Line data Source code
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/fs.h>
56 : #include <linux/init.h>
57 : #include <linux/security.h>
58 : #include <linux/slab.h>
59 : #include <linux/syscalls.h>
60 : #include <linux/time.h>
61 : #include <linux/rcupdate.h>
62 : #include <linux/pid_namespace.h>
63 : #include <linux/hashtable.h>
64 : #include <linux/percpu.h>
65 : #include <linux/sysctl.h>
66 :
67 : #define CREATE_TRACE_POINTS
68 : #include <trace/events/filelock.h>
69 :
70 : #include <linux/uaccess.h>
71 :
72 : #define IS_POSIX(fl) (fl->fl_flags & FL_POSIX)
73 : #define IS_FLOCK(fl) (fl->fl_flags & FL_FLOCK)
74 : #define IS_LEASE(fl) (fl->fl_flags & (FL_LEASE|FL_DELEG|FL_LAYOUT))
75 : #define IS_OFDLCK(fl) (fl->fl_flags & FL_OFDLCK)
76 : #define IS_REMOTELCK(fl) (fl->fl_pid <= 0)
77 :
78 : static bool lease_breaking(struct file_lock *fl)
79 : {
80 0 : return fl->fl_flags & (FL_UNLOCK_PENDING | FL_DOWNGRADE_PENDING);
81 : }
82 :
83 : static int target_leasetype(struct file_lock *fl)
84 : {
85 0 : if (fl->fl_flags & FL_UNLOCK_PENDING)
86 : return F_UNLCK;
87 0 : if (fl->fl_flags & FL_DOWNGRADE_PENDING)
88 : return F_RDLCK;
89 0 : return fl->fl_type;
90 : }
91 :
92 : static int leases_enable = 1;
93 : static int lease_break_time = 45;
94 :
95 : #ifdef CONFIG_SYSCTL
96 : static struct ctl_table locks_sysctls[] = {
97 : {
98 : .procname = "leases-enable",
99 : .data = &leases_enable,
100 : .maxlen = sizeof(int),
101 : .mode = 0644,
102 : .proc_handler = proc_dointvec,
103 : },
104 : #ifdef CONFIG_MMU
105 : {
106 : .procname = "lease-break-time",
107 : .data = &lease_break_time,
108 : .maxlen = sizeof(int),
109 : .mode = 0644,
110 : .proc_handler = proc_dointvec,
111 : },
112 : #endif /* CONFIG_MMU */
113 : {}
114 : };
115 :
116 1 : static int __init init_fs_locks_sysctls(void)
117 : {
118 1 : register_sysctl_init("fs", locks_sysctls);
119 1 : return 0;
120 : }
121 : early_initcall(init_fs_locks_sysctls);
122 : #endif /* CONFIG_SYSCTL */
123 :
124 : /*
125 : * The global file_lock_list is only used for displaying /proc/locks, so we
126 : * keep a list on each CPU, with each list protected by its own spinlock.
127 : * Global serialization is done using file_rwsem.
128 : *
129 : * Note that alterations to the list also require that the relevant flc_lock is
130 : * held.
131 : */
132 : struct file_lock_list_struct {
133 : spinlock_t lock;
134 : struct hlist_head hlist;
135 : };
136 : static DEFINE_PER_CPU(struct file_lock_list_struct, file_lock_list);
137 : DEFINE_STATIC_PERCPU_RWSEM(file_rwsem);
138 :
139 :
140 : /*
141 : * The blocked_hash is used to find POSIX lock loops for deadlock detection.
142 : * It is protected by blocked_lock_lock.
143 : *
144 : * We hash locks by lockowner in order to optimize searching for the lock a
145 : * particular lockowner is waiting on.
146 : *
147 : * FIXME: make this value scale via some heuristic? We generally will want more
148 : * buckets when we have more lockowners holding locks, but that's a little
149 : * difficult to determine without knowing what the workload will look like.
150 : */
151 : #define BLOCKED_HASH_BITS 7
152 : static DEFINE_HASHTABLE(blocked_hash, BLOCKED_HASH_BITS);
153 :
154 : /*
155 : * This lock protects the blocked_hash. Generally, if you're accessing it, you
156 : * want to be holding this lock.
157 : *
158 : * In addition, it also protects the fl->fl_blocked_requests list, and the
159 : * fl->fl_blocker pointer for file_lock structures that are acting as lock
160 : * requests (in contrast to those that are acting as records of acquired locks).
161 : *
162 : * Note that when we acquire this lock in order to change the above fields,
163 : * we often hold the flc_lock as well. In certain cases, when reading the fields
164 : * protected by this lock, we can skip acquiring it iff we already hold the
165 : * flc_lock.
166 : */
167 : static DEFINE_SPINLOCK(blocked_lock_lock);
168 :
169 : static struct kmem_cache *flctx_cache __read_mostly;
170 : static struct kmem_cache *filelock_cache __read_mostly;
171 :
172 : static struct file_lock_context *
173 0 : locks_get_lock_context(struct inode *inode, int type)
174 : {
175 : struct file_lock_context *ctx;
176 :
177 : /* paired with cmpxchg() below */
178 0 : ctx = smp_load_acquire(&inode->i_flctx);
179 0 : if (likely(ctx) || type == F_UNLCK)
180 : goto out;
181 :
182 0 : ctx = kmem_cache_alloc(flctx_cache, GFP_KERNEL);
183 0 : if (!ctx)
184 : goto out;
185 :
186 0 : spin_lock_init(&ctx->flc_lock);
187 0 : INIT_LIST_HEAD(&ctx->flc_flock);
188 0 : INIT_LIST_HEAD(&ctx->flc_posix);
189 0 : INIT_LIST_HEAD(&ctx->flc_lease);
190 :
191 : /*
192 : * Assign the pointer if it's not already assigned. If it is, then
193 : * free the context we just allocated.
194 : */
195 0 : if (cmpxchg(&inode->i_flctx, NULL, ctx)) {
196 0 : kmem_cache_free(flctx_cache, ctx);
197 0 : ctx = smp_load_acquire(&inode->i_flctx);
198 : }
199 : out:
200 0 : trace_locks_get_lock_context(inode, type, ctx);
201 0 : return ctx;
202 : }
203 :
204 : static void
205 0 : locks_dump_ctx_list(struct list_head *list, char *list_type)
206 : {
207 : struct file_lock *fl;
208 :
209 0 : list_for_each_entry(fl, list, fl_list) {
210 0 : 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);
211 : }
212 0 : }
213 :
214 : static void
215 0 : locks_check_ctx_lists(struct inode *inode)
216 : {
217 0 : struct file_lock_context *ctx = inode->i_flctx;
218 :
219 0 : if (unlikely(!list_empty(&ctx->flc_flock) ||
220 : !list_empty(&ctx->flc_posix) ||
221 : !list_empty(&ctx->flc_lease))) {
222 0 : pr_warn("Leaked locks on dev=0x%x:0x%x ino=0x%lx:\n",
223 : MAJOR(inode->i_sb->s_dev), MINOR(inode->i_sb->s_dev),
224 : inode->i_ino);
225 0 : locks_dump_ctx_list(&ctx->flc_flock, "FLOCK");
226 0 : locks_dump_ctx_list(&ctx->flc_posix, "POSIX");
227 0 : locks_dump_ctx_list(&ctx->flc_lease, "LEASE");
228 : }
229 0 : }
230 :
231 : static void
232 0 : locks_check_ctx_file_list(struct file *filp, struct list_head *list,
233 : char *list_type)
234 : {
235 : struct file_lock *fl;
236 0 : struct inode *inode = locks_inode(filp);
237 :
238 0 : list_for_each_entry(fl, list, fl_list)
239 0 : if (fl->fl_file == filp)
240 0 : pr_warn("Leaked %s lock on dev=0x%x:0x%x ino=0x%lx "
241 : " fl_owner=%p fl_flags=0x%x fl_type=0x%x fl_pid=%u\n",
242 : list_type, MAJOR(inode->i_sb->s_dev),
243 : MINOR(inode->i_sb->s_dev), inode->i_ino,
244 : fl->fl_owner, fl->fl_flags, fl->fl_type, fl->fl_pid);
245 0 : }
246 :
247 : void
248 0 : locks_free_lock_context(struct inode *inode)
249 : {
250 0 : struct file_lock_context *ctx = inode->i_flctx;
251 :
252 0 : if (unlikely(ctx)) {
253 0 : locks_check_ctx_lists(inode);
254 0 : kmem_cache_free(flctx_cache, ctx);
255 : }
256 0 : }
257 :
258 : static void locks_init_lock_heads(struct file_lock *fl)
259 : {
260 0 : INIT_HLIST_NODE(&fl->fl_link);
261 0 : INIT_LIST_HEAD(&fl->fl_list);
262 0 : INIT_LIST_HEAD(&fl->fl_blocked_requests);
263 0 : INIT_LIST_HEAD(&fl->fl_blocked_member);
264 0 : init_waitqueue_head(&fl->fl_wait);
265 : }
266 :
267 : /* Allocate an empty lock structure. */
268 0 : struct file_lock *locks_alloc_lock(void)
269 : {
270 0 : struct file_lock *fl = kmem_cache_zalloc(filelock_cache, GFP_KERNEL);
271 :
272 0 : if (fl)
273 : locks_init_lock_heads(fl);
274 :
275 0 : return fl;
276 : }
277 : EXPORT_SYMBOL_GPL(locks_alloc_lock);
278 :
279 0 : void locks_release_private(struct file_lock *fl)
280 : {
281 0 : BUG_ON(waitqueue_active(&fl->fl_wait));
282 0 : BUG_ON(!list_empty(&fl->fl_list));
283 0 : BUG_ON(!list_empty(&fl->fl_blocked_requests));
284 0 : BUG_ON(!list_empty(&fl->fl_blocked_member));
285 0 : BUG_ON(!hlist_unhashed(&fl->fl_link));
286 :
287 0 : if (fl->fl_ops) {
288 0 : if (fl->fl_ops->fl_release_private)
289 0 : fl->fl_ops->fl_release_private(fl);
290 0 : fl->fl_ops = NULL;
291 : }
292 :
293 0 : if (fl->fl_lmops) {
294 0 : if (fl->fl_lmops->lm_put_owner) {
295 0 : fl->fl_lmops->lm_put_owner(fl->fl_owner);
296 0 : fl->fl_owner = NULL;
297 : }
298 0 : fl->fl_lmops = NULL;
299 : }
300 0 : }
301 : EXPORT_SYMBOL_GPL(locks_release_private);
302 :
303 : /* Free a lock which is not in use. */
304 0 : void locks_free_lock(struct file_lock *fl)
305 : {
306 0 : locks_release_private(fl);
307 0 : kmem_cache_free(filelock_cache, fl);
308 0 : }
309 : EXPORT_SYMBOL(locks_free_lock);
310 :
311 : static void
312 0 : locks_dispose_list(struct list_head *dispose)
313 : {
314 : struct file_lock *fl;
315 :
316 0 : while (!list_empty(dispose)) {
317 0 : fl = list_first_entry(dispose, struct file_lock, fl_list);
318 0 : list_del_init(&fl->fl_list);
319 : locks_free_lock(fl);
320 : }
321 0 : }
322 :
323 0 : void locks_init_lock(struct file_lock *fl)
324 : {
325 0 : memset(fl, 0, sizeof(struct file_lock));
326 0 : locks_init_lock_heads(fl);
327 0 : }
328 : EXPORT_SYMBOL(locks_init_lock);
329 :
330 : /*
331 : * Initialize a new lock from an existing file_lock structure.
332 : */
333 0 : void locks_copy_conflock(struct file_lock *new, struct file_lock *fl)
334 : {
335 0 : new->fl_owner = fl->fl_owner;
336 0 : new->fl_pid = fl->fl_pid;
337 0 : new->fl_file = NULL;
338 0 : new->fl_flags = fl->fl_flags;
339 0 : new->fl_type = fl->fl_type;
340 0 : new->fl_start = fl->fl_start;
341 0 : new->fl_end = fl->fl_end;
342 0 : new->fl_lmops = fl->fl_lmops;
343 0 : new->fl_ops = NULL;
344 :
345 0 : if (fl->fl_lmops) {
346 0 : if (fl->fl_lmops->lm_get_owner)
347 0 : fl->fl_lmops->lm_get_owner(fl->fl_owner);
348 : }
349 0 : }
350 : EXPORT_SYMBOL(locks_copy_conflock);
351 :
352 0 : void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
353 : {
354 : /* "new" must be a freshly-initialized lock */
355 0 : WARN_ON_ONCE(new->fl_ops);
356 :
357 0 : locks_copy_conflock(new, fl);
358 :
359 0 : new->fl_file = fl->fl_file;
360 0 : new->fl_ops = fl->fl_ops;
361 :
362 0 : if (fl->fl_ops) {
363 0 : if (fl->fl_ops->fl_copy_lock)
364 0 : fl->fl_ops->fl_copy_lock(new, fl);
365 : }
366 0 : }
367 : EXPORT_SYMBOL(locks_copy_lock);
368 :
369 0 : static void locks_move_blocks(struct file_lock *new, struct file_lock *fl)
370 : {
371 : struct file_lock *f;
372 :
373 : /*
374 : * As ctx->flc_lock is held, new requests cannot be added to
375 : * ->fl_blocked_requests, so we don't need a lock to check if it
376 : * is empty.
377 : */
378 0 : if (list_empty(&fl->fl_blocked_requests))
379 : return;
380 0 : spin_lock(&blocked_lock_lock);
381 0 : list_splice_init(&fl->fl_blocked_requests, &new->fl_blocked_requests);
382 0 : list_for_each_entry(f, &new->fl_blocked_requests, fl_blocked_member)
383 0 : f->fl_blocker = new;
384 : spin_unlock(&blocked_lock_lock);
385 : }
386 :
387 : static inline int flock_translate_cmd(int cmd) {
388 : switch (cmd) {
389 : case LOCK_SH:
390 : return F_RDLCK;
391 : case LOCK_EX:
392 : return F_WRLCK;
393 : case LOCK_UN:
394 : return F_UNLCK;
395 : }
396 : return -EINVAL;
397 : }
398 :
399 : /* Fill in a file_lock structure with an appropriate FLOCK lock. */
400 : static struct file_lock *
401 0 : flock_make_lock(struct file *filp, unsigned int cmd, struct file_lock *fl)
402 : {
403 0 : int type = flock_translate_cmd(cmd);
404 :
405 0 : if (type < 0)
406 0 : return ERR_PTR(type);
407 :
408 0 : if (fl == NULL) {
409 0 : fl = locks_alloc_lock();
410 0 : if (fl == NULL)
411 : return ERR_PTR(-ENOMEM);
412 : } else {
413 0 : locks_init_lock(fl);
414 : }
415 :
416 0 : fl->fl_file = filp;
417 0 : fl->fl_owner = filp;
418 0 : fl->fl_pid = current->tgid;
419 0 : fl->fl_flags = FL_FLOCK;
420 0 : fl->fl_type = type;
421 0 : fl->fl_end = OFFSET_MAX;
422 :
423 0 : return fl;
424 : }
425 :
426 : static int assign_type(struct file_lock *fl, long type)
427 : {
428 0 : switch (type) {
429 : case F_RDLCK:
430 : case F_WRLCK:
431 : case F_UNLCK:
432 0 : fl->fl_type = type;
433 : break;
434 : default:
435 : return -EINVAL;
436 : }
437 : return 0;
438 : }
439 :
440 0 : static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
441 : struct flock64 *l)
442 : {
443 0 : switch (l->l_whence) {
444 : case SEEK_SET:
445 0 : fl->fl_start = 0;
446 0 : break;
447 : case SEEK_CUR:
448 0 : fl->fl_start = filp->f_pos;
449 0 : break;
450 : case SEEK_END:
451 0 : fl->fl_start = i_size_read(file_inode(filp));
452 0 : break;
453 : default:
454 : return -EINVAL;
455 : }
456 0 : if (l->l_start > OFFSET_MAX - fl->fl_start)
457 : return -EOVERFLOW;
458 0 : fl->fl_start += l->l_start;
459 0 : if (fl->fl_start < 0)
460 : return -EINVAL;
461 :
462 : /* POSIX-1996 leaves the case l->l_len < 0 undefined;
463 : POSIX-2001 defines it. */
464 0 : if (l->l_len > 0) {
465 0 : if (l->l_len - 1 > OFFSET_MAX - fl->fl_start)
466 : return -EOVERFLOW;
467 0 : fl->fl_end = fl->fl_start + (l->l_len - 1);
468 :
469 0 : } else if (l->l_len < 0) {
470 0 : if (fl->fl_start + l->l_len < 0)
471 : return -EINVAL;
472 0 : fl->fl_end = fl->fl_start - 1;
473 0 : fl->fl_start += l->l_len;
474 : } else
475 0 : fl->fl_end = OFFSET_MAX;
476 :
477 0 : fl->fl_owner = current->files;
478 0 : fl->fl_pid = current->tgid;
479 0 : fl->fl_file = filp;
480 0 : fl->fl_flags = FL_POSIX;
481 0 : fl->fl_ops = NULL;
482 0 : fl->fl_lmops = NULL;
483 :
484 0 : return assign_type(fl, l->l_type);
485 : }
486 :
487 : /* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
488 : * style lock.
489 : */
490 0 : static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
491 : struct flock *l)
492 : {
493 0 : struct flock64 ll = {
494 0 : .l_type = l->l_type,
495 0 : .l_whence = l->l_whence,
496 0 : .l_start = l->l_start,
497 0 : .l_len = l->l_len,
498 : };
499 :
500 0 : return flock64_to_posix_lock(filp, fl, &ll);
501 : }
502 :
503 : /* default lease lock manager operations */
504 : static bool
505 0 : lease_break_callback(struct file_lock *fl)
506 : {
507 0 : kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
508 0 : return false;
509 : }
510 :
511 : static void
512 0 : lease_setup(struct file_lock *fl, void **priv)
513 : {
514 0 : struct file *filp = fl->fl_file;
515 0 : struct fasync_struct *fa = *priv;
516 :
517 : /*
518 : * fasync_insert_entry() returns the old entry if any. If there was no
519 : * old entry, then it used "priv" and inserted it into the fasync list.
520 : * Clear the pointer to indicate that it shouldn't be freed.
521 : */
522 0 : if (!fasync_insert_entry(fa->fa_fd, filp, &fl->fl_fasync, fa))
523 0 : *priv = NULL;
524 :
525 0 : __f_setown(filp, task_pid(current), PIDTYPE_TGID, 0);
526 0 : }
527 :
528 : static const struct lock_manager_operations lease_manager_ops = {
529 : .lm_break = lease_break_callback,
530 : .lm_change = lease_modify,
531 : .lm_setup = lease_setup,
532 : };
533 :
534 : /*
535 : * Initialize a lease, use the default lock manager operations
536 : */
537 : static int lease_init(struct file *filp, long type, struct file_lock *fl)
538 : {
539 0 : if (assign_type(fl, type) != 0)
540 : return -EINVAL;
541 :
542 0 : fl->fl_owner = filp;
543 0 : fl->fl_pid = current->tgid;
544 :
545 0 : fl->fl_file = filp;
546 0 : fl->fl_flags = FL_LEASE;
547 0 : fl->fl_start = 0;
548 0 : fl->fl_end = OFFSET_MAX;
549 0 : fl->fl_ops = NULL;
550 0 : fl->fl_lmops = &lease_manager_ops;
551 : return 0;
552 : }
553 :
554 : /* Allocate a file_lock initialised to this type of lease */
555 0 : static struct file_lock *lease_alloc(struct file *filp, long type)
556 : {
557 0 : struct file_lock *fl = locks_alloc_lock();
558 0 : int error = -ENOMEM;
559 :
560 0 : if (fl == NULL)
561 : return ERR_PTR(error);
562 :
563 0 : error = lease_init(filp, type, fl);
564 0 : if (error) {
565 0 : locks_free_lock(fl);
566 0 : return ERR_PTR(error);
567 : }
568 : return fl;
569 : }
570 :
571 : /* Check if two locks overlap each other.
572 : */
573 : static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
574 : {
575 0 : return ((fl1->fl_end >= fl2->fl_start) &&
576 0 : (fl2->fl_end >= fl1->fl_start));
577 : }
578 :
579 : /*
580 : * Check whether two locks have the same owner.
581 : */
582 : static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
583 : {
584 : return fl1->fl_owner == fl2->fl_owner;
585 : }
586 :
587 : /* Must be called with the flc_lock held! */
588 : static void locks_insert_global_locks(struct file_lock *fl)
589 : {
590 0 : struct file_lock_list_struct *fll = this_cpu_ptr(&file_lock_list);
591 :
592 : percpu_rwsem_assert_held(&file_rwsem);
593 :
594 0 : spin_lock(&fll->lock);
595 0 : fl->fl_link_cpu = smp_processor_id();
596 0 : hlist_add_head(&fl->fl_link, &fll->hlist);
597 0 : spin_unlock(&fll->lock);
598 : }
599 :
600 : /* Must be called with the flc_lock held! */
601 : static void locks_delete_global_locks(struct file_lock *fl)
602 : {
603 : struct file_lock_list_struct *fll;
604 :
605 : percpu_rwsem_assert_held(&file_rwsem);
606 :
607 : /*
608 : * Avoid taking lock if already unhashed. This is safe since this check
609 : * is done while holding the flc_lock, and new insertions into the list
610 : * also require that it be held.
611 : */
612 0 : if (hlist_unhashed(&fl->fl_link))
613 : return;
614 :
615 0 : fll = per_cpu_ptr(&file_lock_list, fl->fl_link_cpu);
616 0 : spin_lock(&fll->lock);
617 0 : hlist_del_init(&fl->fl_link);
618 0 : spin_unlock(&fll->lock);
619 : }
620 :
621 : static unsigned long
622 : posix_owner_key(struct file_lock *fl)
623 : {
624 0 : return (unsigned long)fl->fl_owner;
625 : }
626 :
627 : static void locks_insert_global_blocked(struct file_lock *waiter)
628 : {
629 : lockdep_assert_held(&blocked_lock_lock);
630 :
631 0 : hash_add(blocked_hash, &waiter->fl_link, posix_owner_key(waiter));
632 : }
633 :
634 : static void locks_delete_global_blocked(struct file_lock *waiter)
635 : {
636 : lockdep_assert_held(&blocked_lock_lock);
637 :
638 0 : hash_del(&waiter->fl_link);
639 : }
640 :
641 : /* Remove waiter from blocker's block list.
642 : * When blocker ends up pointing to itself then the list is empty.
643 : *
644 : * Must be called with blocked_lock_lock held.
645 : */
646 : static void __locks_delete_block(struct file_lock *waiter)
647 : {
648 0 : locks_delete_global_blocked(waiter);
649 0 : list_del_init(&waiter->fl_blocked_member);
650 : }
651 :
652 0 : static void __locks_wake_up_blocks(struct file_lock *blocker)
653 : {
654 0 : while (!list_empty(&blocker->fl_blocked_requests)) {
655 : struct file_lock *waiter;
656 :
657 0 : waiter = list_first_entry(&blocker->fl_blocked_requests,
658 : struct file_lock, fl_blocked_member);
659 0 : __locks_delete_block(waiter);
660 0 : if (waiter->fl_lmops && waiter->fl_lmops->lm_notify)
661 0 : waiter->fl_lmops->lm_notify(waiter);
662 : else
663 0 : wake_up(&waiter->fl_wait);
664 :
665 : /*
666 : * The setting of fl_blocker to NULL marks the "done"
667 : * point in deleting a block. Paired with acquire at the top
668 : * of locks_delete_block().
669 : */
670 0 : smp_store_release(&waiter->fl_blocker, NULL);
671 : }
672 0 : }
673 :
674 : /**
675 : * locks_delete_block - stop waiting for a file lock
676 : * @waiter: the lock which was waiting
677 : *
678 : * lockd/nfsd need to disconnect the lock while working on it.
679 : */
680 0 : int locks_delete_block(struct file_lock *waiter)
681 : {
682 0 : int status = -ENOENT;
683 :
684 : /*
685 : * If fl_blocker is NULL, it won't be set again as this thread "owns"
686 : * the lock and is the only one that might try to claim the lock.
687 : *
688 : * We use acquire/release to manage fl_blocker so that we can
689 : * optimize away taking the blocked_lock_lock in many cases.
690 : *
691 : * The smp_load_acquire guarantees two things:
692 : *
693 : * 1/ that fl_blocked_requests can be tested locklessly. If something
694 : * was recently added to that list it must have been in a locked region
695 : * *before* the locked region when fl_blocker was set to NULL.
696 : *
697 : * 2/ that no other thread is accessing 'waiter', so it is safe to free
698 : * it. __locks_wake_up_blocks is careful not to touch waiter after
699 : * fl_blocker is released.
700 : *
701 : * If a lockless check of fl_blocker shows it to be NULL, we know that
702 : * no new locks can be inserted into its fl_blocked_requests list, and
703 : * can avoid doing anything further if the list is empty.
704 : */
705 0 : if (!smp_load_acquire(&waiter->fl_blocker) &&
706 0 : list_empty(&waiter->fl_blocked_requests))
707 : return status;
708 :
709 0 : spin_lock(&blocked_lock_lock);
710 0 : if (waiter->fl_blocker)
711 0 : status = 0;
712 0 : __locks_wake_up_blocks(waiter);
713 0 : __locks_delete_block(waiter);
714 :
715 : /*
716 : * The setting of fl_blocker to NULL marks the "done" point in deleting
717 : * a block. Paired with acquire at the top of this function.
718 : */
719 0 : smp_store_release(&waiter->fl_blocker, NULL);
720 0 : spin_unlock(&blocked_lock_lock);
721 0 : return status;
722 : }
723 : EXPORT_SYMBOL(locks_delete_block);
724 :
725 : /* Insert waiter into blocker's block list.
726 : * We use a circular list so that processes can be easily woken up in
727 : * the order they blocked. The documentation doesn't require this but
728 : * it seems like the reasonable thing to do.
729 : *
730 : * Must be called with both the flc_lock and blocked_lock_lock held. The
731 : * fl_blocked_requests list itself is protected by the blocked_lock_lock,
732 : * but by ensuring that the flc_lock is also held on insertions we can avoid
733 : * taking the blocked_lock_lock in some cases when we see that the
734 : * fl_blocked_requests list is empty.
735 : *
736 : * Rather than just adding to the list, we check for conflicts with any existing
737 : * waiters, and add beneath any waiter that blocks the new waiter.
738 : * Thus wakeups don't happen until needed.
739 : */
740 0 : static void __locks_insert_block(struct file_lock *blocker,
741 : struct file_lock *waiter,
742 : bool conflict(struct file_lock *,
743 : struct file_lock *))
744 : {
745 : struct file_lock *fl;
746 0 : BUG_ON(!list_empty(&waiter->fl_blocked_member));
747 :
748 : new_blocker:
749 0 : list_for_each_entry(fl, &blocker->fl_blocked_requests, fl_blocked_member)
750 0 : if (conflict(fl, waiter)) {
751 : blocker = fl;
752 : goto new_blocker;
753 : }
754 0 : waiter->fl_blocker = blocker;
755 0 : list_add_tail(&waiter->fl_blocked_member, &blocker->fl_blocked_requests);
756 0 : if (IS_POSIX(blocker) && !IS_OFDLCK(blocker))
757 : locks_insert_global_blocked(waiter);
758 :
759 : /* The requests in waiter->fl_blocked are known to conflict with
760 : * waiter, but might not conflict with blocker, or the requests
761 : * and lock which block it. So they all need to be woken.
762 : */
763 0 : __locks_wake_up_blocks(waiter);
764 0 : }
765 :
766 : /* Must be called with flc_lock held. */
767 : static void locks_insert_block(struct file_lock *blocker,
768 : struct file_lock *waiter,
769 : bool conflict(struct file_lock *,
770 : struct file_lock *))
771 : {
772 0 : spin_lock(&blocked_lock_lock);
773 0 : __locks_insert_block(blocker, waiter, conflict);
774 0 : spin_unlock(&blocked_lock_lock);
775 : }
776 :
777 : /*
778 : * Wake up processes blocked waiting for blocker.
779 : *
780 : * Must be called with the inode->flc_lock held!
781 : */
782 : static void locks_wake_up_blocks(struct file_lock *blocker)
783 : {
784 : /*
785 : * Avoid taking global lock if list is empty. This is safe since new
786 : * blocked requests are only added to the list under the flc_lock, and
787 : * the flc_lock is always held here. Note that removal from the
788 : * fl_blocked_requests list does not require the flc_lock, so we must
789 : * recheck list_empty() after acquiring the blocked_lock_lock.
790 : */
791 0 : if (list_empty(&blocker->fl_blocked_requests))
792 : return;
793 :
794 0 : spin_lock(&blocked_lock_lock);
795 0 : __locks_wake_up_blocks(blocker);
796 : spin_unlock(&blocked_lock_lock);
797 : }
798 :
799 : static void
800 : locks_insert_lock_ctx(struct file_lock *fl, struct list_head *before)
801 : {
802 0 : list_add_tail(&fl->fl_list, before);
803 0 : locks_insert_global_locks(fl);
804 : }
805 :
806 : static void
807 0 : locks_unlink_lock_ctx(struct file_lock *fl)
808 : {
809 0 : locks_delete_global_locks(fl);
810 0 : list_del_init(&fl->fl_list);
811 0 : locks_wake_up_blocks(fl);
812 0 : }
813 :
814 : static void
815 0 : locks_delete_lock_ctx(struct file_lock *fl, struct list_head *dispose)
816 : {
817 0 : locks_unlink_lock_ctx(fl);
818 0 : if (dispose)
819 0 : list_add(&fl->fl_list, dispose);
820 : else
821 : locks_free_lock(fl);
822 0 : }
823 :
824 : /* Determine if lock sys_fl blocks lock caller_fl. Common functionality
825 : * checks for shared/exclusive status of overlapping locks.
826 : */
827 : static bool locks_conflict(struct file_lock *caller_fl,
828 : struct file_lock *sys_fl)
829 : {
830 0 : if (sys_fl->fl_type == F_WRLCK)
831 : return true;
832 0 : if (caller_fl->fl_type == F_WRLCK)
833 : return true;
834 : return false;
835 : }
836 :
837 : /* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
838 : * checking before calling the locks_conflict().
839 : */
840 0 : static bool posix_locks_conflict(struct file_lock *caller_fl,
841 : struct file_lock *sys_fl)
842 : {
843 : /* POSIX locks owned by the same process do not conflict with
844 : * each other.
845 : */
846 0 : if (posix_same_owner(caller_fl, sys_fl))
847 : return false;
848 :
849 : /* Check whether they overlap */
850 0 : if (!locks_overlap(caller_fl, sys_fl))
851 : return false;
852 :
853 0 : return locks_conflict(caller_fl, sys_fl);
854 : }
855 :
856 : /* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
857 : * checking before calling the locks_conflict().
858 : */
859 0 : static bool flock_locks_conflict(struct file_lock *caller_fl,
860 : struct file_lock *sys_fl)
861 : {
862 : /* FLOCK locks referring to the same filp do not conflict with
863 : * each other.
864 : */
865 0 : if (caller_fl->fl_file == sys_fl->fl_file)
866 : return false;
867 :
868 0 : return locks_conflict(caller_fl, sys_fl);
869 : }
870 :
871 : void
872 0 : posix_test_lock(struct file *filp, struct file_lock *fl)
873 : {
874 : struct file_lock *cfl;
875 : struct file_lock_context *ctx;
876 0 : struct inode *inode = locks_inode(filp);
877 :
878 0 : ctx = smp_load_acquire(&inode->i_flctx);
879 0 : if (!ctx || list_empty_careful(&ctx->flc_posix)) {
880 0 : fl->fl_type = F_UNLCK;
881 0 : return;
882 : }
883 :
884 0 : spin_lock(&ctx->flc_lock);
885 0 : list_for_each_entry(cfl, &ctx->flc_posix, fl_list) {
886 0 : if (posix_locks_conflict(fl, cfl)) {
887 0 : locks_copy_conflock(fl, cfl);
888 0 : goto out;
889 : }
890 : }
891 0 : fl->fl_type = F_UNLCK;
892 : out:
893 0 : spin_unlock(&ctx->flc_lock);
894 : return;
895 : }
896 : EXPORT_SYMBOL(posix_test_lock);
897 :
898 : /*
899 : * Deadlock detection:
900 : *
901 : * We attempt to detect deadlocks that are due purely to posix file
902 : * locks.
903 : *
904 : * We assume that a task can be waiting for at most one lock at a time.
905 : * So for any acquired lock, the process holding that lock may be
906 : * waiting on at most one other lock. That lock in turns may be held by
907 : * someone waiting for at most one other lock. Given a requested lock
908 : * caller_fl which is about to wait for a conflicting lock block_fl, we
909 : * follow this chain of waiters to ensure we are not about to create a
910 : * cycle.
911 : *
912 : * Since we do this before we ever put a process to sleep on a lock, we
913 : * are ensured that there is never a cycle; that is what guarantees that
914 : * the while() loop in posix_locks_deadlock() eventually completes.
915 : *
916 : * Note: the above assumption may not be true when handling lock
917 : * requests from a broken NFS client. It may also fail in the presence
918 : * of tasks (such as posix threads) sharing the same open file table.
919 : * To handle those cases, we just bail out after a few iterations.
920 : *
921 : * For FL_OFDLCK locks, the owner is the filp, not the files_struct.
922 : * Because the owner is not even nominally tied to a thread of
923 : * execution, the deadlock detection below can't reasonably work well. Just
924 : * skip it for those.
925 : *
926 : * In principle, we could do a more limited deadlock detection on FL_OFDLCK
927 : * locks that just checks for the case where two tasks are attempting to
928 : * upgrade from read to write locks on the same inode.
929 : */
930 :
931 : #define MAX_DEADLK_ITERATIONS 10
932 :
933 : /* Find a lock that the owner of the given block_fl is blocking on. */
934 0 : static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
935 : {
936 : struct file_lock *fl;
937 :
938 0 : hash_for_each_possible(blocked_hash, fl, fl_link, posix_owner_key(block_fl)) {
939 0 : if (posix_same_owner(fl, block_fl)) {
940 0 : while (fl->fl_blocker)
941 : fl = fl->fl_blocker;
942 : return fl;
943 : }
944 : }
945 : return NULL;
946 : }
947 :
948 : /* Must be called with the blocked_lock_lock held! */
949 0 : static int posix_locks_deadlock(struct file_lock *caller_fl,
950 : struct file_lock *block_fl)
951 : {
952 0 : int i = 0;
953 :
954 : lockdep_assert_held(&blocked_lock_lock);
955 :
956 : /*
957 : * This deadlock detector can't reasonably detect deadlocks with
958 : * FL_OFDLCK locks, since they aren't owned by a process, per-se.
959 : */
960 0 : if (IS_OFDLCK(caller_fl))
961 : return 0;
962 :
963 0 : while ((block_fl = what_owner_is_waiting_for(block_fl))) {
964 0 : if (i++ > MAX_DEADLK_ITERATIONS)
965 : return 0;
966 0 : if (posix_same_owner(caller_fl, block_fl))
967 : return 1;
968 : }
969 : return 0;
970 : }
971 :
972 : /* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
973 : * after any leases, but before any posix locks.
974 : *
975 : * Note that if called with an FL_EXISTS argument, the caller may determine
976 : * whether or not a lock was successfully freed by testing the return
977 : * value for -ENOENT.
978 : */
979 0 : static int flock_lock_inode(struct inode *inode, struct file_lock *request)
980 : {
981 0 : struct file_lock *new_fl = NULL;
982 : struct file_lock *fl;
983 : struct file_lock_context *ctx;
984 0 : int error = 0;
985 0 : bool found = false;
986 0 : LIST_HEAD(dispose);
987 :
988 0 : ctx = locks_get_lock_context(inode, request->fl_type);
989 0 : if (!ctx) {
990 0 : if (request->fl_type != F_UNLCK)
991 : return -ENOMEM;
992 0 : return (request->fl_flags & FL_EXISTS) ? -ENOENT : 0;
993 : }
994 :
995 0 : if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) {
996 0 : new_fl = locks_alloc_lock();
997 0 : if (!new_fl)
998 : return -ENOMEM;
999 : }
1000 :
1001 0 : percpu_down_read(&file_rwsem);
1002 0 : spin_lock(&ctx->flc_lock);
1003 0 : if (request->fl_flags & FL_ACCESS)
1004 : goto find_conflict;
1005 :
1006 0 : list_for_each_entry(fl, &ctx->flc_flock, fl_list) {
1007 0 : if (request->fl_file != fl->fl_file)
1008 0 : continue;
1009 0 : if (request->fl_type == fl->fl_type)
1010 : goto out;
1011 0 : found = true;
1012 : locks_delete_lock_ctx(fl, &dispose);
1013 : break;
1014 : }
1015 :
1016 0 : if (request->fl_type == F_UNLCK) {
1017 0 : if ((request->fl_flags & FL_EXISTS) && !found)
1018 0 : error = -ENOENT;
1019 : goto out;
1020 : }
1021 :
1022 : find_conflict:
1023 0 : list_for_each_entry(fl, &ctx->flc_flock, fl_list) {
1024 0 : if (!flock_locks_conflict(request, fl))
1025 0 : continue;
1026 0 : error = -EAGAIN;
1027 0 : if (!(request->fl_flags & FL_SLEEP))
1028 : goto out;
1029 0 : error = FILE_LOCK_DEFERRED;
1030 : locks_insert_block(fl, request, flock_locks_conflict);
1031 : goto out;
1032 : }
1033 0 : if (request->fl_flags & FL_ACCESS)
1034 : goto out;
1035 0 : locks_copy_lock(new_fl, request);
1036 0 : locks_move_blocks(new_fl, request);
1037 0 : locks_insert_lock_ctx(new_fl, &ctx->flc_flock);
1038 0 : new_fl = NULL;
1039 0 : error = 0;
1040 :
1041 : out:
1042 0 : spin_unlock(&ctx->flc_lock);
1043 0 : percpu_up_read(&file_rwsem);
1044 0 : if (new_fl)
1045 : locks_free_lock(new_fl);
1046 0 : locks_dispose_list(&dispose);
1047 0 : trace_flock_lock_inode(inode, request, error);
1048 0 : return error;
1049 : }
1050 :
1051 0 : static int posix_lock_inode(struct inode *inode, struct file_lock *request,
1052 : struct file_lock *conflock)
1053 : {
1054 : struct file_lock *fl, *tmp;
1055 0 : struct file_lock *new_fl = NULL;
1056 0 : struct file_lock *new_fl2 = NULL;
1057 0 : struct file_lock *left = NULL;
1058 0 : struct file_lock *right = NULL;
1059 : struct file_lock_context *ctx;
1060 : int error;
1061 0 : bool added = false;
1062 0 : LIST_HEAD(dispose);
1063 :
1064 0 : ctx = locks_get_lock_context(inode, request->fl_type);
1065 0 : if (!ctx)
1066 0 : return (request->fl_type == F_UNLCK) ? 0 : -ENOMEM;
1067 :
1068 : /*
1069 : * We may need two file_lock structures for this operation,
1070 : * so we get them in advance to avoid races.
1071 : *
1072 : * In some cases we can be sure, that no new locks will be needed
1073 : */
1074 0 : if (!(request->fl_flags & FL_ACCESS) &&
1075 0 : (request->fl_type != F_UNLCK ||
1076 0 : request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
1077 0 : new_fl = locks_alloc_lock();
1078 0 : new_fl2 = locks_alloc_lock();
1079 : }
1080 :
1081 0 : percpu_down_read(&file_rwsem);
1082 0 : spin_lock(&ctx->flc_lock);
1083 : /*
1084 : * New lock request. Walk all POSIX locks and look for conflicts. If
1085 : * there are any, either return error or put the request on the
1086 : * blocker's list of waiters and the global blocked_hash.
1087 : */
1088 0 : if (request->fl_type != F_UNLCK) {
1089 0 : list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
1090 0 : if (!posix_locks_conflict(request, fl))
1091 0 : continue;
1092 0 : if (conflock)
1093 0 : locks_copy_conflock(conflock, fl);
1094 0 : error = -EAGAIN;
1095 0 : if (!(request->fl_flags & FL_SLEEP))
1096 : goto out;
1097 : /*
1098 : * Deadlock detection and insertion into the blocked
1099 : * locks list must be done while holding the same lock!
1100 : */
1101 0 : error = -EDEADLK;
1102 0 : spin_lock(&blocked_lock_lock);
1103 : /*
1104 : * Ensure that we don't find any locks blocked on this
1105 : * request during deadlock detection.
1106 : */
1107 0 : __locks_wake_up_blocks(request);
1108 0 : if (likely(!posix_locks_deadlock(request, fl))) {
1109 0 : error = FILE_LOCK_DEFERRED;
1110 0 : __locks_insert_block(fl, request,
1111 : posix_locks_conflict);
1112 : }
1113 : spin_unlock(&blocked_lock_lock);
1114 : goto out;
1115 : }
1116 : }
1117 :
1118 : /* If we're just looking for a conflict, we're done. */
1119 0 : error = 0;
1120 0 : if (request->fl_flags & FL_ACCESS)
1121 : goto out;
1122 :
1123 : /* Find the first old lock with the same owner as the new lock */
1124 0 : list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
1125 0 : if (posix_same_owner(request, fl))
1126 : break;
1127 : }
1128 :
1129 : /* Process locks with this owner. */
1130 0 : list_for_each_entry_safe_from(fl, tmp, &ctx->flc_posix, fl_list) {
1131 0 : if (!posix_same_owner(request, fl))
1132 : break;
1133 :
1134 : /* Detect adjacent or overlapping regions (if same lock type) */
1135 0 : if (request->fl_type == fl->fl_type) {
1136 : /* In all comparisons of start vs end, use
1137 : * "start - 1" rather than "end + 1". If end
1138 : * is OFFSET_MAX, end + 1 will become negative.
1139 : */
1140 0 : if (fl->fl_end < request->fl_start - 1)
1141 0 : continue;
1142 : /* If the next lock in the list has entirely bigger
1143 : * addresses than the new one, insert the lock here.
1144 : */
1145 0 : if (fl->fl_start - 1 > request->fl_end)
1146 : break;
1147 :
1148 : /* If we come here, the new and old lock are of the
1149 : * same type and adjacent or overlapping. Make one
1150 : * lock yielding from the lower start address of both
1151 : * locks to the higher end address.
1152 : */
1153 0 : if (fl->fl_start > request->fl_start)
1154 0 : fl->fl_start = request->fl_start;
1155 : else
1156 0 : request->fl_start = fl->fl_start;
1157 0 : if (fl->fl_end < request->fl_end)
1158 0 : fl->fl_end = request->fl_end;
1159 : else
1160 0 : request->fl_end = fl->fl_end;
1161 0 : if (added) {
1162 0 : locks_delete_lock_ctx(fl, &dispose);
1163 0 : continue;
1164 : }
1165 : request = fl;
1166 : added = true;
1167 : } else {
1168 : /* Processing for different lock types is a bit
1169 : * more complex.
1170 : */
1171 0 : if (fl->fl_end < request->fl_start)
1172 0 : continue;
1173 0 : if (fl->fl_start > request->fl_end)
1174 : break;
1175 0 : if (request->fl_type == F_UNLCK)
1176 0 : added = true;
1177 0 : if (fl->fl_start < request->fl_start)
1178 0 : left = fl;
1179 : /* If the next lock in the list has a higher end
1180 : * address than the new one, insert the new one here.
1181 : */
1182 0 : if (fl->fl_end > request->fl_end) {
1183 : right = fl;
1184 : break;
1185 : }
1186 0 : if (fl->fl_start >= request->fl_start) {
1187 : /* The new lock completely replaces an old
1188 : * one (This may happen several times).
1189 : */
1190 0 : if (added) {
1191 0 : locks_delete_lock_ctx(fl, &dispose);
1192 0 : continue;
1193 : }
1194 : /*
1195 : * Replace the old lock with new_fl, and
1196 : * remove the old one. It's safe to do the
1197 : * insert here since we know that we won't be
1198 : * using new_fl later, and that the lock is
1199 : * just replacing an existing lock.
1200 : */
1201 0 : error = -ENOLCK;
1202 0 : if (!new_fl)
1203 : goto out;
1204 0 : locks_copy_lock(new_fl, request);
1205 0 : locks_move_blocks(new_fl, request);
1206 0 : request = new_fl;
1207 0 : new_fl = NULL;
1208 0 : locks_insert_lock_ctx(request, &fl->fl_list);
1209 0 : locks_delete_lock_ctx(fl, &dispose);
1210 0 : added = true;
1211 : }
1212 : }
1213 : }
1214 :
1215 : /*
1216 : * The above code only modifies existing locks in case of merging or
1217 : * replacing. If new lock(s) need to be inserted all modifications are
1218 : * done below this, so it's safe yet to bail out.
1219 : */
1220 0 : error = -ENOLCK; /* "no luck" */
1221 0 : if (right && left == right && !new_fl2)
1222 : goto out;
1223 :
1224 0 : error = 0;
1225 0 : if (!added) {
1226 0 : if (request->fl_type == F_UNLCK) {
1227 0 : if (request->fl_flags & FL_EXISTS)
1228 0 : error = -ENOENT;
1229 : goto out;
1230 : }
1231 :
1232 0 : if (!new_fl) {
1233 : error = -ENOLCK;
1234 : goto out;
1235 : }
1236 0 : locks_copy_lock(new_fl, request);
1237 0 : locks_move_blocks(new_fl, request);
1238 0 : locks_insert_lock_ctx(new_fl, &fl->fl_list);
1239 0 : fl = new_fl;
1240 0 : new_fl = NULL;
1241 : }
1242 0 : if (right) {
1243 0 : if (left == right) {
1244 : /* The new lock breaks the old one in two pieces,
1245 : * so we have to use the second new lock.
1246 : */
1247 0 : left = new_fl2;
1248 0 : new_fl2 = NULL;
1249 0 : locks_copy_lock(left, right);
1250 0 : locks_insert_lock_ctx(left, &fl->fl_list);
1251 : }
1252 0 : right->fl_start = request->fl_end + 1;
1253 : locks_wake_up_blocks(right);
1254 : }
1255 0 : if (left) {
1256 0 : left->fl_end = request->fl_start - 1;
1257 : locks_wake_up_blocks(left);
1258 : }
1259 : out:
1260 0 : spin_unlock(&ctx->flc_lock);
1261 0 : percpu_up_read(&file_rwsem);
1262 : /*
1263 : * Free any unused locks.
1264 : */
1265 0 : if (new_fl)
1266 : locks_free_lock(new_fl);
1267 0 : if (new_fl2)
1268 : locks_free_lock(new_fl2);
1269 0 : locks_dispose_list(&dispose);
1270 0 : trace_posix_lock_inode(inode, request, error);
1271 :
1272 0 : return error;
1273 : }
1274 :
1275 : /**
1276 : * posix_lock_file - Apply a POSIX-style lock to a file
1277 : * @filp: The file to apply the lock to
1278 : * @fl: The lock to be applied
1279 : * @conflock: Place to return a copy of the conflicting lock, if found.
1280 : *
1281 : * Add a POSIX style lock to a file.
1282 : * We merge adjacent & overlapping locks whenever possible.
1283 : * POSIX locks are sorted by owner task, then by starting address
1284 : *
1285 : * Note that if called with an FL_EXISTS argument, the caller may determine
1286 : * whether or not a lock was successfully freed by testing the return
1287 : * value for -ENOENT.
1288 : */
1289 0 : int posix_lock_file(struct file *filp, struct file_lock *fl,
1290 : struct file_lock *conflock)
1291 : {
1292 0 : return posix_lock_inode(locks_inode(filp), fl, conflock);
1293 : }
1294 : EXPORT_SYMBOL(posix_lock_file);
1295 :
1296 : /**
1297 : * posix_lock_inode_wait - Apply a POSIX-style lock to a file
1298 : * @inode: inode of file to which lock request should be applied
1299 : * @fl: The lock to be applied
1300 : *
1301 : * Apply a POSIX style lock request to an inode.
1302 : */
1303 0 : static int posix_lock_inode_wait(struct inode *inode, struct file_lock *fl)
1304 : {
1305 : int error;
1306 : might_sleep ();
1307 : for (;;) {
1308 0 : error = posix_lock_inode(inode, fl, NULL);
1309 0 : if (error != FILE_LOCK_DEFERRED)
1310 : break;
1311 0 : error = wait_event_interruptible(fl->fl_wait,
1312 : list_empty(&fl->fl_blocked_member));
1313 0 : if (error)
1314 : break;
1315 : }
1316 0 : locks_delete_block(fl);
1317 0 : return error;
1318 : }
1319 :
1320 : static void lease_clear_pending(struct file_lock *fl, int arg)
1321 : {
1322 0 : switch (arg) {
1323 : case F_UNLCK:
1324 0 : fl->fl_flags &= ~FL_UNLOCK_PENDING;
1325 : fallthrough;
1326 : case F_RDLCK:
1327 0 : fl->fl_flags &= ~FL_DOWNGRADE_PENDING;
1328 : }
1329 : }
1330 :
1331 : /* We already had a lease on this file; just change its type */
1332 0 : int lease_modify(struct file_lock *fl, int arg, struct list_head *dispose)
1333 : {
1334 0 : int error = assign_type(fl, arg);
1335 :
1336 0 : if (error)
1337 : return error;
1338 0 : lease_clear_pending(fl, arg);
1339 0 : locks_wake_up_blocks(fl);
1340 0 : if (arg == F_UNLCK) {
1341 0 : struct file *filp = fl->fl_file;
1342 :
1343 0 : f_delown(filp);
1344 0 : filp->f_owner.signum = 0;
1345 0 : fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
1346 0 : if (fl->fl_fasync != NULL) {
1347 0 : printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
1348 0 : fl->fl_fasync = NULL;
1349 : }
1350 0 : locks_delete_lock_ctx(fl, dispose);
1351 : }
1352 : return 0;
1353 : }
1354 : EXPORT_SYMBOL(lease_modify);
1355 :
1356 : static bool past_time(unsigned long then)
1357 : {
1358 0 : if (!then)
1359 : /* 0 is a special value meaning "this never expires": */
1360 : return false;
1361 0 : return time_after(jiffies, then);
1362 : }
1363 :
1364 0 : static void time_out_leases(struct inode *inode, struct list_head *dispose)
1365 : {
1366 0 : struct file_lock_context *ctx = inode->i_flctx;
1367 : struct file_lock *fl, *tmp;
1368 :
1369 : lockdep_assert_held(&ctx->flc_lock);
1370 :
1371 0 : list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list) {
1372 0 : trace_time_out_leases(inode, fl);
1373 0 : if (past_time(fl->fl_downgrade_time))
1374 0 : lease_modify(fl, F_RDLCK, dispose);
1375 0 : if (past_time(fl->fl_break_time))
1376 0 : lease_modify(fl, F_UNLCK, dispose);
1377 : }
1378 0 : }
1379 :
1380 0 : static bool leases_conflict(struct file_lock *lease, struct file_lock *breaker)
1381 : {
1382 : bool rc;
1383 :
1384 0 : if (lease->fl_lmops->lm_breaker_owns_lease
1385 0 : && lease->fl_lmops->lm_breaker_owns_lease(lease))
1386 : return false;
1387 0 : if ((breaker->fl_flags & FL_LAYOUT) != (lease->fl_flags & FL_LAYOUT)) {
1388 : rc = false;
1389 : goto trace;
1390 : }
1391 0 : if ((breaker->fl_flags & FL_DELEG) && (lease->fl_flags & FL_LEASE)) {
1392 : rc = false;
1393 : goto trace;
1394 : }
1395 :
1396 0 : rc = locks_conflict(breaker, lease);
1397 : trace:
1398 0 : trace_leases_conflict(rc, lease, breaker);
1399 0 : return rc;
1400 : }
1401 :
1402 : static bool
1403 0 : any_leases_conflict(struct inode *inode, struct file_lock *breaker)
1404 : {
1405 0 : struct file_lock_context *ctx = inode->i_flctx;
1406 : struct file_lock *fl;
1407 :
1408 : lockdep_assert_held(&ctx->flc_lock);
1409 :
1410 0 : list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1411 0 : if (leases_conflict(fl, breaker))
1412 : return true;
1413 : }
1414 : return false;
1415 : }
1416 :
1417 : /**
1418 : * __break_lease - revoke all outstanding leases on file
1419 : * @inode: the inode of the file to return
1420 : * @mode: O_RDONLY: break only write leases; O_WRONLY or O_RDWR:
1421 : * break all leases
1422 : * @type: FL_LEASE: break leases and delegations; FL_DELEG: break
1423 : * only delegations
1424 : *
1425 : * break_lease (inlined for speed) has checked there already is at least
1426 : * some kind of lock (maybe a lease) on this file. Leases are broken on
1427 : * a call to open() or truncate(). This function can sleep unless you
1428 : * specified %O_NONBLOCK to your open().
1429 : */
1430 0 : int __break_lease(struct inode *inode, unsigned int mode, unsigned int type)
1431 : {
1432 0 : int error = 0;
1433 : struct file_lock_context *ctx;
1434 : struct file_lock *new_fl, *fl, *tmp;
1435 : unsigned long break_time;
1436 0 : int want_write = (mode & O_ACCMODE) != O_RDONLY;
1437 0 : LIST_HEAD(dispose);
1438 :
1439 0 : new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK);
1440 0 : if (IS_ERR(new_fl))
1441 0 : return PTR_ERR(new_fl);
1442 0 : new_fl->fl_flags = type;
1443 :
1444 : /* typically we will check that ctx is non-NULL before calling */
1445 0 : ctx = smp_load_acquire(&inode->i_flctx);
1446 0 : if (!ctx) {
1447 0 : WARN_ON_ONCE(1);
1448 : goto free_lock;
1449 : }
1450 :
1451 0 : percpu_down_read(&file_rwsem);
1452 0 : spin_lock(&ctx->flc_lock);
1453 :
1454 0 : time_out_leases(inode, &dispose);
1455 :
1456 0 : if (!any_leases_conflict(inode, new_fl))
1457 : goto out;
1458 :
1459 0 : break_time = 0;
1460 0 : if (lease_break_time > 0) {
1461 0 : break_time = jiffies + lease_break_time * HZ;
1462 0 : if (break_time == 0)
1463 0 : break_time++; /* so that 0 means no break time */
1464 : }
1465 :
1466 0 : list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list) {
1467 0 : if (!leases_conflict(fl, new_fl))
1468 0 : continue;
1469 0 : if (want_write) {
1470 0 : if (fl->fl_flags & FL_UNLOCK_PENDING)
1471 0 : continue;
1472 0 : fl->fl_flags |= FL_UNLOCK_PENDING;
1473 0 : fl->fl_break_time = break_time;
1474 : } else {
1475 0 : if (lease_breaking(fl))
1476 0 : continue;
1477 0 : fl->fl_flags |= FL_DOWNGRADE_PENDING;
1478 0 : fl->fl_downgrade_time = break_time;
1479 : }
1480 0 : if (fl->fl_lmops->lm_break(fl))
1481 : locks_delete_lock_ctx(fl, &dispose);
1482 : }
1483 :
1484 0 : if (list_empty(&ctx->flc_lease))
1485 : goto out;
1486 :
1487 0 : if (mode & O_NONBLOCK) {
1488 : trace_break_lease_noblock(inode, new_fl);
1489 : error = -EWOULDBLOCK;
1490 : goto out;
1491 : }
1492 :
1493 : restart:
1494 0 : fl = list_first_entry(&ctx->flc_lease, struct file_lock, fl_list);
1495 0 : break_time = fl->fl_break_time;
1496 0 : if (break_time != 0)
1497 0 : break_time -= jiffies;
1498 0 : if (break_time == 0)
1499 0 : break_time++;
1500 0 : locks_insert_block(fl, new_fl, leases_conflict);
1501 0 : trace_break_lease_block(inode, new_fl);
1502 0 : spin_unlock(&ctx->flc_lock);
1503 0 : percpu_up_read(&file_rwsem);
1504 :
1505 0 : locks_dispose_list(&dispose);
1506 0 : error = wait_event_interruptible_timeout(new_fl->fl_wait,
1507 : list_empty(&new_fl->fl_blocked_member),
1508 : break_time);
1509 :
1510 0 : percpu_down_read(&file_rwsem);
1511 0 : spin_lock(&ctx->flc_lock);
1512 0 : trace_break_lease_unblock(inode, new_fl);
1513 0 : locks_delete_block(new_fl);
1514 0 : if (error >= 0) {
1515 : /*
1516 : * Wait for the next conflicting lease that has not been
1517 : * broken yet
1518 : */
1519 0 : if (error == 0)
1520 0 : time_out_leases(inode, &dispose);
1521 0 : if (any_leases_conflict(inode, new_fl))
1522 : goto restart;
1523 : error = 0;
1524 : }
1525 : out:
1526 0 : spin_unlock(&ctx->flc_lock);
1527 0 : percpu_up_read(&file_rwsem);
1528 0 : locks_dispose_list(&dispose);
1529 : free_lock:
1530 0 : locks_free_lock(new_fl);
1531 0 : return error;
1532 : }
1533 : EXPORT_SYMBOL(__break_lease);
1534 :
1535 : /**
1536 : * lease_get_mtime - update modified time of an inode with exclusive lease
1537 : * @inode: the inode
1538 : * @time: pointer to a timespec which contains the last modified time
1539 : *
1540 : * This is to force NFS clients to flush their caches for files with
1541 : * exclusive leases. The justification is that if someone has an
1542 : * exclusive lease, then they could be modifying it.
1543 : */
1544 0 : void lease_get_mtime(struct inode *inode, struct timespec64 *time)
1545 : {
1546 0 : bool has_lease = false;
1547 : struct file_lock_context *ctx;
1548 : struct file_lock *fl;
1549 :
1550 0 : ctx = smp_load_acquire(&inode->i_flctx);
1551 0 : if (ctx && !list_empty_careful(&ctx->flc_lease)) {
1552 0 : spin_lock(&ctx->flc_lock);
1553 0 : fl = list_first_entry_or_null(&ctx->flc_lease,
1554 : struct file_lock, fl_list);
1555 0 : if (fl && (fl->fl_type == F_WRLCK))
1556 0 : has_lease = true;
1557 0 : spin_unlock(&ctx->flc_lock);
1558 : }
1559 :
1560 0 : if (has_lease)
1561 0 : *time = current_time(inode);
1562 0 : }
1563 : EXPORT_SYMBOL(lease_get_mtime);
1564 :
1565 : /**
1566 : * fcntl_getlease - Enquire what lease is currently active
1567 : * @filp: the file
1568 : *
1569 : * The value returned by this function will be one of
1570 : * (if no lease break is pending):
1571 : *
1572 : * %F_RDLCK to indicate a shared lease is held.
1573 : *
1574 : * %F_WRLCK to indicate an exclusive lease is held.
1575 : *
1576 : * %F_UNLCK to indicate no lease is held.
1577 : *
1578 : * (if a lease break is pending):
1579 : *
1580 : * %F_RDLCK to indicate an exclusive lease needs to be
1581 : * changed to a shared lease (or removed).
1582 : *
1583 : * %F_UNLCK to indicate the lease needs to be removed.
1584 : *
1585 : * XXX: sfr & willy disagree over whether F_INPROGRESS
1586 : * should be returned to userspace.
1587 : */
1588 0 : int fcntl_getlease(struct file *filp)
1589 : {
1590 : struct file_lock *fl;
1591 0 : struct inode *inode = locks_inode(filp);
1592 : struct file_lock_context *ctx;
1593 0 : int type = F_UNLCK;
1594 0 : LIST_HEAD(dispose);
1595 :
1596 0 : ctx = smp_load_acquire(&inode->i_flctx);
1597 0 : if (ctx && !list_empty_careful(&ctx->flc_lease)) {
1598 0 : percpu_down_read(&file_rwsem);
1599 0 : spin_lock(&ctx->flc_lock);
1600 0 : time_out_leases(inode, &dispose);
1601 0 : list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1602 0 : if (fl->fl_file != filp)
1603 0 : continue;
1604 0 : type = target_leasetype(fl);
1605 : break;
1606 : }
1607 0 : spin_unlock(&ctx->flc_lock);
1608 0 : percpu_up_read(&file_rwsem);
1609 :
1610 0 : locks_dispose_list(&dispose);
1611 : }
1612 0 : return type;
1613 : }
1614 :
1615 : /**
1616 : * check_conflicting_open - see if the given file points to an inode that has
1617 : * an existing open that would conflict with the
1618 : * desired lease.
1619 : * @filp: file to check
1620 : * @arg: type of lease that we're trying to acquire
1621 : * @flags: current lock flags
1622 : *
1623 : * Check to see if there's an existing open fd on this file that would
1624 : * conflict with the lease we're trying to set.
1625 : */
1626 : static int
1627 0 : check_conflicting_open(struct file *filp, const long arg, int flags)
1628 : {
1629 0 : struct inode *inode = locks_inode(filp);
1630 0 : int self_wcount = 0, self_rcount = 0;
1631 :
1632 0 : if (flags & FL_LAYOUT)
1633 : return 0;
1634 0 : if (flags & FL_DELEG)
1635 : /* We leave these checks to the caller */
1636 : return 0;
1637 :
1638 0 : if (arg == F_RDLCK)
1639 0 : return inode_is_open_for_write(inode) ? -EAGAIN : 0;
1640 0 : else if (arg != F_WRLCK)
1641 : return 0;
1642 :
1643 : /*
1644 : * Make sure that only read/write count is from lease requestor.
1645 : * Note that this will result in denying write leases when i_writecount
1646 : * is negative, which is what we want. (We shouldn't grant write leases
1647 : * on files open for execution.)
1648 : */
1649 0 : if (filp->f_mode & FMODE_WRITE)
1650 : self_wcount = 1;
1651 0 : else if (filp->f_mode & FMODE_READ)
1652 0 : self_rcount = 1;
1653 :
1654 0 : if (atomic_read(&inode->i_writecount) != self_wcount ||
1655 0 : atomic_read(&inode->i_readcount) != self_rcount)
1656 : return -EAGAIN;
1657 :
1658 : return 0;
1659 : }
1660 :
1661 : static int
1662 0 : generic_add_lease(struct file *filp, long arg, struct file_lock **flp, void **priv)
1663 : {
1664 0 : struct file_lock *fl, *my_fl = NULL, *lease;
1665 0 : struct inode *inode = locks_inode(filp);
1666 : struct file_lock_context *ctx;
1667 0 : bool is_deleg = (*flp)->fl_flags & FL_DELEG;
1668 : int error;
1669 0 : LIST_HEAD(dispose);
1670 :
1671 0 : lease = *flp;
1672 0 : trace_generic_add_lease(inode, lease);
1673 :
1674 : /* Note that arg is never F_UNLCK here */
1675 0 : ctx = locks_get_lock_context(inode, arg);
1676 0 : if (!ctx)
1677 : return -ENOMEM;
1678 :
1679 : /*
1680 : * In the delegation case we need mutual exclusion with
1681 : * a number of operations that take the i_mutex. We trylock
1682 : * because delegations are an optional optimization, and if
1683 : * there's some chance of a conflict--we'd rather not
1684 : * bother, maybe that's a sign this just isn't a good file to
1685 : * hand out a delegation on.
1686 : */
1687 0 : if (is_deleg && !inode_trylock(inode))
1688 : return -EAGAIN;
1689 :
1690 0 : if (is_deleg && arg == F_WRLCK) {
1691 : /* Write delegations are not currently supported: */
1692 0 : inode_unlock(inode);
1693 0 : WARN_ON_ONCE(1);
1694 : return -EINVAL;
1695 : }
1696 :
1697 0 : percpu_down_read(&file_rwsem);
1698 0 : spin_lock(&ctx->flc_lock);
1699 0 : time_out_leases(inode, &dispose);
1700 0 : error = check_conflicting_open(filp, arg, lease->fl_flags);
1701 0 : if (error)
1702 : goto out;
1703 :
1704 : /*
1705 : * At this point, we know that if there is an exclusive
1706 : * lease on this file, then we hold it on this filp
1707 : * (otherwise our open of this file would have blocked).
1708 : * And if we are trying to acquire an exclusive lease,
1709 : * then the file is not open by anyone (including us)
1710 : * except for this filp.
1711 : */
1712 0 : error = -EAGAIN;
1713 0 : list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1714 0 : if (fl->fl_file == filp &&
1715 0 : fl->fl_owner == lease->fl_owner) {
1716 0 : my_fl = fl;
1717 0 : continue;
1718 : }
1719 :
1720 : /*
1721 : * No exclusive leases if someone else has a lease on
1722 : * this file:
1723 : */
1724 0 : if (arg == F_WRLCK)
1725 : goto out;
1726 : /*
1727 : * Modifying our existing lease is OK, but no getting a
1728 : * new lease if someone else is opening for write:
1729 : */
1730 0 : if (fl->fl_flags & FL_UNLOCK_PENDING)
1731 : goto out;
1732 : }
1733 :
1734 0 : if (my_fl != NULL) {
1735 0 : lease = my_fl;
1736 0 : error = lease->fl_lmops->lm_change(lease, arg, &dispose);
1737 0 : if (error)
1738 : goto out;
1739 : goto out_setup;
1740 : }
1741 :
1742 0 : error = -EINVAL;
1743 0 : if (!leases_enable)
1744 : goto out;
1745 :
1746 0 : locks_insert_lock_ctx(lease, &ctx->flc_lease);
1747 : /*
1748 : * The check in break_lease() is lockless. It's possible for another
1749 : * open to race in after we did the earlier check for a conflicting
1750 : * open but before the lease was inserted. Check again for a
1751 : * conflicting open and cancel the lease if there is one.
1752 : *
1753 : * We also add a barrier here to ensure that the insertion of the lock
1754 : * precedes these checks.
1755 : */
1756 0 : smp_mb();
1757 0 : error = check_conflicting_open(filp, arg, lease->fl_flags);
1758 0 : if (error) {
1759 0 : locks_unlink_lock_ctx(lease);
1760 0 : goto out;
1761 : }
1762 :
1763 : out_setup:
1764 0 : if (lease->fl_lmops->lm_setup)
1765 0 : lease->fl_lmops->lm_setup(lease, priv);
1766 : out:
1767 0 : spin_unlock(&ctx->flc_lock);
1768 0 : percpu_up_read(&file_rwsem);
1769 0 : locks_dispose_list(&dispose);
1770 0 : if (is_deleg)
1771 : inode_unlock(inode);
1772 0 : if (!error && !my_fl)
1773 0 : *flp = NULL;
1774 : return error;
1775 : }
1776 :
1777 0 : static int generic_delete_lease(struct file *filp, void *owner)
1778 : {
1779 0 : int error = -EAGAIN;
1780 0 : struct file_lock *fl, *victim = NULL;
1781 0 : struct inode *inode = locks_inode(filp);
1782 : struct file_lock_context *ctx;
1783 0 : LIST_HEAD(dispose);
1784 :
1785 0 : ctx = smp_load_acquire(&inode->i_flctx);
1786 0 : if (!ctx) {
1787 : trace_generic_delete_lease(inode, NULL);
1788 : return error;
1789 : }
1790 :
1791 0 : percpu_down_read(&file_rwsem);
1792 0 : spin_lock(&ctx->flc_lock);
1793 0 : list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1794 0 : if (fl->fl_file == filp &&
1795 0 : fl->fl_owner == owner) {
1796 : victim = fl;
1797 : break;
1798 : }
1799 : }
1800 0 : trace_generic_delete_lease(inode, victim);
1801 0 : if (victim)
1802 0 : error = fl->fl_lmops->lm_change(victim, F_UNLCK, &dispose);
1803 0 : spin_unlock(&ctx->flc_lock);
1804 0 : percpu_up_read(&file_rwsem);
1805 0 : locks_dispose_list(&dispose);
1806 0 : return error;
1807 : }
1808 :
1809 : /**
1810 : * generic_setlease - sets a lease on an open file
1811 : * @filp: file pointer
1812 : * @arg: type of lease to obtain
1813 : * @flp: input - file_lock to use, output - file_lock inserted
1814 : * @priv: private data for lm_setup (may be NULL if lm_setup
1815 : * doesn't require it)
1816 : *
1817 : * The (input) flp->fl_lmops->lm_break function is required
1818 : * by break_lease().
1819 : */
1820 0 : int generic_setlease(struct file *filp, long arg, struct file_lock **flp,
1821 : void **priv)
1822 : {
1823 0 : struct inode *inode = locks_inode(filp);
1824 : int error;
1825 :
1826 0 : if ((!uid_eq(current_fsuid(), inode->i_uid)) && !capable(CAP_LEASE))
1827 : return -EACCES;
1828 0 : if (!S_ISREG(inode->i_mode))
1829 : return -EINVAL;
1830 0 : error = security_file_lock(filp, arg);
1831 : if (error)
1832 : return error;
1833 :
1834 0 : switch (arg) {
1835 : case F_UNLCK:
1836 0 : return generic_delete_lease(filp, *priv);
1837 : case F_RDLCK:
1838 : case F_WRLCK:
1839 0 : if (!(*flp)->fl_lmops->lm_break) {
1840 0 : WARN_ON_ONCE(1);
1841 : return -ENOLCK;
1842 : }
1843 :
1844 0 : return generic_add_lease(filp, arg, flp, priv);
1845 : default:
1846 : return -EINVAL;
1847 : }
1848 : }
1849 : EXPORT_SYMBOL(generic_setlease);
1850 :
1851 : #if IS_ENABLED(CONFIG_SRCU)
1852 : /*
1853 : * Kernel subsystems can register to be notified on any attempt to set
1854 : * a new lease with the lease_notifier_chain. This is used by (e.g.) nfsd
1855 : * to close files that it may have cached when there is an attempt to set a
1856 : * conflicting lease.
1857 : */
1858 : static struct srcu_notifier_head lease_notifier_chain;
1859 :
1860 : static inline void
1861 : lease_notifier_chain_init(void)
1862 : {
1863 1 : srcu_init_notifier_head(&lease_notifier_chain);
1864 : }
1865 :
1866 : static inline void
1867 : setlease_notifier(long arg, struct file_lock *lease)
1868 : {
1869 0 : if (arg != F_UNLCK)
1870 0 : srcu_notifier_call_chain(&lease_notifier_chain, arg, lease);
1871 : }
1872 :
1873 0 : int lease_register_notifier(struct notifier_block *nb)
1874 : {
1875 0 : return srcu_notifier_chain_register(&lease_notifier_chain, nb);
1876 : }
1877 : EXPORT_SYMBOL_GPL(lease_register_notifier);
1878 :
1879 0 : void lease_unregister_notifier(struct notifier_block *nb)
1880 : {
1881 0 : srcu_notifier_chain_unregister(&lease_notifier_chain, nb);
1882 0 : }
1883 : EXPORT_SYMBOL_GPL(lease_unregister_notifier);
1884 :
1885 : #else /* !IS_ENABLED(CONFIG_SRCU) */
1886 : static inline void
1887 : lease_notifier_chain_init(void)
1888 : {
1889 : }
1890 :
1891 : static inline void
1892 : setlease_notifier(long arg, struct file_lock *lease)
1893 : {
1894 : }
1895 :
1896 : int lease_register_notifier(struct notifier_block *nb)
1897 : {
1898 : return 0;
1899 : }
1900 : EXPORT_SYMBOL_GPL(lease_register_notifier);
1901 :
1902 : void lease_unregister_notifier(struct notifier_block *nb)
1903 : {
1904 : }
1905 : EXPORT_SYMBOL_GPL(lease_unregister_notifier);
1906 :
1907 : #endif /* IS_ENABLED(CONFIG_SRCU) */
1908 :
1909 : /**
1910 : * vfs_setlease - sets a lease on an open file
1911 : * @filp: file pointer
1912 : * @arg: type of lease to obtain
1913 : * @lease: file_lock to use when adding a lease
1914 : * @priv: private info for lm_setup when adding a lease (may be
1915 : * NULL if lm_setup doesn't require it)
1916 : *
1917 : * Call this to establish a lease on the file. The "lease" argument is not
1918 : * used for F_UNLCK requests and may be NULL. For commands that set or alter
1919 : * an existing lease, the ``(*lease)->fl_lmops->lm_break`` operation must be
1920 : * set; if not, this function will return -ENOLCK (and generate a scary-looking
1921 : * stack trace).
1922 : *
1923 : * The "priv" pointer is passed directly to the lm_setup function as-is. It
1924 : * may be NULL if the lm_setup operation doesn't require it.
1925 : */
1926 : int
1927 0 : vfs_setlease(struct file *filp, long arg, struct file_lock **lease, void **priv)
1928 : {
1929 0 : if (lease)
1930 0 : setlease_notifier(arg, *lease);
1931 0 : if (filp->f_op->setlease)
1932 0 : return filp->f_op->setlease(filp, arg, lease, priv);
1933 : else
1934 0 : return generic_setlease(filp, arg, lease, priv);
1935 : }
1936 : EXPORT_SYMBOL_GPL(vfs_setlease);
1937 :
1938 0 : static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg)
1939 : {
1940 : struct file_lock *fl;
1941 : struct fasync_struct *new;
1942 : int error;
1943 :
1944 0 : fl = lease_alloc(filp, arg);
1945 0 : if (IS_ERR(fl))
1946 0 : return PTR_ERR(fl);
1947 :
1948 0 : new = fasync_alloc();
1949 0 : if (!new) {
1950 0 : locks_free_lock(fl);
1951 0 : return -ENOMEM;
1952 : }
1953 0 : new->fa_fd = fd;
1954 :
1955 0 : error = vfs_setlease(filp, arg, &fl, (void **)&new);
1956 0 : if (fl)
1957 0 : locks_free_lock(fl);
1958 0 : if (new)
1959 0 : fasync_free(new);
1960 : return error;
1961 : }
1962 :
1963 : /**
1964 : * fcntl_setlease - sets a lease on an open file
1965 : * @fd: open file descriptor
1966 : * @filp: file pointer
1967 : * @arg: type of lease to obtain
1968 : *
1969 : * Call this fcntl to establish a lease on the file.
1970 : * Note that you also need to call %F_SETSIG to
1971 : * receive a signal when the lease is broken.
1972 : */
1973 0 : int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1974 : {
1975 0 : if (arg == F_UNLCK)
1976 0 : return vfs_setlease(filp, F_UNLCK, NULL, (void **)&filp);
1977 0 : return do_fcntl_add_lease(fd, filp, arg);
1978 : }
1979 :
1980 : /**
1981 : * flock_lock_inode_wait - Apply a FLOCK-style lock to a file
1982 : * @inode: inode of the file to apply to
1983 : * @fl: The lock to be applied
1984 : *
1985 : * Apply a FLOCK style lock request to an inode.
1986 : */
1987 0 : static int flock_lock_inode_wait(struct inode *inode, struct file_lock *fl)
1988 : {
1989 : int error;
1990 : might_sleep();
1991 : for (;;) {
1992 0 : error = flock_lock_inode(inode, fl);
1993 0 : if (error != FILE_LOCK_DEFERRED)
1994 : break;
1995 0 : error = wait_event_interruptible(fl->fl_wait,
1996 : list_empty(&fl->fl_blocked_member));
1997 0 : if (error)
1998 : break;
1999 : }
2000 0 : locks_delete_block(fl);
2001 0 : return error;
2002 : }
2003 :
2004 : /**
2005 : * locks_lock_inode_wait - Apply a lock to an inode
2006 : * @inode: inode of the file to apply to
2007 : * @fl: The lock to be applied
2008 : *
2009 : * Apply a POSIX or FLOCK style lock request to an inode.
2010 : */
2011 0 : int locks_lock_inode_wait(struct inode *inode, struct file_lock *fl)
2012 : {
2013 0 : int res = 0;
2014 0 : switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
2015 : case FL_POSIX:
2016 0 : res = posix_lock_inode_wait(inode, fl);
2017 0 : break;
2018 : case FL_FLOCK:
2019 0 : res = flock_lock_inode_wait(inode, fl);
2020 0 : break;
2021 : default:
2022 0 : BUG();
2023 : }
2024 0 : return res;
2025 : }
2026 : EXPORT_SYMBOL(locks_lock_inode_wait);
2027 :
2028 : /**
2029 : * sys_flock: - flock() system call.
2030 : * @fd: the file descriptor to lock.
2031 : * @cmd: the type of lock to apply.
2032 : *
2033 : * Apply a %FL_FLOCK style lock to an open file descriptor.
2034 : * The @cmd can be one of:
2035 : *
2036 : * - %LOCK_SH -- a shared lock.
2037 : * - %LOCK_EX -- an exclusive lock.
2038 : * - %LOCK_UN -- remove an existing lock.
2039 : * - %LOCK_MAND -- a 'mandatory' flock. (DEPRECATED)
2040 : *
2041 : * %LOCK_MAND support has been removed from the kernel.
2042 : */
2043 0 : SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
2044 : {
2045 0 : struct fd f = fdget(fd);
2046 : struct file_lock *lock;
2047 : int can_sleep, unlock;
2048 : int error;
2049 :
2050 0 : error = -EBADF;
2051 0 : if (!f.file)
2052 : goto out;
2053 :
2054 0 : can_sleep = !(cmd & LOCK_NB);
2055 0 : cmd &= ~LOCK_NB;
2056 0 : unlock = (cmd == LOCK_UN);
2057 :
2058 0 : if (!unlock && !(f.file->f_mode & (FMODE_READ|FMODE_WRITE)))
2059 : goto out_putf;
2060 :
2061 : /*
2062 : * LOCK_MAND locks were broken for a long time in that they never
2063 : * conflicted with one another and didn't prevent any sort of open,
2064 : * read or write activity.
2065 : *
2066 : * Just ignore these requests now, to preserve legacy behavior, but
2067 : * throw a warning to let people know that they don't actually work.
2068 : */
2069 0 : if (cmd & LOCK_MAND) {
2070 0 : pr_warn_once("Attempt to set a LOCK_MAND lock via flock(2). This support has been removed and the request ignored.\n");
2071 : error = 0;
2072 : goto out_putf;
2073 : }
2074 :
2075 0 : lock = flock_make_lock(f.file, cmd, NULL);
2076 0 : if (IS_ERR(lock)) {
2077 0 : error = PTR_ERR(lock);
2078 0 : goto out_putf;
2079 : }
2080 :
2081 0 : if (can_sleep)
2082 0 : lock->fl_flags |= FL_SLEEP;
2083 :
2084 0 : error = security_file_lock(f.file, lock->fl_type);
2085 : if (error)
2086 : goto out_free;
2087 :
2088 0 : if (f.file->f_op->flock)
2089 0 : error = f.file->f_op->flock(f.file,
2090 : (can_sleep) ? F_SETLKW : F_SETLK,
2091 : lock);
2092 : else
2093 0 : error = locks_lock_file_wait(f.file, lock);
2094 :
2095 : out_free:
2096 : locks_free_lock(lock);
2097 :
2098 : out_putf:
2099 0 : fdput(f);
2100 : out:
2101 0 : return error;
2102 : }
2103 :
2104 : /**
2105 : * vfs_test_lock - test file byte range lock
2106 : * @filp: The file to test lock for
2107 : * @fl: The lock to test; also used to hold result
2108 : *
2109 : * Returns -ERRNO on failure. Indicates presence of conflicting lock by
2110 : * setting conf->fl_type to something other than F_UNLCK.
2111 : */
2112 0 : int vfs_test_lock(struct file *filp, struct file_lock *fl)
2113 : {
2114 0 : if (filp->f_op->lock)
2115 0 : return filp->f_op->lock(filp, F_GETLK, fl);
2116 0 : posix_test_lock(filp, fl);
2117 0 : return 0;
2118 : }
2119 : EXPORT_SYMBOL_GPL(vfs_test_lock);
2120 :
2121 : /**
2122 : * locks_translate_pid - translate a file_lock's fl_pid number into a namespace
2123 : * @fl: The file_lock who's fl_pid should be translated
2124 : * @ns: The namespace into which the pid should be translated
2125 : *
2126 : * Used to tranlate a fl_pid into a namespace virtual pid number
2127 : */
2128 0 : static pid_t locks_translate_pid(struct file_lock *fl, struct pid_namespace *ns)
2129 : {
2130 : pid_t vnr;
2131 : struct pid *pid;
2132 :
2133 0 : if (IS_OFDLCK(fl))
2134 : return -1;
2135 0 : if (IS_REMOTELCK(fl))
2136 0 : return fl->fl_pid;
2137 : /*
2138 : * If the flock owner process is dead and its pid has been already
2139 : * freed, the translation below won't work, but we still want to show
2140 : * flock owner pid number in init pidns.
2141 : */
2142 0 : if (ns == &init_pid_ns)
2143 0 : return (pid_t)fl->fl_pid;
2144 :
2145 : rcu_read_lock();
2146 0 : pid = find_pid_ns(fl->fl_pid, &init_pid_ns);
2147 0 : vnr = pid_nr_ns(pid, ns);
2148 : rcu_read_unlock();
2149 : return vnr;
2150 : }
2151 :
2152 0 : static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
2153 : {
2154 0 : flock->l_pid = locks_translate_pid(fl, task_active_pid_ns(current));
2155 : #if BITS_PER_LONG == 32
2156 : /*
2157 : * Make sure we can represent the posix lock via
2158 : * legacy 32bit flock.
2159 : */
2160 : if (fl->fl_start > OFFT_OFFSET_MAX)
2161 : return -EOVERFLOW;
2162 : if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
2163 : return -EOVERFLOW;
2164 : #endif
2165 0 : flock->l_start = fl->fl_start;
2166 0 : flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
2167 0 : fl->fl_end - fl->fl_start + 1;
2168 0 : flock->l_whence = 0;
2169 0 : flock->l_type = fl->fl_type;
2170 0 : return 0;
2171 : }
2172 :
2173 : #if BITS_PER_LONG == 32
2174 : static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
2175 : {
2176 : flock->l_pid = locks_translate_pid(fl, task_active_pid_ns(current));
2177 : flock->l_start = fl->fl_start;
2178 : flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
2179 : fl->fl_end - fl->fl_start + 1;
2180 : flock->l_whence = 0;
2181 : flock->l_type = fl->fl_type;
2182 : }
2183 : #endif
2184 :
2185 : /* Report the first existing lock that would conflict with l.
2186 : * This implements the F_GETLK command of fcntl().
2187 : */
2188 0 : int fcntl_getlk(struct file *filp, unsigned int cmd, struct flock *flock)
2189 : {
2190 : struct file_lock *fl;
2191 : int error;
2192 :
2193 0 : fl = locks_alloc_lock();
2194 0 : if (fl == NULL)
2195 : return -ENOMEM;
2196 0 : error = -EINVAL;
2197 0 : if (flock->l_type != F_RDLCK && flock->l_type != F_WRLCK)
2198 : goto out;
2199 :
2200 0 : error = flock_to_posix_lock(filp, fl, flock);
2201 0 : if (error)
2202 : goto out;
2203 :
2204 0 : if (cmd == F_OFD_GETLK) {
2205 0 : error = -EINVAL;
2206 0 : if (flock->l_pid != 0)
2207 : goto out;
2208 :
2209 0 : fl->fl_flags |= FL_OFDLCK;
2210 0 : fl->fl_owner = filp;
2211 : }
2212 :
2213 0 : error = vfs_test_lock(filp, fl);
2214 0 : if (error)
2215 : goto out;
2216 :
2217 0 : flock->l_type = fl->fl_type;
2218 0 : if (fl->fl_type != F_UNLCK) {
2219 0 : error = posix_lock_to_flock(flock, fl);
2220 : if (error)
2221 : goto out;
2222 : }
2223 : out:
2224 0 : locks_free_lock(fl);
2225 0 : return error;
2226 : }
2227 :
2228 : /**
2229 : * vfs_lock_file - file byte range lock
2230 : * @filp: The file to apply the lock to
2231 : * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
2232 : * @fl: The lock to be applied
2233 : * @conf: Place to return a copy of the conflicting lock, if found.
2234 : *
2235 : * A caller that doesn't care about the conflicting lock may pass NULL
2236 : * as the final argument.
2237 : *
2238 : * If the filesystem defines a private ->lock() method, then @conf will
2239 : * be left unchanged; so a caller that cares should initialize it to
2240 : * some acceptable default.
2241 : *
2242 : * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
2243 : * locks, the ->lock() interface may return asynchronously, before the lock has
2244 : * been granted or denied by the underlying filesystem, if (and only if)
2245 : * lm_grant is set. Callers expecting ->lock() to return asynchronously
2246 : * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
2247 : * the request is for a blocking lock. When ->lock() does return asynchronously,
2248 : * it must return FILE_LOCK_DEFERRED, and call ->lm_grant() when the lock
2249 : * request completes.
2250 : * If the request is for non-blocking lock the file system should return
2251 : * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
2252 : * with the result. If the request timed out the callback routine will return a
2253 : * nonzero return code and the file system should release the lock. The file
2254 : * system is also responsible to keep a corresponding posix lock when it
2255 : * grants a lock so the VFS can find out which locks are locally held and do
2256 : * the correct lock cleanup when required.
2257 : * The underlying filesystem must not drop the kernel lock or call
2258 : * ->lm_grant() before returning to the caller with a FILE_LOCK_DEFERRED
2259 : * return code.
2260 : */
2261 0 : int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
2262 : {
2263 0 : if (filp->f_op->lock)
2264 0 : return filp->f_op->lock(filp, cmd, fl);
2265 : else
2266 0 : return posix_lock_file(filp, fl, conf);
2267 : }
2268 : EXPORT_SYMBOL_GPL(vfs_lock_file);
2269 :
2270 0 : static int do_lock_file_wait(struct file *filp, unsigned int cmd,
2271 : struct file_lock *fl)
2272 : {
2273 : int error;
2274 :
2275 0 : error = security_file_lock(filp, fl->fl_type);
2276 : if (error)
2277 : return error;
2278 :
2279 : for (;;) {
2280 0 : error = vfs_lock_file(filp, cmd, fl, NULL);
2281 0 : if (error != FILE_LOCK_DEFERRED)
2282 : break;
2283 0 : error = wait_event_interruptible(fl->fl_wait,
2284 : list_empty(&fl->fl_blocked_member));
2285 0 : if (error)
2286 : break;
2287 : }
2288 0 : locks_delete_block(fl);
2289 :
2290 : return error;
2291 : }
2292 :
2293 : /* Ensure that fl->fl_file has compatible f_mode for F_SETLK calls */
2294 : static int
2295 : check_fmode_for_setlk(struct file_lock *fl)
2296 : {
2297 0 : switch (fl->fl_type) {
2298 : case F_RDLCK:
2299 0 : if (!(fl->fl_file->f_mode & FMODE_READ))
2300 : return -EBADF;
2301 : break;
2302 : case F_WRLCK:
2303 0 : if (!(fl->fl_file->f_mode & FMODE_WRITE))
2304 : return -EBADF;
2305 : }
2306 : return 0;
2307 : }
2308 :
2309 : /* Apply the lock described by l to an open file descriptor.
2310 : * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2311 : */
2312 0 : int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
2313 : struct flock *flock)
2314 : {
2315 0 : struct file_lock *file_lock = locks_alloc_lock();
2316 0 : struct inode *inode = locks_inode(filp);
2317 : struct file *f;
2318 : int error;
2319 :
2320 0 : if (file_lock == NULL)
2321 : return -ENOLCK;
2322 :
2323 0 : error = flock_to_posix_lock(filp, file_lock, flock);
2324 0 : if (error)
2325 : goto out;
2326 :
2327 0 : error = check_fmode_for_setlk(file_lock);
2328 0 : if (error)
2329 : goto out;
2330 :
2331 : /*
2332 : * If the cmd is requesting file-private locks, then set the
2333 : * FL_OFDLCK flag and override the owner.
2334 : */
2335 0 : switch (cmd) {
2336 : case F_OFD_SETLK:
2337 0 : error = -EINVAL;
2338 0 : if (flock->l_pid != 0)
2339 : goto out;
2340 :
2341 0 : cmd = F_SETLK;
2342 0 : file_lock->fl_flags |= FL_OFDLCK;
2343 0 : file_lock->fl_owner = filp;
2344 0 : break;
2345 : case F_OFD_SETLKW:
2346 0 : error = -EINVAL;
2347 0 : if (flock->l_pid != 0)
2348 : goto out;
2349 :
2350 0 : cmd = F_SETLKW;
2351 0 : file_lock->fl_flags |= FL_OFDLCK;
2352 0 : file_lock->fl_owner = filp;
2353 : fallthrough;
2354 : case F_SETLKW:
2355 0 : file_lock->fl_flags |= FL_SLEEP;
2356 : }
2357 :
2358 0 : error = do_lock_file_wait(filp, cmd, file_lock);
2359 :
2360 : /*
2361 : * Attempt to detect a close/fcntl race and recover by releasing the
2362 : * lock that was just acquired. There is no need to do that when we're
2363 : * unlocking though, or for OFD locks.
2364 : */
2365 0 : if (!error && file_lock->fl_type != F_UNLCK &&
2366 0 : !(file_lock->fl_flags & FL_OFDLCK)) {
2367 0 : struct files_struct *files = current->files;
2368 : /*
2369 : * We need that spin_lock here - it prevents reordering between
2370 : * update of i_flctx->flc_posix and check for it done in
2371 : * close(). rcu_read_lock() wouldn't do.
2372 : */
2373 0 : spin_lock(&files->file_lock);
2374 0 : f = files_lookup_fd_locked(files, fd);
2375 0 : spin_unlock(&files->file_lock);
2376 0 : if (f != filp) {
2377 0 : file_lock->fl_type = F_UNLCK;
2378 0 : error = do_lock_file_wait(filp, cmd, file_lock);
2379 0 : WARN_ON_ONCE(error);
2380 : error = -EBADF;
2381 : }
2382 : }
2383 : out:
2384 0 : trace_fcntl_setlk(inode, file_lock, error);
2385 0 : locks_free_lock(file_lock);
2386 0 : return error;
2387 : }
2388 :
2389 : #if BITS_PER_LONG == 32
2390 : /* Report the first existing lock that would conflict with l.
2391 : * This implements the F_GETLK command of fcntl().
2392 : */
2393 : int fcntl_getlk64(struct file *filp, unsigned int cmd, struct flock64 *flock)
2394 : {
2395 : struct file_lock *fl;
2396 : int error;
2397 :
2398 : fl = locks_alloc_lock();
2399 : if (fl == NULL)
2400 : return -ENOMEM;
2401 :
2402 : error = -EINVAL;
2403 : if (flock->l_type != F_RDLCK && flock->l_type != F_WRLCK)
2404 : goto out;
2405 :
2406 : error = flock64_to_posix_lock(filp, fl, flock);
2407 : if (error)
2408 : goto out;
2409 :
2410 : if (cmd == F_OFD_GETLK) {
2411 : error = -EINVAL;
2412 : if (flock->l_pid != 0)
2413 : goto out;
2414 :
2415 : cmd = F_GETLK64;
2416 : fl->fl_flags |= FL_OFDLCK;
2417 : fl->fl_owner = filp;
2418 : }
2419 :
2420 : error = vfs_test_lock(filp, fl);
2421 : if (error)
2422 : goto out;
2423 :
2424 : flock->l_type = fl->fl_type;
2425 : if (fl->fl_type != F_UNLCK)
2426 : posix_lock_to_flock64(flock, fl);
2427 :
2428 : out:
2429 : locks_free_lock(fl);
2430 : return error;
2431 : }
2432 :
2433 : /* Apply the lock described by l to an open file descriptor.
2434 : * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2435 : */
2436 : int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
2437 : struct flock64 *flock)
2438 : {
2439 : struct file_lock *file_lock = locks_alloc_lock();
2440 : struct file *f;
2441 : int error;
2442 :
2443 : if (file_lock == NULL)
2444 : return -ENOLCK;
2445 :
2446 : error = flock64_to_posix_lock(filp, file_lock, flock);
2447 : if (error)
2448 : goto out;
2449 :
2450 : error = check_fmode_for_setlk(file_lock);
2451 : if (error)
2452 : goto out;
2453 :
2454 : /*
2455 : * If the cmd is requesting file-private locks, then set the
2456 : * FL_OFDLCK flag and override the owner.
2457 : */
2458 : switch (cmd) {
2459 : case F_OFD_SETLK:
2460 : error = -EINVAL;
2461 : if (flock->l_pid != 0)
2462 : goto out;
2463 :
2464 : cmd = F_SETLK64;
2465 : file_lock->fl_flags |= FL_OFDLCK;
2466 : file_lock->fl_owner = filp;
2467 : break;
2468 : case F_OFD_SETLKW:
2469 : error = -EINVAL;
2470 : if (flock->l_pid != 0)
2471 : goto out;
2472 :
2473 : cmd = F_SETLKW64;
2474 : file_lock->fl_flags |= FL_OFDLCK;
2475 : file_lock->fl_owner = filp;
2476 : fallthrough;
2477 : case F_SETLKW64:
2478 : file_lock->fl_flags |= FL_SLEEP;
2479 : }
2480 :
2481 : error = do_lock_file_wait(filp, cmd, file_lock);
2482 :
2483 : /*
2484 : * Attempt to detect a close/fcntl race and recover by releasing the
2485 : * lock that was just acquired. There is no need to do that when we're
2486 : * unlocking though, or for OFD locks.
2487 : */
2488 : if (!error && file_lock->fl_type != F_UNLCK &&
2489 : !(file_lock->fl_flags & FL_OFDLCK)) {
2490 : struct files_struct *files = current->files;
2491 : /*
2492 : * We need that spin_lock here - it prevents reordering between
2493 : * update of i_flctx->flc_posix and check for it done in
2494 : * close(). rcu_read_lock() wouldn't do.
2495 : */
2496 : spin_lock(&files->file_lock);
2497 : f = files_lookup_fd_locked(files, fd);
2498 : spin_unlock(&files->file_lock);
2499 : if (f != filp) {
2500 : file_lock->fl_type = F_UNLCK;
2501 : error = do_lock_file_wait(filp, cmd, file_lock);
2502 : WARN_ON_ONCE(error);
2503 : error = -EBADF;
2504 : }
2505 : }
2506 : out:
2507 : locks_free_lock(file_lock);
2508 : return error;
2509 : }
2510 : #endif /* BITS_PER_LONG == 32 */
2511 :
2512 : /*
2513 : * This function is called when the file is being removed
2514 : * from the task's fd array. POSIX locks belonging to this task
2515 : * are deleted at this time.
2516 : */
2517 0 : void locks_remove_posix(struct file *filp, fl_owner_t owner)
2518 : {
2519 : int error;
2520 0 : struct inode *inode = locks_inode(filp);
2521 : struct file_lock lock;
2522 : struct file_lock_context *ctx;
2523 :
2524 : /*
2525 : * If there are no locks held on this file, we don't need to call
2526 : * posix_lock_file(). Another process could be setting a lock on this
2527 : * file at the same time, but we wouldn't remove that lock anyway.
2528 : */
2529 0 : ctx = smp_load_acquire(&inode->i_flctx);
2530 0 : if (!ctx || list_empty(&ctx->flc_posix))
2531 0 : return;
2532 :
2533 0 : locks_init_lock(&lock);
2534 0 : lock.fl_type = F_UNLCK;
2535 0 : lock.fl_flags = FL_POSIX | FL_CLOSE;
2536 0 : lock.fl_start = 0;
2537 0 : lock.fl_end = OFFSET_MAX;
2538 0 : lock.fl_owner = owner;
2539 0 : lock.fl_pid = current->tgid;
2540 0 : lock.fl_file = filp;
2541 0 : lock.fl_ops = NULL;
2542 0 : lock.fl_lmops = NULL;
2543 :
2544 0 : error = vfs_lock_file(filp, F_SETLK, &lock, NULL);
2545 :
2546 0 : if (lock.fl_ops && lock.fl_ops->fl_release_private)
2547 0 : lock.fl_ops->fl_release_private(&lock);
2548 0 : trace_locks_remove_posix(inode, &lock, error);
2549 : }
2550 : EXPORT_SYMBOL(locks_remove_posix);
2551 :
2552 : /* The i_flctx must be valid when calling into here */
2553 : static void
2554 0 : locks_remove_flock(struct file *filp, struct file_lock_context *flctx)
2555 : {
2556 : struct file_lock fl;
2557 0 : struct inode *inode = locks_inode(filp);
2558 :
2559 0 : if (list_empty(&flctx->flc_flock))
2560 0 : return;
2561 :
2562 0 : flock_make_lock(filp, LOCK_UN, &fl);
2563 0 : fl.fl_flags |= FL_CLOSE;
2564 :
2565 0 : if (filp->f_op->flock)
2566 0 : filp->f_op->flock(filp, F_SETLKW, &fl);
2567 : else
2568 0 : flock_lock_inode(inode, &fl);
2569 :
2570 0 : if (fl.fl_ops && fl.fl_ops->fl_release_private)
2571 0 : fl.fl_ops->fl_release_private(&fl);
2572 : }
2573 :
2574 : /* The i_flctx must be valid when calling into here */
2575 : static void
2576 0 : locks_remove_lease(struct file *filp, struct file_lock_context *ctx)
2577 : {
2578 : struct file_lock *fl, *tmp;
2579 0 : LIST_HEAD(dispose);
2580 :
2581 0 : if (list_empty(&ctx->flc_lease))
2582 0 : return;
2583 :
2584 0 : percpu_down_read(&file_rwsem);
2585 0 : spin_lock(&ctx->flc_lock);
2586 0 : list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list)
2587 0 : if (filp == fl->fl_file)
2588 0 : lease_modify(fl, F_UNLCK, &dispose);
2589 0 : spin_unlock(&ctx->flc_lock);
2590 0 : percpu_up_read(&file_rwsem);
2591 :
2592 0 : locks_dispose_list(&dispose);
2593 : }
2594 :
2595 : /*
2596 : * This function is called on the last close of an open file.
2597 : */
2598 0 : void locks_remove_file(struct file *filp)
2599 : {
2600 : struct file_lock_context *ctx;
2601 :
2602 0 : ctx = smp_load_acquire(&locks_inode(filp)->i_flctx);
2603 0 : if (!ctx)
2604 : return;
2605 :
2606 : /* remove any OFD locks */
2607 0 : locks_remove_posix(filp, filp);
2608 :
2609 : /* remove flock locks */
2610 0 : locks_remove_flock(filp, ctx);
2611 :
2612 : /* remove any leases */
2613 0 : locks_remove_lease(filp, ctx);
2614 :
2615 0 : spin_lock(&ctx->flc_lock);
2616 0 : locks_check_ctx_file_list(filp, &ctx->flc_posix, "POSIX");
2617 0 : locks_check_ctx_file_list(filp, &ctx->flc_flock, "FLOCK");
2618 0 : locks_check_ctx_file_list(filp, &ctx->flc_lease, "LEASE");
2619 0 : spin_unlock(&ctx->flc_lock);
2620 : }
2621 :
2622 : /**
2623 : * vfs_cancel_lock - file byte range unblock lock
2624 : * @filp: The file to apply the unblock to
2625 : * @fl: The lock to be unblocked
2626 : *
2627 : * Used by lock managers to cancel blocked requests
2628 : */
2629 0 : int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2630 : {
2631 0 : if (filp->f_op->lock)
2632 0 : return filp->f_op->lock(filp, F_CANCELLK, fl);
2633 : return 0;
2634 : }
2635 : EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2636 :
2637 : #ifdef CONFIG_PROC_FS
2638 : #include <linux/proc_fs.h>
2639 : #include <linux/seq_file.h>
2640 :
2641 : struct locks_iterator {
2642 : int li_cpu;
2643 : loff_t li_pos;
2644 : };
2645 :
2646 0 : static void lock_get_status(struct seq_file *f, struct file_lock *fl,
2647 : loff_t id, char *pfx, int repeat)
2648 : {
2649 0 : struct inode *inode = NULL;
2650 : unsigned int fl_pid;
2651 0 : struct pid_namespace *proc_pidns = proc_pid_ns(file_inode(f->file)->i_sb);
2652 : int type;
2653 :
2654 0 : fl_pid = locks_translate_pid(fl, proc_pidns);
2655 : /*
2656 : * If lock owner is dead (and pid is freed) or not visible in current
2657 : * pidns, zero is shown as a pid value. Check lock info from
2658 : * init_pid_ns to get saved lock pid value.
2659 : */
2660 :
2661 0 : if (fl->fl_file != NULL)
2662 0 : inode = locks_inode(fl->fl_file);
2663 :
2664 0 : seq_printf(f, "%lld: ", id);
2665 :
2666 0 : if (repeat)
2667 0 : seq_printf(f, "%*s", repeat - 1 + (int)strlen(pfx), pfx);
2668 :
2669 0 : if (IS_POSIX(fl)) {
2670 0 : if (fl->fl_flags & FL_ACCESS)
2671 0 : seq_puts(f, "ACCESS");
2672 0 : else if (IS_OFDLCK(fl))
2673 0 : seq_puts(f, "OFDLCK");
2674 : else
2675 0 : seq_puts(f, "POSIX ");
2676 :
2677 0 : seq_printf(f, " %s ",
2678 : (inode == NULL) ? "*NOINODE*" : "ADVISORY ");
2679 0 : } else if (IS_FLOCK(fl)) {
2680 0 : seq_puts(f, "FLOCK ADVISORY ");
2681 0 : } else if (IS_LEASE(fl)) {
2682 0 : if (fl->fl_flags & FL_DELEG)
2683 0 : seq_puts(f, "DELEG ");
2684 : else
2685 0 : seq_puts(f, "LEASE ");
2686 :
2687 0 : if (lease_breaking(fl))
2688 0 : seq_puts(f, "BREAKING ");
2689 0 : else if (fl->fl_file)
2690 0 : seq_puts(f, "ACTIVE ");
2691 : else
2692 0 : seq_puts(f, "BREAKER ");
2693 : } else {
2694 0 : seq_puts(f, "UNKNOWN UNKNOWN ");
2695 : }
2696 0 : type = IS_LEASE(fl) ? target_leasetype(fl) : fl->fl_type;
2697 :
2698 0 : seq_printf(f, "%s ", (type == F_WRLCK) ? "WRITE" :
2699 0 : (type == F_RDLCK) ? "READ" : "UNLCK");
2700 0 : if (inode) {
2701 : /* userspace relies on this representation of dev_t */
2702 0 : seq_printf(f, "%d %02x:%02x:%lu ", fl_pid,
2703 0 : MAJOR(inode->i_sb->s_dev),
2704 0 : MINOR(inode->i_sb->s_dev), inode->i_ino);
2705 : } else {
2706 0 : seq_printf(f, "%d <none>:0 ", fl_pid);
2707 : }
2708 0 : if (IS_POSIX(fl)) {
2709 0 : if (fl->fl_end == OFFSET_MAX)
2710 0 : seq_printf(f, "%Ld EOF\n", fl->fl_start);
2711 : else
2712 0 : seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
2713 : } else {
2714 0 : seq_puts(f, "0 EOF\n");
2715 : }
2716 0 : }
2717 :
2718 : static struct file_lock *get_next_blocked_member(struct file_lock *node)
2719 : {
2720 : struct file_lock *tmp;
2721 :
2722 : /* NULL node or root node */
2723 0 : if (node == NULL || node->fl_blocker == NULL)
2724 : return NULL;
2725 :
2726 : /* Next member in the linked list could be itself */
2727 0 : tmp = list_next_entry(node, fl_blocked_member);
2728 0 : if (list_entry_is_head(tmp, &node->fl_blocker->fl_blocked_requests, fl_blocked_member)
2729 0 : || tmp == node) {
2730 : return NULL;
2731 : }
2732 :
2733 : return tmp;
2734 : }
2735 :
2736 0 : static int locks_show(struct seq_file *f, void *v)
2737 : {
2738 0 : struct locks_iterator *iter = f->private;
2739 : struct file_lock *cur, *tmp;
2740 0 : struct pid_namespace *proc_pidns = proc_pid_ns(file_inode(f->file)->i_sb);
2741 0 : int level = 0;
2742 :
2743 0 : cur = hlist_entry(v, struct file_lock, fl_link);
2744 :
2745 0 : if (locks_translate_pid(cur, proc_pidns) == 0)
2746 : return 0;
2747 :
2748 : /* View this crossed linked list as a binary tree, the first member of fl_blocked_requests
2749 : * is the left child of current node, the next silibing in fl_blocked_member is the
2750 : * right child, we can alse get the parent of current node from fl_blocker, so this
2751 : * question becomes traversal of a binary tree
2752 : */
2753 0 : while (cur != NULL) {
2754 0 : if (level)
2755 0 : lock_get_status(f, cur, iter->li_pos, "-> ", level);
2756 : else
2757 0 : lock_get_status(f, cur, iter->li_pos, "", level);
2758 :
2759 0 : if (!list_empty(&cur->fl_blocked_requests)) {
2760 : /* Turn left */
2761 0 : cur = list_first_entry_or_null(&cur->fl_blocked_requests,
2762 : struct file_lock, fl_blocked_member);
2763 0 : level++;
2764 : } else {
2765 : /* Turn right */
2766 : tmp = get_next_blocked_member(cur);
2767 : /* Fall back to parent node */
2768 0 : while (tmp == NULL && cur->fl_blocker != NULL) {
2769 0 : cur = cur->fl_blocker;
2770 0 : level--;
2771 : tmp = get_next_blocked_member(cur);
2772 : }
2773 : cur = tmp;
2774 : }
2775 : }
2776 :
2777 : return 0;
2778 : }
2779 :
2780 0 : static void __show_fd_locks(struct seq_file *f,
2781 : struct list_head *head, int *id,
2782 : struct file *filp, struct files_struct *files)
2783 : {
2784 : struct file_lock *fl;
2785 :
2786 0 : list_for_each_entry(fl, head, fl_list) {
2787 :
2788 0 : if (filp != fl->fl_file)
2789 0 : continue;
2790 0 : if (fl->fl_owner != files &&
2791 : fl->fl_owner != filp)
2792 0 : continue;
2793 :
2794 0 : (*id)++;
2795 0 : seq_puts(f, "lock:\t");
2796 0 : lock_get_status(f, fl, *id, "", 0);
2797 : }
2798 0 : }
2799 :
2800 0 : void show_fd_locks(struct seq_file *f,
2801 : struct file *filp, struct files_struct *files)
2802 : {
2803 0 : struct inode *inode = locks_inode(filp);
2804 : struct file_lock_context *ctx;
2805 0 : int id = 0;
2806 :
2807 0 : ctx = smp_load_acquire(&inode->i_flctx);
2808 0 : if (!ctx)
2809 0 : return;
2810 :
2811 0 : spin_lock(&ctx->flc_lock);
2812 0 : __show_fd_locks(f, &ctx->flc_flock, &id, filp, files);
2813 0 : __show_fd_locks(f, &ctx->flc_posix, &id, filp, files);
2814 0 : __show_fd_locks(f, &ctx->flc_lease, &id, filp, files);
2815 0 : spin_unlock(&ctx->flc_lock);
2816 : }
2817 :
2818 0 : static void *locks_start(struct seq_file *f, loff_t *pos)
2819 : __acquires(&blocked_lock_lock)
2820 : {
2821 0 : struct locks_iterator *iter = f->private;
2822 :
2823 0 : iter->li_pos = *pos + 1;
2824 0 : percpu_down_write(&file_rwsem);
2825 0 : spin_lock(&blocked_lock_lock);
2826 0 : return seq_hlist_start_percpu(&file_lock_list.hlist, &iter->li_cpu, *pos);
2827 : }
2828 :
2829 0 : static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2830 : {
2831 0 : struct locks_iterator *iter = f->private;
2832 :
2833 0 : ++iter->li_pos;
2834 0 : return seq_hlist_next_percpu(v, &file_lock_list.hlist, &iter->li_cpu, pos);
2835 : }
2836 :
2837 0 : static void locks_stop(struct seq_file *f, void *v)
2838 : __releases(&blocked_lock_lock)
2839 : {
2840 0 : spin_unlock(&blocked_lock_lock);
2841 0 : percpu_up_write(&file_rwsem);
2842 0 : }
2843 :
2844 : static const struct seq_operations locks_seq_operations = {
2845 : .start = locks_start,
2846 : .next = locks_next,
2847 : .stop = locks_stop,
2848 : .show = locks_show,
2849 : };
2850 :
2851 1 : static int __init proc_locks_init(void)
2852 : {
2853 1 : proc_create_seq_private("locks", 0, NULL, &locks_seq_operations,
2854 : sizeof(struct locks_iterator), NULL);
2855 1 : return 0;
2856 : }
2857 : fs_initcall(proc_locks_init);
2858 : #endif
2859 :
2860 1 : static int __init filelock_init(void)
2861 : {
2862 : int i;
2863 :
2864 1 : flctx_cache = kmem_cache_create("file_lock_ctx",
2865 : sizeof(struct file_lock_context), 0, SLAB_PANIC, NULL);
2866 :
2867 1 : filelock_cache = kmem_cache_create("file_lock_cache",
2868 : sizeof(struct file_lock), 0, SLAB_PANIC, NULL);
2869 :
2870 2 : for_each_possible_cpu(i) {
2871 1 : struct file_lock_list_struct *fll = per_cpu_ptr(&file_lock_list, i);
2872 :
2873 1 : spin_lock_init(&fll->lock);
2874 1 : INIT_HLIST_HEAD(&fll->hlist);
2875 : }
2876 :
2877 : lease_notifier_chain_init();
2878 1 : return 0;
2879 : }
2880 : core_initcall(filelock_init);
|
