[ARM] 3599/1: AT91RM9200 remove global variables
[linux-2.6/mini2440.git] / fs / locks.c
blobab61a8b548292c025c86a4ba87db4dcab08015d8
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/fs.h>
120 #include <linux/init.h>
121 #include <linux/module.h>
122 #include <linux/security.h>
123 #include <linux/slab.h>
124 #include <linux/smp_lock.h>
125 #include <linux/syscalls.h>
126 #include <linux/time.h>
127 #include <linux/rcupdate.h>
129 #include <asm/semaphore.h>
130 #include <asm/uaccess.h>
132 #define IS_POSIX(fl) (fl->fl_flags & FL_POSIX)
133 #define IS_FLOCK(fl) (fl->fl_flags & FL_FLOCK)
134 #define IS_LEASE(fl) (fl->fl_flags & FL_LEASE)
136 int leases_enable = 1;
137 int lease_break_time = 45;
139 #define for_each_lock(inode, lockp) \
140 for (lockp = &inode->i_flock; *lockp != NULL; lockp = &(*lockp)->fl_next)
142 static LIST_HEAD(file_lock_list);
143 static LIST_HEAD(blocked_list);
145 static kmem_cache_t *filelock_cache __read_mostly;
147 /* Allocate an empty lock structure. */
148 static struct file_lock *locks_alloc_lock(void)
150 return kmem_cache_alloc(filelock_cache, SLAB_KERNEL);
153 static void locks_release_private(struct file_lock *fl)
155 if (fl->fl_ops) {
156 if (fl->fl_ops->fl_release_private)
157 fl->fl_ops->fl_release_private(fl);
158 fl->fl_ops = NULL;
160 if (fl->fl_lmops) {
161 if (fl->fl_lmops->fl_release_private)
162 fl->fl_lmops->fl_release_private(fl);
163 fl->fl_lmops = NULL;
168 /* Free a lock which is not in use. */
169 static void locks_free_lock(struct file_lock *fl)
171 BUG_ON(waitqueue_active(&fl->fl_wait));
172 BUG_ON(!list_empty(&fl->fl_block));
173 BUG_ON(!list_empty(&fl->fl_link));
175 locks_release_private(fl);
176 kmem_cache_free(filelock_cache, fl);
179 void locks_init_lock(struct file_lock *fl)
181 INIT_LIST_HEAD(&fl->fl_link);
182 INIT_LIST_HEAD(&fl->fl_block);
183 init_waitqueue_head(&fl->fl_wait);
184 fl->fl_next = NULL;
185 fl->fl_fasync = NULL;
186 fl->fl_owner = NULL;
187 fl->fl_pid = 0;
188 fl->fl_file = NULL;
189 fl->fl_flags = 0;
190 fl->fl_type = 0;
191 fl->fl_start = fl->fl_end = 0;
192 fl->fl_ops = NULL;
193 fl->fl_lmops = NULL;
196 EXPORT_SYMBOL(locks_init_lock);
199 * Initialises the fields of the file lock which are invariant for
200 * free file_locks.
202 static void init_once(void *foo, kmem_cache_t *cache, unsigned long flags)
204 struct file_lock *lock = (struct file_lock *) foo;
206 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) !=
207 SLAB_CTOR_CONSTRUCTOR)
208 return;
210 locks_init_lock(lock);
213 static void locks_copy_private(struct file_lock *new, struct file_lock *fl)
215 if (fl->fl_ops) {
216 if (fl->fl_ops->fl_copy_lock)
217 fl->fl_ops->fl_copy_lock(new, fl);
218 new->fl_ops = fl->fl_ops;
220 if (fl->fl_lmops) {
221 if (fl->fl_lmops->fl_copy_lock)
222 fl->fl_lmops->fl_copy_lock(new, fl);
223 new->fl_lmops = fl->fl_lmops;
228 * Initialize a new lock from an existing file_lock structure.
230 static void __locks_copy_lock(struct file_lock *new, const struct file_lock *fl)
232 new->fl_owner = fl->fl_owner;
233 new->fl_pid = fl->fl_pid;
234 new->fl_file = NULL;
235 new->fl_flags = fl->fl_flags;
236 new->fl_type = fl->fl_type;
237 new->fl_start = fl->fl_start;
238 new->fl_end = fl->fl_end;
239 new->fl_ops = NULL;
240 new->fl_lmops = NULL;
243 void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
245 locks_release_private(new);
247 __locks_copy_lock(new, fl);
248 new->fl_file = fl->fl_file;
249 new->fl_ops = fl->fl_ops;
250 new->fl_lmops = fl->fl_lmops;
252 locks_copy_private(new, fl);
255 EXPORT_SYMBOL(locks_copy_lock);
257 static inline int flock_translate_cmd(int cmd) {
258 if (cmd & LOCK_MAND)
259 return cmd & (LOCK_MAND | LOCK_RW);
260 switch (cmd) {
261 case LOCK_SH:
262 return F_RDLCK;
263 case LOCK_EX:
264 return F_WRLCK;
265 case LOCK_UN:
266 return F_UNLCK;
268 return -EINVAL;
271 /* Fill in a file_lock structure with an appropriate FLOCK lock. */
272 static int flock_make_lock(struct file *filp, struct file_lock **lock,
273 unsigned int cmd)
275 struct file_lock *fl;
276 int type = flock_translate_cmd(cmd);
277 if (type < 0)
278 return type;
280 fl = locks_alloc_lock();
281 if (fl == NULL)
282 return -ENOMEM;
284 fl->fl_file = filp;
285 fl->fl_pid = current->tgid;
286 fl->fl_flags = FL_FLOCK;
287 fl->fl_type = type;
288 fl->fl_end = OFFSET_MAX;
290 *lock = fl;
291 return 0;
294 static int assign_type(struct file_lock *fl, int type)
296 switch (type) {
297 case F_RDLCK:
298 case F_WRLCK:
299 case F_UNLCK:
300 fl->fl_type = type;
301 break;
302 default:
303 return -EINVAL;
305 return 0;
308 /* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
309 * style lock.
311 static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
312 struct flock *l)
314 off_t start, end;
316 switch (l->l_whence) {
317 case 0: /*SEEK_SET*/
318 start = 0;
319 break;
320 case 1: /*SEEK_CUR*/
321 start = filp->f_pos;
322 break;
323 case 2: /*SEEK_END*/
324 start = i_size_read(filp->f_dentry->d_inode);
325 break;
326 default:
327 return -EINVAL;
330 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
331 POSIX-2001 defines it. */
332 start += l->l_start;
333 if (start < 0)
334 return -EINVAL;
335 fl->fl_end = OFFSET_MAX;
336 if (l->l_len > 0) {
337 end = start + l->l_len - 1;
338 fl->fl_end = end;
339 } else if (l->l_len < 0) {
340 end = start - 1;
341 fl->fl_end = end;
342 start += l->l_len;
343 if (start < 0)
344 return -EINVAL;
346 fl->fl_start = start; /* we record the absolute position */
347 if (fl->fl_end < fl->fl_start)
348 return -EOVERFLOW;
350 fl->fl_owner = current->files;
351 fl->fl_pid = current->tgid;
352 fl->fl_file = filp;
353 fl->fl_flags = FL_POSIX;
354 fl->fl_ops = NULL;
355 fl->fl_lmops = NULL;
357 return assign_type(fl, l->l_type);
360 #if BITS_PER_LONG == 32
361 static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
362 struct flock64 *l)
364 loff_t start;
366 switch (l->l_whence) {
367 case 0: /*SEEK_SET*/
368 start = 0;
369 break;
370 case 1: /*SEEK_CUR*/
371 start = filp->f_pos;
372 break;
373 case 2: /*SEEK_END*/
374 start = i_size_read(filp->f_dentry->d_inode);
375 break;
376 default:
377 return -EINVAL;
380 start += l->l_start;
381 if (start < 0)
382 return -EINVAL;
383 fl->fl_end = OFFSET_MAX;
384 if (l->l_len > 0) {
385 fl->fl_end = start + l->l_len - 1;
386 } else if (l->l_len < 0) {
387 fl->fl_end = start - 1;
388 start += l->l_len;
389 if (start < 0)
390 return -EINVAL;
392 fl->fl_start = start; /* we record the absolute position */
393 if (fl->fl_end < fl->fl_start)
394 return -EOVERFLOW;
396 fl->fl_owner = current->files;
397 fl->fl_pid = current->tgid;
398 fl->fl_file = filp;
399 fl->fl_flags = FL_POSIX;
400 fl->fl_ops = NULL;
401 fl->fl_lmops = NULL;
403 switch (l->l_type) {
404 case F_RDLCK:
405 case F_WRLCK:
406 case F_UNLCK:
407 fl->fl_type = l->l_type;
408 break;
409 default:
410 return -EINVAL;
413 return (0);
415 #endif
417 /* default lease lock manager operations */
418 static void lease_break_callback(struct file_lock *fl)
420 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
423 static void lease_release_private_callback(struct file_lock *fl)
425 if (!fl->fl_file)
426 return;
428 f_delown(fl->fl_file);
429 fl->fl_file->f_owner.signum = 0;
432 static int lease_mylease_callback(struct file_lock *fl, struct file_lock *try)
434 return fl->fl_file == try->fl_file;
437 static struct lock_manager_operations lease_manager_ops = {
438 .fl_break = lease_break_callback,
439 .fl_release_private = lease_release_private_callback,
440 .fl_mylease = lease_mylease_callback,
441 .fl_change = lease_modify,
445 * Initialize a lease, use the default lock manager operations
447 static int lease_init(struct file *filp, int type, struct file_lock *fl)
449 if (assign_type(fl, type) != 0)
450 return -EINVAL;
452 fl->fl_owner = current->files;
453 fl->fl_pid = current->tgid;
455 fl->fl_file = filp;
456 fl->fl_flags = FL_LEASE;
457 fl->fl_start = 0;
458 fl->fl_end = OFFSET_MAX;
459 fl->fl_ops = NULL;
460 fl->fl_lmops = &lease_manager_ops;
461 return 0;
464 /* Allocate a file_lock initialised to this type of lease */
465 static int lease_alloc(struct file *filp, int type, struct file_lock **flp)
467 struct file_lock *fl = locks_alloc_lock();
468 int error = -ENOMEM;
470 if (fl == NULL)
471 goto out;
473 error = lease_init(filp, type, fl);
474 if (error) {
475 locks_free_lock(fl);
476 fl = NULL;
478 out:
479 *flp = fl;
480 return error;
483 /* Check if two locks overlap each other.
485 static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
487 return ((fl1->fl_end >= fl2->fl_start) &&
488 (fl2->fl_end >= fl1->fl_start));
492 * Check whether two locks have the same owner.
494 static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
496 if (fl1->fl_lmops && fl1->fl_lmops->fl_compare_owner)
497 return fl2->fl_lmops == fl1->fl_lmops &&
498 fl1->fl_lmops->fl_compare_owner(fl1, fl2);
499 return fl1->fl_owner == fl2->fl_owner;
502 /* Remove waiter from blocker's block list.
503 * When blocker ends up pointing to itself then the list is empty.
505 static void __locks_delete_block(struct file_lock *waiter)
507 list_del_init(&waiter->fl_block);
508 list_del_init(&waiter->fl_link);
509 waiter->fl_next = NULL;
514 static void locks_delete_block(struct file_lock *waiter)
516 lock_kernel();
517 __locks_delete_block(waiter);
518 unlock_kernel();
521 /* Insert waiter into blocker's block list.
522 * We use a circular list so that processes can be easily woken up in
523 * the order they blocked. The documentation doesn't require this but
524 * it seems like the reasonable thing to do.
526 static void locks_insert_block(struct file_lock *blocker,
527 struct file_lock *waiter)
529 BUG_ON(!list_empty(&waiter->fl_block));
530 list_add_tail(&waiter->fl_block, &blocker->fl_block);
531 waiter->fl_next = blocker;
532 if (IS_POSIX(blocker))
533 list_add(&waiter->fl_link, &blocked_list);
536 /* Wake up processes blocked waiting for blocker.
537 * If told to wait then schedule the processes until the block list
538 * is empty, otherwise empty the block list ourselves.
540 static void locks_wake_up_blocks(struct file_lock *blocker)
542 while (!list_empty(&blocker->fl_block)) {
543 struct file_lock *waiter = list_entry(blocker->fl_block.next,
544 struct file_lock, fl_block);
545 __locks_delete_block(waiter);
546 if (waiter->fl_lmops && waiter->fl_lmops->fl_notify)
547 waiter->fl_lmops->fl_notify(waiter);
548 else
549 wake_up(&waiter->fl_wait);
553 /* Insert file lock fl into an inode's lock list at the position indicated
554 * by pos. At the same time add the lock to the global file lock list.
556 static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
558 list_add(&fl->fl_link, &file_lock_list);
560 /* insert into file's list */
561 fl->fl_next = *pos;
562 *pos = fl;
564 if (fl->fl_ops && fl->fl_ops->fl_insert)
565 fl->fl_ops->fl_insert(fl);
569 * Delete a lock and then free it.
570 * Wake up processes that are blocked waiting for this lock,
571 * notify the FS that the lock has been cleared and
572 * finally free the lock.
574 static void locks_delete_lock(struct file_lock **thisfl_p)
576 struct file_lock *fl = *thisfl_p;
578 *thisfl_p = fl->fl_next;
579 fl->fl_next = NULL;
580 list_del_init(&fl->fl_link);
582 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
583 if (fl->fl_fasync != NULL) {
584 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
585 fl->fl_fasync = NULL;
588 if (fl->fl_ops && fl->fl_ops->fl_remove)
589 fl->fl_ops->fl_remove(fl);
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 static int interruptible_sleep_on_locked(wait_queue_head_t *fl_wait, int timeout)
643 int result = 0;
644 DECLARE_WAITQUEUE(wait, current);
646 __set_current_state(TASK_INTERRUPTIBLE);
647 add_wait_queue(fl_wait, &wait);
648 if (timeout == 0)
649 schedule();
650 else
651 result = schedule_timeout(timeout);
652 if (signal_pending(current))
653 result = -ERESTARTSYS;
654 remove_wait_queue(fl_wait, &wait);
655 __set_current_state(TASK_RUNNING);
656 return result;
659 static int locks_block_on_timeout(struct file_lock *blocker, struct file_lock *waiter, int time)
661 int result;
662 locks_insert_block(blocker, waiter);
663 result = interruptible_sleep_on_locked(&waiter->fl_wait, time);
664 __locks_delete_block(waiter);
665 return result;
669 posix_test_lock(struct file *filp, struct file_lock *fl,
670 struct file_lock *conflock)
672 struct file_lock *cfl;
674 lock_kernel();
675 for (cfl = filp->f_dentry->d_inode->i_flock; cfl; cfl = cfl->fl_next) {
676 if (!IS_POSIX(cfl))
677 continue;
678 if (posix_locks_conflict(cfl, fl))
679 break;
681 if (cfl) {
682 __locks_copy_lock(conflock, cfl);
683 unlock_kernel();
684 return 1;
686 unlock_kernel();
687 return 0;
690 EXPORT_SYMBOL(posix_test_lock);
692 /* This function tests for deadlock condition before putting a process to
693 * sleep. The detection scheme is no longer recursive. Recursive was neat,
694 * but dangerous - we risked stack corruption if the lock data was bad, or
695 * if the recursion was too deep for any other reason.
697 * We rely on the fact that a task can only be on one lock's wait queue
698 * at a time. When we find blocked_task on a wait queue we can re-search
699 * with blocked_task equal to that queue's owner, until either blocked_task
700 * isn't found, or blocked_task is found on a queue owned by my_task.
702 * Note: the above assumption may not be true when handling lock requests
703 * from a broken NFS client. But broken NFS clients have a lot more to
704 * worry about than proper deadlock detection anyway... --okir
706 int posix_locks_deadlock(struct file_lock *caller_fl,
707 struct file_lock *block_fl)
709 struct list_head *tmp;
711 next_task:
712 if (posix_same_owner(caller_fl, block_fl))
713 return 1;
714 list_for_each(tmp, &blocked_list) {
715 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
716 if (posix_same_owner(fl, block_fl)) {
717 fl = fl->fl_next;
718 block_fl = fl;
719 goto next_task;
722 return 0;
725 EXPORT_SYMBOL(posix_locks_deadlock);
727 /* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
728 * at the head of the list, but that's secret knowledge known only to
729 * flock_lock_file and posix_lock_file.
731 static int flock_lock_file(struct file *filp, struct file_lock *request)
733 struct file_lock *new_fl = NULL;
734 struct file_lock **before;
735 struct inode * inode = filp->f_dentry->d_inode;
736 int error = 0;
737 int found = 0;
739 lock_kernel();
740 for_each_lock(inode, before) {
741 struct file_lock *fl = *before;
742 if (IS_POSIX(fl))
743 break;
744 if (IS_LEASE(fl))
745 continue;
746 if (filp != fl->fl_file)
747 continue;
748 if (request->fl_type == fl->fl_type)
749 goto out;
750 found = 1;
751 locks_delete_lock(before);
752 break;
755 if (request->fl_type == F_UNLCK)
756 goto out;
758 error = -ENOMEM;
759 new_fl = locks_alloc_lock();
760 if (new_fl == NULL)
761 goto out;
763 * If a higher-priority process was blocked on the old file lock,
764 * give it the opportunity to lock the file.
766 if (found)
767 cond_resched();
769 for_each_lock(inode, before) {
770 struct file_lock *fl = *before;
771 if (IS_POSIX(fl))
772 break;
773 if (IS_LEASE(fl))
774 continue;
775 if (!flock_locks_conflict(request, fl))
776 continue;
777 error = -EAGAIN;
778 if (request->fl_flags & FL_SLEEP)
779 locks_insert_block(fl, request);
780 goto out;
782 locks_copy_lock(new_fl, request);
783 locks_insert_lock(&inode->i_flock, new_fl);
784 new_fl = NULL;
785 error = 0;
787 out:
788 unlock_kernel();
789 if (new_fl)
790 locks_free_lock(new_fl);
791 return error;
794 static int __posix_lock_file_conf(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
796 struct file_lock *fl;
797 struct file_lock *new_fl, *new_fl2;
798 struct file_lock *left = NULL;
799 struct file_lock *right = NULL;
800 struct file_lock **before;
801 int error, added = 0;
804 * We may need two file_lock structures for this operation,
805 * so we get them in advance to avoid races.
807 new_fl = locks_alloc_lock();
808 new_fl2 = locks_alloc_lock();
810 lock_kernel();
811 if (request->fl_type != F_UNLCK) {
812 for_each_lock(inode, before) {
813 struct file_lock *fl = *before;
814 if (!IS_POSIX(fl))
815 continue;
816 if (!posix_locks_conflict(request, fl))
817 continue;
818 if (conflock)
819 locks_copy_lock(conflock, fl);
820 error = -EAGAIN;
821 if (!(request->fl_flags & FL_SLEEP))
822 goto out;
823 error = -EDEADLK;
824 if (posix_locks_deadlock(request, fl))
825 goto out;
826 error = -EAGAIN;
827 locks_insert_block(fl, request);
828 goto out;
832 /* If we're just looking for a conflict, we're done. */
833 error = 0;
834 if (request->fl_flags & FL_ACCESS)
835 goto out;
837 error = -ENOLCK; /* "no luck" */
838 if (!(new_fl && new_fl2))
839 goto out;
842 * We've allocated the new locks in advance, so there are no
843 * errors possible (and no blocking operations) from here on.
845 * Find the first old lock with the same owner as the new lock.
848 before = &inode->i_flock;
850 /* First skip locks owned by other processes. */
851 while ((fl = *before) && (!IS_POSIX(fl) ||
852 !posix_same_owner(request, fl))) {
853 before = &fl->fl_next;
856 /* Process locks with this owner. */
857 while ((fl = *before) && posix_same_owner(request, fl)) {
858 /* Detect adjacent or overlapping regions (if same lock type)
860 if (request->fl_type == fl->fl_type) {
861 /* In all comparisons of start vs end, use
862 * "start - 1" rather than "end + 1". If end
863 * is OFFSET_MAX, end + 1 will become negative.
865 if (fl->fl_end < request->fl_start - 1)
866 goto next_lock;
867 /* If the next lock in the list has entirely bigger
868 * addresses than the new one, insert the lock here.
870 if (fl->fl_start - 1 > request->fl_end)
871 break;
873 /* If we come here, the new and old lock are of the
874 * same type and adjacent or overlapping. Make one
875 * lock yielding from the lower start address of both
876 * locks to the higher end address.
878 if (fl->fl_start > request->fl_start)
879 fl->fl_start = request->fl_start;
880 else
881 request->fl_start = fl->fl_start;
882 if (fl->fl_end < request->fl_end)
883 fl->fl_end = request->fl_end;
884 else
885 request->fl_end = fl->fl_end;
886 if (added) {
887 locks_delete_lock(before);
888 continue;
890 request = fl;
891 added = 1;
893 else {
894 /* Processing for different lock types is a bit
895 * more complex.
897 if (fl->fl_end < request->fl_start)
898 goto next_lock;
899 if (fl->fl_start > request->fl_end)
900 break;
901 if (request->fl_type == F_UNLCK)
902 added = 1;
903 if (fl->fl_start < request->fl_start)
904 left = fl;
905 /* If the next lock in the list has a higher end
906 * address than the new one, insert the new one here.
908 if (fl->fl_end > request->fl_end) {
909 right = fl;
910 break;
912 if (fl->fl_start >= request->fl_start) {
913 /* The new lock completely replaces an old
914 * one (This may happen several times).
916 if (added) {
917 locks_delete_lock(before);
918 continue;
920 /* Replace the old lock with the new one.
921 * Wake up anybody waiting for the old one,
922 * as the change in lock type might satisfy
923 * their needs.
925 locks_wake_up_blocks(fl);
926 fl->fl_start = request->fl_start;
927 fl->fl_end = request->fl_end;
928 fl->fl_type = request->fl_type;
929 locks_release_private(fl);
930 locks_copy_private(fl, request);
931 request = fl;
932 added = 1;
935 /* Go on to next lock.
937 next_lock:
938 before = &fl->fl_next;
941 error = 0;
942 if (!added) {
943 if (request->fl_type == F_UNLCK)
944 goto out;
945 locks_copy_lock(new_fl, request);
946 locks_insert_lock(before, new_fl);
947 new_fl = NULL;
949 if (right) {
950 if (left == right) {
951 /* The new lock breaks the old one in two pieces,
952 * so we have to use the second new lock.
954 left = new_fl2;
955 new_fl2 = NULL;
956 locks_copy_lock(left, right);
957 locks_insert_lock(before, left);
959 right->fl_start = request->fl_end + 1;
960 locks_wake_up_blocks(right);
962 if (left) {
963 left->fl_end = request->fl_start - 1;
964 locks_wake_up_blocks(left);
966 out:
967 unlock_kernel();
969 * Free any unused locks.
971 if (new_fl)
972 locks_free_lock(new_fl);
973 if (new_fl2)
974 locks_free_lock(new_fl2);
975 return error;
979 * posix_lock_file - Apply a POSIX-style lock to a file
980 * @filp: The file to apply the lock to
981 * @fl: The lock to be applied
983 * Add a POSIX style lock to a file.
984 * We merge adjacent & overlapping locks whenever possible.
985 * POSIX locks are sorted by owner task, then by starting address
987 int posix_lock_file(struct file *filp, struct file_lock *fl)
989 return __posix_lock_file_conf(filp->f_dentry->d_inode, fl, NULL);
991 EXPORT_SYMBOL(posix_lock_file);
994 * posix_lock_file_conf - Apply a POSIX-style lock to a file
995 * @filp: The file to apply the lock to
996 * @fl: The lock to be applied
997 * @conflock: Place to return a copy of the conflicting lock, if found.
999 * Except for the conflock parameter, acts just like posix_lock_file.
1001 int posix_lock_file_conf(struct file *filp, struct file_lock *fl,
1002 struct file_lock *conflock)
1004 return __posix_lock_file_conf(filp->f_dentry->d_inode, fl, conflock);
1006 EXPORT_SYMBOL(posix_lock_file_conf);
1009 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1010 * @filp: The file to apply the lock to
1011 * @fl: The lock to be applied
1013 * Add a POSIX style lock to a file.
1014 * We merge adjacent & overlapping locks whenever possible.
1015 * POSIX locks are sorted by owner task, then by starting address
1017 int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1019 int error;
1020 might_sleep ();
1021 for (;;) {
1022 error = posix_lock_file(filp, fl);
1023 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1024 break;
1025 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1026 if (!error)
1027 continue;
1029 locks_delete_block(fl);
1030 break;
1032 return error;
1034 EXPORT_SYMBOL(posix_lock_file_wait);
1037 * locks_mandatory_locked - Check for an active lock
1038 * @inode: the file to check
1040 * Searches the inode's list of locks to find any POSIX locks which conflict.
1041 * This function is called from locks_verify_locked() only.
1043 int locks_mandatory_locked(struct inode *inode)
1045 fl_owner_t owner = current->files;
1046 struct file_lock *fl;
1049 * Search the lock list for this inode for any POSIX locks.
1051 lock_kernel();
1052 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1053 if (!IS_POSIX(fl))
1054 continue;
1055 if (fl->fl_owner != owner)
1056 break;
1058 unlock_kernel();
1059 return fl ? -EAGAIN : 0;
1063 * locks_mandatory_area - Check for a conflicting lock
1064 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1065 * for shared
1066 * @inode: the file to check
1067 * @filp: how the file was opened (if it was)
1068 * @offset: start of area to check
1069 * @count: length of area to check
1071 * Searches the inode's list of locks to find any POSIX locks which conflict.
1072 * This function is called from rw_verify_area() and
1073 * locks_verify_truncate().
1075 int locks_mandatory_area(int read_write, struct inode *inode,
1076 struct file *filp, loff_t offset,
1077 size_t count)
1079 struct file_lock fl;
1080 int error;
1082 locks_init_lock(&fl);
1083 fl.fl_owner = current->files;
1084 fl.fl_pid = current->tgid;
1085 fl.fl_file = filp;
1086 fl.fl_flags = FL_POSIX | FL_ACCESS;
1087 if (filp && !(filp->f_flags & O_NONBLOCK))
1088 fl.fl_flags |= FL_SLEEP;
1089 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1090 fl.fl_start = offset;
1091 fl.fl_end = offset + count - 1;
1093 for (;;) {
1094 error = __posix_lock_file_conf(inode, &fl, NULL);
1095 if (error != -EAGAIN)
1096 break;
1097 if (!(fl.fl_flags & FL_SLEEP))
1098 break;
1099 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1100 if (!error) {
1102 * If we've been sleeping someone might have
1103 * changed the permissions behind our back.
1105 if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID)
1106 continue;
1109 locks_delete_block(&fl);
1110 break;
1113 return error;
1116 EXPORT_SYMBOL(locks_mandatory_area);
1118 /* We already had a lease on this file; just change its type */
1119 int lease_modify(struct file_lock **before, int arg)
1121 struct file_lock *fl = *before;
1122 int error = assign_type(fl, arg);
1124 if (error)
1125 return error;
1126 locks_wake_up_blocks(fl);
1127 if (arg == F_UNLCK)
1128 locks_delete_lock(before);
1129 return 0;
1132 EXPORT_SYMBOL(lease_modify);
1134 static void time_out_leases(struct inode *inode)
1136 struct file_lock **before;
1137 struct file_lock *fl;
1139 before = &inode->i_flock;
1140 while ((fl = *before) && IS_LEASE(fl) && (fl->fl_type & F_INPROGRESS)) {
1141 if ((fl->fl_break_time == 0)
1142 || time_before(jiffies, fl->fl_break_time)) {
1143 before = &fl->fl_next;
1144 continue;
1146 lease_modify(before, fl->fl_type & ~F_INPROGRESS);
1147 if (fl == *before) /* lease_modify may have freed fl */
1148 before = &fl->fl_next;
1153 * __break_lease - revoke all outstanding leases on file
1154 * @inode: the inode of the file to return
1155 * @mode: the open mode (read or write)
1157 * break_lease (inlined for speed) has checked there already
1158 * is a lease on this file. Leases are broken on a call to open()
1159 * or truncate(). This function can sleep unless you
1160 * specified %O_NONBLOCK to your open().
1162 int __break_lease(struct inode *inode, unsigned int mode)
1164 int error = 0, future;
1165 struct file_lock *new_fl, *flock;
1166 struct file_lock *fl;
1167 int alloc_err;
1168 unsigned long break_time;
1169 int i_have_this_lease = 0;
1171 alloc_err = lease_alloc(NULL, mode & FMODE_WRITE ? F_WRLCK : F_RDLCK,
1172 &new_fl);
1174 lock_kernel();
1176 time_out_leases(inode);
1178 flock = inode->i_flock;
1179 if ((flock == NULL) || !IS_LEASE(flock))
1180 goto out;
1182 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1183 if (fl->fl_owner == current->files)
1184 i_have_this_lease = 1;
1186 if (mode & FMODE_WRITE) {
1187 /* If we want write access, we have to revoke any lease. */
1188 future = F_UNLCK | F_INPROGRESS;
1189 } else if (flock->fl_type & F_INPROGRESS) {
1190 /* If the lease is already being broken, we just leave it */
1191 future = flock->fl_type;
1192 } else if (flock->fl_type & F_WRLCK) {
1193 /* Downgrade the exclusive lease to a read-only lease. */
1194 future = F_RDLCK | F_INPROGRESS;
1195 } else {
1196 /* the existing lease was read-only, so we can read too. */
1197 goto out;
1200 if (alloc_err && !i_have_this_lease && ((mode & O_NONBLOCK) == 0)) {
1201 error = alloc_err;
1202 goto out;
1205 break_time = 0;
1206 if (lease_break_time > 0) {
1207 break_time = jiffies + lease_break_time * HZ;
1208 if (break_time == 0)
1209 break_time++; /* so that 0 means no break time */
1212 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1213 if (fl->fl_type != future) {
1214 fl->fl_type = future;
1215 fl->fl_break_time = break_time;
1216 /* lease must have lmops break callback */
1217 fl->fl_lmops->fl_break(fl);
1221 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1222 error = -EWOULDBLOCK;
1223 goto out;
1226 restart:
1227 break_time = flock->fl_break_time;
1228 if (break_time != 0) {
1229 break_time -= jiffies;
1230 if (break_time == 0)
1231 break_time++;
1233 error = locks_block_on_timeout(flock, new_fl, break_time);
1234 if (error >= 0) {
1235 if (error == 0)
1236 time_out_leases(inode);
1237 /* Wait for the next lease that has not been broken yet */
1238 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1239 flock = flock->fl_next) {
1240 if (flock->fl_type & F_INPROGRESS)
1241 goto restart;
1243 error = 0;
1246 out:
1247 unlock_kernel();
1248 if (!alloc_err)
1249 locks_free_lock(new_fl);
1250 return error;
1253 EXPORT_SYMBOL(__break_lease);
1256 * lease_get_mtime
1257 * @inode: the inode
1258 * @time: pointer to a timespec which will contain the last modified time
1260 * This is to force NFS clients to flush their caches for files with
1261 * exclusive leases. The justification is that if someone has an
1262 * exclusive lease, then they could be modifiying it.
1264 void lease_get_mtime(struct inode *inode, struct timespec *time)
1266 struct file_lock *flock = inode->i_flock;
1267 if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK))
1268 *time = current_fs_time(inode->i_sb);
1269 else
1270 *time = inode->i_mtime;
1273 EXPORT_SYMBOL(lease_get_mtime);
1276 * fcntl_getlease - Enquire what lease is currently active
1277 * @filp: the file
1279 * The value returned by this function will be one of
1280 * (if no lease break is pending):
1282 * %F_RDLCK to indicate a shared lease is held.
1284 * %F_WRLCK to indicate an exclusive lease is held.
1286 * %F_UNLCK to indicate no lease is held.
1288 * (if a lease break is pending):
1290 * %F_RDLCK to indicate an exclusive lease needs to be
1291 * changed to a shared lease (or removed).
1293 * %F_UNLCK to indicate the lease needs to be removed.
1295 * XXX: sfr & willy disagree over whether F_INPROGRESS
1296 * should be returned to userspace.
1298 int fcntl_getlease(struct file *filp)
1300 struct file_lock *fl;
1301 int type = F_UNLCK;
1303 lock_kernel();
1304 time_out_leases(filp->f_dentry->d_inode);
1305 for (fl = filp->f_dentry->d_inode->i_flock; fl && IS_LEASE(fl);
1306 fl = fl->fl_next) {
1307 if (fl->fl_file == filp) {
1308 type = fl->fl_type & ~F_INPROGRESS;
1309 break;
1312 unlock_kernel();
1313 return type;
1317 * __setlease - sets a lease on an open file
1318 * @filp: file pointer
1319 * @arg: type of lease to obtain
1320 * @flp: input - file_lock to use, output - file_lock inserted
1322 * The (input) flp->fl_lmops->fl_break function is required
1323 * by break_lease().
1325 * Called with kernel lock held.
1327 static int __setlease(struct file *filp, long arg, struct file_lock **flp)
1329 struct file_lock *fl, **before, **my_before = NULL, *lease;
1330 struct dentry *dentry = filp->f_dentry;
1331 struct inode *inode = dentry->d_inode;
1332 int error, rdlease_count = 0, wrlease_count = 0;
1334 time_out_leases(inode);
1336 error = -EINVAL;
1337 if (!flp || !(*flp) || !(*flp)->fl_lmops || !(*flp)->fl_lmops->fl_break)
1338 goto out;
1340 lease = *flp;
1342 error = -EAGAIN;
1343 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1344 goto out;
1345 if ((arg == F_WRLCK)
1346 && ((atomic_read(&dentry->d_count) > 1)
1347 || (atomic_read(&inode->i_count) > 1)))
1348 goto out;
1351 * At this point, we know that if there is an exclusive
1352 * lease on this file, then we hold it on this filp
1353 * (otherwise our open of this file would have blocked).
1354 * And if we are trying to acquire an exclusive lease,
1355 * then the file is not open by anyone (including us)
1356 * except for this filp.
1358 for (before = &inode->i_flock;
1359 ((fl = *before) != NULL) && IS_LEASE(fl);
1360 before = &fl->fl_next) {
1361 if (lease->fl_lmops->fl_mylease(fl, lease))
1362 my_before = before;
1363 else if (fl->fl_type == (F_INPROGRESS | F_UNLCK))
1365 * Someone is in the process of opening this
1366 * file for writing so we may not take an
1367 * exclusive lease on it.
1369 wrlease_count++;
1370 else
1371 rdlease_count++;
1374 if ((arg == F_RDLCK && (wrlease_count > 0)) ||
1375 (arg == F_WRLCK && ((rdlease_count + wrlease_count) > 0)))
1376 goto out;
1378 if (my_before != NULL) {
1379 *flp = *my_before;
1380 error = lease->fl_lmops->fl_change(my_before, arg);
1381 goto out;
1384 error = 0;
1385 if (arg == F_UNLCK)
1386 goto out;
1388 error = -EINVAL;
1389 if (!leases_enable)
1390 goto out;
1392 error = lease_alloc(filp, arg, &fl);
1393 if (error)
1394 goto out;
1396 locks_copy_lock(fl, lease);
1398 locks_insert_lock(before, fl);
1400 *flp = fl;
1401 out:
1402 return error;
1406 * setlease - sets a lease on an open file
1407 * @filp: file pointer
1408 * @arg: type of lease to obtain
1409 * @lease: file_lock to use
1411 * Call this to establish a lease on the file.
1412 * The fl_lmops fl_break function is required by break_lease
1415 int setlease(struct file *filp, long arg, struct file_lock **lease)
1417 struct dentry *dentry = filp->f_dentry;
1418 struct inode *inode = dentry->d_inode;
1419 int error;
1421 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1422 return -EACCES;
1423 if (!S_ISREG(inode->i_mode))
1424 return -EINVAL;
1425 error = security_file_lock(filp, arg);
1426 if (error)
1427 return error;
1429 lock_kernel();
1430 error = __setlease(filp, arg, lease);
1431 unlock_kernel();
1433 return error;
1436 EXPORT_SYMBOL(setlease);
1439 * fcntl_setlease - sets a lease on an open file
1440 * @fd: open file descriptor
1441 * @filp: file pointer
1442 * @arg: type of lease to obtain
1444 * Call this fcntl to establish a lease on the file.
1445 * Note that you also need to call %F_SETSIG to
1446 * receive a signal when the lease is broken.
1448 int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1450 struct file_lock fl, *flp = &fl;
1451 struct dentry *dentry = filp->f_dentry;
1452 struct inode *inode = dentry->d_inode;
1453 int error;
1455 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1456 return -EACCES;
1457 if (!S_ISREG(inode->i_mode))
1458 return -EINVAL;
1459 error = security_file_lock(filp, arg);
1460 if (error)
1461 return error;
1463 locks_init_lock(&fl);
1464 error = lease_init(filp, arg, &fl);
1465 if (error)
1466 return error;
1468 lock_kernel();
1470 error = __setlease(filp, arg, &flp);
1471 if (error || arg == F_UNLCK)
1472 goto out_unlock;
1474 error = fasync_helper(fd, filp, 1, &flp->fl_fasync);
1475 if (error < 0) {
1476 /* remove lease just inserted by __setlease */
1477 flp->fl_type = F_UNLCK | F_INPROGRESS;
1478 flp->fl_break_time = jiffies- 10;
1479 time_out_leases(inode);
1480 goto out_unlock;
1483 error = f_setown(filp, current->pid, 0);
1484 out_unlock:
1485 unlock_kernel();
1486 return error;
1490 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1491 * @filp: The file to apply the lock to
1492 * @fl: The lock to be applied
1494 * Add a FLOCK style lock to a file.
1496 int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1498 int error;
1499 might_sleep();
1500 for (;;) {
1501 error = flock_lock_file(filp, fl);
1502 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1503 break;
1504 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1505 if (!error)
1506 continue;
1508 locks_delete_block(fl);
1509 break;
1511 return error;
1514 EXPORT_SYMBOL(flock_lock_file_wait);
1517 * sys_flock: - flock() system call.
1518 * @fd: the file descriptor to lock.
1519 * @cmd: the type of lock to apply.
1521 * Apply a %FL_FLOCK style lock to an open file descriptor.
1522 * The @cmd can be one of
1524 * %LOCK_SH -- a shared lock.
1526 * %LOCK_EX -- an exclusive lock.
1528 * %LOCK_UN -- remove an existing lock.
1530 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1532 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1533 * processes read and write access respectively.
1535 asmlinkage long sys_flock(unsigned int fd, unsigned int cmd)
1537 struct file *filp;
1538 struct file_lock *lock;
1539 int can_sleep, unlock;
1540 int error;
1542 error = -EBADF;
1543 filp = fget(fd);
1544 if (!filp)
1545 goto out;
1547 can_sleep = !(cmd & LOCK_NB);
1548 cmd &= ~LOCK_NB;
1549 unlock = (cmd == LOCK_UN);
1551 if (!unlock && !(cmd & LOCK_MAND) && !(filp->f_mode & 3))
1552 goto out_putf;
1554 error = flock_make_lock(filp, &lock, cmd);
1555 if (error)
1556 goto out_putf;
1557 if (can_sleep)
1558 lock->fl_flags |= FL_SLEEP;
1560 error = security_file_lock(filp, cmd);
1561 if (error)
1562 goto out_free;
1564 if (filp->f_op && filp->f_op->flock)
1565 error = filp->f_op->flock(filp,
1566 (can_sleep) ? F_SETLKW : F_SETLK,
1567 lock);
1568 else
1569 error = flock_lock_file_wait(filp, lock);
1571 out_free:
1572 locks_free_lock(lock);
1574 out_putf:
1575 fput(filp);
1576 out:
1577 return error;
1580 /* Report the first existing lock that would conflict with l.
1581 * This implements the F_GETLK command of fcntl().
1583 int fcntl_getlk(struct file *filp, struct flock __user *l)
1585 struct file_lock *fl, cfl, file_lock;
1586 struct flock flock;
1587 int error;
1589 error = -EFAULT;
1590 if (copy_from_user(&flock, l, sizeof(flock)))
1591 goto out;
1592 error = -EINVAL;
1593 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1594 goto out;
1596 error = flock_to_posix_lock(filp, &file_lock, &flock);
1597 if (error)
1598 goto out;
1600 if (filp->f_op && filp->f_op->lock) {
1601 error = filp->f_op->lock(filp, F_GETLK, &file_lock);
1602 if (file_lock.fl_ops && file_lock.fl_ops->fl_release_private)
1603 file_lock.fl_ops->fl_release_private(&file_lock);
1604 if (error < 0)
1605 goto out;
1606 else
1607 fl = (file_lock.fl_type == F_UNLCK ? NULL : &file_lock);
1608 } else {
1609 fl = (posix_test_lock(filp, &file_lock, &cfl) ? &cfl : NULL);
1612 flock.l_type = F_UNLCK;
1613 if (fl != NULL) {
1614 flock.l_pid = fl->fl_pid;
1615 #if BITS_PER_LONG == 32
1617 * Make sure we can represent the posix lock via
1618 * legacy 32bit flock.
1620 error = -EOVERFLOW;
1621 if (fl->fl_start > OFFT_OFFSET_MAX)
1622 goto out;
1623 if ((fl->fl_end != OFFSET_MAX)
1624 && (fl->fl_end > OFFT_OFFSET_MAX))
1625 goto out;
1626 #endif
1627 flock.l_start = fl->fl_start;
1628 flock.l_len = fl->fl_end == OFFSET_MAX ? 0 :
1629 fl->fl_end - fl->fl_start + 1;
1630 flock.l_whence = 0;
1631 flock.l_type = fl->fl_type;
1633 error = -EFAULT;
1634 if (!copy_to_user(l, &flock, sizeof(flock)))
1635 error = 0;
1636 out:
1637 return error;
1640 /* Apply the lock described by l to an open file descriptor.
1641 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1643 int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1644 struct flock __user *l)
1646 struct file_lock *file_lock = locks_alloc_lock();
1647 struct flock flock;
1648 struct inode *inode;
1649 int error;
1651 if (file_lock == NULL)
1652 return -ENOLCK;
1655 * This might block, so we do it before checking the inode.
1657 error = -EFAULT;
1658 if (copy_from_user(&flock, l, sizeof(flock)))
1659 goto out;
1661 inode = filp->f_dentry->d_inode;
1663 /* Don't allow mandatory locks on files that may be memory mapped
1664 * and shared.
1666 if (IS_MANDLOCK(inode) &&
1667 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1668 mapping_writably_mapped(filp->f_mapping)) {
1669 error = -EAGAIN;
1670 goto out;
1673 again:
1674 error = flock_to_posix_lock(filp, file_lock, &flock);
1675 if (error)
1676 goto out;
1677 if (cmd == F_SETLKW) {
1678 file_lock->fl_flags |= FL_SLEEP;
1681 error = -EBADF;
1682 switch (flock.l_type) {
1683 case F_RDLCK:
1684 if (!(filp->f_mode & FMODE_READ))
1685 goto out;
1686 break;
1687 case F_WRLCK:
1688 if (!(filp->f_mode & FMODE_WRITE))
1689 goto out;
1690 break;
1691 case F_UNLCK:
1692 break;
1693 default:
1694 error = -EINVAL;
1695 goto out;
1698 error = security_file_lock(filp, file_lock->fl_type);
1699 if (error)
1700 goto out;
1702 if (filp->f_op && filp->f_op->lock != NULL)
1703 error = filp->f_op->lock(filp, cmd, file_lock);
1704 else {
1705 for (;;) {
1706 error = posix_lock_file(filp, file_lock);
1707 if ((error != -EAGAIN) || (cmd == F_SETLK))
1708 break;
1709 error = wait_event_interruptible(file_lock->fl_wait,
1710 !file_lock->fl_next);
1711 if (!error)
1712 continue;
1714 locks_delete_block(file_lock);
1715 break;
1720 * Attempt to detect a close/fcntl race and recover by
1721 * releasing the lock that was just acquired.
1723 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1724 flock.l_type = F_UNLCK;
1725 goto again;
1728 out:
1729 locks_free_lock(file_lock);
1730 return error;
1733 #if BITS_PER_LONG == 32
1734 /* Report the first existing lock that would conflict with l.
1735 * This implements the F_GETLK command of fcntl().
1737 int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1739 struct file_lock *fl, cfl, file_lock;
1740 struct flock64 flock;
1741 int error;
1743 error = -EFAULT;
1744 if (copy_from_user(&flock, l, sizeof(flock)))
1745 goto out;
1746 error = -EINVAL;
1747 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1748 goto out;
1750 error = flock64_to_posix_lock(filp, &file_lock, &flock);
1751 if (error)
1752 goto out;
1754 if (filp->f_op && filp->f_op->lock) {
1755 error = filp->f_op->lock(filp, F_GETLK, &file_lock);
1756 if (file_lock.fl_ops && file_lock.fl_ops->fl_release_private)
1757 file_lock.fl_ops->fl_release_private(&file_lock);
1758 if (error < 0)
1759 goto out;
1760 else
1761 fl = (file_lock.fl_type == F_UNLCK ? NULL : &file_lock);
1762 } else {
1763 fl = (posix_test_lock(filp, &file_lock, &cfl) ? &cfl : NULL);
1766 flock.l_type = F_UNLCK;
1767 if (fl != NULL) {
1768 flock.l_pid = fl->fl_pid;
1769 flock.l_start = fl->fl_start;
1770 flock.l_len = fl->fl_end == OFFSET_MAX ? 0 :
1771 fl->fl_end - fl->fl_start + 1;
1772 flock.l_whence = 0;
1773 flock.l_type = fl->fl_type;
1775 error = -EFAULT;
1776 if (!copy_to_user(l, &flock, sizeof(flock)))
1777 error = 0;
1779 out:
1780 return error;
1783 /* Apply the lock described by l to an open file descriptor.
1784 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1786 int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
1787 struct flock64 __user *l)
1789 struct file_lock *file_lock = locks_alloc_lock();
1790 struct flock64 flock;
1791 struct inode *inode;
1792 int error;
1794 if (file_lock == NULL)
1795 return -ENOLCK;
1798 * This might block, so we do it before checking the inode.
1800 error = -EFAULT;
1801 if (copy_from_user(&flock, l, sizeof(flock)))
1802 goto out;
1804 inode = filp->f_dentry->d_inode;
1806 /* Don't allow mandatory locks on files that may be memory mapped
1807 * and shared.
1809 if (IS_MANDLOCK(inode) &&
1810 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1811 mapping_writably_mapped(filp->f_mapping)) {
1812 error = -EAGAIN;
1813 goto out;
1816 again:
1817 error = flock64_to_posix_lock(filp, file_lock, &flock);
1818 if (error)
1819 goto out;
1820 if (cmd == F_SETLKW64) {
1821 file_lock->fl_flags |= FL_SLEEP;
1824 error = -EBADF;
1825 switch (flock.l_type) {
1826 case F_RDLCK:
1827 if (!(filp->f_mode & FMODE_READ))
1828 goto out;
1829 break;
1830 case F_WRLCK:
1831 if (!(filp->f_mode & FMODE_WRITE))
1832 goto out;
1833 break;
1834 case F_UNLCK:
1835 break;
1836 default:
1837 error = -EINVAL;
1838 goto out;
1841 error = security_file_lock(filp, file_lock->fl_type);
1842 if (error)
1843 goto out;
1845 if (filp->f_op && filp->f_op->lock != NULL)
1846 error = filp->f_op->lock(filp, cmd, file_lock);
1847 else {
1848 for (;;) {
1849 error = posix_lock_file(filp, file_lock);
1850 if ((error != -EAGAIN) || (cmd == F_SETLK64))
1851 break;
1852 error = wait_event_interruptible(file_lock->fl_wait,
1853 !file_lock->fl_next);
1854 if (!error)
1855 continue;
1857 locks_delete_block(file_lock);
1858 break;
1863 * Attempt to detect a close/fcntl race and recover by
1864 * releasing the lock that was just acquired.
1866 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1867 flock.l_type = F_UNLCK;
1868 goto again;
1871 out:
1872 locks_free_lock(file_lock);
1873 return error;
1875 #endif /* BITS_PER_LONG == 32 */
1878 * This function is called when the file is being removed
1879 * from the task's fd array. POSIX locks belonging to this task
1880 * are deleted at this time.
1882 void locks_remove_posix(struct file *filp, fl_owner_t owner)
1884 struct file_lock lock, **before;
1887 * If there are no locks held on this file, we don't need to call
1888 * posix_lock_file(). Another process could be setting a lock on this
1889 * file at the same time, but we wouldn't remove that lock anyway.
1891 before = &filp->f_dentry->d_inode->i_flock;
1892 if (*before == NULL)
1893 return;
1895 lock.fl_type = F_UNLCK;
1896 lock.fl_flags = FL_POSIX;
1897 lock.fl_start = 0;
1898 lock.fl_end = OFFSET_MAX;
1899 lock.fl_owner = owner;
1900 lock.fl_pid = current->tgid;
1901 lock.fl_file = filp;
1902 lock.fl_ops = NULL;
1903 lock.fl_lmops = NULL;
1905 if (filp->f_op && filp->f_op->lock != NULL) {
1906 filp->f_op->lock(filp, F_SETLK, &lock);
1907 goto out;
1910 /* Can't use posix_lock_file here; we need to remove it no matter
1911 * which pid we have.
1913 lock_kernel();
1914 while (*before != NULL) {
1915 struct file_lock *fl = *before;
1916 if (IS_POSIX(fl) && posix_same_owner(fl, &lock)) {
1917 locks_delete_lock(before);
1918 continue;
1920 before = &fl->fl_next;
1922 unlock_kernel();
1923 out:
1924 if (lock.fl_ops && lock.fl_ops->fl_release_private)
1925 lock.fl_ops->fl_release_private(&lock);
1928 EXPORT_SYMBOL(locks_remove_posix);
1931 * This function is called on the last close of an open file.
1933 void locks_remove_flock(struct file *filp)
1935 struct inode * inode = filp->f_dentry->d_inode;
1936 struct file_lock *fl;
1937 struct file_lock **before;
1939 if (!inode->i_flock)
1940 return;
1942 if (filp->f_op && filp->f_op->flock) {
1943 struct file_lock fl = {
1944 .fl_pid = current->tgid,
1945 .fl_file = filp,
1946 .fl_flags = FL_FLOCK,
1947 .fl_type = F_UNLCK,
1948 .fl_end = OFFSET_MAX,
1950 filp->f_op->flock(filp, F_SETLKW, &fl);
1951 if (fl.fl_ops && fl.fl_ops->fl_release_private)
1952 fl.fl_ops->fl_release_private(&fl);
1955 lock_kernel();
1956 before = &inode->i_flock;
1958 while ((fl = *before) != NULL) {
1959 if (fl->fl_file == filp) {
1960 if (IS_FLOCK(fl)) {
1961 locks_delete_lock(before);
1962 continue;
1964 if (IS_LEASE(fl)) {
1965 lease_modify(before, F_UNLCK);
1966 continue;
1968 /* What? */
1969 BUG();
1971 before = &fl->fl_next;
1973 unlock_kernel();
1977 * posix_unblock_lock - stop waiting for a file lock
1978 * @filp: how the file was opened
1979 * @waiter: the lock which was waiting
1981 * lockd needs to block waiting for locks.
1984 posix_unblock_lock(struct file *filp, struct file_lock *waiter)
1986 int status = 0;
1988 lock_kernel();
1989 if (waiter->fl_next)
1990 __locks_delete_block(waiter);
1991 else
1992 status = -ENOENT;
1993 unlock_kernel();
1994 return status;
1997 EXPORT_SYMBOL(posix_unblock_lock);
1999 static void lock_get_status(char* out, struct file_lock *fl, int id, char *pfx)
2001 struct inode *inode = NULL;
2003 if (fl->fl_file != NULL)
2004 inode = fl->fl_file->f_dentry->d_inode;
2006 out += sprintf(out, "%d:%s ", id, pfx);
2007 if (IS_POSIX(fl)) {
2008 out += sprintf(out, "%6s %s ",
2009 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2010 (inode == NULL) ? "*NOINODE*" :
2011 (IS_MANDLOCK(inode) &&
2012 (inode->i_mode & (S_IXGRP | S_ISGID)) == S_ISGID) ?
2013 "MANDATORY" : "ADVISORY ");
2014 } else if (IS_FLOCK(fl)) {
2015 if (fl->fl_type & LOCK_MAND) {
2016 out += sprintf(out, "FLOCK MSNFS ");
2017 } else {
2018 out += sprintf(out, "FLOCK ADVISORY ");
2020 } else if (IS_LEASE(fl)) {
2021 out += sprintf(out, "LEASE ");
2022 if (fl->fl_type & F_INPROGRESS)
2023 out += sprintf(out, "BREAKING ");
2024 else if (fl->fl_file)
2025 out += sprintf(out, "ACTIVE ");
2026 else
2027 out += sprintf(out, "BREAKER ");
2028 } else {
2029 out += sprintf(out, "UNKNOWN UNKNOWN ");
2031 if (fl->fl_type & LOCK_MAND) {
2032 out += sprintf(out, "%s ",
2033 (fl->fl_type & LOCK_READ)
2034 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2035 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2036 } else {
2037 out += sprintf(out, "%s ",
2038 (fl->fl_type & F_INPROGRESS)
2039 ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ "
2040 : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ ");
2042 if (inode) {
2043 #ifdef WE_CAN_BREAK_LSLK_NOW
2044 out += sprintf(out, "%d %s:%ld ", fl->fl_pid,
2045 inode->i_sb->s_id, inode->i_ino);
2046 #else
2047 /* userspace relies on this representation of dev_t ;-( */
2048 out += sprintf(out, "%d %02x:%02x:%ld ", fl->fl_pid,
2049 MAJOR(inode->i_sb->s_dev),
2050 MINOR(inode->i_sb->s_dev), inode->i_ino);
2051 #endif
2052 } else {
2053 out += sprintf(out, "%d <none>:0 ", fl->fl_pid);
2055 if (IS_POSIX(fl)) {
2056 if (fl->fl_end == OFFSET_MAX)
2057 out += sprintf(out, "%Ld EOF\n", fl->fl_start);
2058 else
2059 out += sprintf(out, "%Ld %Ld\n", fl->fl_start,
2060 fl->fl_end);
2061 } else {
2062 out += sprintf(out, "0 EOF\n");
2066 static void move_lock_status(char **p, off_t* pos, off_t offset)
2068 int len;
2069 len = strlen(*p);
2070 if(*pos >= offset) {
2071 /* the complete line is valid */
2072 *p += len;
2073 *pos += len;
2074 return;
2076 if(*pos+len > offset) {
2077 /* use the second part of the line */
2078 int i = offset-*pos;
2079 memmove(*p,*p+i,len-i);
2080 *p += len-i;
2081 *pos += len;
2082 return;
2084 /* discard the complete line */
2085 *pos += len;
2089 * get_locks_status - reports lock usage in /proc/locks
2090 * @buffer: address in userspace to write into
2091 * @start: ?
2092 * @offset: how far we are through the buffer
2093 * @length: how much to read
2096 int get_locks_status(char *buffer, char **start, off_t offset, int length)
2098 struct list_head *tmp;
2099 char *q = buffer;
2100 off_t pos = 0;
2101 int i = 0;
2103 lock_kernel();
2104 list_for_each(tmp, &file_lock_list) {
2105 struct list_head *btmp;
2106 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
2107 lock_get_status(q, fl, ++i, "");
2108 move_lock_status(&q, &pos, offset);
2110 if(pos >= offset+length)
2111 goto done;
2113 list_for_each(btmp, &fl->fl_block) {
2114 struct file_lock *bfl = list_entry(btmp,
2115 struct file_lock, fl_block);
2116 lock_get_status(q, bfl, i, " ->");
2117 move_lock_status(&q, &pos, offset);
2119 if(pos >= offset+length)
2120 goto done;
2123 done:
2124 unlock_kernel();
2125 *start = buffer;
2126 if(q-buffer < length)
2127 return (q-buffer);
2128 return length;
2132 * lock_may_read - checks that the region is free of locks
2133 * @inode: the inode that is being read
2134 * @start: the first byte to read
2135 * @len: the number of bytes to read
2137 * Emulates Windows locking requirements. Whole-file
2138 * mandatory locks (share modes) can prohibit a read and
2139 * byte-range POSIX locks can prohibit a read if they overlap.
2141 * N.B. this function is only ever called
2142 * from knfsd and ownership of locks is never checked.
2144 int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2146 struct file_lock *fl;
2147 int result = 1;
2148 lock_kernel();
2149 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2150 if (IS_POSIX(fl)) {
2151 if (fl->fl_type == F_RDLCK)
2152 continue;
2153 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2154 continue;
2155 } else if (IS_FLOCK(fl)) {
2156 if (!(fl->fl_type & LOCK_MAND))
2157 continue;
2158 if (fl->fl_type & LOCK_READ)
2159 continue;
2160 } else
2161 continue;
2162 result = 0;
2163 break;
2165 unlock_kernel();
2166 return result;
2169 EXPORT_SYMBOL(lock_may_read);
2172 * lock_may_write - checks that the region is free of locks
2173 * @inode: the inode that is being written
2174 * @start: the first byte to write
2175 * @len: the number of bytes to write
2177 * Emulates Windows locking requirements. Whole-file
2178 * mandatory locks (share modes) can prohibit a write and
2179 * byte-range POSIX locks can prohibit a write if they overlap.
2181 * N.B. this function is only ever called
2182 * from knfsd and ownership of locks is never checked.
2184 int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2186 struct file_lock *fl;
2187 int result = 1;
2188 lock_kernel();
2189 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2190 if (IS_POSIX(fl)) {
2191 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2192 continue;
2193 } else if (IS_FLOCK(fl)) {
2194 if (!(fl->fl_type & LOCK_MAND))
2195 continue;
2196 if (fl->fl_type & LOCK_WRITE)
2197 continue;
2198 } else
2199 continue;
2200 result = 0;
2201 break;
2203 unlock_kernel();
2204 return result;
2207 EXPORT_SYMBOL(lock_may_write);
2209 static inline void __steal_locks(struct file *file, fl_owner_t from)
2211 struct inode *inode = file->f_dentry->d_inode;
2212 struct file_lock *fl = inode->i_flock;
2214 while (fl) {
2215 if (fl->fl_file == file && fl->fl_owner == from)
2216 fl->fl_owner = current->files;
2217 fl = fl->fl_next;
2221 /* When getting ready for executing a binary, we make sure that current
2222 * has a files_struct on its own. Before dropping the old files_struct,
2223 * we take over ownership of all locks for all file descriptors we own.
2224 * Note that we may accidentally steal a lock for a file that a sibling
2225 * has created since the unshare_files() call.
2227 void steal_locks(fl_owner_t from)
2229 struct files_struct *files = current->files;
2230 int i, j;
2231 struct fdtable *fdt;
2233 if (from == files)
2234 return;
2236 lock_kernel();
2237 j = 0;
2240 * We are not taking a ref to the file structures, so
2241 * we need to acquire ->file_lock.
2243 spin_lock(&files->file_lock);
2244 fdt = files_fdtable(files);
2245 for (;;) {
2246 unsigned long set;
2247 i = j * __NFDBITS;
2248 if (i >= fdt->max_fdset || i >= fdt->max_fds)
2249 break;
2250 set = fdt->open_fds->fds_bits[j++];
2251 while (set) {
2252 if (set & 1) {
2253 struct file *file = fdt->fd[i];
2254 if (file)
2255 __steal_locks(file, from);
2257 i++;
2258 set >>= 1;
2261 spin_unlock(&files->file_lock);
2262 unlock_kernel();
2264 EXPORT_SYMBOL(steal_locks);
2266 static int __init filelock_init(void)
2268 filelock_cache = kmem_cache_create("file_lock_cache",
2269 sizeof(struct file_lock), 0, SLAB_PANIC,
2270 init_once, NULL);
2271 return 0;
2274 core_initcall(filelock_init);