 | percpu_up_read(sem: &file_rwsem); |
1855 | locks_dispose_list(dispose: &dispose); |
1856 | return error; |
1857 | } |
1858 | |
1859 | /** |
1860 | * generic_setlease - sets a lease on an open file |
1861 | * @filp: file pointer |
1862 | * @arg: type of lease to obtain |
1863 | * @flp: input - file_lock to use, output - file_lock inserted |
1864 | * @priv: private data for lm_setup (may be NULL if lm_setup |
1865 | * doesn't require it) |
1866 | * |
1867 | * The (input) flp->fl_lmops->lm_break function is required |
1868 | * by break_lease(). |
1869 | */ |
1870 | int generic_setlease(struct file *filp, int arg, struct file_lock **flp, |
1871 | void **priv) |
1872 | { |
1873 | struct inode *inode = file_inode(f: filp); |
1874 | vfsuid_t vfsuid = i_uid_into_vfsuid(idmap: file_mnt_idmap(file: filp), inode); |
1875 | int error; |
1876 | |
1877 | if ((!vfsuid_eq_kuid(vfsuid, current_fsuid())) && !capable(CAP_LEASE)) |
1878 | return -EACCES; |
1879 | if (!S_ISREG(inode->i_mode)) |
1880 | return -EINVAL; |
1881 | error = security_file_lock(file: filp, cmd: arg); |
1882 | if (error) |
1883 | return error; |
1884 | |
1885 | switch (arg) { |
1886 | case F_UNLCK: |
1887 | return generic_delete_lease(filp, owner: *priv); |
1888 | case F_RDLCK: |
1889 | case F_WRLCK: |
1890 | if (!(*flp)->fl_lmops->lm_break) { |
1891 | WARN_ON_ONCE(1); |
1892 | return -ENOLCK; |
1893 | } |
1894 | |
1895 | return generic_add_lease(filp, arg, flp, priv); |
1896 | default: |
1897 | return -EINVAL; |
1898 | } |
1899 | } |
1900 | EXPORT_SYMBOL(generic_setlease); |
1901 | |
1902 | /* |
1903 | * Kernel subsystems can register to be notified on any attempt to set |
1904 | * a new lease with the lease_notifier_chain. This is used by (e.g.) nfsd |
1905 | * to close files that it may have cached when there is an attempt to set a |
1906 | * conflicting lease. |
1907 | */ |
1908 | static struct srcu_notifier_head lease_notifier_chain; |
1909 | |
1910 | static inline void |
1911 | lease_notifier_chain_init(void) |
1912 | { |
1913 | srcu_init_notifier_head(nh: &lease_notifier_chain); |
1914 | } |
1915 | |
1916 | static inline void |
1917 | setlease_notifier(int arg, struct file_lock *lease) |
1918 | { |
1919 | if (arg != F_UNLCK) |
1920 | srcu_notifier_call_chain(nh: &lease_notifier_chain, val: arg, v: lease); |
1921 | } |
1922 | |
1923 | int lease_register_notifier(struct notifier_block *nb) |
1924 | { |
1925 | return srcu_notifier_chain_register(nh: &lease_notifier_chain, nb); |
1926 | } |
1927 | EXPORT_SYMBOL_GPL(lease_register_notifier); |
1928 | |
1929 | void lease_unregister_notifier(struct notifier_block *nb) |
1930 | { |
1931 | srcu_notifier_chain_unregister(nh: &lease_notifier_chain, nb); |
1932 | } |
1933 | EXPORT_SYMBOL_GPL(lease_unregister_notifier); |
1934 | |
1935 | /** |
1936 | * vfs_setlease - sets a lease on an open file |
1937 | * @filp: file pointer |
1938 | * @arg: type of lease to obtain |
1939 | * @lease: file_lock to use when adding a lease |
1940 | * @priv: private info for lm_setup when adding a lease (may be |
1941 | * NULL if lm_setup doesn't require it) |
1942 | * |
1943 | * Call this to establish a lease on the file. The "lease" argument is not |
1944 | * used for F_UNLCK requests and may be NULL. For commands that set or alter |
1945 | * an existing lease, the ``(*lease)->fl_lmops->lm_break`` operation must be |
1946 | * set; if not, this function will return -ENOLCK (and generate a scary-looking |
1947 | * stack trace). |
1948 | * |
1949 | * The "priv" pointer is passed directly to the lm_setup function as-is. It |
1950 | * may be NULL if the lm_setup operation doesn't require it. |
1951 | */ |
1952 | int |
1953 | vfs_setlease(struct file *filp, int arg, struct file_lock **lease, void **priv) |
1954 | { |
1955 | if (lease) |
1956 | setlease_notifier(arg, lease: *lease); |
1957 | if (filp->f_op->setlease) |
1958 | return filp->f_op->setlease(filp, arg, lease, priv); |
1959 | else |
1960 | return generic_setlease(filp, arg, lease, priv); |
1961 | } |
1962 | EXPORT_SYMBOL_GPL(vfs_setlease); |
1963 | |
1964 | static int do_fcntl_add_lease(unsigned int fd, struct file *filp, int arg) |
1965 | { |
1966 | struct file_lock *fl; |
1967 | struct fasync_struct *new; |
1968 | int error; |
1969 | |
1970 | fl = lease_alloc(filp, type: arg); |
1971 | if (IS_ERR(ptr: fl)) |
1972 | return PTR_ERR(ptr: fl); |
1973 | |
1974 | new = fasync_alloc(); |
1975 | if (!new) { |
1976 | locks_free_lock(fl); |
1977 | return -ENOMEM; |
1978 | } |
1979 | new->fa_fd = fd; |
1980 | |
1981 | error = vfs_setlease(filp, arg, &fl, (void **)&new); |
1982 | if (fl) |
1983 | locks_free_lock(fl); |
1984 | if (new) |
1985 | fasync_free(new); |
1986 | return error; |
1987 | } |
1988 | |
1989 | /** |
1990 | * fcntl_setlease - sets a lease on an open file |
1991 | * @fd: open file descriptor |
1992 | * @filp: file pointer |
1993 | * @arg: type of lease to obtain |
1994 | * |
1995 | * Call this fcntl to establish a lease on the file. |
1996 | * Note that you also need to call %F_SETSIG to |
1997 | * receive a signal when the lease is broken. |
1998 | */ |
1999 | int fcntl_setlease(unsigned int fd, struct file *filp, int arg) |
2000 | { |
2001 | if (arg == F_UNLCK) |
2002 | return vfs_setlease(filp, F_UNLCK, NULL, (void **)&filp); |
2003 | return do_fcntl_add_lease(fd, filp, arg); |
2004 | } |
2005 | |
2006 | /** |
2007 | * flock_lock_inode_wait - Apply a FLOCK-style lock to a file |
2008 | * @inode: inode of the file to apply to |
2009 | * @fl: The lock to be applied |
2010 | * |
2011 | * Apply a FLOCK style lock request to an inode. |
2012 | */ |
2013 | static int flock_lock_inode_wait(struct inode *inode, struct file_lock *fl) |
2014 | { |
2015 | int error; |
2016 | might_sleep(); |
2017 | for (;;) { |
2018 | error = flock_lock_inode(inode, request: fl); |
2019 | if (error != FILE_LOCK_DEFERRED) |
2020 | break; |
2021 | error = wait_event_interruptible(fl->fl_wait, |
2022 | list_empty(&fl->fl_blocked_member)); |
2023 | if (error) |
2024 | break; |
2025 | } |
2026 | locks_delete_block(fl); |
2027 | return error; |
2028 | } |
2029 | |
2030 | /** |
2031 | * locks_lock_inode_wait - Apply a lock to an inode |
2032 | * @inode: inode of the file to apply to |
2033 | * @fl: The lock to be applied |
2034 | * |
2035 | * Apply a POSIX or FLOCK style lock request to an inode. |
2036 | */ |
2037 | int locks_lock_inode_wait(struct inode *inode, struct file_lock *fl) |
2038 | { |
2039 | int res = 0; |
2040 | switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) { |
2041 | case FL_POSIX: |
2042 | res = posix_lock_inode_wait(inode, fl); |
2043 | break; |
2044 | case FL_FLOCK: |
2045 | res = flock_lock_inode_wait(inode, fl); |
2046 | break; |
2047 | default: |
2048 | BUG(); |
2049 | } |
2050 | return res; |
2051 | } |
2052 | EXPORT_SYMBOL(locks_lock_inode_wait); |
2053 | |
2054 | /** |
2055 | * sys_flock: - flock() system call. |
2056 | * @fd: the file descriptor to lock. |
2057 | * @cmd: the type of lock to apply. |
2058 | * |
2059 | * Apply a %FL_FLOCK style lock to an open file descriptor. |
2060 | * The @cmd can be one of: |
2061 | * |
2062 | * - %LOCK_SH -- a shared lock. |
2063 | * - %LOCK_EX -- an exclusive lock. |
2064 | * - %LOCK_UN -- remove an existing lock. |
2065 | * - %LOCK_MAND -- a 'mandatory' flock. (DEPRECATED) |
2066 | * |
2067 | * %LOCK_MAND support has been removed from the kernel. |
2068 | */ |
2069 | SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd) |
2070 | { |
2071 | int can_sleep, error, type; |
2072 | struct file_lock fl; |
2073 | struct fd f; |
2074 | |
2075 | /* |
2076 | * LOCK_MAND locks were broken for a long time in that they never |
2077 | * conflicted with one another and didn't prevent any sort of open, |
2078 | * read or write activity. |
2079 | * |
2080 | * Just ignore these requests now, to preserve legacy behavior, but |
2081 | * throw a warning to let people know that they don't actually work. |
2082 | */ |
2083 | if (cmd & LOCK_MAND) { |
2084 | pr_warn_once("%s(%d): Attempt to set a LOCK_MAND lock via flock(2). This support has been removed and the request ignored.\n" , current->comm, current->pid); |
2085 | return 0; |
2086 | } |
2087 | |
2088 | type = flock_translate_cmd(cmd: cmd & ~LOCK_NB); |
2089 | if (type < 0) |
2090 | return type; |
2091 | |
2092 | error = -EBADF; |
2093 | f = fdget(fd); |
2094 | if (!f.file) |
2095 | return error; |
2096 | |
2097 | if (type != F_UNLCK && !(f.file->f_mode & (FMODE_READ | FMODE_WRITE))) |
2098 | goto out_putf; |
2099 | |
2100 | flock_make_lock(filp: f.file, fl: &fl, type); |
2101 | |
2102 | error = security_file_lock(file: f.file, cmd: fl.fl_type); |
2103 | if (error) |
2104 | goto out_putf; |
2105 | |
2106 | can_sleep = !(cmd & LOCK_NB); |
2107 | if (can_sleep) |
2108 | fl.fl_flags |= FL_SLEEP; |
2109 | |
2110 | if (f.file->f_op->flock) |
2111 | error = f.file->f_op->flock(f.file, |
2112 | (can_sleep) ? F_SETLKW : F_SETLK, |
2113 | &fl); |
2114 | else |
2115 | error = locks_lock_file_wait(filp: f.file, fl: &fl); |
2116 | |
2117 | locks_release_private(&fl); |
2118 | out_putf: |
2119 | fdput(fd: f); |
2120 | |
2121 | return error; |
2122 | } |
2123 | |
2124 | /** |
2125 | * vfs_test_lock - test file byte range lock |
2126 | * @filp: The file to test lock for |
2127 | * @fl: The lock to test; also used to hold result |
2128 | * |
2129 | * Returns -ERRNO on failure. Indicates presence of conflicting lock by |
2130 | * setting conf->fl_type to something other than F_UNLCK. |
2131 | */ |
2132 | int vfs_test_lock(struct file *filp, struct file_lock *fl) |
2133 | { |
2134 | WARN_ON_ONCE(filp != fl->fl_file); |
2135 | if (filp->f_op->lock) |
2136 | return filp->f_op->lock(filp, F_GETLK, fl); |
2137 | posix_test_lock(filp, fl); |
2138 | return 0; |
2139 | } |
2140 | EXPORT_SYMBOL_GPL(vfs_test_lock); |
2141 | |
2142 | /** |
2143 | * locks_translate_pid - translate a file_lock's fl_pid number into a namespace |
2144 | * @fl: The file_lock who's fl_pid should be translated |
2145 | * @ns: The namespace into which the pid should be translated |
2146 | * |
2147 | * Used to translate a fl_pid into a namespace virtual pid number |
2148 | */ |
2149 | static pid_t locks_translate_pid(struct file_lock *fl, struct pid_namespace *ns) |
2150 | { |
2151 | pid_t vnr; |
2152 | struct pid *pid; |
2153 | |
2154 | if (IS_OFDLCK(fl)) |
2155 | return -1; |
2156 | if (IS_REMOTELCK(fl)) |
2157 | return fl->fl_pid; |
2158 | /* |
2159 | * If the flock owner process is dead and its pid has been already |
2160 | * freed, the translation below won't work, but we still want to show |
2161 | * flock owner pid number in init pidns. |
2162 | */ |
2163 | if (ns == &init_pid_ns) |
2164 | return (pid_t)fl->fl_pid; |
2165 | |
2166 | rcu_read_lock(); |
2167 | pid = find_pid_ns(nr: fl->fl_pid, ns: &init_pid_ns); |
2168 | vnr = pid_nr_ns(pid, ns); |
2169 | rcu_read_unlock(); |
2170 | return vnr; |
2171 | } |
2172 | |
2173 | static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl) |
2174 | { |
2175 | flock->l_pid = locks_translate_pid(fl, ns: task_active_pid_ns(current)); |
2176 | #if BITS_PER_LONG == 32 |
2177 | /* |
2178 | * Make sure we can represent the posix lock via |
2179 | * legacy 32bit flock. |
2180 | */ |
2181 | if (fl->fl_start > OFFT_OFFSET_MAX) |
2182 | return -EOVERFLOW; |
2183 | if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX) |
2184 | return -EOVERFLOW; |
2185 | #endif |
2186 | flock->l_start = fl->fl_start; |
2187 | flock->l_len = fl->fl_end == OFFSET_MAX ? 0 : |
2188 | fl->fl_end - fl->fl_start + 1; |
2189 | flock->l_whence = 0; |
2190 | flock->l_type = fl->fl_type; |
2191 | return 0; |
2192 | } |
2193 | |
2194 | #if BITS_PER_LONG == 32 |
2195 | static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl) |
2196 | { |
2197 | flock->l_pid = locks_translate_pid(fl, task_active_pid_ns(current)); |
2198 | flock->l_start = fl->fl_start; |
2199 | flock->l_len = fl->fl_end == OFFSET_MAX ? 0 : |
2200 | fl->fl_end - fl->fl_start + 1; |
2201 | flock->l_whence = 0; |
2202 | flock->l_type = fl->fl_type; |
2203 | } |
2204 | #endif |
2205 | |
2206 | /* Report the first existing lock that would conflict with l. |
2207 | * This implements the F_GETLK command of fcntl(). |
2208 | */ |
2209 | int fcntl_getlk(struct file *filp, unsigned int cmd, struct flock *flock) |
2210 | { |
2211 | struct file_lock *fl; |
2212 | int error; |
2213 | |
2214 | fl = locks_alloc_lock(); |
2215 | if (fl == NULL) |
2216 | return -ENOMEM; |
2217 | error = -EINVAL; |
2218 | if (cmd != F_OFD_GETLK && flock->l_type != F_RDLCK |
2219 | && flock->l_type != F_WRLCK) |
2220 | goto out; |
2221 | |
2222 | error = flock_to_posix_lock(filp, fl, l: flock); |
2223 | if (error) |
2224 | goto out; |
2225 | |
2226 | if (cmd == F_OFD_GETLK) { |
2227 | error = -EINVAL; |
2228 | if (flock->l_pid != 0) |
2229 | goto out; |
2230 | |
2231 | fl->fl_flags |= FL_OFDLCK; |
2232 | fl->fl_owner = filp; |
2233 | } |
2234 | |
2235 | error = vfs_test_lock(filp, fl); |
2236 | if (error) |
2237 | goto out; |
2238 | |
2239 | flock->l_type = fl->fl_type; |
2240 | if (fl->fl_type != F_UNLCK) { |
2241 | error = posix_lock_to_flock(flock, fl); |
2242 | if (error) |
2243 | goto out; |
2244 | } |
2245 | out: |
2246 | locks_free_lock(fl); |
2247 | return error; |
2248 | } |
2249 | |
2250 | /** |
2251 | * vfs_lock_file - file byte range lock |
2252 | * @filp: The file to apply the lock to |
2253 | * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.) |
2254 | * @fl: The lock to be applied |
2255 | * @conf: Place to return a copy of the conflicting lock, if found. |
2256 | * |
2257 | * A caller that doesn't care about the conflicting lock may pass NULL |
2258 | * as the final argument. |
2259 | * |
2260 | * If the filesystem defines a private ->lock() method, then @conf will |
2261 | * be left unchanged; so a caller that cares should initialize it to |
2262 | * some acceptable default. |
2263 | * |
2264 | * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX |
2265 | * locks, the ->lock() interface may return asynchronously, before the lock has |
2266 | * been granted or denied by the underlying filesystem, if (and only if) |
2267 | * lm_grant is set. Additionally EXPORT_OP_ASYNC_LOCK in export_operations |
2268 | * flags need to be set. |
2269 | * |
2270 | * Callers expecting ->lock() to return asynchronously will only use F_SETLK, |
2271 | * not F_SETLKW; they will set FL_SLEEP if (and only if) the request is for a |
2272 | * blocking lock. When ->lock() does return asynchronously, it must return |
2273 | * FILE_LOCK_DEFERRED, and call ->lm_grant() when the lock request completes. |
2274 | * If the request is for non-blocking lock the file system should return |
2275 | * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine |
2276 | * with the result. If the request timed out the callback routine will return a |
2277 | * nonzero return code and the file system should release the lock. The file |
2278 | * system is also responsible to keep a corresponding posix lock when it |
2279 | * grants a lock so the VFS can find out which locks are locally held and do |
2280 | * the correct lock cleanup when required. |
2281 | * The underlying filesystem must not drop the kernel lock or call |
2282 | * ->lm_grant() before returning to the caller with a FILE_LOCK_DEFERRED |
2283 | * return code. |
2284 | */ |
2285 | int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf) |
2286 | { |
2287 | WARN_ON_ONCE(filp != fl->fl_file); |
2288 | if (filp->f_op->lock) |
2289 | return filp->f_op->lock(filp, cmd, fl); |
2290 | else |
2291 | return posix_lock_file(filp, fl, conf); |
2292 | } |
2293 | EXPORT_SYMBOL_GPL(vfs_lock_file); |
2294 | |
2295 | static int do_lock_file_wait(struct file *filp, unsigned int cmd, |
2296 | struct file_lock *fl) |
2297 | { |
2298 | int error; |
2299 | |
2300 | error = security_file_lock(file: filp, cmd: fl->fl_type); |
2301 | if (error) |
2302 | return error; |
2303 | |
2304 | for (;;) { |
2305 | error = vfs_lock_file(filp, cmd, fl, NULL); |
2306 | if (error != FILE_LOCK_DEFERRED) |
2307 | break; |
2308 | error = wait_event_interruptible(fl->fl_wait, |
2309 | list_empty(&fl->fl_blocked_member)); |
2310 | if (error) |
2311 | break; |
2312 | } |
2313 | locks_delete_block(fl); |
2314 | |
2315 | return error; |
2316 | } |
2317 | |
2318 | /* Ensure that fl->fl_file has compatible f_mode for F_SETLK calls */ |
2319 | static int |
2320 | check_fmode_for_setlk(struct file_lock *fl) |
2321 | { |
2322 | switch (fl->fl_type) { |
2323 | case F_RDLCK: |
2324 | if (!(fl->fl_file->f_mode & FMODE_READ)) |
2325 | return -EBADF; |
2326 | break; |
2327 | case F_WRLCK: |
2328 | if (!(fl->fl_file->f_mode & FMODE_WRITE)) |
2329 | return -EBADF; |
2330 | } |
2331 | return 0; |
2332 | } |
2333 | |
2334 | /* Apply the lock described by l to an open file descriptor. |
2335 | * This implements both the F_SETLK and F_SETLKW commands of fcntl(). |
2336 | */ |
2337 | int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd, |
2338 | struct flock *flock) |
2339 | { |
2340 | struct file_lock *file_lock = locks_alloc_lock(); |
2341 | struct inode *inode = file_inode(f: filp); |
2342 | struct file *f; |
2343 | int error; |
2344 | |
2345 | if (file_lock == NULL) |
2346 | return -ENOLCK; |
2347 | |
2348 | error = flock_to_posix_lock(filp, fl: file_lock, l: flock); |
2349 | if (error) |
2350 | goto out; |
2351 | |
2352 | error = check_fmode_for_setlk(fl: file_lock); |
2353 | if (error) |
2354 | goto out; |
2355 | |
2356 | /* |
2357 | * If the cmd is requesting file-private locks, then set the |
2358 | * FL_OFDLCK flag and override the owner. |
2359 | */ |
2360 | switch (cmd) { |
2361 | case F_OFD_SETLK: |
2362 | error = -EINVAL; |
2363 | if (flock->l_pid != 0) |
2364 | goto out; |
2365 | |
2366 | cmd = F_SETLK; |
2367 | file_lock->fl_flags |= FL_OFDLCK; |
2368 | file_lock->fl_owner = filp; |
2369 | break; |
2370 | case F_OFD_SETLKW: |
2371 | error = -EINVAL; |
2372 | if (flock->l_pid != 0) |
2373 | goto out; |
2374 | |
2375 | cmd = F_SETLKW; |
2376 | file_lock->fl_flags |= FL_OFDLCK; |
2377 | file_lock->fl_owner = filp; |
2378 | fallthrough; |
2379 | case F_SETLKW: |
2380 | file_lock->fl_flags |= FL_SLEEP; |
2381 | } |
2382 | |
2383 | error = do_lock_file_wait(filp, cmd, fl: file_lock); |
2384 | |
2385 | /* |
2386 | * Attempt to detect a close/fcntl race and recover by releasing the |
2387 | * lock that was just acquired. There is no need to do that when we're |
2388 | * unlocking though, or for OFD locks. |
2389 | */ |
2390 | if (!error && file_lock->fl_type != F_UNLCK && |
2391 | !(file_lock->fl_flags & FL_OFDLCK)) { |
2392 | struct files_struct *files = current->files; |
2393 | /* |
2394 | * We need that spin_lock here - it prevents reordering between |
2395 | * update of i_flctx->flc_posix and check for it done in |
2396 | * close(). rcu_read_lock() wouldn't do. |
2397 | */ |
2398 | spin_lock(lock: &files->file_lock); |
2399 | f = files_lookup_fd_locked(files, fd); |
2400 | spin_unlock(lock: &files->file_lock); |
2401 | if (f != filp) { |
2402 | file_lock->fl_type = F_UNLCK; |
2403 | error = do_lock_file_wait(filp, cmd, fl: file_lock); |
2404 | WARN_ON_ONCE(error); |
2405 | error = -EBADF; |
2406 | } |
2407 | } |
2408 | out: |
2409 | trace_fcntl_setlk(inode, fl: file_lock, ret: error); |
2410 | locks_free_lock(file_lock); |
2411 | return error; |
2412 | } |
2413 | |
2414 | #if BITS_PER_LONG == 32 |
2415 | /* Report the first existing lock that would conflict with l. |
2416 | * This implements the F_GETLK command of fcntl(). |
2417 | */ |
2418 | int fcntl_getlk64(struct file *filp, unsigned int cmd, struct flock64 *flock) |
2419 | { |
2420 | struct file_lock *fl; |
2421 | int error; |
2422 | |
2423 | fl = locks_alloc_lock(); |
2424 | if (fl == NULL) |
2425 | return -ENOMEM; |
2426 | |
2427 | error = -EINVAL; |
2428 | if (cmd != F_OFD_GETLK && flock->l_type != F_RDLCK |
2429 | && flock->l_type != F_WRLCK) |
2430 | goto out; |
2431 | |
2432 | error = flock64_to_posix_lock(filp, fl, flock); |
2433 | if (error) |
2434 | goto out; |
2435 | |
2436 | if (cmd == F_OFD_GETLK) { |
2437 | error = -EINVAL; |
2438 | if (flock->l_pid != 0) |
2439 | goto out; |
2440 | |
2441 | fl->fl_flags |= FL_OFDLCK; |
2442 | fl->fl_owner = filp; |
2443 | } |
2444 | |
2445 | error = vfs_test_lock(filp, fl); |
2446 | if (error) |
2447 | goto out; |
2448 | |
2449 | flock->l_type = fl->fl_type; |
2450 | if (fl->fl_type != F_UNLCK) |
2451 | posix_lock_to_flock64(flock, fl); |
2452 | |
2453 | out: |
2454 | locks_free_lock(fl); |
2455 | return error; |
2456 | } |
2457 | |
2458 | /* Apply the lock described by l to an open file descriptor. |
2459 | * This implements both the F_SETLK and F_SETLKW commands of fcntl(). |
2460 | */ |
2461 | int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd, |
2462 | struct flock64 *flock) |
2463 | { |
2464 | struct file_lock *file_lock = locks_alloc_lock(); |
2465 | struct file *f; |
2466 | int error; |
2467 | |
2468 | if (file_lock == NULL) |
2469 | return -ENOLCK; |
2470 | |
2471 | error = flock64_to_posix_lock(filp, file_lock, flock); |
2472 | if (error) |
2473 | goto out; |
2474 | |
2475 | error = check_fmode_for_setlk(file_lock); |
2476 | if (error) |
2477 | goto out; |
2478 | |
2479 | /* |
2480 | * If the cmd is requesting file-private locks, then set the |
2481 | * FL_OFDLCK flag and override the owner. |
2482 | */ |
2483 | switch (cmd) { |
2484 | case F_OFD_SETLK: |
2485 | error = -EINVAL; |
2486 | if (flock->l_pid != 0) |
2487 | goto out; |
2488 | |
2489 | cmd = F_SETLK64; |
2490 | file_lock->fl_flags |= FL_OFDLCK; |
2491 | file_lock->fl_owner = filp; |
2492 | break; |
2493 | case F_OFD_SETLKW: |
2494 | error = -EINVAL; |
2495 | if (flock->l_pid != 0) |
2496 | goto out; |
2497 | |
2498 | cmd = F_SETLKW64; |
2499 | file_lock->fl_flags |= FL_OFDLCK; |
2500 | file_lock->fl_owner = filp; |
2501 | fallthrough; |
2502 | case F_SETLKW64: |
2503 | file_lock->fl_flags |= FL_SLEEP; |
2504 | } |
2505 | |
2506 | error = do_lock_file_wait(filp, cmd, file_lock); |
2507 | |
2508 | /* |
2509 | * Attempt to detect a close/fcntl race and recover by releasing the |
2510 | * lock that was just acquired. There is no need to do that when we're |
2511 | * unlocking though, or for OFD locks. |
2512 | */ |
2513 | if (!error && file_lock->fl_type != F_UNLCK && |
2514 | !(file_lock->fl_flags & FL_OFDLCK)) { |
2515 | struct files_struct *files = current->files; |
2516 | /* |
2517 | * We need that spin_lock here - it prevents reordering between |
2518 | * update of i_flctx->flc_posix and check for it done in |
2519 | * close(). rcu_read_lock() wouldn't do. |
2520 | */ |
2521 | spin_lock(&files->file_lock); |
2522 | f = files_lookup_fd_locked(files, fd); |
2523 | spin_unlock(&files->file_lock); |
2524 | if (f != filp) { |
2525 | file_lock->fl_type = F_UNLCK; |
2526 | error = do_lock_file_wait(filp, cmd, file_lock); |
2527 | WARN_ON_ONCE(error); |
2528 | error = -EBADF; |
2529 | } |
2530 | } |
2531 | out: |
2532 | locks_free_lock(file_lock); |
2533 | return error; |
2534 | } |
2535 | #endif /* BITS_PER_LONG == 32 */ |
2536 | |
2537 | /* |
2538 | * This function is called when the file is being removed |
2539 | * from the task's fd array. POSIX locks belonging to this task |
2540 | * are deleted at this time. |
2541 | */ |
2542 | void locks_remove_posix(struct file *filp, fl_owner_t owner) |
2543 | { |
2544 | int error; |
2545 | struct inode *inode = file_inode(f: filp); |
2546 | struct file_lock lock; |
2547 | struct file_lock_context *ctx; |
2548 | |
2549 | /* |
2550 | * If there are no locks held on this file, we don't need to call |
2551 | * posix_lock_file(). Another process could be setting a lock on this |
2552 | * file at the same time, but we wouldn't remove that lock anyway. |
2553 | */ |
2554 | ctx = locks_inode_context(inode); |
2555 | if (!ctx || list_empty(head: &ctx->flc_posix)) |
2556 | return; |
2557 | |
2558 | locks_init_lock(&lock); |
2559 | lock.fl_type = F_UNLCK; |
2560 | lock.fl_flags = FL_POSIX | FL_CLOSE; |
2561 | lock.fl_start = 0; |
2562 | lock.fl_end = OFFSET_MAX; |
2563 | lock.fl_owner = owner; |
2564 | lock.fl_pid = current->tgid; |
2565 | lock.fl_file = filp; |
2566 | lock.fl_ops = NULL; |
2567 | lock.fl_lmops = NULL; |
2568 | |
2569 | error = vfs_lock_file(filp, F_SETLK, &lock, NULL); |
2570 | |
2571 | if (lock.fl_ops && lock.fl_ops->fl_release_private) |
2572 | lock.fl_ops->fl_release_private(&lock); |
2573 | trace_locks_remove_posix(inode, fl: &lock, ret: error); |
2574 | } |
2575 | EXPORT_SYMBOL(locks_remove_posix); |
2576 | |
2577 | /* The i_flctx must be valid when calling into here */ |
2578 | static void |
2579 | locks_remove_flock(struct file *filp, struct file_lock_context *flctx) |
2580 | { |
2581 | struct file_lock fl; |
2582 | struct inode *inode = file_inode(f: filp); |
2583 | |
2584 | if (list_empty(head: &flctx->flc_flock)) |
2585 | return; |
2586 | |
2587 | flock_make_lock(filp, fl: &fl, F_UNLCK); |
2588 | fl.fl_flags |= FL_CLOSE; |
2589 | |
2590 | if (filp->f_op->flock) |
2591 | filp->f_op->flock(filp, F_SETLKW, &fl); |
2592 | else |
2593 | flock_lock_inode(inode, request: &fl); |
2594 | |
2595 | if (fl.fl_ops && fl.fl_ops->fl_release_private) |
2596 | fl.fl_ops->fl_release_private(&fl); |
2597 | } |
2598 | |
2599 | /* The i_flctx must be valid when calling into here */ |
2600 | static void |
2601 | locks_remove_lease(struct file *filp, struct file_lock_context *ctx) |
2602 | { |
2603 | struct file_lock *fl, *tmp; |
2604 | LIST_HEAD(dispose); |
2605 | |
2606 | if (list_empty(head: &ctx->flc_lease)) |
2607 | return; |
2608 | |
2609 | percpu_down_read(sem: &file_rwsem); |
2610 | spin_lock(lock: &ctx->flc_lock); |
2611 | list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list) |
2612 | if (filp == fl->fl_file) |
2613 | lease_modify(fl, F_UNLCK, &dispose); |
2614 | spin_unlock(lock: &ctx->flc_lock); |
2615 | percpu_up_read(sem: &file_rwsem); |
2616 | |
2617 | locks_dispose_list(dispose: &dispose); |
2618 | } |
2619 | |
2620 | /* |
2621 | * This function is called on the last close of an open file. |
2622 | */ |
2623 | void locks_remove_file(struct file *filp) |
2624 | { |
2625 | struct file_lock_context *ctx; |
2626 | |
2627 | ctx = locks_inode_context(inode: file_inode(f: filp)); |
2628 | if (!ctx) |
2629 | return; |
2630 | |
2631 | /* remove any OFD locks */ |
2632 | locks_remove_posix(filp, filp); |
2633 | |
2634 | /* remove flock locks */ |
2635 | locks_remove_flock(filp, flctx: ctx); |
2636 | |
2637 | /* remove any leases */ |
2638 | locks_remove_lease(filp, ctx); |
2639 | |
2640 | spin_lock(lock: &ctx->flc_lock); |
2641 | locks_check_ctx_file_list(filp, list: &ctx->flc_posix, list_type: "POSIX" ); |
2642 | locks_check_ctx_file_list(filp, list: &ctx->flc_flock, list_type: "FLOCK" ); |
2643 | locks_check_ctx_file_list(filp, list: &ctx->flc_lease, list_type: "LEASE" ); |
2644 | spin_unlock(lock: &ctx->flc_lock); |
2645 | } |
2646 | |
2647 | /** |
2648 | * vfs_cancel_lock - file byte range unblock lock |
2649 | * @filp: The file to apply the unblock to |
2650 | * @fl: The lock to be unblocked |
2651 | * |
2652 | * Used by lock managers to cancel blocked requests |
2653 | */ |
2654 | int vfs_cancel_lock(struct file *filp, struct file_lock *fl) |
2655 | { |
2656 | WARN_ON_ONCE(filp != fl->fl_file); |
2657 | if (filp->f_op->lock) |
2658 | return filp->f_op->lock(filp, F_CANCELLK, fl); |
2659 | return 0; |
2660 | } |
2661 | EXPORT_SYMBOL_GPL(vfs_cancel_lock); |
2662 | |
2663 | /** |
2664 | * vfs_inode_has_locks - are any file locks held on @inode? |
2665 | * @inode: inode to check for locks |
2666 | * |
2667 | * Return true if there are any FL_POSIX or FL_FLOCK locks currently |
2668 | * set on @inode. |
2669 | */ |
2670 | bool vfs_inode_has_locks(struct inode *inode) |
2671 | { |
2672 | struct file_lock_context *ctx; |
2673 | bool ret; |
2674 | |
2675 | ctx = locks_inode_context(inode); |
2676 | if (!ctx) |
2677 | return false; |
2678 | |
2679 | spin_lock(lock: &ctx->flc_lock); |
2680 | ret = !list_empty(head: &ctx->flc_posix) || !list_empty(head: &ctx->flc_flock); |
2681 | spin_unlock(lock: &ctx->flc_lock); |
2682 | return ret; |
2683 | } |
2684 | EXPORT_SYMBOL_GPL(vfs_inode_has_locks); |
2685 | |
2686 | #ifdef CONFIG_PROC_FS |
2687 | #include <linux/proc_fs.h> |
2688 | #include <linux/seq_file.h> |
2689 | |
2690 | struct locks_iterator { |
2691 | int li_cpu; |
2692 | loff_t li_pos; |
2693 | }; |
2694 | |
2695 | static void lock_get_status(struct seq_file *f, struct file_lock *fl, |
2696 | loff_t id, char *pfx, int repeat) |
2697 | { |
2698 | struct inode *inode = NULL; |
2699 | unsigned int fl_pid; |
2700 | struct pid_namespace *proc_pidns = proc_pid_ns(sb: file_inode(f: f->file)->i_sb); |
2701 | int type; |
2702 | |
2703 | fl_pid = locks_translate_pid(fl, ns: proc_pidns); |
2704 | /* |
2705 | * If lock owner is dead (and pid is freed) or not visible in current |
2706 | * pidns, zero is shown as a pid value. Check lock info from |
2707 | * init_pid_ns to get saved lock pid value. |
2708 | */ |
2709 | |
2710 | if (fl->fl_file != NULL) |
2711 | inode = file_inode(f: fl->fl_file); |
2712 | |
2713 | seq_printf(m: f, fmt: "%lld: " , id); |
2714 | |
2715 | if (repeat) |
2716 | seq_printf(m: f, fmt: "%*s" , repeat - 1 + (int)strlen(pfx), pfx); |
2717 | |
2718 | if (IS_POSIX(fl)) { |
2719 | if (fl->fl_flags & FL_ACCESS) |
2720 | seq_puts(m: f, s: "ACCESS" ); |
2721 | else if (IS_OFDLCK(fl)) |
2722 | seq_puts(m: f, s: "OFDLCK" ); |
2723 | else |
2724 | seq_puts(m: f, s: "POSIX " ); |
2725 | |
2726 | seq_printf(m: f, fmt: " %s " , |
2727 | (inode == NULL) ? "*NOINODE*" : "ADVISORY " ); |
2728 | } else if (IS_FLOCK(fl)) { |
2729 | seq_puts(m: f, s: "FLOCK ADVISORY " ); |
2730 | } else if (IS_LEASE(fl)) { |
2731 | if (fl->fl_flags & FL_DELEG) |
2732 | seq_puts(m: f, s: "DELEG " ); |
2733 | else |
2734 | seq_puts(m: f, s: "LEASE " ); |
2735 | |
2736 | if (lease_breaking(fl)) |
2737 | seq_puts(m: f, s: "BREAKING " ); |
2738 | else if (fl->fl_file) |
2739 | seq_puts(m: f, s: "ACTIVE " ); |
2740 | else |
2741 | seq_puts(m: f, s: "BREAKER " ); |
2742 | } else { |
2743 | seq_puts(m: f, s: "UNKNOWN UNKNOWN " ); |
2744 | } |
2745 | type = IS_LEASE(fl) ? target_leasetype(fl) : fl->fl_type; |
2746 | |
2747 | seq_printf(m: f, fmt: "%s " , (type == F_WRLCK) ? "WRITE" : |
2748 | (type == F_RDLCK) ? "READ" : "UNLCK" ); |
2749 | if (inode) { |
2750 | /* userspace relies on this representation of dev_t */ |
2751 | seq_printf(m: f, fmt: "%d %02x:%02x:%lu " , fl_pid, |
2752 | MAJOR(inode->i_sb->s_dev), |
2753 | MINOR(inode->i_sb->s_dev), inode->i_ino); |
2754 | } else { |
2755 | seq_printf(m: f, fmt: "%d <none>:0 " , fl_pid); |
2756 | } |
2757 | if (IS_POSIX(fl)) { |
2758 | if (fl->fl_end == OFFSET_MAX) |
2759 | seq_printf(m: f, fmt: "%Ld EOF\n" , fl->fl_start); |
2760 | else |
2761 | seq_printf(m: f, fmt: "%Ld %Ld\n" , fl->fl_start, fl->fl_end); |
2762 | } else { |
2763 | seq_puts(m: f, s: "0 EOF\n" ); |
2764 | } |
2765 | } |
2766 | |
2767 | static struct file_lock *get_next_blocked_member(struct file_lock *node) |
2768 | { |
2769 | struct file_lock *tmp; |
2770 | |
2771 | /* NULL node or root node */ |
2772 | if (node == NULL || node->fl_blocker == NULL) |
2773 | return NULL; |
2774 | |
2775 | /* Next member in the linked list could be itself */ |
2776 | tmp = list_next_entry(node, fl_blocked_member); |
2777 | if (list_entry_is_head(tmp, &node->fl_blocker->fl_blocked_requests, fl_blocked_member) |
2778 | || tmp == node) { |
2779 | return NULL; |
2780 | } |
2781 | |
2782 | return tmp; |
2783 | } |
2784 | |
2785 | static int locks_show(struct seq_file *f, void *v) |
2786 | { |
2787 | struct locks_iterator *iter = f->private; |
2788 | struct file_lock *cur, *tmp; |
2789 | struct pid_namespace *proc_pidns = proc_pid_ns(sb: file_inode(f: f->file)->i_sb); |
2790 | int level = 0; |
2791 | |
2792 | cur = hlist_entry(v, struct file_lock, fl_link); |
2793 | |
2794 | if (locks_translate_pid(fl: cur, ns: proc_pidns) == 0) |
2795 | return 0; |
2796 | |
2797 | /* View this crossed linked list as a binary tree, the first member of fl_blocked_requests |
2798 | * is the left child of current node, the next silibing in fl_blocked_member is the |
2799 | * right child, we can alse get the parent of current node from fl_blocker, so this |
2800 | * question becomes traversal of a binary tree |
2801 | */ |
2802 | while (cur != NULL) { |
2803 | if (level) |
2804 | lock_get_status(f, fl: cur, id: iter->li_pos, pfx: "-> " , repeat: level); |
2805 | else |
2806 | lock_get_status(f, fl: cur, id: iter->li_pos, pfx: "" , repeat: level); |
2807 | |
2808 | if (!list_empty(head: &cur->fl_blocked_requests)) { |
2809 | /* Turn left */ |
2810 | cur = list_first_entry_or_null(&cur->fl_blocked_requests, |
2811 | struct file_lock, fl_blocked_member); |
2812 | level++; |
2813 | } else { |
2814 | /* Turn right */ |
2815 | tmp = get_next_blocked_member(node: cur); |
2816 | /* Fall back to parent node */ |
2817 | while (tmp == NULL && cur->fl_blocker != NULL) { |
2818 | cur = cur->fl_blocker; |
2819 | level--; |
2820 | tmp = get_next_blocked_member(node: cur); |
2821 | } |
2822 | cur = tmp; |
2823 | } |
2824 | } |
2825 | |
2826 | return 0; |
2827 | } |
2828 | |
2829 | static void __show_fd_locks(struct seq_file *f, |
2830 | struct list_head *head, int *id, |
2831 | struct file *filp, struct files_struct *files) |
2832 | { |
2833 | struct file_lock *fl; |
2834 | |
2835 | list_for_each_entry(fl, head, fl_list) { |
2836 | |
2837 | if (filp != fl->fl_file) |
2838 | continue; |
2839 | if (fl->fl_owner != files && |
2840 | fl->fl_owner != filp) |
2841 | continue; |
2842 | |
2843 | (*id)++; |
2844 | seq_puts(m: f, s: "lock:\t" ); |
2845 | lock_get_status(f, fl, id: *id, pfx: "" , repeat: 0); |
2846 | } |
2847 | } |
2848 | |
2849 | void show_fd_locks(struct seq_file *f, |
2850 | struct file *filp, struct files_struct *files) |
2851 | { |
2852 | struct inode *inode = file_inode(f: filp); |
2853 | struct file_lock_context *ctx; |
2854 | int id = 0; |
2855 | |
2856 | ctx = locks_inode_context(inode); |
2857 | if (!ctx) |
2858 | return; |
2859 | |
2860 | spin_lock(lock: &ctx->flc_lock); |
2861 | __show_fd_locks(f, head: &ctx->flc_flock, id: &id, filp, files); |
2862 | __show_fd_locks(f, head: &ctx->flc_posix, id: &id, filp, files); |
2863 | __show_fd_locks(f, head: &ctx->flc_lease, id: &id, filp, files); |
2864 | spin_unlock(lock: &ctx->flc_lock); |
2865 | } |
2866 | |
2867 | static void *locks_start(struct seq_file *f, loff_t *pos) |
2868 | __acquires(&blocked_lock_lock) |
2869 | { |
2870 | struct locks_iterator *iter = f->private; |
2871 | |
2872 | iter->li_pos = *pos + 1; |
2873 | percpu_down_write(&file_rwsem); |
2874 | spin_lock(lock: &blocked_lock_lock); |
2875 | return seq_hlist_start_percpu(head: &file_lock_list.hlist, cpu: &iter->li_cpu, pos: *pos); |
2876 | } |
2877 | |
2878 | static void *locks_next(struct seq_file *f, void *v, loff_t *pos) |
2879 | { |
2880 | struct locks_iterator *iter = f->private; |
2881 | |
2882 | ++iter->li_pos; |
2883 | return seq_hlist_next_percpu(v, head: &file_lock_list.hlist, cpu: &iter->li_cpu, pos); |
2884 | } |
2885 | |
2886 | static void locks_stop(struct seq_file *f, void *v) |
2887 | __releases(&blocked_lock_lock) |
2888 | { |
2889 | spin_unlock(lock: &blocked_lock_lock); |
2890 | percpu_up_write(&file_rwsem); |
2891 | } |
2892 | |
2893 | static const struct seq_operations locks_seq_operations = { |
2894 | .start = locks_start, |
2895 | .next = locks_next, |
2896 | .stop = locks_stop, |
2897 | .show = locks_show, |
2898 | }; |
2899 | |
2900 | static int __init proc_locks_init(void) |
2901 | { |
2902 | proc_create_seq_private(name: "locks" , mode: 0, NULL, ops: &locks_seq_operations, |
2903 | state_size: sizeof(struct locks_iterator), NULL); |
2904 | return 0; |
2905 | } |
2906 | fs_initcall(proc_locks_init); |
2907 | #endif |
2908 | |
2909 | static int __init filelock_init(void) |
2910 | { |
2911 | int i; |
2912 | |
2913 | flctx_cache = kmem_cache_create(name: "file_lock_ctx" , |
2914 | size: sizeof(struct file_lock_context), align: 0, SLAB_PANIC, NULL); |
2915 | |
2916 | filelock_cache = kmem_cache_create(name: "file_lock_cache" , |
2917 | size: sizeof(struct file_lock), align: 0, SLAB_PANIC, NULL); |
2918 | |
2919 | for_each_possible_cpu(i) { |
2920 | struct file_lock_list_struct *fll = per_cpu_ptr(&file_lock_list, i); |
2921 | |
2922 | spin_lock_init(&fll->lock); |
2923 | INIT_HLIST_HEAD(&fll->hlist); |
2924 | } |
2925 | |
2926 | lease_notifier_chain_init(); |
2927 | return 0; |
2928 | } |
2929 | core_initcall(filelock_init); |
2930 | |