x86, PAT: Remove page granularity tracking for vm_insert_pfn maps
[linux-2.6/mini2440.git] / fs / locks.c
blobec3deea29e37cbd581488087e94053708dd100b1
1 /*
2 * linux/fs/locks.c
4 * Provide support for fcntl()'s F_GETLK, F_SETLK, and F_SETLKW calls.
5 * Doug Evans (dje@spiff.uucp), August 07, 1992
7 * Deadlock detection added.
8 * FIXME: one thing isn't handled yet:
9 * - mandatory locks (requires lots of changes elsewhere)
10 * Kelly Carmichael (kelly@[142.24.8.65]), September 17, 1994.
12 * Miscellaneous edits, and a total rewrite of posix_lock_file() code.
13 * Kai Petzke (wpp@marie.physik.tu-berlin.de), 1994
15 * Converted file_lock_table to a linked list from an array, which eliminates
16 * the limits on how many active file locks are open.
17 * Chad Page (pageone@netcom.com), November 27, 1994
19 * Removed dependency on file descriptors. dup()'ed file descriptors now
20 * get the same locks as the original file descriptors, and a close() on
21 * any file descriptor removes ALL the locks on the file for the current
22 * process. Since locks still depend on the process id, locks are inherited
23 * after an exec() but not after a fork(). This agrees with POSIX, and both
24 * BSD and SVR4 practice.
25 * Andy Walker (andy@lysaker.kvaerner.no), February 14, 1995
27 * Scrapped free list which is redundant now that we allocate locks
28 * dynamically with kmalloc()/kfree().
29 * Andy Walker (andy@lysaker.kvaerner.no), February 21, 1995
31 * Implemented two lock personalities - FL_FLOCK and FL_POSIX.
33 * FL_POSIX locks are created with calls to fcntl() and lockf() through the
34 * fcntl() system call. They have the semantics described above.
36 * FL_FLOCK locks are created with calls to flock(), through the flock()
37 * system call, which is new. Old C libraries implement flock() via fcntl()
38 * and will continue to use the old, broken implementation.
40 * FL_FLOCK locks follow the 4.4 BSD flock() semantics. They are associated
41 * with a file pointer (filp). As a result they can be shared by a parent
42 * process and its children after a fork(). They are removed when the last
43 * file descriptor referring to the file pointer is closed (unless explicitly
44 * unlocked).
46 * FL_FLOCK locks never deadlock, an existing lock is always removed before
47 * upgrading from shared to exclusive (or vice versa). When this happens
48 * any processes blocked by the current lock are woken up and allowed to
49 * run before the new lock is applied.
50 * Andy Walker (andy@lysaker.kvaerner.no), June 09, 1995
52 * Removed some race conditions in flock_lock_file(), marked other possible
53 * races. Just grep for FIXME to see them.
54 * Dmitry Gorodchanin (pgmdsg@ibi.com), February 09, 1996.
56 * Addressed Dmitry's concerns. Deadlock checking no longer recursive.
57 * Lock allocation changed to GFP_ATOMIC as we can't afford to sleep
58 * once we've checked for blocking and deadlocking.
59 * Andy Walker (andy@lysaker.kvaerner.no), April 03, 1996.
61 * Initial implementation of mandatory locks. SunOS turned out to be
62 * a rotten model, so I implemented the "obvious" semantics.
63 * See 'Documentation/mandatory.txt' for details.
64 * Andy Walker (andy@lysaker.kvaerner.no), April 06, 1996.
66 * Don't allow mandatory locks on mmap()'ed files. Added simple functions to
67 * check if a file has mandatory locks, used by mmap(), open() and creat() to
68 * see if system call should be rejected. Ref. HP-UX/SunOS/Solaris Reference
69 * Manual, Section 2.
70 * Andy Walker (andy@lysaker.kvaerner.no), April 09, 1996.
72 * Tidied up block list handling. Added '/proc/locks' interface.
73 * Andy Walker (andy@lysaker.kvaerner.no), April 24, 1996.
75 * Fixed deadlock condition for pathological code that mixes calls to
76 * flock() and fcntl().
77 * Andy Walker (andy@lysaker.kvaerner.no), April 29, 1996.
79 * Allow only one type of locking scheme (FL_POSIX or FL_FLOCK) to be in use
80 * for a given file at a time. Changed the CONFIG_LOCK_MANDATORY scheme to
81 * guarantee sensible behaviour in the case where file system modules might
82 * be compiled with different options than the kernel itself.
83 * Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996.
85 * Added a couple of missing wake_up() calls. Thanks to Thomas Meckel
86 * (Thomas.Meckel@mni.fh-giessen.de) for spotting this.
87 * Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996.
89 * Changed FL_POSIX locks to use the block list in the same way as FL_FLOCK
90 * locks. Changed process synchronisation to avoid dereferencing locks that
91 * have already been freed.
92 * Andy Walker (andy@lysaker.kvaerner.no), Sep 21, 1996.
94 * Made the block list a circular list to minimise searching in the list.
95 * Andy Walker (andy@lysaker.kvaerner.no), Sep 25, 1996.
97 * Made mandatory locking a mount option. Default is not to allow mandatory
98 * locking.
99 * Andy Walker (andy@lysaker.kvaerner.no), Oct 04, 1996.
101 * Some adaptations for NFS support.
102 * Olaf Kirch (okir@monad.swb.de), Dec 1996,
104 * Fixed /proc/locks interface so that we can't overrun the buffer we are handed.
105 * Andy Walker (andy@lysaker.kvaerner.no), May 12, 1997.
107 * Use slab allocator instead of kmalloc/kfree.
108 * Use generic list implementation from <linux/list.h>.
109 * Sped up posix_locks_deadlock by only considering blocked locks.
110 * Matthew Wilcox <willy@debian.org>, March, 2000.
112 * Leases and LOCK_MAND
113 * Matthew Wilcox <willy@debian.org>, June, 2000.
114 * Stephen Rothwell <sfr@canb.auug.org.au>, June, 2000.
117 #include <linux/capability.h>
118 #include <linux/file.h>
119 #include <linux/fdtable.h>
120 #include <linux/fs.h>
121 #include <linux/init.h>
122 #include <linux/module.h>
123 #include <linux/security.h>
124 #include <linux/slab.h>
125 #include <linux/smp_lock.h>
126 #include <linux/syscalls.h>
127 #include <linux/time.h>
128 #include <linux/rcupdate.h>
129 #include <linux/pid_namespace.h>
131 #include <asm/uaccess.h>
133 #define IS_POSIX(fl) (fl->fl_flags & FL_POSIX)
134 #define IS_FLOCK(fl) (fl->fl_flags & FL_FLOCK)
135 #define IS_LEASE(fl) (fl->fl_flags & FL_LEASE)
137 int leases_enable = 1;
138 int lease_break_time = 45;
140 #define for_each_lock(inode, lockp) \
141 for (lockp = &inode->i_flock; *lockp != NULL; lockp = &(*lockp)->fl_next)
143 static LIST_HEAD(file_lock_list);
144 static LIST_HEAD(blocked_list);
146 static struct kmem_cache *filelock_cache __read_mostly;
148 /* Allocate an empty lock structure. */
149 static struct file_lock *locks_alloc_lock(void)
151 return kmem_cache_alloc(filelock_cache, GFP_KERNEL);
154 static void locks_release_private(struct file_lock *fl)
156 if (fl->fl_ops) {
157 if (fl->fl_ops->fl_release_private)
158 fl->fl_ops->fl_release_private(fl);
159 fl->fl_ops = NULL;
161 if (fl->fl_lmops) {
162 if (fl->fl_lmops->fl_release_private)
163 fl->fl_lmops->fl_release_private(fl);
164 fl->fl_lmops = NULL;
169 /* Free a lock which is not in use. */
170 static void locks_free_lock(struct file_lock *fl)
172 BUG_ON(waitqueue_active(&fl->fl_wait));
173 BUG_ON(!list_empty(&fl->fl_block));
174 BUG_ON(!list_empty(&fl->fl_link));
176 locks_release_private(fl);
177 kmem_cache_free(filelock_cache, fl);
180 void locks_init_lock(struct file_lock *fl)
182 INIT_LIST_HEAD(&fl->fl_link);
183 INIT_LIST_HEAD(&fl->fl_block);
184 init_waitqueue_head(&fl->fl_wait);
185 fl->fl_next = NULL;
186 fl->fl_fasync = NULL;
187 fl->fl_owner = NULL;
188 fl->fl_pid = 0;
189 fl->fl_nspid = NULL;
190 fl->fl_file = NULL;
191 fl->fl_flags = 0;
192 fl->fl_type = 0;
193 fl->fl_start = fl->fl_end = 0;
194 fl->fl_ops = NULL;
195 fl->fl_lmops = NULL;
198 EXPORT_SYMBOL(locks_init_lock);
201 * Initialises the fields of the file lock which are invariant for
202 * free file_locks.
204 static void init_once(void *foo)
206 struct file_lock *lock = (struct file_lock *) foo;
208 locks_init_lock(lock);
211 static void locks_copy_private(struct file_lock *new, struct file_lock *fl)
213 if (fl->fl_ops) {
214 if (fl->fl_ops->fl_copy_lock)
215 fl->fl_ops->fl_copy_lock(new, fl);
216 new->fl_ops = fl->fl_ops;
218 if (fl->fl_lmops) {
219 if (fl->fl_lmops->fl_copy_lock)
220 fl->fl_lmops->fl_copy_lock(new, fl);
221 new->fl_lmops = fl->fl_lmops;
226 * Initialize a new lock from an existing file_lock structure.
228 void __locks_copy_lock(struct file_lock *new, const struct file_lock *fl)
230 new->fl_owner = fl->fl_owner;
231 new->fl_pid = fl->fl_pid;
232 new->fl_file = NULL;
233 new->fl_flags = fl->fl_flags;
234 new->fl_type = fl->fl_type;
235 new->fl_start = fl->fl_start;
236 new->fl_end = fl->fl_end;
237 new->fl_ops = NULL;
238 new->fl_lmops = NULL;
240 EXPORT_SYMBOL(__locks_copy_lock);
242 void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
244 locks_release_private(new);
246 __locks_copy_lock(new, fl);
247 new->fl_file = fl->fl_file;
248 new->fl_ops = fl->fl_ops;
249 new->fl_lmops = fl->fl_lmops;
251 locks_copy_private(new, fl);
254 EXPORT_SYMBOL(locks_copy_lock);
256 static inline int flock_translate_cmd(int cmd) {
257 if (cmd & LOCK_MAND)
258 return cmd & (LOCK_MAND | LOCK_RW);
259 switch (cmd) {
260 case LOCK_SH:
261 return F_RDLCK;
262 case LOCK_EX:
263 return F_WRLCK;
264 case LOCK_UN:
265 return F_UNLCK;
267 return -EINVAL;
270 /* Fill in a file_lock structure with an appropriate FLOCK lock. */
271 static int flock_make_lock(struct file *filp, struct file_lock **lock,
272 unsigned int cmd)
274 struct file_lock *fl;
275 int type = flock_translate_cmd(cmd);
276 if (type < 0)
277 return type;
279 fl = locks_alloc_lock();
280 if (fl == NULL)
281 return -ENOMEM;
283 fl->fl_file = filp;
284 fl->fl_pid = current->tgid;
285 fl->fl_flags = FL_FLOCK;
286 fl->fl_type = type;
287 fl->fl_end = OFFSET_MAX;
289 *lock = fl;
290 return 0;
293 static int assign_type(struct file_lock *fl, int type)
295 switch (type) {
296 case F_RDLCK:
297 case F_WRLCK:
298 case F_UNLCK:
299 fl->fl_type = type;
300 break;
301 default:
302 return -EINVAL;
304 return 0;
307 /* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
308 * style lock.
310 static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
311 struct flock *l)
313 off_t start, end;
315 switch (l->l_whence) {
316 case SEEK_SET:
317 start = 0;
318 break;
319 case SEEK_CUR:
320 start = filp->f_pos;
321 break;
322 case SEEK_END:
323 start = i_size_read(filp->f_path.dentry->d_inode);
324 break;
325 default:
326 return -EINVAL;
329 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
330 POSIX-2001 defines it. */
331 start += l->l_start;
332 if (start < 0)
333 return -EINVAL;
334 fl->fl_end = OFFSET_MAX;
335 if (l->l_len > 0) {
336 end = start + l->l_len - 1;
337 fl->fl_end = end;
338 } else if (l->l_len < 0) {
339 end = start - 1;
340 fl->fl_end = end;
341 start += l->l_len;
342 if (start < 0)
343 return -EINVAL;
345 fl->fl_start = start; /* we record the absolute position */
346 if (fl->fl_end < fl->fl_start)
347 return -EOVERFLOW;
349 fl->fl_owner = current->files;
350 fl->fl_pid = current->tgid;
351 fl->fl_file = filp;
352 fl->fl_flags = FL_POSIX;
353 fl->fl_ops = NULL;
354 fl->fl_lmops = NULL;
356 return assign_type(fl, l->l_type);
359 #if BITS_PER_LONG == 32
360 static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
361 struct flock64 *l)
363 loff_t start;
365 switch (l->l_whence) {
366 case SEEK_SET:
367 start = 0;
368 break;
369 case SEEK_CUR:
370 start = filp->f_pos;
371 break;
372 case SEEK_END:
373 start = i_size_read(filp->f_path.dentry->d_inode);
374 break;
375 default:
376 return -EINVAL;
379 start += l->l_start;
380 if (start < 0)
381 return -EINVAL;
382 fl->fl_end = OFFSET_MAX;
383 if (l->l_len > 0) {
384 fl->fl_end = start + l->l_len - 1;
385 } else if (l->l_len < 0) {
386 fl->fl_end = start - 1;
387 start += l->l_len;
388 if (start < 0)
389 return -EINVAL;
391 fl->fl_start = start; /* we record the absolute position */
392 if (fl->fl_end < fl->fl_start)
393 return -EOVERFLOW;
395 fl->fl_owner = current->files;
396 fl->fl_pid = current->tgid;
397 fl->fl_file = filp;
398 fl->fl_flags = FL_POSIX;
399 fl->fl_ops = NULL;
400 fl->fl_lmops = NULL;
402 switch (l->l_type) {
403 case F_RDLCK:
404 case F_WRLCK:
405 case F_UNLCK:
406 fl->fl_type = l->l_type;
407 break;
408 default:
409 return -EINVAL;
412 return (0);
414 #endif
416 /* default lease lock manager operations */
417 static void lease_break_callback(struct file_lock *fl)
419 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
422 static void lease_release_private_callback(struct file_lock *fl)
424 if (!fl->fl_file)
425 return;
427 f_delown(fl->fl_file);
428 fl->fl_file->f_owner.signum = 0;
431 static int lease_mylease_callback(struct file_lock *fl, struct file_lock *try)
433 return fl->fl_file == try->fl_file;
436 static struct lock_manager_operations lease_manager_ops = {
437 .fl_break = lease_break_callback,
438 .fl_release_private = lease_release_private_callback,
439 .fl_mylease = lease_mylease_callback,
440 .fl_change = lease_modify,
444 * Initialize a lease, use the default lock manager operations
446 static int lease_init(struct file *filp, int type, struct file_lock *fl)
448 if (assign_type(fl, type) != 0)
449 return -EINVAL;
451 fl->fl_owner = current->files;
452 fl->fl_pid = current->tgid;
454 fl->fl_file = filp;
455 fl->fl_flags = FL_LEASE;
456 fl->fl_start = 0;
457 fl->fl_end = OFFSET_MAX;
458 fl->fl_ops = NULL;
459 fl->fl_lmops = &lease_manager_ops;
460 return 0;
463 /* Allocate a file_lock initialised to this type of lease */
464 static struct file_lock *lease_alloc(struct file *filp, int type)
466 struct file_lock *fl = locks_alloc_lock();
467 int error = -ENOMEM;
469 if (fl == NULL)
470 return ERR_PTR(error);
472 error = lease_init(filp, type, fl);
473 if (error) {
474 locks_free_lock(fl);
475 return ERR_PTR(error);
477 return fl;
480 /* Check if two locks overlap each other.
482 static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
484 return ((fl1->fl_end >= fl2->fl_start) &&
485 (fl2->fl_end >= fl1->fl_start));
489 * Check whether two locks have the same owner.
491 static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
493 if (fl1->fl_lmops && fl1->fl_lmops->fl_compare_owner)
494 return fl2->fl_lmops == fl1->fl_lmops &&
495 fl1->fl_lmops->fl_compare_owner(fl1, fl2);
496 return fl1->fl_owner == fl2->fl_owner;
499 /* Remove waiter from blocker's block list.
500 * When blocker ends up pointing to itself then the list is empty.
502 static void __locks_delete_block(struct file_lock *waiter)
504 list_del_init(&waiter->fl_block);
505 list_del_init(&waiter->fl_link);
506 waiter->fl_next = NULL;
511 static void locks_delete_block(struct file_lock *waiter)
513 lock_kernel();
514 __locks_delete_block(waiter);
515 unlock_kernel();
518 /* Insert waiter into blocker's block list.
519 * We use a circular list so that processes can be easily woken up in
520 * the order they blocked. The documentation doesn't require this but
521 * it seems like the reasonable thing to do.
523 static void locks_insert_block(struct file_lock *blocker,
524 struct file_lock *waiter)
526 BUG_ON(!list_empty(&waiter->fl_block));
527 list_add_tail(&waiter->fl_block, &blocker->fl_block);
528 waiter->fl_next = blocker;
529 if (IS_POSIX(blocker))
530 list_add(&waiter->fl_link, &blocked_list);
533 /* Wake up processes blocked waiting for blocker.
534 * If told to wait then schedule the processes until the block list
535 * is empty, otherwise empty the block list ourselves.
537 static void locks_wake_up_blocks(struct file_lock *blocker)
539 while (!list_empty(&blocker->fl_block)) {
540 struct file_lock *waiter;
542 waiter = list_first_entry(&blocker->fl_block,
543 struct file_lock, fl_block);
544 __locks_delete_block(waiter);
545 if (waiter->fl_lmops && waiter->fl_lmops->fl_notify)
546 waiter->fl_lmops->fl_notify(waiter);
547 else
548 wake_up(&waiter->fl_wait);
552 /* Insert file lock fl into an inode's lock list at the position indicated
553 * by pos. At the same time add the lock to the global file lock list.
555 static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
557 list_add(&fl->fl_link, &file_lock_list);
559 fl->fl_nspid = get_pid(task_tgid(current));
561 /* insert into file's list */
562 fl->fl_next = *pos;
563 *pos = fl;
567 * Delete a lock and then free it.
568 * Wake up processes that are blocked waiting for this lock,
569 * notify the FS that the lock has been cleared and
570 * finally free the lock.
572 static void locks_delete_lock(struct file_lock **thisfl_p)
574 struct file_lock *fl = *thisfl_p;
576 *thisfl_p = fl->fl_next;
577 fl->fl_next = NULL;
578 list_del_init(&fl->fl_link);
580 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
581 if (fl->fl_fasync != NULL) {
582 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
583 fl->fl_fasync = NULL;
586 if (fl->fl_nspid) {
587 put_pid(fl->fl_nspid);
588 fl->fl_nspid = NULL;
591 locks_wake_up_blocks(fl);
592 locks_free_lock(fl);
595 /* Determine if lock sys_fl blocks lock caller_fl. Common functionality
596 * checks for shared/exclusive status of overlapping locks.
598 static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
600 if (sys_fl->fl_type == F_WRLCK)
601 return 1;
602 if (caller_fl->fl_type == F_WRLCK)
603 return 1;
604 return 0;
607 /* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
608 * checking before calling the locks_conflict().
610 static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
612 /* POSIX locks owned by the same process do not conflict with
613 * each other.
615 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
616 return (0);
618 /* Check whether they overlap */
619 if (!locks_overlap(caller_fl, sys_fl))
620 return 0;
622 return (locks_conflict(caller_fl, sys_fl));
625 /* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
626 * checking before calling the locks_conflict().
628 static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
630 /* FLOCK locks referring to the same filp do not conflict with
631 * each other.
633 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
634 return (0);
635 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
636 return 0;
638 return (locks_conflict(caller_fl, sys_fl));
641 void
642 posix_test_lock(struct file *filp, struct file_lock *fl)
644 struct file_lock *cfl;
646 lock_kernel();
647 for (cfl = filp->f_path.dentry->d_inode->i_flock; cfl; cfl = cfl->fl_next) {
648 if (!IS_POSIX(cfl))
649 continue;
650 if (posix_locks_conflict(fl, cfl))
651 break;
653 if (cfl) {
654 __locks_copy_lock(fl, cfl);
655 if (cfl->fl_nspid)
656 fl->fl_pid = pid_vnr(cfl->fl_nspid);
657 } else
658 fl->fl_type = F_UNLCK;
659 unlock_kernel();
660 return;
662 EXPORT_SYMBOL(posix_test_lock);
665 * Deadlock detection:
667 * We attempt to detect deadlocks that are due purely to posix file
668 * locks.
670 * We assume that a task can be waiting for at most one lock at a time.
671 * So for any acquired lock, the process holding that lock may be
672 * waiting on at most one other lock. That lock in turns may be held by
673 * someone waiting for at most one other lock. Given a requested lock
674 * caller_fl which is about to wait for a conflicting lock block_fl, we
675 * follow this chain of waiters to ensure we are not about to create a
676 * cycle.
678 * Since we do this before we ever put a process to sleep on a lock, we
679 * are ensured that there is never a cycle; that is what guarantees that
680 * the while() loop in posix_locks_deadlock() eventually completes.
682 * Note: the above assumption may not be true when handling lock
683 * requests from a broken NFS client. It may also fail in the presence
684 * of tasks (such as posix threads) sharing the same open file table.
686 * To handle those cases, we just bail out after a few iterations.
689 #define MAX_DEADLK_ITERATIONS 10
691 /* Find a lock that the owner of the given block_fl is blocking on. */
692 static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
694 struct file_lock *fl;
696 list_for_each_entry(fl, &blocked_list, fl_link) {
697 if (posix_same_owner(fl, block_fl))
698 return fl->fl_next;
700 return NULL;
703 static int posix_locks_deadlock(struct file_lock *caller_fl,
704 struct file_lock *block_fl)
706 int i = 0;
708 while ((block_fl = what_owner_is_waiting_for(block_fl))) {
709 if (i++ > MAX_DEADLK_ITERATIONS)
710 return 0;
711 if (posix_same_owner(caller_fl, block_fl))
712 return 1;
714 return 0;
717 /* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
718 * after any leases, but before any posix locks.
720 * Note that if called with an FL_EXISTS argument, the caller may determine
721 * whether or not a lock was successfully freed by testing the return
722 * value for -ENOENT.
724 static int flock_lock_file(struct file *filp, struct file_lock *request)
726 struct file_lock *new_fl = NULL;
727 struct file_lock **before;
728 struct inode * inode = filp->f_path.dentry->d_inode;
729 int error = 0;
730 int found = 0;
732 lock_kernel();
733 if (request->fl_flags & FL_ACCESS)
734 goto find_conflict;
736 if (request->fl_type != F_UNLCK) {
737 error = -ENOMEM;
738 new_fl = locks_alloc_lock();
739 if (new_fl == NULL)
740 goto out;
741 error = 0;
744 for_each_lock(inode, before) {
745 struct file_lock *fl = *before;
746 if (IS_POSIX(fl))
747 break;
748 if (IS_LEASE(fl))
749 continue;
750 if (filp != fl->fl_file)
751 continue;
752 if (request->fl_type == fl->fl_type)
753 goto out;
754 found = 1;
755 locks_delete_lock(before);
756 break;
759 if (request->fl_type == F_UNLCK) {
760 if ((request->fl_flags & FL_EXISTS) && !found)
761 error = -ENOENT;
762 goto out;
766 * If a higher-priority process was blocked on the old file lock,
767 * give it the opportunity to lock the file.
769 if (found)
770 cond_resched_bkl();
772 find_conflict:
773 for_each_lock(inode, before) {
774 struct file_lock *fl = *before;
775 if (IS_POSIX(fl))
776 break;
777 if (IS_LEASE(fl))
778 continue;
779 if (!flock_locks_conflict(request, fl))
780 continue;
781 error = -EAGAIN;
782 if (!(request->fl_flags & FL_SLEEP))
783 goto out;
784 error = FILE_LOCK_DEFERRED;
785 locks_insert_block(fl, request);
786 goto out;
788 if (request->fl_flags & FL_ACCESS)
789 goto out;
790 locks_copy_lock(new_fl, request);
791 locks_insert_lock(before, new_fl);
792 new_fl = NULL;
793 error = 0;
795 out:
796 unlock_kernel();
797 if (new_fl)
798 locks_free_lock(new_fl);
799 return error;
802 static int __posix_lock_file(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
804 struct file_lock *fl;
805 struct file_lock *new_fl = NULL;
806 struct file_lock *new_fl2 = NULL;
807 struct file_lock *left = NULL;
808 struct file_lock *right = NULL;
809 struct file_lock **before;
810 int error, added = 0;
813 * We may need two file_lock structures for this operation,
814 * so we get them in advance to avoid races.
816 * In some cases we can be sure, that no new locks will be needed
818 if (!(request->fl_flags & FL_ACCESS) &&
819 (request->fl_type != F_UNLCK ||
820 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
821 new_fl = locks_alloc_lock();
822 new_fl2 = locks_alloc_lock();
825 lock_kernel();
826 if (request->fl_type != F_UNLCK) {
827 for_each_lock(inode, before) {
828 fl = *before;
829 if (!IS_POSIX(fl))
830 continue;
831 if (!posix_locks_conflict(request, fl))
832 continue;
833 if (conflock)
834 __locks_copy_lock(conflock, fl);
835 error = -EAGAIN;
836 if (!(request->fl_flags & FL_SLEEP))
837 goto out;
838 error = -EDEADLK;
839 if (posix_locks_deadlock(request, fl))
840 goto out;
841 error = FILE_LOCK_DEFERRED;
842 locks_insert_block(fl, request);
843 goto out;
847 /* If we're just looking for a conflict, we're done. */
848 error = 0;
849 if (request->fl_flags & FL_ACCESS)
850 goto out;
853 * Find the first old lock with the same owner as the new lock.
856 before = &inode->i_flock;
858 /* First skip locks owned by other processes. */
859 while ((fl = *before) && (!IS_POSIX(fl) ||
860 !posix_same_owner(request, fl))) {
861 before = &fl->fl_next;
864 /* Process locks with this owner. */
865 while ((fl = *before) && posix_same_owner(request, fl)) {
866 /* Detect adjacent or overlapping regions (if same lock type)
868 if (request->fl_type == fl->fl_type) {
869 /* In all comparisons of start vs end, use
870 * "start - 1" rather than "end + 1". If end
871 * is OFFSET_MAX, end + 1 will become negative.
873 if (fl->fl_end < request->fl_start - 1)
874 goto next_lock;
875 /* If the next lock in the list has entirely bigger
876 * addresses than the new one, insert the lock here.
878 if (fl->fl_start - 1 > request->fl_end)
879 break;
881 /* If we come here, the new and old lock are of the
882 * same type and adjacent or overlapping. Make one
883 * lock yielding from the lower start address of both
884 * locks to the higher end address.
886 if (fl->fl_start > request->fl_start)
887 fl->fl_start = request->fl_start;
888 else
889 request->fl_start = fl->fl_start;
890 if (fl->fl_end < request->fl_end)
891 fl->fl_end = request->fl_end;
892 else
893 request->fl_end = fl->fl_end;
894 if (added) {
895 locks_delete_lock(before);
896 continue;
898 request = fl;
899 added = 1;
901 else {
902 /* Processing for different lock types is a bit
903 * more complex.
905 if (fl->fl_end < request->fl_start)
906 goto next_lock;
907 if (fl->fl_start > request->fl_end)
908 break;
909 if (request->fl_type == F_UNLCK)
910 added = 1;
911 if (fl->fl_start < request->fl_start)
912 left = fl;
913 /* If the next lock in the list has a higher end
914 * address than the new one, insert the new one here.
916 if (fl->fl_end > request->fl_end) {
917 right = fl;
918 break;
920 if (fl->fl_start >= request->fl_start) {
921 /* The new lock completely replaces an old
922 * one (This may happen several times).
924 if (added) {
925 locks_delete_lock(before);
926 continue;
928 /* Replace the old lock with the new one.
929 * Wake up anybody waiting for the old one,
930 * as the change in lock type might satisfy
931 * their needs.
933 locks_wake_up_blocks(fl);
934 fl->fl_start = request->fl_start;
935 fl->fl_end = request->fl_end;
936 fl->fl_type = request->fl_type;
937 locks_release_private(fl);
938 locks_copy_private(fl, request);
939 request = fl;
940 added = 1;
943 /* Go on to next lock.
945 next_lock:
946 before = &fl->fl_next;
950 * The above code only modifies existing locks in case of
951 * merging or replacing. If new lock(s) need to be inserted
952 * all modifications are done bellow this, so it's safe yet to
953 * bail out.
955 error = -ENOLCK; /* "no luck" */
956 if (right && left == right && !new_fl2)
957 goto out;
959 error = 0;
960 if (!added) {
961 if (request->fl_type == F_UNLCK) {
962 if (request->fl_flags & FL_EXISTS)
963 error = -ENOENT;
964 goto out;
967 if (!new_fl) {
968 error = -ENOLCK;
969 goto out;
971 locks_copy_lock(new_fl, request);
972 locks_insert_lock(before, new_fl);
973 new_fl = NULL;
975 if (right) {
976 if (left == right) {
977 /* The new lock breaks the old one in two pieces,
978 * so we have to use the second new lock.
980 left = new_fl2;
981 new_fl2 = NULL;
982 locks_copy_lock(left, right);
983 locks_insert_lock(before, left);
985 right->fl_start = request->fl_end + 1;
986 locks_wake_up_blocks(right);
988 if (left) {
989 left->fl_end = request->fl_start - 1;
990 locks_wake_up_blocks(left);
992 out:
993 unlock_kernel();
995 * Free any unused locks.
997 if (new_fl)
998 locks_free_lock(new_fl);
999 if (new_fl2)
1000 locks_free_lock(new_fl2);
1001 return error;
1005 * posix_lock_file - Apply a POSIX-style lock to a file
1006 * @filp: The file to apply the lock to
1007 * @fl: The lock to be applied
1008 * @conflock: Place to return a copy of the conflicting lock, if found.
1010 * Add a POSIX style lock to a file.
1011 * We merge adjacent & overlapping locks whenever possible.
1012 * POSIX locks are sorted by owner task, then by starting address
1014 * Note that if called with an FL_EXISTS argument, the caller may determine
1015 * whether or not a lock was successfully freed by testing the return
1016 * value for -ENOENT.
1018 int posix_lock_file(struct file *filp, struct file_lock *fl,
1019 struct file_lock *conflock)
1021 return __posix_lock_file(filp->f_path.dentry->d_inode, fl, conflock);
1023 EXPORT_SYMBOL(posix_lock_file);
1026 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1027 * @filp: The file to apply the lock to
1028 * @fl: The lock to be applied
1030 * Add a POSIX style lock to a file.
1031 * We merge adjacent & overlapping locks whenever possible.
1032 * POSIX locks are sorted by owner task, then by starting address
1034 int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1036 int error;
1037 might_sleep ();
1038 for (;;) {
1039 error = posix_lock_file(filp, fl, NULL);
1040 if (error != FILE_LOCK_DEFERRED)
1041 break;
1042 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1043 if (!error)
1044 continue;
1046 locks_delete_block(fl);
1047 break;
1049 return error;
1051 EXPORT_SYMBOL(posix_lock_file_wait);
1054 * locks_mandatory_locked - Check for an active lock
1055 * @inode: the file to check
1057 * Searches the inode's list of locks to find any POSIX locks which conflict.
1058 * This function is called from locks_verify_locked() only.
1060 int locks_mandatory_locked(struct inode *inode)
1062 fl_owner_t owner = current->files;
1063 struct file_lock *fl;
1066 * Search the lock list for this inode for any POSIX locks.
1068 lock_kernel();
1069 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1070 if (!IS_POSIX(fl))
1071 continue;
1072 if (fl->fl_owner != owner)
1073 break;
1075 unlock_kernel();
1076 return fl ? -EAGAIN : 0;
1080 * locks_mandatory_area - Check for a conflicting lock
1081 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1082 * for shared
1083 * @inode: the file to check
1084 * @filp: how the file was opened (if it was)
1085 * @offset: start of area to check
1086 * @count: length of area to check
1088 * Searches the inode's list of locks to find any POSIX locks which conflict.
1089 * This function is called from rw_verify_area() and
1090 * locks_verify_truncate().
1092 int locks_mandatory_area(int read_write, struct inode *inode,
1093 struct file *filp, loff_t offset,
1094 size_t count)
1096 struct file_lock fl;
1097 int error;
1099 locks_init_lock(&fl);
1100 fl.fl_owner = current->files;
1101 fl.fl_pid = current->tgid;
1102 fl.fl_file = filp;
1103 fl.fl_flags = FL_POSIX | FL_ACCESS;
1104 if (filp && !(filp->f_flags & O_NONBLOCK))
1105 fl.fl_flags |= FL_SLEEP;
1106 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1107 fl.fl_start = offset;
1108 fl.fl_end = offset + count - 1;
1110 for (;;) {
1111 error = __posix_lock_file(inode, &fl, NULL);
1112 if (error != FILE_LOCK_DEFERRED)
1113 break;
1114 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1115 if (!error) {
1117 * If we've been sleeping someone might have
1118 * changed the permissions behind our back.
1120 if (__mandatory_lock(inode))
1121 continue;
1124 locks_delete_block(&fl);
1125 break;
1128 return error;
1131 EXPORT_SYMBOL(locks_mandatory_area);
1133 /* We already had a lease on this file; just change its type */
1134 int lease_modify(struct file_lock **before, int arg)
1136 struct file_lock *fl = *before;
1137 int error = assign_type(fl, arg);
1139 if (error)
1140 return error;
1141 locks_wake_up_blocks(fl);
1142 if (arg == F_UNLCK)
1143 locks_delete_lock(before);
1144 return 0;
1147 EXPORT_SYMBOL(lease_modify);
1149 static void time_out_leases(struct inode *inode)
1151 struct file_lock **before;
1152 struct file_lock *fl;
1154 before = &inode->i_flock;
1155 while ((fl = *before) && IS_LEASE(fl) && (fl->fl_type & F_INPROGRESS)) {
1156 if ((fl->fl_break_time == 0)
1157 || time_before(jiffies, fl->fl_break_time)) {
1158 before = &fl->fl_next;
1159 continue;
1161 lease_modify(before, fl->fl_type & ~F_INPROGRESS);
1162 if (fl == *before) /* lease_modify may have freed fl */
1163 before = &fl->fl_next;
1168 * __break_lease - revoke all outstanding leases on file
1169 * @inode: the inode of the file to return
1170 * @mode: the open mode (read or write)
1172 * break_lease (inlined for speed) has checked there already is at least
1173 * some kind of lock (maybe a lease) on this file. Leases are broken on
1174 * a call to open() or truncate(). This function can sleep unless you
1175 * specified %O_NONBLOCK to your open().
1177 int __break_lease(struct inode *inode, unsigned int mode)
1179 int error = 0, future;
1180 struct file_lock *new_fl, *flock;
1181 struct file_lock *fl;
1182 unsigned long break_time;
1183 int i_have_this_lease = 0;
1185 new_fl = lease_alloc(NULL, mode & FMODE_WRITE ? F_WRLCK : F_RDLCK);
1187 lock_kernel();
1189 time_out_leases(inode);
1191 flock = inode->i_flock;
1192 if ((flock == NULL) || !IS_LEASE(flock))
1193 goto out;
1195 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1196 if (fl->fl_owner == current->files)
1197 i_have_this_lease = 1;
1199 if (mode & FMODE_WRITE) {
1200 /* If we want write access, we have to revoke any lease. */
1201 future = F_UNLCK | F_INPROGRESS;
1202 } else if (flock->fl_type & F_INPROGRESS) {
1203 /* If the lease is already being broken, we just leave it */
1204 future = flock->fl_type;
1205 } else if (flock->fl_type & F_WRLCK) {
1206 /* Downgrade the exclusive lease to a read-only lease. */
1207 future = F_RDLCK | F_INPROGRESS;
1208 } else {
1209 /* the existing lease was read-only, so we can read too. */
1210 goto out;
1213 if (IS_ERR(new_fl) && !i_have_this_lease
1214 && ((mode & O_NONBLOCK) == 0)) {
1215 error = PTR_ERR(new_fl);
1216 goto out;
1219 break_time = 0;
1220 if (lease_break_time > 0) {
1221 break_time = jiffies + lease_break_time * HZ;
1222 if (break_time == 0)
1223 break_time++; /* so that 0 means no break time */
1226 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1227 if (fl->fl_type != future) {
1228 fl->fl_type = future;
1229 fl->fl_break_time = break_time;
1230 /* lease must have lmops break callback */
1231 fl->fl_lmops->fl_break(fl);
1235 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1236 error = -EWOULDBLOCK;
1237 goto out;
1240 restart:
1241 break_time = flock->fl_break_time;
1242 if (break_time != 0) {
1243 break_time -= jiffies;
1244 if (break_time == 0)
1245 break_time++;
1247 locks_insert_block(flock, new_fl);
1248 error = wait_event_interruptible_timeout(new_fl->fl_wait,
1249 !new_fl->fl_next, break_time);
1250 __locks_delete_block(new_fl);
1251 if (error >= 0) {
1252 if (error == 0)
1253 time_out_leases(inode);
1254 /* Wait for the next lease that has not been broken yet */
1255 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1256 flock = flock->fl_next) {
1257 if (flock->fl_type & F_INPROGRESS)
1258 goto restart;
1260 error = 0;
1263 out:
1264 unlock_kernel();
1265 if (!IS_ERR(new_fl))
1266 locks_free_lock(new_fl);
1267 return error;
1270 EXPORT_SYMBOL(__break_lease);
1273 * lease_get_mtime - get the last modified time of an inode
1274 * @inode: the inode
1275 * @time: pointer to a timespec which will contain the last modified time
1277 * This is to force NFS clients to flush their caches for files with
1278 * exclusive leases. The justification is that if someone has an
1279 * exclusive lease, then they could be modifying it.
1281 void lease_get_mtime(struct inode *inode, struct timespec *time)
1283 struct file_lock *flock = inode->i_flock;
1284 if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK))
1285 *time = current_fs_time(inode->i_sb);
1286 else
1287 *time = inode->i_mtime;
1290 EXPORT_SYMBOL(lease_get_mtime);
1293 * fcntl_getlease - Enquire what lease is currently active
1294 * @filp: the file
1296 * The value returned by this function will be one of
1297 * (if no lease break is pending):
1299 * %F_RDLCK to indicate a shared lease is held.
1301 * %F_WRLCK to indicate an exclusive lease is held.
1303 * %F_UNLCK to indicate no lease is held.
1305 * (if a lease break is pending):
1307 * %F_RDLCK to indicate an exclusive lease needs to be
1308 * changed to a shared lease (or removed).
1310 * %F_UNLCK to indicate the lease needs to be removed.
1312 * XXX: sfr & willy disagree over whether F_INPROGRESS
1313 * should be returned to userspace.
1315 int fcntl_getlease(struct file *filp)
1317 struct file_lock *fl;
1318 int type = F_UNLCK;
1320 lock_kernel();
1321 time_out_leases(filp->f_path.dentry->d_inode);
1322 for (fl = filp->f_path.dentry->d_inode->i_flock; fl && IS_LEASE(fl);
1323 fl = fl->fl_next) {
1324 if (fl->fl_file == filp) {
1325 type = fl->fl_type & ~F_INPROGRESS;
1326 break;
1329 unlock_kernel();
1330 return type;
1334 * generic_setlease - sets a lease on an open file
1335 * @filp: file pointer
1336 * @arg: type of lease to obtain
1337 * @flp: input - file_lock to use, output - file_lock inserted
1339 * The (input) flp->fl_lmops->fl_break function is required
1340 * by break_lease().
1342 * Called with kernel lock held.
1344 int generic_setlease(struct file *filp, long arg, struct file_lock **flp)
1346 struct file_lock *fl, **before, **my_before = NULL, *lease;
1347 struct file_lock *new_fl = NULL;
1348 struct dentry *dentry = filp->f_path.dentry;
1349 struct inode *inode = dentry->d_inode;
1350 int error, rdlease_count = 0, wrlease_count = 0;
1352 if ((current_fsuid() != inode->i_uid) && !capable(CAP_LEASE))
1353 return -EACCES;
1354 if (!S_ISREG(inode->i_mode))
1355 return -EINVAL;
1356 error = security_file_lock(filp, arg);
1357 if (error)
1358 return error;
1360 time_out_leases(inode);
1362 BUG_ON(!(*flp)->fl_lmops->fl_break);
1364 lease = *flp;
1366 if (arg != F_UNLCK) {
1367 error = -ENOMEM;
1368 new_fl = locks_alloc_lock();
1369 if (new_fl == NULL)
1370 goto out;
1372 error = -EAGAIN;
1373 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1374 goto out;
1375 if ((arg == F_WRLCK)
1376 && ((atomic_read(&dentry->d_count) > 1)
1377 || (atomic_read(&inode->i_count) > 1)))
1378 goto out;
1382 * At this point, we know that if there is an exclusive
1383 * lease on this file, then we hold it on this filp
1384 * (otherwise our open of this file would have blocked).
1385 * And if we are trying to acquire an exclusive lease,
1386 * then the file is not open by anyone (including us)
1387 * except for this filp.
1389 for (before = &inode->i_flock;
1390 ((fl = *before) != NULL) && IS_LEASE(fl);
1391 before = &fl->fl_next) {
1392 if (lease->fl_lmops->fl_mylease(fl, lease))
1393 my_before = before;
1394 else if (fl->fl_type == (F_INPROGRESS | F_UNLCK))
1396 * Someone is in the process of opening this
1397 * file for writing so we may not take an
1398 * exclusive lease on it.
1400 wrlease_count++;
1401 else
1402 rdlease_count++;
1405 error = -EAGAIN;
1406 if ((arg == F_RDLCK && (wrlease_count > 0)) ||
1407 (arg == F_WRLCK && ((rdlease_count + wrlease_count) > 0)))
1408 goto out;
1410 if (my_before != NULL) {
1411 *flp = *my_before;
1412 error = lease->fl_lmops->fl_change(my_before, arg);
1413 goto out;
1416 error = 0;
1417 if (arg == F_UNLCK)
1418 goto out;
1420 error = -EINVAL;
1421 if (!leases_enable)
1422 goto out;
1424 locks_copy_lock(new_fl, lease);
1425 locks_insert_lock(before, new_fl);
1427 *flp = new_fl;
1428 return 0;
1430 out:
1431 if (new_fl != NULL)
1432 locks_free_lock(new_fl);
1433 return error;
1435 EXPORT_SYMBOL(generic_setlease);
1438 * vfs_setlease - sets a lease on an open file
1439 * @filp: file pointer
1440 * @arg: type of lease to obtain
1441 * @lease: file_lock to use
1443 * Call this to establish a lease on the file.
1444 * The (*lease)->fl_lmops->fl_break operation must be set; if not,
1445 * break_lease will oops!
1447 * This will call the filesystem's setlease file method, if
1448 * defined. Note that there is no getlease method; instead, the
1449 * filesystem setlease method should call back to setlease() to
1450 * add a lease to the inode's lease list, where fcntl_getlease() can
1451 * find it. Since fcntl_getlease() only reports whether the current
1452 * task holds a lease, a cluster filesystem need only do this for
1453 * leases held by processes on this node.
1455 * There is also no break_lease method; filesystems that
1456 * handle their own leases shoud break leases themselves from the
1457 * filesystem's open, create, and (on truncate) setattr methods.
1459 * Warning: the only current setlease methods exist only to disable
1460 * leases in certain cases. More vfs changes may be required to
1461 * allow a full filesystem lease implementation.
1464 int vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
1466 int error;
1468 lock_kernel();
1469 if (filp->f_op && filp->f_op->setlease)
1470 error = filp->f_op->setlease(filp, arg, lease);
1471 else
1472 error = generic_setlease(filp, arg, lease);
1473 unlock_kernel();
1475 return error;
1477 EXPORT_SYMBOL_GPL(vfs_setlease);
1480 * fcntl_setlease - sets a lease on an open file
1481 * @fd: open file descriptor
1482 * @filp: file pointer
1483 * @arg: type of lease to obtain
1485 * Call this fcntl to establish a lease on the file.
1486 * Note that you also need to call %F_SETSIG to
1487 * receive a signal when the lease is broken.
1489 int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1491 struct file_lock fl, *flp = &fl;
1492 struct inode *inode = filp->f_path.dentry->d_inode;
1493 int error;
1495 locks_init_lock(&fl);
1496 error = lease_init(filp, arg, &fl);
1497 if (error)
1498 return error;
1500 lock_kernel();
1502 error = vfs_setlease(filp, arg, &flp);
1503 if (error || arg == F_UNLCK)
1504 goto out_unlock;
1506 error = fasync_helper(fd, filp, 1, &flp->fl_fasync);
1507 if (error < 0) {
1508 /* remove lease just inserted by setlease */
1509 flp->fl_type = F_UNLCK | F_INPROGRESS;
1510 flp->fl_break_time = jiffies - 10;
1511 time_out_leases(inode);
1512 goto out_unlock;
1515 error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
1516 out_unlock:
1517 unlock_kernel();
1518 return error;
1522 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1523 * @filp: The file to apply the lock to
1524 * @fl: The lock to be applied
1526 * Add a FLOCK style lock to a file.
1528 int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1530 int error;
1531 might_sleep();
1532 for (;;) {
1533 error = flock_lock_file(filp, fl);
1534 if (error != FILE_LOCK_DEFERRED)
1535 break;
1536 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1537 if (!error)
1538 continue;
1540 locks_delete_block(fl);
1541 break;
1543 return error;
1546 EXPORT_SYMBOL(flock_lock_file_wait);
1549 * sys_flock: - flock() system call.
1550 * @fd: the file descriptor to lock.
1551 * @cmd: the type of lock to apply.
1553 * Apply a %FL_FLOCK style lock to an open file descriptor.
1554 * The @cmd can be one of
1556 * %LOCK_SH -- a shared lock.
1558 * %LOCK_EX -- an exclusive lock.
1560 * %LOCK_UN -- remove an existing lock.
1562 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1564 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1565 * processes read and write access respectively.
1567 SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
1569 struct file *filp;
1570 struct file_lock *lock;
1571 int can_sleep, unlock;
1572 int error;
1574 error = -EBADF;
1575 filp = fget(fd);
1576 if (!filp)
1577 goto out;
1579 can_sleep = !(cmd & LOCK_NB);
1580 cmd &= ~LOCK_NB;
1581 unlock = (cmd == LOCK_UN);
1583 if (!unlock && !(cmd & LOCK_MAND) &&
1584 !(filp->f_mode & (FMODE_READ|FMODE_WRITE)))
1585 goto out_putf;
1587 error = flock_make_lock(filp, &lock, cmd);
1588 if (error)
1589 goto out_putf;
1590 if (can_sleep)
1591 lock->fl_flags |= FL_SLEEP;
1593 error = security_file_lock(filp, cmd);
1594 if (error)
1595 goto out_free;
1597 if (filp->f_op && filp->f_op->flock)
1598 error = filp->f_op->flock(filp,
1599 (can_sleep) ? F_SETLKW : F_SETLK,
1600 lock);
1601 else
1602 error = flock_lock_file_wait(filp, lock);
1604 out_free:
1605 locks_free_lock(lock);
1607 out_putf:
1608 fput(filp);
1609 out:
1610 return error;
1614 * vfs_test_lock - test file byte range lock
1615 * @filp: The file to test lock for
1616 * @fl: The lock to test; also used to hold result
1618 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
1619 * setting conf->fl_type to something other than F_UNLCK.
1621 int vfs_test_lock(struct file *filp, struct file_lock *fl)
1623 if (filp->f_op && filp->f_op->lock)
1624 return filp->f_op->lock(filp, F_GETLK, fl);
1625 posix_test_lock(filp, fl);
1626 return 0;
1628 EXPORT_SYMBOL_GPL(vfs_test_lock);
1630 static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
1632 flock->l_pid = fl->fl_pid;
1633 #if BITS_PER_LONG == 32
1635 * Make sure we can represent the posix lock via
1636 * legacy 32bit flock.
1638 if (fl->fl_start > OFFT_OFFSET_MAX)
1639 return -EOVERFLOW;
1640 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
1641 return -EOVERFLOW;
1642 #endif
1643 flock->l_start = fl->fl_start;
1644 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1645 fl->fl_end - fl->fl_start + 1;
1646 flock->l_whence = 0;
1647 flock->l_type = fl->fl_type;
1648 return 0;
1651 #if BITS_PER_LONG == 32
1652 static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
1654 flock->l_pid = fl->fl_pid;
1655 flock->l_start = fl->fl_start;
1656 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1657 fl->fl_end - fl->fl_start + 1;
1658 flock->l_whence = 0;
1659 flock->l_type = fl->fl_type;
1661 #endif
1663 /* Report the first existing lock that would conflict with l.
1664 * This implements the F_GETLK command of fcntl().
1666 int fcntl_getlk(struct file *filp, struct flock __user *l)
1668 struct file_lock file_lock;
1669 struct flock flock;
1670 int error;
1672 error = -EFAULT;
1673 if (copy_from_user(&flock, l, sizeof(flock)))
1674 goto out;
1675 error = -EINVAL;
1676 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1677 goto out;
1679 error = flock_to_posix_lock(filp, &file_lock, &flock);
1680 if (error)
1681 goto out;
1683 error = vfs_test_lock(filp, &file_lock);
1684 if (error)
1685 goto out;
1687 flock.l_type = file_lock.fl_type;
1688 if (file_lock.fl_type != F_UNLCK) {
1689 error = posix_lock_to_flock(&flock, &file_lock);
1690 if (error)
1691 goto out;
1693 error = -EFAULT;
1694 if (!copy_to_user(l, &flock, sizeof(flock)))
1695 error = 0;
1696 out:
1697 return error;
1701 * vfs_lock_file - file byte range lock
1702 * @filp: The file to apply the lock to
1703 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
1704 * @fl: The lock to be applied
1705 * @conf: Place to return a copy of the conflicting lock, if found.
1707 * A caller that doesn't care about the conflicting lock may pass NULL
1708 * as the final argument.
1710 * If the filesystem defines a private ->lock() method, then @conf will
1711 * be left unchanged; so a caller that cares should initialize it to
1712 * some acceptable default.
1714 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
1715 * locks, the ->lock() interface may return asynchronously, before the lock has
1716 * been granted or denied by the underlying filesystem, if (and only if)
1717 * fl_grant is set. Callers expecting ->lock() to return asynchronously
1718 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
1719 * the request is for a blocking lock. When ->lock() does return asynchronously,
1720 * it must return FILE_LOCK_DEFERRED, and call ->fl_grant() when the lock
1721 * request completes.
1722 * If the request is for non-blocking lock the file system should return
1723 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
1724 * with the result. If the request timed out the callback routine will return a
1725 * nonzero return code and the file system should release the lock. The file
1726 * system is also responsible to keep a corresponding posix lock when it
1727 * grants a lock so the VFS can find out which locks are locally held and do
1728 * the correct lock cleanup when required.
1729 * The underlying filesystem must not drop the kernel lock or call
1730 * ->fl_grant() before returning to the caller with a FILE_LOCK_DEFERRED
1731 * return code.
1733 int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
1735 if (filp->f_op && filp->f_op->lock)
1736 return filp->f_op->lock(filp, cmd, fl);
1737 else
1738 return posix_lock_file(filp, fl, conf);
1740 EXPORT_SYMBOL_GPL(vfs_lock_file);
1742 static int do_lock_file_wait(struct file *filp, unsigned int cmd,
1743 struct file_lock *fl)
1745 int error;
1747 error = security_file_lock(filp, fl->fl_type);
1748 if (error)
1749 return error;
1751 for (;;) {
1752 error = vfs_lock_file(filp, cmd, fl, NULL);
1753 if (error != FILE_LOCK_DEFERRED)
1754 break;
1755 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1756 if (!error)
1757 continue;
1759 locks_delete_block(fl);
1760 break;
1763 return error;
1766 /* Apply the lock described by l to an open file descriptor.
1767 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1769 int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1770 struct flock __user *l)
1772 struct file_lock *file_lock = locks_alloc_lock();
1773 struct flock flock;
1774 struct inode *inode;
1775 struct file *f;
1776 int error;
1778 if (file_lock == NULL)
1779 return -ENOLCK;
1782 * This might block, so we do it before checking the inode.
1784 error = -EFAULT;
1785 if (copy_from_user(&flock, l, sizeof(flock)))
1786 goto out;
1788 inode = filp->f_path.dentry->d_inode;
1790 /* Don't allow mandatory locks on files that may be memory mapped
1791 * and shared.
1793 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
1794 error = -EAGAIN;
1795 goto out;
1798 again:
1799 error = flock_to_posix_lock(filp, file_lock, &flock);
1800 if (error)
1801 goto out;
1802 if (cmd == F_SETLKW) {
1803 file_lock->fl_flags |= FL_SLEEP;
1806 error = -EBADF;
1807 switch (flock.l_type) {
1808 case F_RDLCK:
1809 if (!(filp->f_mode & FMODE_READ))
1810 goto out;
1811 break;
1812 case F_WRLCK:
1813 if (!(filp->f_mode & FMODE_WRITE))
1814 goto out;
1815 break;
1816 case F_UNLCK:
1817 break;
1818 default:
1819 error = -EINVAL;
1820 goto out;
1823 error = do_lock_file_wait(filp, cmd, file_lock);
1826 * Attempt to detect a close/fcntl race and recover by
1827 * releasing the lock that was just acquired.
1830 * we need that spin_lock here - it prevents reordering between
1831 * update of inode->i_flock and check for it done in close().
1832 * rcu_read_lock() wouldn't do.
1834 spin_lock(&current->files->file_lock);
1835 f = fcheck(fd);
1836 spin_unlock(&current->files->file_lock);
1837 if (!error && f != filp && flock.l_type != F_UNLCK) {
1838 flock.l_type = F_UNLCK;
1839 goto again;
1842 out:
1843 locks_free_lock(file_lock);
1844 return error;
1847 #if BITS_PER_LONG == 32
1848 /* Report the first existing lock that would conflict with l.
1849 * This implements the F_GETLK command of fcntl().
1851 int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1853 struct file_lock file_lock;
1854 struct flock64 flock;
1855 int error;
1857 error = -EFAULT;
1858 if (copy_from_user(&flock, l, sizeof(flock)))
1859 goto out;
1860 error = -EINVAL;
1861 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1862 goto out;
1864 error = flock64_to_posix_lock(filp, &file_lock, &flock);
1865 if (error)
1866 goto out;
1868 error = vfs_test_lock(filp, &file_lock);
1869 if (error)
1870 goto out;
1872 flock.l_type = file_lock.fl_type;
1873 if (file_lock.fl_type != F_UNLCK)
1874 posix_lock_to_flock64(&flock, &file_lock);
1876 error = -EFAULT;
1877 if (!copy_to_user(l, &flock, sizeof(flock)))
1878 error = 0;
1880 out:
1881 return error;
1884 /* Apply the lock described by l to an open file descriptor.
1885 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1887 int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
1888 struct flock64 __user *l)
1890 struct file_lock *file_lock = locks_alloc_lock();
1891 struct flock64 flock;
1892 struct inode *inode;
1893 struct file *f;
1894 int error;
1896 if (file_lock == NULL)
1897 return -ENOLCK;
1900 * This might block, so we do it before checking the inode.
1902 error = -EFAULT;
1903 if (copy_from_user(&flock, l, sizeof(flock)))
1904 goto out;
1906 inode = filp->f_path.dentry->d_inode;
1908 /* Don't allow mandatory locks on files that may be memory mapped
1909 * and shared.
1911 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
1912 error = -EAGAIN;
1913 goto out;
1916 again:
1917 error = flock64_to_posix_lock(filp, file_lock, &flock);
1918 if (error)
1919 goto out;
1920 if (cmd == F_SETLKW64) {
1921 file_lock->fl_flags |= FL_SLEEP;
1924 error = -EBADF;
1925 switch (flock.l_type) {
1926 case F_RDLCK:
1927 if (!(filp->f_mode & FMODE_READ))
1928 goto out;
1929 break;
1930 case F_WRLCK:
1931 if (!(filp->f_mode & FMODE_WRITE))
1932 goto out;
1933 break;
1934 case F_UNLCK:
1935 break;
1936 default:
1937 error = -EINVAL;
1938 goto out;
1941 error = do_lock_file_wait(filp, cmd, file_lock);
1944 * Attempt to detect a close/fcntl race and recover by
1945 * releasing the lock that was just acquired.
1947 spin_lock(&current->files->file_lock);
1948 f = fcheck(fd);
1949 spin_unlock(&current->files->file_lock);
1950 if (!error && f != filp && flock.l_type != F_UNLCK) {
1951 flock.l_type = F_UNLCK;
1952 goto again;
1955 out:
1956 locks_free_lock(file_lock);
1957 return error;
1959 #endif /* BITS_PER_LONG == 32 */
1962 * This function is called when the file is being removed
1963 * from the task's fd array. POSIX locks belonging to this task
1964 * are deleted at this time.
1966 void locks_remove_posix(struct file *filp, fl_owner_t owner)
1968 struct file_lock lock;
1971 * If there are no locks held on this file, we don't need to call
1972 * posix_lock_file(). Another process could be setting a lock on this
1973 * file at the same time, but we wouldn't remove that lock anyway.
1975 if (!filp->f_path.dentry->d_inode->i_flock)
1976 return;
1978 lock.fl_type = F_UNLCK;
1979 lock.fl_flags = FL_POSIX | FL_CLOSE;
1980 lock.fl_start = 0;
1981 lock.fl_end = OFFSET_MAX;
1982 lock.fl_owner = owner;
1983 lock.fl_pid = current->tgid;
1984 lock.fl_file = filp;
1985 lock.fl_ops = NULL;
1986 lock.fl_lmops = NULL;
1988 vfs_lock_file(filp, F_SETLK, &lock, NULL);
1990 if (lock.fl_ops && lock.fl_ops->fl_release_private)
1991 lock.fl_ops->fl_release_private(&lock);
1994 EXPORT_SYMBOL(locks_remove_posix);
1997 * This function is called on the last close of an open file.
1999 void locks_remove_flock(struct file *filp)
2001 struct inode * inode = filp->f_path.dentry->d_inode;
2002 struct file_lock *fl;
2003 struct file_lock **before;
2005 if (!inode->i_flock)
2006 return;
2008 if (filp->f_op && filp->f_op->flock) {
2009 struct file_lock fl = {
2010 .fl_pid = current->tgid,
2011 .fl_file = filp,
2012 .fl_flags = FL_FLOCK,
2013 .fl_type = F_UNLCK,
2014 .fl_end = OFFSET_MAX,
2016 filp->f_op->flock(filp, F_SETLKW, &fl);
2017 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2018 fl.fl_ops->fl_release_private(&fl);
2021 lock_kernel();
2022 before = &inode->i_flock;
2024 while ((fl = *before) != NULL) {
2025 if (fl->fl_file == filp) {
2026 if (IS_FLOCK(fl)) {
2027 locks_delete_lock(before);
2028 continue;
2030 if (IS_LEASE(fl)) {
2031 lease_modify(before, F_UNLCK);
2032 continue;
2034 /* What? */
2035 BUG();
2037 before = &fl->fl_next;
2039 unlock_kernel();
2043 * posix_unblock_lock - stop waiting for a file lock
2044 * @filp: how the file was opened
2045 * @waiter: the lock which was waiting
2047 * lockd needs to block waiting for locks.
2050 posix_unblock_lock(struct file *filp, struct file_lock *waiter)
2052 int status = 0;
2054 lock_kernel();
2055 if (waiter->fl_next)
2056 __locks_delete_block(waiter);
2057 else
2058 status = -ENOENT;
2059 unlock_kernel();
2060 return status;
2063 EXPORT_SYMBOL(posix_unblock_lock);
2066 * vfs_cancel_lock - file byte range unblock lock
2067 * @filp: The file to apply the unblock to
2068 * @fl: The lock to be unblocked
2070 * Used by lock managers to cancel blocked requests
2072 int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2074 if (filp->f_op && filp->f_op->lock)
2075 return filp->f_op->lock(filp, F_CANCELLK, fl);
2076 return 0;
2079 EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2081 #ifdef CONFIG_PROC_FS
2082 #include <linux/proc_fs.h>
2083 #include <linux/seq_file.h>
2085 static void lock_get_status(struct seq_file *f, struct file_lock *fl,
2086 int id, char *pfx)
2088 struct inode *inode = NULL;
2089 unsigned int fl_pid;
2091 if (fl->fl_nspid)
2092 fl_pid = pid_vnr(fl->fl_nspid);
2093 else
2094 fl_pid = fl->fl_pid;
2096 if (fl->fl_file != NULL)
2097 inode = fl->fl_file->f_path.dentry->d_inode;
2099 seq_printf(f, "%d:%s ", id, pfx);
2100 if (IS_POSIX(fl)) {
2101 seq_printf(f, "%6s %s ",
2102 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2103 (inode == NULL) ? "*NOINODE*" :
2104 mandatory_lock(inode) ? "MANDATORY" : "ADVISORY ");
2105 } else if (IS_FLOCK(fl)) {
2106 if (fl->fl_type & LOCK_MAND) {
2107 seq_printf(f, "FLOCK MSNFS ");
2108 } else {
2109 seq_printf(f, "FLOCK ADVISORY ");
2111 } else if (IS_LEASE(fl)) {
2112 seq_printf(f, "LEASE ");
2113 if (fl->fl_type & F_INPROGRESS)
2114 seq_printf(f, "BREAKING ");
2115 else if (fl->fl_file)
2116 seq_printf(f, "ACTIVE ");
2117 else
2118 seq_printf(f, "BREAKER ");
2119 } else {
2120 seq_printf(f, "UNKNOWN UNKNOWN ");
2122 if (fl->fl_type & LOCK_MAND) {
2123 seq_printf(f, "%s ",
2124 (fl->fl_type & LOCK_READ)
2125 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2126 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2127 } else {
2128 seq_printf(f, "%s ",
2129 (fl->fl_type & F_INPROGRESS)
2130 ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ "
2131 : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ ");
2133 if (inode) {
2134 #ifdef WE_CAN_BREAK_LSLK_NOW
2135 seq_printf(f, "%d %s:%ld ", fl_pid,
2136 inode->i_sb->s_id, inode->i_ino);
2137 #else
2138 /* userspace relies on this representation of dev_t ;-( */
2139 seq_printf(f, "%d %02x:%02x:%ld ", fl_pid,
2140 MAJOR(inode->i_sb->s_dev),
2141 MINOR(inode->i_sb->s_dev), inode->i_ino);
2142 #endif
2143 } else {
2144 seq_printf(f, "%d <none>:0 ", fl_pid);
2146 if (IS_POSIX(fl)) {
2147 if (fl->fl_end == OFFSET_MAX)
2148 seq_printf(f, "%Ld EOF\n", fl->fl_start);
2149 else
2150 seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
2151 } else {
2152 seq_printf(f, "0 EOF\n");
2156 static int locks_show(struct seq_file *f, void *v)
2158 struct file_lock *fl, *bfl;
2160 fl = list_entry(v, struct file_lock, fl_link);
2162 lock_get_status(f, fl, (long)f->private, "");
2164 list_for_each_entry(bfl, &fl->fl_block, fl_block)
2165 lock_get_status(f, bfl, (long)f->private, " ->");
2167 f->private++;
2168 return 0;
2171 static void *locks_start(struct seq_file *f, loff_t *pos)
2173 lock_kernel();
2174 f->private = (void *)1;
2175 return seq_list_start(&file_lock_list, *pos);
2178 static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2180 return seq_list_next(v, &file_lock_list, pos);
2183 static void locks_stop(struct seq_file *f, void *v)
2185 unlock_kernel();
2188 static const struct seq_operations locks_seq_operations = {
2189 .start = locks_start,
2190 .next = locks_next,
2191 .stop = locks_stop,
2192 .show = locks_show,
2195 static int locks_open(struct inode *inode, struct file *filp)
2197 return seq_open(filp, &locks_seq_operations);
2200 static const struct file_operations proc_locks_operations = {
2201 .open = locks_open,
2202 .read = seq_read,
2203 .llseek = seq_lseek,
2204 .release = seq_release,
2207 static int __init proc_locks_init(void)
2209 proc_create("locks", 0, NULL, &proc_locks_operations);
2210 return 0;
2212 module_init(proc_locks_init);
2213 #endif
2216 * lock_may_read - checks that the region is free of locks
2217 * @inode: the inode that is being read
2218 * @start: the first byte to read
2219 * @len: the number of bytes to read
2221 * Emulates Windows locking requirements. Whole-file
2222 * mandatory locks (share modes) can prohibit a read and
2223 * byte-range POSIX locks can prohibit a read if they overlap.
2225 * N.B. this function is only ever called
2226 * from knfsd and ownership of locks is never checked.
2228 int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2230 struct file_lock *fl;
2231 int result = 1;
2232 lock_kernel();
2233 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2234 if (IS_POSIX(fl)) {
2235 if (fl->fl_type == F_RDLCK)
2236 continue;
2237 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2238 continue;
2239 } else if (IS_FLOCK(fl)) {
2240 if (!(fl->fl_type & LOCK_MAND))
2241 continue;
2242 if (fl->fl_type & LOCK_READ)
2243 continue;
2244 } else
2245 continue;
2246 result = 0;
2247 break;
2249 unlock_kernel();
2250 return result;
2253 EXPORT_SYMBOL(lock_may_read);
2256 * lock_may_write - checks that the region is free of locks
2257 * @inode: the inode that is being written
2258 * @start: the first byte to write
2259 * @len: the number of bytes to write
2261 * Emulates Windows locking requirements. Whole-file
2262 * mandatory locks (share modes) can prohibit a write and
2263 * byte-range POSIX locks can prohibit a write if they overlap.
2265 * N.B. this function is only ever called
2266 * from knfsd and ownership of locks is never checked.
2268 int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2270 struct file_lock *fl;
2271 int result = 1;
2272 lock_kernel();
2273 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2274 if (IS_POSIX(fl)) {
2275 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2276 continue;
2277 } else if (IS_FLOCK(fl)) {
2278 if (!(fl->fl_type & LOCK_MAND))
2279 continue;
2280 if (fl->fl_type & LOCK_WRITE)
2281 continue;
2282 } else
2283 continue;
2284 result = 0;
2285 break;
2287 unlock_kernel();
2288 return result;
2291 EXPORT_SYMBOL(lock_may_write);
2293 static int __init filelock_init(void)
2295 filelock_cache = kmem_cache_create("file_lock_cache",
2296 sizeof(struct file_lock), 0, SLAB_PANIC,
2297 init_once);
2298 return 0;
2301 core_initcall(filelock_init);