Linux 2.6.26-rc5
[linux-2.6/openmoko-kernel/knife-kernel.git] / fs / locks.c
blob11dbf08651b70861d31530ef73436dc2e3beffc1
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(struct kmem_cache *cache, 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;
565 if (fl->fl_ops && fl->fl_ops->fl_insert)
566 fl->fl_ops->fl_insert(fl);
570 * Delete a lock and then free it.
571 * Wake up processes that are blocked waiting for this lock,
572 * notify the FS that the lock has been cleared and
573 * finally free the lock.
575 static void locks_delete_lock(struct file_lock **thisfl_p)
577 struct file_lock *fl = *thisfl_p;
579 *thisfl_p = fl->fl_next;
580 fl->fl_next = NULL;
581 list_del_init(&fl->fl_link);
583 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
584 if (fl->fl_fasync != NULL) {
585 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
586 fl->fl_fasync = NULL;
589 if (fl->fl_ops && fl->fl_ops->fl_remove)
590 fl->fl_ops->fl_remove(fl);
592 if (fl->fl_nspid) {
593 put_pid(fl->fl_nspid);
594 fl->fl_nspid = NULL;
597 locks_wake_up_blocks(fl);
598 locks_free_lock(fl);
601 /* Determine if lock sys_fl blocks lock caller_fl. Common functionality
602 * checks for shared/exclusive status of overlapping locks.
604 static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
606 if (sys_fl->fl_type == F_WRLCK)
607 return 1;
608 if (caller_fl->fl_type == F_WRLCK)
609 return 1;
610 return 0;
613 /* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
614 * checking before calling the locks_conflict().
616 static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
618 /* POSIX locks owned by the same process do not conflict with
619 * each other.
621 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
622 return (0);
624 /* Check whether they overlap */
625 if (!locks_overlap(caller_fl, sys_fl))
626 return 0;
628 return (locks_conflict(caller_fl, sys_fl));
631 /* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
632 * checking before calling the locks_conflict().
634 static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
636 /* FLOCK locks referring to the same filp do not conflict with
637 * each other.
639 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
640 return (0);
641 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
642 return 0;
644 return (locks_conflict(caller_fl, sys_fl));
647 void
648 posix_test_lock(struct file *filp, struct file_lock *fl)
650 struct file_lock *cfl;
652 lock_kernel();
653 for (cfl = filp->f_path.dentry->d_inode->i_flock; cfl; cfl = cfl->fl_next) {
654 if (!IS_POSIX(cfl))
655 continue;
656 if (posix_locks_conflict(fl, cfl))
657 break;
659 if (cfl) {
660 __locks_copy_lock(fl, cfl);
661 if (cfl->fl_nspid)
662 fl->fl_pid = pid_vnr(cfl->fl_nspid);
663 } else
664 fl->fl_type = F_UNLCK;
665 unlock_kernel();
666 return;
668 EXPORT_SYMBOL(posix_test_lock);
671 * Deadlock detection:
673 * We attempt to detect deadlocks that are due purely to posix file
674 * locks.
676 * We assume that a task can be waiting for at most one lock at a time.
677 * So for any acquired lock, the process holding that lock may be
678 * waiting on at most one other lock. That lock in turns may be held by
679 * someone waiting for at most one other lock. Given a requested lock
680 * caller_fl which is about to wait for a conflicting lock block_fl, we
681 * follow this chain of waiters to ensure we are not about to create a
682 * cycle.
684 * Since we do this before we ever put a process to sleep on a lock, we
685 * are ensured that there is never a cycle; that is what guarantees that
686 * the while() loop in posix_locks_deadlock() eventually completes.
688 * Note: the above assumption may not be true when handling lock
689 * requests from a broken NFS client. It may also fail in the presence
690 * of tasks (such as posix threads) sharing the same open file table.
692 * To handle those cases, we just bail out after a few iterations.
695 #define MAX_DEADLK_ITERATIONS 10
697 /* Find a lock that the owner of the given block_fl is blocking on. */
698 static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
700 struct file_lock *fl;
702 list_for_each_entry(fl, &blocked_list, fl_link) {
703 if (posix_same_owner(fl, block_fl))
704 return fl->fl_next;
706 return NULL;
709 static int posix_locks_deadlock(struct file_lock *caller_fl,
710 struct file_lock *block_fl)
712 int i = 0;
714 while ((block_fl = what_owner_is_waiting_for(block_fl))) {
715 if (i++ > MAX_DEADLK_ITERATIONS)
716 return 0;
717 if (posix_same_owner(caller_fl, block_fl))
718 return 1;
720 return 0;
723 /* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
724 * after any leases, but before any posix locks.
726 * Note that if called with an FL_EXISTS argument, the caller may determine
727 * whether or not a lock was successfully freed by testing the return
728 * value for -ENOENT.
730 static int flock_lock_file(struct file *filp, struct file_lock *request)
732 struct file_lock *new_fl = NULL;
733 struct file_lock **before;
734 struct inode * inode = filp->f_path.dentry->d_inode;
735 int error = 0;
736 int found = 0;
738 lock_kernel();
739 if (request->fl_flags & FL_ACCESS)
740 goto find_conflict;
742 if (request->fl_type != F_UNLCK) {
743 error = -ENOMEM;
744 new_fl = locks_alloc_lock();
745 if (new_fl == NULL)
746 goto out;
747 error = 0;
750 for_each_lock(inode, before) {
751 struct file_lock *fl = *before;
752 if (IS_POSIX(fl))
753 break;
754 if (IS_LEASE(fl))
755 continue;
756 if (filp != fl->fl_file)
757 continue;
758 if (request->fl_type == fl->fl_type)
759 goto out;
760 found = 1;
761 locks_delete_lock(before);
762 break;
765 if (request->fl_type == F_UNLCK) {
766 if ((request->fl_flags & FL_EXISTS) && !found)
767 error = -ENOENT;
768 goto out;
772 * If a higher-priority process was blocked on the old file lock,
773 * give it the opportunity to lock the file.
775 if (found)
776 cond_resched_bkl();
778 find_conflict:
779 for_each_lock(inode, before) {
780 struct file_lock *fl = *before;
781 if (IS_POSIX(fl))
782 break;
783 if (IS_LEASE(fl))
784 continue;
785 if (!flock_locks_conflict(request, fl))
786 continue;
787 error = -EAGAIN;
788 if (request->fl_flags & FL_SLEEP)
789 locks_insert_block(fl, request);
790 goto out;
792 if (request->fl_flags & FL_ACCESS)
793 goto out;
794 locks_copy_lock(new_fl, request);
795 locks_insert_lock(before, new_fl);
796 new_fl = NULL;
797 error = 0;
799 out:
800 unlock_kernel();
801 if (new_fl)
802 locks_free_lock(new_fl);
803 return error;
806 static int __posix_lock_file(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
808 struct file_lock *fl;
809 struct file_lock *new_fl = NULL;
810 struct file_lock *new_fl2 = NULL;
811 struct file_lock *left = NULL;
812 struct file_lock *right = NULL;
813 struct file_lock **before;
814 int error, added = 0;
817 * We may need two file_lock structures for this operation,
818 * so we get them in advance to avoid races.
820 * In some cases we can be sure, that no new locks will be needed
822 if (!(request->fl_flags & FL_ACCESS) &&
823 (request->fl_type != F_UNLCK ||
824 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
825 new_fl = locks_alloc_lock();
826 new_fl2 = locks_alloc_lock();
829 lock_kernel();
830 if (request->fl_type != F_UNLCK) {
831 for_each_lock(inode, before) {
832 fl = *before;
833 if (!IS_POSIX(fl))
834 continue;
835 if (!posix_locks_conflict(request, fl))
836 continue;
837 if (conflock)
838 __locks_copy_lock(conflock, fl);
839 error = -EAGAIN;
840 if (!(request->fl_flags & FL_SLEEP))
841 goto out;
842 error = -EDEADLK;
843 if (posix_locks_deadlock(request, fl))
844 goto out;
845 error = -EAGAIN;
846 locks_insert_block(fl, request);
847 goto out;
851 /* If we're just looking for a conflict, we're done. */
852 error = 0;
853 if (request->fl_flags & FL_ACCESS)
854 goto out;
857 * Find the first old lock with the same owner as the new lock.
860 before = &inode->i_flock;
862 /* First skip locks owned by other processes. */
863 while ((fl = *before) && (!IS_POSIX(fl) ||
864 !posix_same_owner(request, fl))) {
865 before = &fl->fl_next;
868 /* Process locks with this owner. */
869 while ((fl = *before) && posix_same_owner(request, fl)) {
870 /* Detect adjacent or overlapping regions (if same lock type)
872 if (request->fl_type == fl->fl_type) {
873 /* In all comparisons of start vs end, use
874 * "start - 1" rather than "end + 1". If end
875 * is OFFSET_MAX, end + 1 will become negative.
877 if (fl->fl_end < request->fl_start - 1)
878 goto next_lock;
879 /* If the next lock in the list has entirely bigger
880 * addresses than the new one, insert the lock here.
882 if (fl->fl_start - 1 > request->fl_end)
883 break;
885 /* If we come here, the new and old lock are of the
886 * same type and adjacent or overlapping. Make one
887 * lock yielding from the lower start address of both
888 * locks to the higher end address.
890 if (fl->fl_start > request->fl_start)
891 fl->fl_start = request->fl_start;
892 else
893 request->fl_start = fl->fl_start;
894 if (fl->fl_end < request->fl_end)
895 fl->fl_end = request->fl_end;
896 else
897 request->fl_end = fl->fl_end;
898 if (added) {
899 locks_delete_lock(before);
900 continue;
902 request = fl;
903 added = 1;
905 else {
906 /* Processing for different lock types is a bit
907 * more complex.
909 if (fl->fl_end < request->fl_start)
910 goto next_lock;
911 if (fl->fl_start > request->fl_end)
912 break;
913 if (request->fl_type == F_UNLCK)
914 added = 1;
915 if (fl->fl_start < request->fl_start)
916 left = fl;
917 /* If the next lock in the list has a higher end
918 * address than the new one, insert the new one here.
920 if (fl->fl_end > request->fl_end) {
921 right = fl;
922 break;
924 if (fl->fl_start >= request->fl_start) {
925 /* The new lock completely replaces an old
926 * one (This may happen several times).
928 if (added) {
929 locks_delete_lock(before);
930 continue;
932 /* Replace the old lock with the new one.
933 * Wake up anybody waiting for the old one,
934 * as the change in lock type might satisfy
935 * their needs.
937 locks_wake_up_blocks(fl);
938 fl->fl_start = request->fl_start;
939 fl->fl_end = request->fl_end;
940 fl->fl_type = request->fl_type;
941 locks_release_private(fl);
942 locks_copy_private(fl, request);
943 request = fl;
944 added = 1;
947 /* Go on to next lock.
949 next_lock:
950 before = &fl->fl_next;
954 * The above code only modifies existing locks in case of
955 * merging or replacing. If new lock(s) need to be inserted
956 * all modifications are done bellow this, so it's safe yet to
957 * bail out.
959 error = -ENOLCK; /* "no luck" */
960 if (right && left == right && !new_fl2)
961 goto out;
963 error = 0;
964 if (!added) {
965 if (request->fl_type == F_UNLCK) {
966 if (request->fl_flags & FL_EXISTS)
967 error = -ENOENT;
968 goto out;
971 if (!new_fl) {
972 error = -ENOLCK;
973 goto out;
975 locks_copy_lock(new_fl, request);
976 locks_insert_lock(before, new_fl);
977 new_fl = NULL;
979 if (right) {
980 if (left == right) {
981 /* The new lock breaks the old one in two pieces,
982 * so we have to use the second new lock.
984 left = new_fl2;
985 new_fl2 = NULL;
986 locks_copy_lock(left, right);
987 locks_insert_lock(before, left);
989 right->fl_start = request->fl_end + 1;
990 locks_wake_up_blocks(right);
992 if (left) {
993 left->fl_end = request->fl_start - 1;
994 locks_wake_up_blocks(left);
996 out:
997 unlock_kernel();
999 * Free any unused locks.
1001 if (new_fl)
1002 locks_free_lock(new_fl);
1003 if (new_fl2)
1004 locks_free_lock(new_fl2);
1005 return error;
1009 * posix_lock_file - Apply a POSIX-style lock to a file
1010 * @filp: The file to apply the lock to
1011 * @fl: The lock to be applied
1012 * @conflock: Place to return a copy of the conflicting lock, if found.
1014 * Add a POSIX style lock to a file.
1015 * We merge adjacent & overlapping locks whenever possible.
1016 * POSIX locks are sorted by owner task, then by starting address
1018 * Note that if called with an FL_EXISTS argument, the caller may determine
1019 * whether or not a lock was successfully freed by testing the return
1020 * value for -ENOENT.
1022 int posix_lock_file(struct file *filp, struct file_lock *fl,
1023 struct file_lock *conflock)
1025 return __posix_lock_file(filp->f_path.dentry->d_inode, fl, conflock);
1027 EXPORT_SYMBOL(posix_lock_file);
1030 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1031 * @filp: The file to apply the lock to
1032 * @fl: The lock to be applied
1034 * Add a POSIX style lock to a file.
1035 * We merge adjacent & overlapping locks whenever possible.
1036 * POSIX locks are sorted by owner task, then by starting address
1038 int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1040 int error;
1041 might_sleep ();
1042 for (;;) {
1043 error = posix_lock_file(filp, fl, NULL);
1044 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1045 break;
1046 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1047 if (!error)
1048 continue;
1050 locks_delete_block(fl);
1051 break;
1053 return error;
1055 EXPORT_SYMBOL(posix_lock_file_wait);
1058 * locks_mandatory_locked - Check for an active lock
1059 * @inode: the file to check
1061 * Searches the inode's list of locks to find any POSIX locks which conflict.
1062 * This function is called from locks_verify_locked() only.
1064 int locks_mandatory_locked(struct inode *inode)
1066 fl_owner_t owner = current->files;
1067 struct file_lock *fl;
1070 * Search the lock list for this inode for any POSIX locks.
1072 lock_kernel();
1073 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1074 if (!IS_POSIX(fl))
1075 continue;
1076 if (fl->fl_owner != owner)
1077 break;
1079 unlock_kernel();
1080 return fl ? -EAGAIN : 0;
1084 * locks_mandatory_area - Check for a conflicting lock
1085 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1086 * for shared
1087 * @inode: the file to check
1088 * @filp: how the file was opened (if it was)
1089 * @offset: start of area to check
1090 * @count: length of area to check
1092 * Searches the inode's list of locks to find any POSIX locks which conflict.
1093 * This function is called from rw_verify_area() and
1094 * locks_verify_truncate().
1096 int locks_mandatory_area(int read_write, struct inode *inode,
1097 struct file *filp, loff_t offset,
1098 size_t count)
1100 struct file_lock fl;
1101 int error;
1103 locks_init_lock(&fl);
1104 fl.fl_owner = current->files;
1105 fl.fl_pid = current->tgid;
1106 fl.fl_file = filp;
1107 fl.fl_flags = FL_POSIX | FL_ACCESS;
1108 if (filp && !(filp->f_flags & O_NONBLOCK))
1109 fl.fl_flags |= FL_SLEEP;
1110 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1111 fl.fl_start = offset;
1112 fl.fl_end = offset + count - 1;
1114 for (;;) {
1115 error = __posix_lock_file(inode, &fl, NULL);
1116 if (error != -EAGAIN)
1117 break;
1118 if (!(fl.fl_flags & FL_SLEEP))
1119 break;
1120 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1121 if (!error) {
1123 * If we've been sleeping someone might have
1124 * changed the permissions behind our back.
1126 if (__mandatory_lock(inode))
1127 continue;
1130 locks_delete_block(&fl);
1131 break;
1134 return error;
1137 EXPORT_SYMBOL(locks_mandatory_area);
1139 /* We already had a lease on this file; just change its type */
1140 int lease_modify(struct file_lock **before, int arg)
1142 struct file_lock *fl = *before;
1143 int error = assign_type(fl, arg);
1145 if (error)
1146 return error;
1147 locks_wake_up_blocks(fl);
1148 if (arg == F_UNLCK)
1149 locks_delete_lock(before);
1150 return 0;
1153 EXPORT_SYMBOL(lease_modify);
1155 static void time_out_leases(struct inode *inode)
1157 struct file_lock **before;
1158 struct file_lock *fl;
1160 before = &inode->i_flock;
1161 while ((fl = *before) && IS_LEASE(fl) && (fl->fl_type & F_INPROGRESS)) {
1162 if ((fl->fl_break_time == 0)
1163 || time_before(jiffies, fl->fl_break_time)) {
1164 before = &fl->fl_next;
1165 continue;
1167 lease_modify(before, fl->fl_type & ~F_INPROGRESS);
1168 if (fl == *before) /* lease_modify may have freed fl */
1169 before = &fl->fl_next;
1174 * __break_lease - revoke all outstanding leases on file
1175 * @inode: the inode of the file to return
1176 * @mode: the open mode (read or write)
1178 * break_lease (inlined for speed) has checked there already is at least
1179 * some kind of lock (maybe a lease) on this file. Leases are broken on
1180 * a call to open() or truncate(). This function can sleep unless you
1181 * specified %O_NONBLOCK to your open().
1183 int __break_lease(struct inode *inode, unsigned int mode)
1185 int error = 0, future;
1186 struct file_lock *new_fl, *flock;
1187 struct file_lock *fl;
1188 unsigned long break_time;
1189 int i_have_this_lease = 0;
1191 new_fl = lease_alloc(NULL, mode & FMODE_WRITE ? F_WRLCK : F_RDLCK);
1193 lock_kernel();
1195 time_out_leases(inode);
1197 flock = inode->i_flock;
1198 if ((flock == NULL) || !IS_LEASE(flock))
1199 goto out;
1201 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1202 if (fl->fl_owner == current->files)
1203 i_have_this_lease = 1;
1205 if (mode & FMODE_WRITE) {
1206 /* If we want write access, we have to revoke any lease. */
1207 future = F_UNLCK | F_INPROGRESS;
1208 } else if (flock->fl_type & F_INPROGRESS) {
1209 /* If the lease is already being broken, we just leave it */
1210 future = flock->fl_type;
1211 } else if (flock->fl_type & F_WRLCK) {
1212 /* Downgrade the exclusive lease to a read-only lease. */
1213 future = F_RDLCK | F_INPROGRESS;
1214 } else {
1215 /* the existing lease was read-only, so we can read too. */
1216 goto out;
1219 if (IS_ERR(new_fl) && !i_have_this_lease
1220 && ((mode & O_NONBLOCK) == 0)) {
1221 error = PTR_ERR(new_fl);
1222 goto out;
1225 break_time = 0;
1226 if (lease_break_time > 0) {
1227 break_time = jiffies + lease_break_time * HZ;
1228 if (break_time == 0)
1229 break_time++; /* so that 0 means no break time */
1232 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1233 if (fl->fl_type != future) {
1234 fl->fl_type = future;
1235 fl->fl_break_time = break_time;
1236 /* lease must have lmops break callback */
1237 fl->fl_lmops->fl_break(fl);
1241 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1242 error = -EWOULDBLOCK;
1243 goto out;
1246 restart:
1247 break_time = flock->fl_break_time;
1248 if (break_time != 0) {
1249 break_time -= jiffies;
1250 if (break_time == 0)
1251 break_time++;
1253 locks_insert_block(flock, new_fl);
1254 error = wait_event_interruptible_timeout(new_fl->fl_wait,
1255 !new_fl->fl_next, break_time);
1256 __locks_delete_block(new_fl);
1257 if (error >= 0) {
1258 if (error == 0)
1259 time_out_leases(inode);
1260 /* Wait for the next lease that has not been broken yet */
1261 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1262 flock = flock->fl_next) {
1263 if (flock->fl_type & F_INPROGRESS)
1264 goto restart;
1266 error = 0;
1269 out:
1270 unlock_kernel();
1271 if (!IS_ERR(new_fl))
1272 locks_free_lock(new_fl);
1273 return error;
1276 EXPORT_SYMBOL(__break_lease);
1279 * lease_get_mtime - get the last modified time of an inode
1280 * @inode: the inode
1281 * @time: pointer to a timespec which will contain the last modified time
1283 * This is to force NFS clients to flush their caches for files with
1284 * exclusive leases. The justification is that if someone has an
1285 * exclusive lease, then they could be modifying it.
1287 void lease_get_mtime(struct inode *inode, struct timespec *time)
1289 struct file_lock *flock = inode->i_flock;
1290 if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK))
1291 *time = current_fs_time(inode->i_sb);
1292 else
1293 *time = inode->i_mtime;
1296 EXPORT_SYMBOL(lease_get_mtime);
1299 * fcntl_getlease - Enquire what lease is currently active
1300 * @filp: the file
1302 * The value returned by this function will be one of
1303 * (if no lease break is pending):
1305 * %F_RDLCK to indicate a shared lease is held.
1307 * %F_WRLCK to indicate an exclusive lease is held.
1309 * %F_UNLCK to indicate no lease is held.
1311 * (if a lease break is pending):
1313 * %F_RDLCK to indicate an exclusive lease needs to be
1314 * changed to a shared lease (or removed).
1316 * %F_UNLCK to indicate the lease needs to be removed.
1318 * XXX: sfr & willy disagree over whether F_INPROGRESS
1319 * should be returned to userspace.
1321 int fcntl_getlease(struct file *filp)
1323 struct file_lock *fl;
1324 int type = F_UNLCK;
1326 lock_kernel();
1327 time_out_leases(filp->f_path.dentry->d_inode);
1328 for (fl = filp->f_path.dentry->d_inode->i_flock; fl && IS_LEASE(fl);
1329 fl = fl->fl_next) {
1330 if (fl->fl_file == filp) {
1331 type = fl->fl_type & ~F_INPROGRESS;
1332 break;
1335 unlock_kernel();
1336 return type;
1340 * generic_setlease - sets a lease on an open file
1341 * @filp: file pointer
1342 * @arg: type of lease to obtain
1343 * @flp: input - file_lock to use, output - file_lock inserted
1345 * The (input) flp->fl_lmops->fl_break function is required
1346 * by break_lease().
1348 * Called with kernel lock held.
1350 int generic_setlease(struct file *filp, long arg, struct file_lock **flp)
1352 struct file_lock *fl, **before, **my_before = NULL, *lease;
1353 struct file_lock *new_fl = NULL;
1354 struct dentry *dentry = filp->f_path.dentry;
1355 struct inode *inode = dentry->d_inode;
1356 int error, rdlease_count = 0, wrlease_count = 0;
1358 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1359 return -EACCES;
1360 if (!S_ISREG(inode->i_mode))
1361 return -EINVAL;
1362 error = security_file_lock(filp, arg);
1363 if (error)
1364 return error;
1366 time_out_leases(inode);
1368 BUG_ON(!(*flp)->fl_lmops->fl_break);
1370 lease = *flp;
1372 if (arg != F_UNLCK) {
1373 error = -ENOMEM;
1374 new_fl = locks_alloc_lock();
1375 if (new_fl == NULL)
1376 goto out;
1378 error = -EAGAIN;
1379 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1380 goto out;
1381 if ((arg == F_WRLCK)
1382 && ((atomic_read(&dentry->d_count) > 1)
1383 || (atomic_read(&inode->i_count) > 1)))
1384 goto out;
1388 * At this point, we know that if there is an exclusive
1389 * lease on this file, then we hold it on this filp
1390 * (otherwise our open of this file would have blocked).
1391 * And if we are trying to acquire an exclusive lease,
1392 * then the file is not open by anyone (including us)
1393 * except for this filp.
1395 for (before = &inode->i_flock;
1396 ((fl = *before) != NULL) && IS_LEASE(fl);
1397 before = &fl->fl_next) {
1398 if (lease->fl_lmops->fl_mylease(fl, lease))
1399 my_before = before;
1400 else if (fl->fl_type == (F_INPROGRESS | F_UNLCK))
1402 * Someone is in the process of opening this
1403 * file for writing so we may not take an
1404 * exclusive lease on it.
1406 wrlease_count++;
1407 else
1408 rdlease_count++;
1411 error = -EAGAIN;
1412 if ((arg == F_RDLCK && (wrlease_count > 0)) ||
1413 (arg == F_WRLCK && ((rdlease_count + wrlease_count) > 0)))
1414 goto out;
1416 if (my_before != NULL) {
1417 *flp = *my_before;
1418 error = lease->fl_lmops->fl_change(my_before, arg);
1419 goto out;
1422 error = 0;
1423 if (arg == F_UNLCK)
1424 goto out;
1426 error = -EINVAL;
1427 if (!leases_enable)
1428 goto out;
1430 locks_copy_lock(new_fl, lease);
1431 locks_insert_lock(before, new_fl);
1433 *flp = new_fl;
1434 return 0;
1436 out:
1437 if (new_fl != NULL)
1438 locks_free_lock(new_fl);
1439 return error;
1441 EXPORT_SYMBOL(generic_setlease);
1444 * vfs_setlease - sets a lease on an open file
1445 * @filp: file pointer
1446 * @arg: type of lease to obtain
1447 * @lease: file_lock to use
1449 * Call this to establish a lease on the file.
1450 * The (*lease)->fl_lmops->fl_break operation must be set; if not,
1451 * break_lease will oops!
1453 * This will call the filesystem's setlease file method, if
1454 * defined. Note that there is no getlease method; instead, the
1455 * filesystem setlease method should call back to setlease() to
1456 * add a lease to the inode's lease list, where fcntl_getlease() can
1457 * find it. Since fcntl_getlease() only reports whether the current
1458 * task holds a lease, a cluster filesystem need only do this for
1459 * leases held by processes on this node.
1461 * There is also no break_lease method; filesystems that
1462 * handle their own leases shoud break leases themselves from the
1463 * filesystem's open, create, and (on truncate) setattr methods.
1465 * Warning: the only current setlease methods exist only to disable
1466 * leases in certain cases. More vfs changes may be required to
1467 * allow a full filesystem lease implementation.
1470 int vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
1472 int error;
1474 lock_kernel();
1475 if (filp->f_op && filp->f_op->setlease)
1476 error = filp->f_op->setlease(filp, arg, lease);
1477 else
1478 error = generic_setlease(filp, arg, lease);
1479 unlock_kernel();
1481 return error;
1483 EXPORT_SYMBOL_GPL(vfs_setlease);
1486 * fcntl_setlease - sets a lease on an open file
1487 * @fd: open file descriptor
1488 * @filp: file pointer
1489 * @arg: type of lease to obtain
1491 * Call this fcntl to establish a lease on the file.
1492 * Note that you also need to call %F_SETSIG to
1493 * receive a signal when the lease is broken.
1495 int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1497 struct file_lock fl, *flp = &fl;
1498 struct inode *inode = filp->f_path.dentry->d_inode;
1499 int error;
1501 locks_init_lock(&fl);
1502 error = lease_init(filp, arg, &fl);
1503 if (error)
1504 return error;
1506 lock_kernel();
1508 error = vfs_setlease(filp, arg, &flp);
1509 if (error || arg == F_UNLCK)
1510 goto out_unlock;
1512 error = fasync_helper(fd, filp, 1, &flp->fl_fasync);
1513 if (error < 0) {
1514 /* remove lease just inserted by setlease */
1515 flp->fl_type = F_UNLCK | F_INPROGRESS;
1516 flp->fl_break_time = jiffies - 10;
1517 time_out_leases(inode);
1518 goto out_unlock;
1521 error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
1522 out_unlock:
1523 unlock_kernel();
1524 return error;
1528 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1529 * @filp: The file to apply the lock to
1530 * @fl: The lock to be applied
1532 * Add a FLOCK style lock to a file.
1534 int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1536 int error;
1537 might_sleep();
1538 for (;;) {
1539 error = flock_lock_file(filp, fl);
1540 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1541 break;
1542 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1543 if (!error)
1544 continue;
1546 locks_delete_block(fl);
1547 break;
1549 return error;
1552 EXPORT_SYMBOL(flock_lock_file_wait);
1555 * sys_flock: - flock() system call.
1556 * @fd: the file descriptor to lock.
1557 * @cmd: the type of lock to apply.
1559 * Apply a %FL_FLOCK style lock to an open file descriptor.
1560 * The @cmd can be one of
1562 * %LOCK_SH -- a shared lock.
1564 * %LOCK_EX -- an exclusive lock.
1566 * %LOCK_UN -- remove an existing lock.
1568 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1570 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1571 * processes read and write access respectively.
1573 asmlinkage long sys_flock(unsigned int fd, unsigned int cmd)
1575 struct file *filp;
1576 struct file_lock *lock;
1577 int can_sleep, unlock;
1578 int error;
1580 error = -EBADF;
1581 filp = fget(fd);
1582 if (!filp)
1583 goto out;
1585 can_sleep = !(cmd & LOCK_NB);
1586 cmd &= ~LOCK_NB;
1587 unlock = (cmd == LOCK_UN);
1589 if (!unlock && !(cmd & LOCK_MAND) && !(filp->f_mode & 3))
1590 goto out_putf;
1592 error = flock_make_lock(filp, &lock, cmd);
1593 if (error)
1594 goto out_putf;
1595 if (can_sleep)
1596 lock->fl_flags |= FL_SLEEP;
1598 error = security_file_lock(filp, cmd);
1599 if (error)
1600 goto out_free;
1602 if (filp->f_op && filp->f_op->flock)
1603 error = filp->f_op->flock(filp,
1604 (can_sleep) ? F_SETLKW : F_SETLK,
1605 lock);
1606 else
1607 error = flock_lock_file_wait(filp, lock);
1609 out_free:
1610 locks_free_lock(lock);
1612 out_putf:
1613 fput(filp);
1614 out:
1615 return error;
1619 * vfs_test_lock - test file byte range lock
1620 * @filp: The file to test lock for
1621 * @fl: The lock to test; also used to hold result
1623 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
1624 * setting conf->fl_type to something other than F_UNLCK.
1626 int vfs_test_lock(struct file *filp, struct file_lock *fl)
1628 if (filp->f_op && filp->f_op->lock)
1629 return filp->f_op->lock(filp, F_GETLK, fl);
1630 posix_test_lock(filp, fl);
1631 return 0;
1633 EXPORT_SYMBOL_GPL(vfs_test_lock);
1635 static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
1637 flock->l_pid = fl->fl_pid;
1638 #if BITS_PER_LONG == 32
1640 * Make sure we can represent the posix lock via
1641 * legacy 32bit flock.
1643 if (fl->fl_start > OFFT_OFFSET_MAX)
1644 return -EOVERFLOW;
1645 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
1646 return -EOVERFLOW;
1647 #endif
1648 flock->l_start = fl->fl_start;
1649 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1650 fl->fl_end - fl->fl_start + 1;
1651 flock->l_whence = 0;
1652 flock->l_type = fl->fl_type;
1653 return 0;
1656 #if BITS_PER_LONG == 32
1657 static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
1659 flock->l_pid = fl->fl_pid;
1660 flock->l_start = fl->fl_start;
1661 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1662 fl->fl_end - fl->fl_start + 1;
1663 flock->l_whence = 0;
1664 flock->l_type = fl->fl_type;
1666 #endif
1668 /* Report the first existing lock that would conflict with l.
1669 * This implements the F_GETLK command of fcntl().
1671 int fcntl_getlk(struct file *filp, struct flock __user *l)
1673 struct file_lock file_lock;
1674 struct flock flock;
1675 int error;
1677 error = -EFAULT;
1678 if (copy_from_user(&flock, l, sizeof(flock)))
1679 goto out;
1680 error = -EINVAL;
1681 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1682 goto out;
1684 error = flock_to_posix_lock(filp, &file_lock, &flock);
1685 if (error)
1686 goto out;
1688 error = vfs_test_lock(filp, &file_lock);
1689 if (error)
1690 goto out;
1692 flock.l_type = file_lock.fl_type;
1693 if (file_lock.fl_type != F_UNLCK) {
1694 error = posix_lock_to_flock(&flock, &file_lock);
1695 if (error)
1696 goto out;
1698 error = -EFAULT;
1699 if (!copy_to_user(l, &flock, sizeof(flock)))
1700 error = 0;
1701 out:
1702 return error;
1706 * vfs_lock_file - file byte range lock
1707 * @filp: The file to apply the lock to
1708 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
1709 * @fl: The lock to be applied
1710 * @conf: Place to return a copy of the conflicting lock, if found.
1712 * A caller that doesn't care about the conflicting lock may pass NULL
1713 * as the final argument.
1715 * If the filesystem defines a private ->lock() method, then @conf will
1716 * be left unchanged; so a caller that cares should initialize it to
1717 * some acceptable default.
1719 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
1720 * locks, the ->lock() interface may return asynchronously, before the lock has
1721 * been granted or denied by the underlying filesystem, if (and only if)
1722 * fl_grant is set. Callers expecting ->lock() to return asynchronously
1723 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
1724 * the request is for a blocking lock. When ->lock() does return asynchronously,
1725 * it must return -EINPROGRESS, and call ->fl_grant() when the lock
1726 * request completes.
1727 * If the request is for non-blocking lock the file system should return
1728 * -EINPROGRESS then try to get the lock and call the callback routine with
1729 * the result. If the request timed out the callback routine will return a
1730 * nonzero return code and the file system should release the lock. The file
1731 * system is also responsible to keep a corresponding posix lock when it
1732 * grants a lock so the VFS can find out which locks are locally held and do
1733 * the correct lock cleanup when required.
1734 * The underlying filesystem must not drop the kernel lock or call
1735 * ->fl_grant() before returning to the caller with a -EINPROGRESS
1736 * return code.
1738 int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
1740 if (filp->f_op && filp->f_op->lock)
1741 return filp->f_op->lock(filp, cmd, fl);
1742 else
1743 return posix_lock_file(filp, fl, conf);
1745 EXPORT_SYMBOL_GPL(vfs_lock_file);
1747 /* Apply the lock described by l to an open file descriptor.
1748 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1750 int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1751 struct flock __user *l)
1753 struct file_lock *file_lock = locks_alloc_lock();
1754 struct flock flock;
1755 struct inode *inode;
1756 struct file *f;
1757 int error;
1759 if (file_lock == NULL)
1760 return -ENOLCK;
1763 * This might block, so we do it before checking the inode.
1765 error = -EFAULT;
1766 if (copy_from_user(&flock, l, sizeof(flock)))
1767 goto out;
1769 inode = filp->f_path.dentry->d_inode;
1771 /* Don't allow mandatory locks on files that may be memory mapped
1772 * and shared.
1774 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
1775 error = -EAGAIN;
1776 goto out;
1779 again:
1780 error = flock_to_posix_lock(filp, file_lock, &flock);
1781 if (error)
1782 goto out;
1783 if (cmd == F_SETLKW) {
1784 file_lock->fl_flags |= FL_SLEEP;
1787 error = -EBADF;
1788 switch (flock.l_type) {
1789 case F_RDLCK:
1790 if (!(filp->f_mode & FMODE_READ))
1791 goto out;
1792 break;
1793 case F_WRLCK:
1794 if (!(filp->f_mode & FMODE_WRITE))
1795 goto out;
1796 break;
1797 case F_UNLCK:
1798 break;
1799 default:
1800 error = -EINVAL;
1801 goto out;
1804 error = security_file_lock(filp, file_lock->fl_type);
1805 if (error)
1806 goto out;
1808 if (filp->f_op && filp->f_op->lock != NULL)
1809 error = filp->f_op->lock(filp, cmd, file_lock);
1810 else {
1811 for (;;) {
1812 error = posix_lock_file(filp, file_lock, NULL);
1813 if (error != -EAGAIN || cmd == F_SETLK)
1814 break;
1815 error = wait_event_interruptible(file_lock->fl_wait,
1816 !file_lock->fl_next);
1817 if (!error)
1818 continue;
1820 locks_delete_block(file_lock);
1821 break;
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 = security_file_lock(filp, file_lock->fl_type);
1942 if (error)
1943 goto out;
1945 if (filp->f_op && filp->f_op->lock != NULL)
1946 error = filp->f_op->lock(filp, cmd, file_lock);
1947 else {
1948 for (;;) {
1949 error = posix_lock_file(filp, file_lock, NULL);
1950 if (error != -EAGAIN || cmd == F_SETLK64)
1951 break;
1952 error = wait_event_interruptible(file_lock->fl_wait,
1953 !file_lock->fl_next);
1954 if (!error)
1955 continue;
1957 locks_delete_block(file_lock);
1958 break;
1963 * Attempt to detect a close/fcntl race and recover by
1964 * releasing the lock that was just acquired.
1966 spin_lock(&current->files->file_lock);
1967 f = fcheck(fd);
1968 spin_unlock(&current->files->file_lock);
1969 if (!error && f != filp && flock.l_type != F_UNLCK) {
1970 flock.l_type = F_UNLCK;
1971 goto again;
1974 out:
1975 locks_free_lock(file_lock);
1976 return error;
1978 #endif /* BITS_PER_LONG == 32 */
1981 * This function is called when the file is being removed
1982 * from the task's fd array. POSIX locks belonging to this task
1983 * are deleted at this time.
1985 void locks_remove_posix(struct file *filp, fl_owner_t owner)
1987 struct file_lock lock;
1990 * If there are no locks held on this file, we don't need to call
1991 * posix_lock_file(). Another process could be setting a lock on this
1992 * file at the same time, but we wouldn't remove that lock anyway.
1994 if (!filp->f_path.dentry->d_inode->i_flock)
1995 return;
1997 lock.fl_type = F_UNLCK;
1998 lock.fl_flags = FL_POSIX | FL_CLOSE;
1999 lock.fl_start = 0;
2000 lock.fl_end = OFFSET_MAX;
2001 lock.fl_owner = owner;
2002 lock.fl_pid = current->tgid;
2003 lock.fl_file = filp;
2004 lock.fl_ops = NULL;
2005 lock.fl_lmops = NULL;
2007 vfs_lock_file(filp, F_SETLK, &lock, NULL);
2009 if (lock.fl_ops && lock.fl_ops->fl_release_private)
2010 lock.fl_ops->fl_release_private(&lock);
2013 EXPORT_SYMBOL(locks_remove_posix);
2016 * This function is called on the last close of an open file.
2018 void locks_remove_flock(struct file *filp)
2020 struct inode * inode = filp->f_path.dentry->d_inode;
2021 struct file_lock *fl;
2022 struct file_lock **before;
2024 if (!inode->i_flock)
2025 return;
2027 if (filp->f_op && filp->f_op->flock) {
2028 struct file_lock fl = {
2029 .fl_pid = current->tgid,
2030 .fl_file = filp,
2031 .fl_flags = FL_FLOCK,
2032 .fl_type = F_UNLCK,
2033 .fl_end = OFFSET_MAX,
2035 filp->f_op->flock(filp, F_SETLKW, &fl);
2036 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2037 fl.fl_ops->fl_release_private(&fl);
2040 lock_kernel();
2041 before = &inode->i_flock;
2043 while ((fl = *before) != NULL) {
2044 if (fl->fl_file == filp) {
2045 if (IS_FLOCK(fl)) {
2046 locks_delete_lock(before);
2047 continue;
2049 if (IS_LEASE(fl)) {
2050 lease_modify(before, F_UNLCK);
2051 continue;
2053 /* What? */
2054 BUG();
2056 before = &fl->fl_next;
2058 unlock_kernel();
2062 * posix_unblock_lock - stop waiting for a file lock
2063 * @filp: how the file was opened
2064 * @waiter: the lock which was waiting
2066 * lockd needs to block waiting for locks.
2069 posix_unblock_lock(struct file *filp, struct file_lock *waiter)
2071 int status = 0;
2073 lock_kernel();
2074 if (waiter->fl_next)
2075 __locks_delete_block(waiter);
2076 else
2077 status = -ENOENT;
2078 unlock_kernel();
2079 return status;
2082 EXPORT_SYMBOL(posix_unblock_lock);
2085 * vfs_cancel_lock - file byte range unblock lock
2086 * @filp: The file to apply the unblock to
2087 * @fl: The lock to be unblocked
2089 * Used by lock managers to cancel blocked requests
2091 int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2093 if (filp->f_op && filp->f_op->lock)
2094 return filp->f_op->lock(filp, F_CANCELLK, fl);
2095 return 0;
2098 EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2100 #ifdef CONFIG_PROC_FS
2101 #include <linux/seq_file.h>
2103 static void lock_get_status(struct seq_file *f, struct file_lock *fl,
2104 int id, char *pfx)
2106 struct inode *inode = NULL;
2107 unsigned int fl_pid;
2109 if (fl->fl_nspid)
2110 fl_pid = pid_vnr(fl->fl_nspid);
2111 else
2112 fl_pid = fl->fl_pid;
2114 if (fl->fl_file != NULL)
2115 inode = fl->fl_file->f_path.dentry->d_inode;
2117 seq_printf(f, "%d:%s ", id, pfx);
2118 if (IS_POSIX(fl)) {
2119 seq_printf(f, "%6s %s ",
2120 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2121 (inode == NULL) ? "*NOINODE*" :
2122 mandatory_lock(inode) ? "MANDATORY" : "ADVISORY ");
2123 } else if (IS_FLOCK(fl)) {
2124 if (fl->fl_type & LOCK_MAND) {
2125 seq_printf(f, "FLOCK MSNFS ");
2126 } else {
2127 seq_printf(f, "FLOCK ADVISORY ");
2129 } else if (IS_LEASE(fl)) {
2130 seq_printf(f, "LEASE ");
2131 if (fl->fl_type & F_INPROGRESS)
2132 seq_printf(f, "BREAKING ");
2133 else if (fl->fl_file)
2134 seq_printf(f, "ACTIVE ");
2135 else
2136 seq_printf(f, "BREAKER ");
2137 } else {
2138 seq_printf(f, "UNKNOWN UNKNOWN ");
2140 if (fl->fl_type & LOCK_MAND) {
2141 seq_printf(f, "%s ",
2142 (fl->fl_type & LOCK_READ)
2143 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2144 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2145 } else {
2146 seq_printf(f, "%s ",
2147 (fl->fl_type & F_INPROGRESS)
2148 ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ "
2149 : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ ");
2151 if (inode) {
2152 #ifdef WE_CAN_BREAK_LSLK_NOW
2153 seq_printf(f, "%d %s:%ld ", fl_pid,
2154 inode->i_sb->s_id, inode->i_ino);
2155 #else
2156 /* userspace relies on this representation of dev_t ;-( */
2157 seq_printf(f, "%d %02x:%02x:%ld ", fl_pid,
2158 MAJOR(inode->i_sb->s_dev),
2159 MINOR(inode->i_sb->s_dev), inode->i_ino);
2160 #endif
2161 } else {
2162 seq_printf(f, "%d <none>:0 ", fl_pid);
2164 if (IS_POSIX(fl)) {
2165 if (fl->fl_end == OFFSET_MAX)
2166 seq_printf(f, "%Ld EOF\n", fl->fl_start);
2167 else
2168 seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
2169 } else {
2170 seq_printf(f, "0 EOF\n");
2174 static int locks_show(struct seq_file *f, void *v)
2176 struct file_lock *fl, *bfl;
2178 fl = list_entry(v, struct file_lock, fl_link);
2180 lock_get_status(f, fl, (long)f->private, "");
2182 list_for_each_entry(bfl, &fl->fl_block, fl_block)
2183 lock_get_status(f, bfl, (long)f->private, " ->");
2185 f->private++;
2186 return 0;
2189 static void *locks_start(struct seq_file *f, loff_t *pos)
2191 lock_kernel();
2192 f->private = (void *)1;
2193 return seq_list_start(&file_lock_list, *pos);
2196 static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2198 return seq_list_next(v, &file_lock_list, pos);
2201 static void locks_stop(struct seq_file *f, void *v)
2203 unlock_kernel();
2206 struct seq_operations locks_seq_operations = {
2207 .start = locks_start,
2208 .next = locks_next,
2209 .stop = locks_stop,
2210 .show = locks_show,
2212 #endif
2215 * lock_may_read - checks that the region is free of locks
2216 * @inode: the inode that is being read
2217 * @start: the first byte to read
2218 * @len: the number of bytes to read
2220 * Emulates Windows locking requirements. Whole-file
2221 * mandatory locks (share modes) can prohibit a read and
2222 * byte-range POSIX locks can prohibit a read if they overlap.
2224 * N.B. this function is only ever called
2225 * from knfsd and ownership of locks is never checked.
2227 int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2229 struct file_lock *fl;
2230 int result = 1;
2231 lock_kernel();
2232 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2233 if (IS_POSIX(fl)) {
2234 if (fl->fl_type == F_RDLCK)
2235 continue;
2236 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2237 continue;
2238 } else if (IS_FLOCK(fl)) {
2239 if (!(fl->fl_type & LOCK_MAND))
2240 continue;
2241 if (fl->fl_type & LOCK_READ)
2242 continue;
2243 } else
2244 continue;
2245 result = 0;
2246 break;
2248 unlock_kernel();
2249 return result;
2252 EXPORT_SYMBOL(lock_may_read);
2255 * lock_may_write - checks that the region is free of locks
2256 * @inode: the inode that is being written
2257 * @start: the first byte to write
2258 * @len: the number of bytes to write
2260 * Emulates Windows locking requirements. Whole-file
2261 * mandatory locks (share modes) can prohibit a write and
2262 * byte-range POSIX locks can prohibit a write if they overlap.
2264 * N.B. this function is only ever called
2265 * from knfsd and ownership of locks is never checked.
2267 int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2269 struct file_lock *fl;
2270 int result = 1;
2271 lock_kernel();
2272 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2273 if (IS_POSIX(fl)) {
2274 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2275 continue;
2276 } else if (IS_FLOCK(fl)) {
2277 if (!(fl->fl_type & LOCK_MAND))
2278 continue;
2279 if (fl->fl_type & LOCK_WRITE)
2280 continue;
2281 } else
2282 continue;
2283 result = 0;
2284 break;
2286 unlock_kernel();
2287 return result;
2290 EXPORT_SYMBOL(lock_may_write);
2292 static int __init filelock_init(void)
2294 filelock_cache = kmem_cache_create("file_lock_cache",
2295 sizeof(struct file_lock), 0, SLAB_PANIC,
2296 init_once);
2297 return 0;
2300 core_initcall(filelock_init);