blob: efad798824dc19913b7aace09dbee47b8c6ca6a7 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/locks.c
3 *
4 * Provide support for fcntl()'s F_GETLK, F_SETLK, and F_SETLKW calls.
5 * Doug Evans (dje@spiff.uucp), August 07, 1992
6 *
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.
11 *
12 * Miscellaneous edits, and a total rewrite of posix_lock_file() code.
13 * Kai Petzke (wpp@marie.physik.tu-berlin.de), 1994
14 *
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
18 *
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
26 *
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
30 *
31 * Implemented two lock personalities - FL_FLOCK and FL_POSIX.
32 *
33 * FL_POSIX locks are created with calls to fcntl() and lockf() through the
34 * fcntl() system call. They have the semantics described above.
35 *
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.
39 *
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).
45 *
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
51 *
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.
55 *
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.
60 *
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.
65 *
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.
71 *
72 * Tidied up block list handling. Added '/proc/locks' interface.
73 * Andy Walker (andy@lysaker.kvaerner.no), April 24, 1996.
74 *
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.
78 *
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.
84 *
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.
88 *
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.
93 *
94 * Made the block list a circular list to minimise searching in the list.
95 * Andy Walker (andy@lysaker.kvaerner.no), Sep 25, 1996.
96 *
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.
100 *
101 * Some adaptations for NFS support.
102 * Olaf Kirch (okir@monad.swb.de), Dec 1996,
103 *
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.
106 *
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.
111 *
112 * Leases and LOCK_MAND
113 * Matthew Wilcox <willy@debian.org>, June, 2000.
114 * Stephen Rothwell <sfr@canb.auug.org.au>, June, 2000.
115 */
116
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>
Dipankar Sarma4fb3a532005-09-16 19:28:13 -0700127#include <linux/rcupdate.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128
129#include <asm/semaphore.h>
130#include <asm/uaccess.h>
131
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)
135
136int leases_enable = 1;
137int lease_break_time = 45;
138
139#define for_each_lock(inode, lockp) \
140 for (lockp = &inode->i_flock; *lockp != NULL; lockp = &(*lockp)->fl_next)
141
Christoph Hellwig26bcbf92006-03-20 13:44:40 -0500142static LIST_HEAD(file_lock_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143static LIST_HEAD(blocked_list);
144
Eric Dumazetfa3536c2006-03-26 01:37:24 -0800145static kmem_cache_t *filelock_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146
147/* Allocate an empty lock structure. */
148static struct file_lock *locks_alloc_lock(void)
149{
150 return kmem_cache_alloc(filelock_cache, SLAB_KERNEL);
151}
152
Trond Myklebust47831f32006-03-20 13:44:05 -0500153static void locks_release_private(struct file_lock *fl)
154{
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;
159 }
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;
164 }
165
166}
167
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168/* Free a lock which is not in use. */
Matt Mackall33443c42006-01-08 01:05:22 -0800169static void locks_free_lock(struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170{
Miklos Szeredi5ce29642006-03-31 02:30:29 -0800171 BUG_ON(waitqueue_active(&fl->fl_wait));
172 BUG_ON(!list_empty(&fl->fl_block));
173 BUG_ON(!list_empty(&fl->fl_link));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174
Trond Myklebust47831f32006-03-20 13:44:05 -0500175 locks_release_private(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 kmem_cache_free(filelock_cache, fl);
177}
178
179void locks_init_lock(struct file_lock *fl)
180{
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;
194}
195
196EXPORT_SYMBOL(locks_init_lock);
197
198/*
199 * Initialises the fields of the file lock which are invariant for
200 * free file_locks.
201 */
202static void init_once(void *foo, kmem_cache_t *cache, unsigned long flags)
203{
204 struct file_lock *lock = (struct file_lock *) foo;
205
206 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) !=
207 SLAB_CTOR_CONSTRUCTOR)
208 return;
209
210 locks_init_lock(lock);
211}
212
Trond Myklebust47831f32006-03-20 13:44:05 -0500213static void locks_copy_private(struct file_lock *new, struct file_lock *fl)
214{
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;
219 }
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;
224 }
225}
226
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227/*
228 * Initialize a new lock from an existing file_lock structure.
229 */
Trond Myklebust09969052006-03-20 13:44:38 -0500230static void __locks_copy_lock(struct file_lock *new, const struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231{
232 new->fl_owner = fl->fl_owner;
233 new->fl_pid = fl->fl_pid;
Trond Myklebust09969052006-03-20 13:44:38 -0500234 new->fl_file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 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;
Trond Myklebust09969052006-03-20 13:44:38 -0500239 new->fl_ops = NULL;
240 new->fl_lmops = NULL;
241}
242
243void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
244{
245 locks_release_private(new);
246
247 __locks_copy_lock(new, fl);
248 new->fl_file = fl->fl_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 new->fl_ops = fl->fl_ops;
250 new->fl_lmops = fl->fl_lmops;
Trond Myklebust47831f32006-03-20 13:44:05 -0500251
252 locks_copy_private(new, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253}
254
255EXPORT_SYMBOL(locks_copy_lock);
256
257static 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;
267 }
268 return -EINVAL;
269}
270
271/* Fill in a file_lock structure with an appropriate FLOCK lock. */
272static int flock_make_lock(struct file *filp, struct file_lock **lock,
273 unsigned int cmd)
274{
275 struct file_lock *fl;
276 int type = flock_translate_cmd(cmd);
277 if (type < 0)
278 return type;
279
280 fl = locks_alloc_lock();
281 if (fl == NULL)
282 return -ENOMEM;
283
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;
289
290 *lock = fl;
291 return 0;
292}
293
294static int assign_type(struct file_lock *fl, int type)
295{
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;
304 }
305 return 0;
306}
307
308/* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
309 * style lock.
310 */
311static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
312 struct flock *l)
313{
314 off_t start, end;
315
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;
328 }
329
330 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
331 POSIX-2001 defines it. */
332 start += l->l_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333 if (start < 0)
334 return -EINVAL;
Trond Myklebust4c780a42005-10-18 14:20:21 -0700335 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;
345 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 fl->fl_start = start; /* we record the absolute position */
Trond Myklebust4c780a42005-10-18 14:20:21 -0700347 if (fl->fl_end < fl->fl_start)
348 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349
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;
356
357 return assign_type(fl, l->l_type);
358}
359
360#if BITS_PER_LONG == 32
361static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
362 struct flock64 *l)
363{
364 loff_t start;
365
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;
378 }
379
Trond Myklebust4c780a42005-10-18 14:20:21 -0700380 start += l->l_start;
381 if (start < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 return -EINVAL;
Trond Myklebust4c780a42005-10-18 14:20:21 -0700383 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;
391 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392 fl->fl_start = start; /* we record the absolute position */
Trond Myklebust4c780a42005-10-18 14:20:21 -0700393 if (fl->fl_end < fl->fl_start)
394 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395
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;
402
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;
411 }
412
413 return (0);
414}
415#endif
416
417/* default lease lock manager operations */
418static void lease_break_callback(struct file_lock *fl)
419{
420 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
421}
422
423static void lease_release_private_callback(struct file_lock *fl)
424{
425 if (!fl->fl_file)
426 return;
427
428 f_delown(fl->fl_file);
429 fl->fl_file->f_owner.signum = 0;
430}
431
Adrian Bunk75c96f82005-05-05 16:16:09 -0700432static int lease_mylease_callback(struct file_lock *fl, struct file_lock *try)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433{
434 return fl->fl_file == try->fl_file;
435}
436
Adrian Bunk75c96f82005-05-05 16:16:09 -0700437static struct lock_manager_operations lease_manager_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 .fl_break = lease_break_callback,
439 .fl_release_private = lease_release_private_callback,
440 .fl_mylease = lease_mylease_callback,
441 .fl_change = lease_modify,
442};
443
444/*
445 * Initialize a lease, use the default lock manager operations
446 */
447static int lease_init(struct file *filp, int type, struct file_lock *fl)
448 {
449 fl->fl_owner = current->files;
450 fl->fl_pid = current->tgid;
451
452 fl->fl_file = filp;
453 fl->fl_flags = FL_LEASE;
454 if (assign_type(fl, type) != 0) {
455 locks_free_lock(fl);
456 return -EINVAL;
457 }
458 fl->fl_start = 0;
459 fl->fl_end = OFFSET_MAX;
460 fl->fl_ops = NULL;
461 fl->fl_lmops = &lease_manager_ops;
462 return 0;
463}
464
465/* Allocate a file_lock initialised to this type of lease */
466static int lease_alloc(struct file *filp, int type, struct file_lock **flp)
467{
468 struct file_lock *fl = locks_alloc_lock();
469 int error;
470
471 if (fl == NULL)
472 return -ENOMEM;
473
474 error = lease_init(filp, type, fl);
475 if (error)
476 return error;
477 *flp = fl;
478 return 0;
479}
480
481/* Check if two locks overlap each other.
482 */
483static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
484{
485 return ((fl1->fl_end >= fl2->fl_start) &&
486 (fl2->fl_end >= fl1->fl_start));
487}
488
489/*
490 * Check whether two locks have the same owner.
491 */
Matt Mackall33443c42006-01-08 01:05:22 -0800492static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493{
494 if (fl1->fl_lmops && fl1->fl_lmops->fl_compare_owner)
495 return fl2->fl_lmops == fl1->fl_lmops &&
496 fl1->fl_lmops->fl_compare_owner(fl1, fl2);
497 return fl1->fl_owner == fl2->fl_owner;
498}
499
500/* Remove waiter from blocker's block list.
501 * When blocker ends up pointing to itself then the list is empty.
502 */
Matt Mackall33443c42006-01-08 01:05:22 -0800503static void __locks_delete_block(struct file_lock *waiter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504{
505 list_del_init(&waiter->fl_block);
506 list_del_init(&waiter->fl_link);
507 waiter->fl_next = NULL;
508}
509
510/*
511 */
512static void locks_delete_block(struct file_lock *waiter)
513{
514 lock_kernel();
515 __locks_delete_block(waiter);
516 unlock_kernel();
517}
518
519/* Insert waiter into blocker's block list.
520 * We use a circular list so that processes can be easily woken up in
521 * the order they blocked. The documentation doesn't require this but
522 * it seems like the reasonable thing to do.
523 */
524static void locks_insert_block(struct file_lock *blocker,
525 struct file_lock *waiter)
526{
J. Bruce Fields6dc0fe82006-03-26 01:37:24 -0800527 BUG_ON(!list_empty(&waiter->fl_block));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 list_add_tail(&waiter->fl_block, &blocker->fl_block);
529 waiter->fl_next = blocker;
530 if (IS_POSIX(blocker))
531 list_add(&waiter->fl_link, &blocked_list);
532}
533
534/* Wake up processes blocked waiting for blocker.
535 * If told to wait then schedule the processes until the block list
536 * is empty, otherwise empty the block list ourselves.
537 */
538static void locks_wake_up_blocks(struct file_lock *blocker)
539{
540 while (!list_empty(&blocker->fl_block)) {
541 struct file_lock *waiter = list_entry(blocker->fl_block.next,
542 struct file_lock, fl_block);
543 __locks_delete_block(waiter);
544 if (waiter->fl_lmops && waiter->fl_lmops->fl_notify)
545 waiter->fl_lmops->fl_notify(waiter);
546 else
547 wake_up(&waiter->fl_wait);
548 }
549}
550
551/* Insert file lock fl into an inode's lock list at the position indicated
552 * by pos. At the same time add the lock to the global file lock list.
553 */
554static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
555{
556 list_add(&fl->fl_link, &file_lock_list);
557
558 /* insert into file's list */
559 fl->fl_next = *pos;
560 *pos = fl;
561
562 if (fl->fl_ops && fl->fl_ops->fl_insert)
563 fl->fl_ops->fl_insert(fl);
564}
565
566/*
567 * Delete a lock and then free it.
568 * Wake up processes that are blocked waiting for this lock,
569 * notify the FS that the lock has been cleared and
570 * finally free the lock.
571 */
572static void locks_delete_lock(struct file_lock **thisfl_p)
573{
574 struct file_lock *fl = *thisfl_p;
575
576 *thisfl_p = fl->fl_next;
577 fl->fl_next = NULL;
578 list_del_init(&fl->fl_link);
579
580 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
581 if (fl->fl_fasync != NULL) {
582 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
583 fl->fl_fasync = NULL;
584 }
585
586 if (fl->fl_ops && fl->fl_ops->fl_remove)
587 fl->fl_ops->fl_remove(fl);
588
589 locks_wake_up_blocks(fl);
590 locks_free_lock(fl);
591}
592
593/* Determine if lock sys_fl blocks lock caller_fl. Common functionality
594 * checks for shared/exclusive status of overlapping locks.
595 */
596static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
597{
598 if (sys_fl->fl_type == F_WRLCK)
599 return 1;
600 if (caller_fl->fl_type == F_WRLCK)
601 return 1;
602 return 0;
603}
604
605/* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
606 * checking before calling the locks_conflict().
607 */
608static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
609{
610 /* POSIX locks owned by the same process do not conflict with
611 * each other.
612 */
613 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
614 return (0);
615
616 /* Check whether they overlap */
617 if (!locks_overlap(caller_fl, sys_fl))
618 return 0;
619
620 return (locks_conflict(caller_fl, sys_fl));
621}
622
623/* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
624 * checking before calling the locks_conflict().
625 */
626static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
627{
628 /* FLOCK locks referring to the same filp do not conflict with
629 * each other.
630 */
631 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
632 return (0);
633 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
634 return 0;
635
636 return (locks_conflict(caller_fl, sys_fl));
637}
638
639static int interruptible_sleep_on_locked(wait_queue_head_t *fl_wait, int timeout)
640{
641 int result = 0;
642 DECLARE_WAITQUEUE(wait, current);
643
644 __set_current_state(TASK_INTERRUPTIBLE);
645 add_wait_queue(fl_wait, &wait);
646 if (timeout == 0)
647 schedule();
648 else
649 result = schedule_timeout(timeout);
650 if (signal_pending(current))
651 result = -ERESTARTSYS;
652 remove_wait_queue(fl_wait, &wait);
653 __set_current_state(TASK_RUNNING);
654 return result;
655}
656
657static int locks_block_on_timeout(struct file_lock *blocker, struct file_lock *waiter, int time)
658{
659 int result;
660 locks_insert_block(blocker, waiter);
661 result = interruptible_sleep_on_locked(&waiter->fl_wait, time);
662 __locks_delete_block(waiter);
663 return result;
664}
665
Andy Adamson8dc7c312006-03-20 13:44:26 -0500666int
667posix_test_lock(struct file *filp, struct file_lock *fl,
668 struct file_lock *conflock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669{
670 struct file_lock *cfl;
671
672 lock_kernel();
673 for (cfl = filp->f_dentry->d_inode->i_flock; cfl; cfl = cfl->fl_next) {
674 if (!IS_POSIX(cfl))
675 continue;
676 if (posix_locks_conflict(cfl, fl))
677 break;
678 }
Andy Adamson8dc7c312006-03-20 13:44:26 -0500679 if (cfl) {
Trond Myklebust09969052006-03-20 13:44:38 -0500680 __locks_copy_lock(conflock, cfl);
Andy Adamson8dc7c312006-03-20 13:44:26 -0500681 unlock_kernel();
682 return 1;
683 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 unlock_kernel();
Andy Adamson8dc7c312006-03-20 13:44:26 -0500685 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686}
687
688EXPORT_SYMBOL(posix_test_lock);
689
690/* This function tests for deadlock condition before putting a process to
691 * sleep. The detection scheme is no longer recursive. Recursive was neat,
692 * but dangerous - we risked stack corruption if the lock data was bad, or
693 * if the recursion was too deep for any other reason.
694 *
695 * We rely on the fact that a task can only be on one lock's wait queue
696 * at a time. When we find blocked_task on a wait queue we can re-search
697 * with blocked_task equal to that queue's owner, until either blocked_task
698 * isn't found, or blocked_task is found on a queue owned by my_task.
699 *
700 * Note: the above assumption may not be true when handling lock requests
701 * from a broken NFS client. But broken NFS clients have a lot more to
702 * worry about than proper deadlock detection anyway... --okir
703 */
704int posix_locks_deadlock(struct file_lock *caller_fl,
705 struct file_lock *block_fl)
706{
707 struct list_head *tmp;
708
709next_task:
710 if (posix_same_owner(caller_fl, block_fl))
711 return 1;
712 list_for_each(tmp, &blocked_list) {
713 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
714 if (posix_same_owner(fl, block_fl)) {
715 fl = fl->fl_next;
716 block_fl = fl;
717 goto next_task;
718 }
719 }
720 return 0;
721}
722
723EXPORT_SYMBOL(posix_locks_deadlock);
724
725/* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
726 * at the head of the list, but that's secret knowledge known only to
727 * flock_lock_file and posix_lock_file.
728 */
Trond Myklebust993dfa82006-03-31 02:30:55 -0800729static int flock_lock_file(struct file *filp, struct file_lock *request)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730{
Trond Myklebust993dfa82006-03-31 02:30:55 -0800731 struct file_lock *new_fl = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 struct file_lock **before;
733 struct inode * inode = filp->f_dentry->d_inode;
734 int error = 0;
735 int found = 0;
736
737 lock_kernel();
738 for_each_lock(inode, before) {
739 struct file_lock *fl = *before;
740 if (IS_POSIX(fl))
741 break;
742 if (IS_LEASE(fl))
743 continue;
744 if (filp != fl->fl_file)
745 continue;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800746 if (request->fl_type == fl->fl_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 goto out;
748 found = 1;
749 locks_delete_lock(before);
750 break;
751 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752
Trond Myklebust993dfa82006-03-31 02:30:55 -0800753 if (request->fl_type == F_UNLCK)
754 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755
Trond Myklebust993dfa82006-03-31 02:30:55 -0800756 new_fl = locks_alloc_lock();
757 if (new_fl == NULL)
758 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 /*
760 * If a higher-priority process was blocked on the old file lock,
761 * give it the opportunity to lock the file.
762 */
763 if (found)
764 cond_resched();
765
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 for_each_lock(inode, before) {
767 struct file_lock *fl = *before;
768 if (IS_POSIX(fl))
769 break;
770 if (IS_LEASE(fl))
771 continue;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800772 if (!flock_locks_conflict(request, fl))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 continue;
774 error = -EAGAIN;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800775 if (request->fl_flags & FL_SLEEP)
776 locks_insert_block(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 goto out;
778 }
Trond Myklebust993dfa82006-03-31 02:30:55 -0800779 locks_copy_lock(new_fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 locks_insert_lock(&inode->i_flock, new_fl);
Trond Myklebust993dfa82006-03-31 02:30:55 -0800781 new_fl = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782
783out:
784 unlock_kernel();
Trond Myklebust993dfa82006-03-31 02:30:55 -0800785 if (new_fl)
786 locks_free_lock(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 return error;
788}
789
Andy Adamson5842add2006-03-26 01:37:26 -0800790static int __posix_lock_file_conf(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791{
792 struct file_lock *fl;
793 struct file_lock *new_fl, *new_fl2;
794 struct file_lock *left = NULL;
795 struct file_lock *right = NULL;
796 struct file_lock **before;
797 int error, added = 0;
798
799 /*
800 * We may need two file_lock structures for this operation,
801 * so we get them in advance to avoid races.
802 */
803 new_fl = locks_alloc_lock();
804 new_fl2 = locks_alloc_lock();
805
806 lock_kernel();
807 if (request->fl_type != F_UNLCK) {
808 for_each_lock(inode, before) {
809 struct file_lock *fl = *before;
810 if (!IS_POSIX(fl))
811 continue;
812 if (!posix_locks_conflict(request, fl))
813 continue;
Andy Adamson5842add2006-03-26 01:37:26 -0800814 if (conflock)
815 locks_copy_lock(conflock, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 error = -EAGAIN;
817 if (!(request->fl_flags & FL_SLEEP))
818 goto out;
819 error = -EDEADLK;
820 if (posix_locks_deadlock(request, fl))
821 goto out;
822 error = -EAGAIN;
823 locks_insert_block(fl, request);
824 goto out;
825 }
826 }
827
828 /* If we're just looking for a conflict, we're done. */
829 error = 0;
830 if (request->fl_flags & FL_ACCESS)
831 goto out;
832
833 error = -ENOLCK; /* "no luck" */
834 if (!(new_fl && new_fl2))
835 goto out;
836
837 /*
838 * We've allocated the new locks in advance, so there are no
839 * errors possible (and no blocking operations) from here on.
840 *
841 * Find the first old lock with the same owner as the new lock.
842 */
843
844 before = &inode->i_flock;
845
846 /* First skip locks owned by other processes. */
847 while ((fl = *before) && (!IS_POSIX(fl) ||
848 !posix_same_owner(request, fl))) {
849 before = &fl->fl_next;
850 }
851
852 /* Process locks with this owner. */
853 while ((fl = *before) && posix_same_owner(request, fl)) {
854 /* Detect adjacent or overlapping regions (if same lock type)
855 */
856 if (request->fl_type == fl->fl_type) {
Olaf Kirch449231d2005-08-25 16:25:35 -0700857 /* In all comparisons of start vs end, use
858 * "start - 1" rather than "end + 1". If end
859 * is OFFSET_MAX, end + 1 will become negative.
860 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 if (fl->fl_end < request->fl_start - 1)
862 goto next_lock;
863 /* If the next lock in the list has entirely bigger
864 * addresses than the new one, insert the lock here.
865 */
Olaf Kirch449231d2005-08-25 16:25:35 -0700866 if (fl->fl_start - 1 > request->fl_end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 break;
868
869 /* If we come here, the new and old lock are of the
870 * same type and adjacent or overlapping. Make one
871 * lock yielding from the lower start address of both
872 * locks to the higher end address.
873 */
874 if (fl->fl_start > request->fl_start)
875 fl->fl_start = request->fl_start;
876 else
877 request->fl_start = fl->fl_start;
878 if (fl->fl_end < request->fl_end)
879 fl->fl_end = request->fl_end;
880 else
881 request->fl_end = fl->fl_end;
882 if (added) {
883 locks_delete_lock(before);
884 continue;
885 }
886 request = fl;
887 added = 1;
888 }
889 else {
890 /* Processing for different lock types is a bit
891 * more complex.
892 */
893 if (fl->fl_end < request->fl_start)
894 goto next_lock;
895 if (fl->fl_start > request->fl_end)
896 break;
897 if (request->fl_type == F_UNLCK)
898 added = 1;
899 if (fl->fl_start < request->fl_start)
900 left = fl;
901 /* If the next lock in the list has a higher end
902 * address than the new one, insert the new one here.
903 */
904 if (fl->fl_end > request->fl_end) {
905 right = fl;
906 break;
907 }
908 if (fl->fl_start >= request->fl_start) {
909 /* The new lock completely replaces an old
910 * one (This may happen several times).
911 */
912 if (added) {
913 locks_delete_lock(before);
914 continue;
915 }
916 /* Replace the old lock with the new one.
917 * Wake up anybody waiting for the old one,
918 * as the change in lock type might satisfy
919 * their needs.
920 */
921 locks_wake_up_blocks(fl);
922 fl->fl_start = request->fl_start;
923 fl->fl_end = request->fl_end;
924 fl->fl_type = request->fl_type;
Trond Myklebust47831f32006-03-20 13:44:05 -0500925 locks_release_private(fl);
926 locks_copy_private(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927 request = fl;
928 added = 1;
929 }
930 }
931 /* Go on to next lock.
932 */
933 next_lock:
934 before = &fl->fl_next;
935 }
936
937 error = 0;
938 if (!added) {
939 if (request->fl_type == F_UNLCK)
940 goto out;
941 locks_copy_lock(new_fl, request);
942 locks_insert_lock(before, new_fl);
943 new_fl = NULL;
944 }
945 if (right) {
946 if (left == right) {
947 /* The new lock breaks the old one in two pieces,
948 * so we have to use the second new lock.
949 */
950 left = new_fl2;
951 new_fl2 = NULL;
952 locks_copy_lock(left, right);
953 locks_insert_lock(before, left);
954 }
955 right->fl_start = request->fl_end + 1;
956 locks_wake_up_blocks(right);
957 }
958 if (left) {
959 left->fl_end = request->fl_start - 1;
960 locks_wake_up_blocks(left);
961 }
962 out:
963 unlock_kernel();
964 /*
965 * Free any unused locks.
966 */
967 if (new_fl)
968 locks_free_lock(new_fl);
969 if (new_fl2)
970 locks_free_lock(new_fl2);
971 return error;
972}
973
974/**
975 * posix_lock_file - Apply a POSIX-style lock to a file
976 * @filp: The file to apply the lock to
977 * @fl: The lock to be applied
978 *
979 * Add a POSIX style lock to a file.
980 * We merge adjacent & overlapping locks whenever possible.
981 * POSIX locks are sorted by owner task, then by starting address
982 */
983int posix_lock_file(struct file *filp, struct file_lock *fl)
984{
Andy Adamson5842add2006-03-26 01:37:26 -0800985 return __posix_lock_file_conf(filp->f_dentry->d_inode, fl, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986}
Andy Adamson5842add2006-03-26 01:37:26 -0800987EXPORT_SYMBOL(posix_lock_file);
988
989/**
990 * posix_lock_file_conf - Apply a POSIX-style lock to a file
991 * @filp: The file to apply the lock to
992 * @fl: The lock to be applied
993 * @conflock: Place to return a copy of the conflicting lock, if found.
994 *
995 * Except for the conflock parameter, acts just like posix_lock_file.
996 */
997int posix_lock_file_conf(struct file *filp, struct file_lock *fl,
998 struct file_lock *conflock)
999{
1000 return __posix_lock_file_conf(filp->f_dentry->d_inode, fl, conflock);
1001}
1002EXPORT_SYMBOL(posix_lock_file_conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003
1004/**
1005 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1006 * @filp: The file to apply the lock to
1007 * @fl: The lock to be applied
1008 *
1009 * Add a POSIX style lock to a file.
1010 * We merge adjacent & overlapping locks whenever possible.
1011 * POSIX locks are sorted by owner task, then by starting address
1012 */
1013int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1014{
1015 int error;
1016 might_sleep ();
1017 for (;;) {
Andy Adamson5842add2006-03-26 01:37:26 -08001018 error = posix_lock_file(filp, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1020 break;
1021 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1022 if (!error)
1023 continue;
1024
1025 locks_delete_block(fl);
1026 break;
1027 }
1028 return error;
1029}
1030EXPORT_SYMBOL(posix_lock_file_wait);
1031
1032/**
1033 * locks_mandatory_locked - Check for an active lock
1034 * @inode: the file to check
1035 *
1036 * Searches the inode's list of locks to find any POSIX locks which conflict.
1037 * This function is called from locks_verify_locked() only.
1038 */
1039int locks_mandatory_locked(struct inode *inode)
1040{
1041 fl_owner_t owner = current->files;
1042 struct file_lock *fl;
1043
1044 /*
1045 * Search the lock list for this inode for any POSIX locks.
1046 */
1047 lock_kernel();
1048 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1049 if (!IS_POSIX(fl))
1050 continue;
1051 if (fl->fl_owner != owner)
1052 break;
1053 }
1054 unlock_kernel();
1055 return fl ? -EAGAIN : 0;
1056}
1057
1058/**
1059 * locks_mandatory_area - Check for a conflicting lock
1060 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1061 * for shared
1062 * @inode: the file to check
1063 * @filp: how the file was opened (if it was)
1064 * @offset: start of area to check
1065 * @count: length of area to check
1066 *
1067 * Searches the inode's list of locks to find any POSIX locks which conflict.
1068 * This function is called from rw_verify_area() and
1069 * locks_verify_truncate().
1070 */
1071int locks_mandatory_area(int read_write, struct inode *inode,
1072 struct file *filp, loff_t offset,
1073 size_t count)
1074{
1075 struct file_lock fl;
1076 int error;
1077
1078 locks_init_lock(&fl);
1079 fl.fl_owner = current->files;
1080 fl.fl_pid = current->tgid;
1081 fl.fl_file = filp;
1082 fl.fl_flags = FL_POSIX | FL_ACCESS;
1083 if (filp && !(filp->f_flags & O_NONBLOCK))
1084 fl.fl_flags |= FL_SLEEP;
1085 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1086 fl.fl_start = offset;
1087 fl.fl_end = offset + count - 1;
1088
1089 for (;;) {
Andy Adamson5842add2006-03-26 01:37:26 -08001090 error = __posix_lock_file_conf(inode, &fl, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 if (error != -EAGAIN)
1092 break;
1093 if (!(fl.fl_flags & FL_SLEEP))
1094 break;
1095 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1096 if (!error) {
1097 /*
1098 * If we've been sleeping someone might have
1099 * changed the permissions behind our back.
1100 */
1101 if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID)
1102 continue;
1103 }
1104
1105 locks_delete_block(&fl);
1106 break;
1107 }
1108
1109 return error;
1110}
1111
1112EXPORT_SYMBOL(locks_mandatory_area);
1113
1114/* We already had a lease on this file; just change its type */
1115int lease_modify(struct file_lock **before, int arg)
1116{
1117 struct file_lock *fl = *before;
1118 int error = assign_type(fl, arg);
1119
1120 if (error)
1121 return error;
1122 locks_wake_up_blocks(fl);
1123 if (arg == F_UNLCK)
1124 locks_delete_lock(before);
1125 return 0;
1126}
1127
1128EXPORT_SYMBOL(lease_modify);
1129
1130static void time_out_leases(struct inode *inode)
1131{
1132 struct file_lock **before;
1133 struct file_lock *fl;
1134
1135 before = &inode->i_flock;
1136 while ((fl = *before) && IS_LEASE(fl) && (fl->fl_type & F_INPROGRESS)) {
1137 if ((fl->fl_break_time == 0)
1138 || time_before(jiffies, fl->fl_break_time)) {
1139 before = &fl->fl_next;
1140 continue;
1141 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142 lease_modify(before, fl->fl_type & ~F_INPROGRESS);
1143 if (fl == *before) /* lease_modify may have freed fl */
1144 before = &fl->fl_next;
1145 }
1146}
1147
1148/**
1149 * __break_lease - revoke all outstanding leases on file
1150 * @inode: the inode of the file to return
1151 * @mode: the open mode (read or write)
1152 *
1153 * break_lease (inlined for speed) has checked there already
1154 * is a lease on this file. Leases are broken on a call to open()
1155 * or truncate(). This function can sleep unless you
1156 * specified %O_NONBLOCK to your open().
1157 */
1158int __break_lease(struct inode *inode, unsigned int mode)
1159{
1160 int error = 0, future;
1161 struct file_lock *new_fl, *flock;
1162 struct file_lock *fl;
1163 int alloc_err;
1164 unsigned long break_time;
1165 int i_have_this_lease = 0;
1166
1167 alloc_err = lease_alloc(NULL, mode & FMODE_WRITE ? F_WRLCK : F_RDLCK,
1168 &new_fl);
1169
1170 lock_kernel();
1171
1172 time_out_leases(inode);
1173
1174 flock = inode->i_flock;
1175 if ((flock == NULL) || !IS_LEASE(flock))
1176 goto out;
1177
1178 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1179 if (fl->fl_owner == current->files)
1180 i_have_this_lease = 1;
1181
1182 if (mode & FMODE_WRITE) {
1183 /* If we want write access, we have to revoke any lease. */
1184 future = F_UNLCK | F_INPROGRESS;
1185 } else if (flock->fl_type & F_INPROGRESS) {
1186 /* If the lease is already being broken, we just leave it */
1187 future = flock->fl_type;
1188 } else if (flock->fl_type & F_WRLCK) {
1189 /* Downgrade the exclusive lease to a read-only lease. */
1190 future = F_RDLCK | F_INPROGRESS;
1191 } else {
1192 /* the existing lease was read-only, so we can read too. */
1193 goto out;
1194 }
1195
1196 if (alloc_err && !i_have_this_lease && ((mode & O_NONBLOCK) == 0)) {
1197 error = alloc_err;
1198 goto out;
1199 }
1200
1201 break_time = 0;
1202 if (lease_break_time > 0) {
1203 break_time = jiffies + lease_break_time * HZ;
1204 if (break_time == 0)
1205 break_time++; /* so that 0 means no break time */
1206 }
1207
1208 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1209 if (fl->fl_type != future) {
1210 fl->fl_type = future;
1211 fl->fl_break_time = break_time;
1212 /* lease must have lmops break callback */
1213 fl->fl_lmops->fl_break(fl);
1214 }
1215 }
1216
1217 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1218 error = -EWOULDBLOCK;
1219 goto out;
1220 }
1221
1222restart:
1223 break_time = flock->fl_break_time;
1224 if (break_time != 0) {
1225 break_time -= jiffies;
1226 if (break_time == 0)
1227 break_time++;
1228 }
1229 error = locks_block_on_timeout(flock, new_fl, break_time);
1230 if (error >= 0) {
1231 if (error == 0)
1232 time_out_leases(inode);
1233 /* Wait for the next lease that has not been broken yet */
1234 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1235 flock = flock->fl_next) {
1236 if (flock->fl_type & F_INPROGRESS)
1237 goto restart;
1238 }
1239 error = 0;
1240 }
1241
1242out:
1243 unlock_kernel();
1244 if (!alloc_err)
1245 locks_free_lock(new_fl);
1246 return error;
1247}
1248
1249EXPORT_SYMBOL(__break_lease);
1250
1251/**
1252 * lease_get_mtime
1253 * @inode: the inode
1254 * @time: pointer to a timespec which will contain the last modified time
1255 *
1256 * This is to force NFS clients to flush their caches for files with
1257 * exclusive leases. The justification is that if someone has an
1258 * exclusive lease, then they could be modifiying it.
1259 */
1260void lease_get_mtime(struct inode *inode, struct timespec *time)
1261{
1262 struct file_lock *flock = inode->i_flock;
1263 if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK))
1264 *time = current_fs_time(inode->i_sb);
1265 else
1266 *time = inode->i_mtime;
1267}
1268
1269EXPORT_SYMBOL(lease_get_mtime);
1270
1271/**
1272 * fcntl_getlease - Enquire what lease is currently active
1273 * @filp: the file
1274 *
1275 * The value returned by this function will be one of
1276 * (if no lease break is pending):
1277 *
1278 * %F_RDLCK to indicate a shared lease is held.
1279 *
1280 * %F_WRLCK to indicate an exclusive lease is held.
1281 *
1282 * %F_UNLCK to indicate no lease is held.
1283 *
1284 * (if a lease break is pending):
1285 *
1286 * %F_RDLCK to indicate an exclusive lease needs to be
1287 * changed to a shared lease (or removed).
1288 *
1289 * %F_UNLCK to indicate the lease needs to be removed.
1290 *
1291 * XXX: sfr & willy disagree over whether F_INPROGRESS
1292 * should be returned to userspace.
1293 */
1294int fcntl_getlease(struct file *filp)
1295{
1296 struct file_lock *fl;
1297 int type = F_UNLCK;
1298
1299 lock_kernel();
1300 time_out_leases(filp->f_dentry->d_inode);
1301 for (fl = filp->f_dentry->d_inode->i_flock; fl && IS_LEASE(fl);
1302 fl = fl->fl_next) {
1303 if (fl->fl_file == filp) {
1304 type = fl->fl_type & ~F_INPROGRESS;
1305 break;
1306 }
1307 }
1308 unlock_kernel();
1309 return type;
1310}
1311
1312/**
1313 * __setlease - sets a lease on an open file
1314 * @filp: file pointer
1315 * @arg: type of lease to obtain
1316 * @flp: input - file_lock to use, output - file_lock inserted
1317 *
1318 * The (input) flp->fl_lmops->fl_break function is required
1319 * by break_lease().
1320 *
1321 * Called with kernel lock held.
1322 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07001323static int __setlease(struct file *filp, long arg, struct file_lock **flp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324{
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001325 struct file_lock *fl, **before, **my_before = NULL, *lease;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326 struct dentry *dentry = filp->f_dentry;
1327 struct inode *inode = dentry->d_inode;
1328 int error, rdlease_count = 0, wrlease_count = 0;
1329
1330 time_out_leases(inode);
1331
1332 error = -EINVAL;
1333 if (!flp || !(*flp) || !(*flp)->fl_lmops || !(*flp)->fl_lmops->fl_break)
1334 goto out;
1335
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001336 lease = *flp;
1337
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338 error = -EAGAIN;
1339 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1340 goto out;
1341 if ((arg == F_WRLCK)
1342 && ((atomic_read(&dentry->d_count) > 1)
1343 || (atomic_read(&inode->i_count) > 1)))
1344 goto out;
1345
1346 /*
1347 * At this point, we know that if there is an exclusive
1348 * lease on this file, then we hold it on this filp
1349 * (otherwise our open of this file would have blocked).
1350 * And if we are trying to acquire an exclusive lease,
1351 * then the file is not open by anyone (including us)
1352 * except for this filp.
1353 */
1354 for (before = &inode->i_flock;
1355 ((fl = *before) != NULL) && IS_LEASE(fl);
1356 before = &fl->fl_next) {
1357 if (lease->fl_lmops->fl_mylease(fl, lease))
1358 my_before = before;
1359 else if (fl->fl_type == (F_INPROGRESS | F_UNLCK))
1360 /*
1361 * Someone is in the process of opening this
1362 * file for writing so we may not take an
1363 * exclusive lease on it.
1364 */
1365 wrlease_count++;
1366 else
1367 rdlease_count++;
1368 }
1369
1370 if ((arg == F_RDLCK && (wrlease_count > 0)) ||
1371 (arg == F_WRLCK && ((rdlease_count + wrlease_count) > 0)))
1372 goto out;
1373
1374 if (my_before != NULL) {
1375 error = lease->fl_lmops->fl_change(my_before, arg);
1376 goto out;
1377 }
1378
1379 error = 0;
1380 if (arg == F_UNLCK)
1381 goto out;
1382
1383 error = -EINVAL;
1384 if (!leases_enable)
1385 goto out;
1386
1387 error = lease_alloc(filp, arg, &fl);
1388 if (error)
1389 goto out;
1390
1391 locks_copy_lock(fl, lease);
1392
1393 locks_insert_lock(before, fl);
1394
1395 *flp = fl;
1396out:
1397 return error;
1398}
1399
1400 /**
1401 * setlease - sets a lease on an open file
1402 * @filp: file pointer
1403 * @arg: type of lease to obtain
1404 * @lease: file_lock to use
1405 *
1406 * Call this to establish a lease on the file.
1407 * The fl_lmops fl_break function is required by break_lease
1408 */
1409
1410int setlease(struct file *filp, long arg, struct file_lock **lease)
1411{
1412 struct dentry *dentry = filp->f_dentry;
1413 struct inode *inode = dentry->d_inode;
1414 int error;
1415
1416 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1417 return -EACCES;
1418 if (!S_ISREG(inode->i_mode))
1419 return -EINVAL;
1420 error = security_file_lock(filp, arg);
1421 if (error)
1422 return error;
1423
1424 lock_kernel();
1425 error = __setlease(filp, arg, lease);
1426 unlock_kernel();
1427
1428 return error;
1429}
1430
1431EXPORT_SYMBOL(setlease);
1432
1433/**
1434 * fcntl_setlease - sets a lease on an open file
1435 * @fd: open file descriptor
1436 * @filp: file pointer
1437 * @arg: type of lease to obtain
1438 *
1439 * Call this fcntl to establish a lease on the file.
1440 * Note that you also need to call %F_SETSIG to
1441 * receive a signal when the lease is broken.
1442 */
1443int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1444{
1445 struct file_lock fl, *flp = &fl;
1446 struct dentry *dentry = filp->f_dentry;
1447 struct inode *inode = dentry->d_inode;
1448 int error;
1449
1450 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1451 return -EACCES;
1452 if (!S_ISREG(inode->i_mode))
1453 return -EINVAL;
1454 error = security_file_lock(filp, arg);
1455 if (error)
1456 return error;
1457
1458 locks_init_lock(&fl);
1459 error = lease_init(filp, arg, &fl);
1460 if (error)
1461 return error;
1462
1463 lock_kernel();
1464
1465 error = __setlease(filp, arg, &flp);
J. Bruce Fieldsdc15ae12005-11-10 19:08:00 -05001466 if (error || arg == F_UNLCK)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 goto out_unlock;
1468
1469 error = fasync_helper(fd, filp, 1, &flp->fl_fasync);
1470 if (error < 0) {
1471 /* remove lease just inserted by __setlease */
1472 flp->fl_type = F_UNLCK | F_INPROGRESS;
1473 flp->fl_break_time = jiffies- 10;
1474 time_out_leases(inode);
1475 goto out_unlock;
1476 }
1477
1478 error = f_setown(filp, current->pid, 0);
1479out_unlock:
1480 unlock_kernel();
1481 return error;
1482}
1483
1484/**
1485 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1486 * @filp: The file to apply the lock to
1487 * @fl: The lock to be applied
1488 *
1489 * Add a FLOCK style lock to a file.
1490 */
1491int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1492{
1493 int error;
1494 might_sleep();
1495 for (;;) {
1496 error = flock_lock_file(filp, fl);
1497 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1498 break;
1499 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1500 if (!error)
1501 continue;
1502
1503 locks_delete_block(fl);
1504 break;
1505 }
1506 return error;
1507}
1508
1509EXPORT_SYMBOL(flock_lock_file_wait);
1510
1511/**
1512 * sys_flock: - flock() system call.
1513 * @fd: the file descriptor to lock.
1514 * @cmd: the type of lock to apply.
1515 *
1516 * Apply a %FL_FLOCK style lock to an open file descriptor.
1517 * The @cmd can be one of
1518 *
1519 * %LOCK_SH -- a shared lock.
1520 *
1521 * %LOCK_EX -- an exclusive lock.
1522 *
1523 * %LOCK_UN -- remove an existing lock.
1524 *
1525 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1526 *
1527 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1528 * processes read and write access respectively.
1529 */
1530asmlinkage long sys_flock(unsigned int fd, unsigned int cmd)
1531{
1532 struct file *filp;
1533 struct file_lock *lock;
1534 int can_sleep, unlock;
1535 int error;
1536
1537 error = -EBADF;
1538 filp = fget(fd);
1539 if (!filp)
1540 goto out;
1541
1542 can_sleep = !(cmd & LOCK_NB);
1543 cmd &= ~LOCK_NB;
1544 unlock = (cmd == LOCK_UN);
1545
1546 if (!unlock && !(cmd & LOCK_MAND) && !(filp->f_mode & 3))
1547 goto out_putf;
1548
1549 error = flock_make_lock(filp, &lock, cmd);
1550 if (error)
1551 goto out_putf;
1552 if (can_sleep)
1553 lock->fl_flags |= FL_SLEEP;
1554
1555 error = security_file_lock(filp, cmd);
1556 if (error)
1557 goto out_free;
1558
1559 if (filp->f_op && filp->f_op->flock)
1560 error = filp->f_op->flock(filp,
1561 (can_sleep) ? F_SETLKW : F_SETLK,
1562 lock);
1563 else
1564 error = flock_lock_file_wait(filp, lock);
1565
1566 out_free:
Trond Myklebust993dfa82006-03-31 02:30:55 -08001567 locks_free_lock(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568
1569 out_putf:
1570 fput(filp);
1571 out:
1572 return error;
1573}
1574
1575/* Report the first existing lock that would conflict with l.
1576 * This implements the F_GETLK command of fcntl().
1577 */
1578int fcntl_getlk(struct file *filp, struct flock __user *l)
1579{
Andy Adamson8dc7c312006-03-20 13:44:26 -05001580 struct file_lock *fl, cfl, file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 struct flock flock;
1582 int error;
1583
1584 error = -EFAULT;
1585 if (copy_from_user(&flock, l, sizeof(flock)))
1586 goto out;
1587 error = -EINVAL;
1588 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1589 goto out;
1590
1591 error = flock_to_posix_lock(filp, &file_lock, &flock);
1592 if (error)
1593 goto out;
1594
1595 if (filp->f_op && filp->f_op->lock) {
1596 error = filp->f_op->lock(filp, F_GETLK, &file_lock);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00001597 if (file_lock.fl_ops && file_lock.fl_ops->fl_release_private)
1598 file_lock.fl_ops->fl_release_private(&file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599 if (error < 0)
1600 goto out;
1601 else
1602 fl = (file_lock.fl_type == F_UNLCK ? NULL : &file_lock);
1603 } else {
Andy Adamson8dc7c312006-03-20 13:44:26 -05001604 fl = (posix_test_lock(filp, &file_lock, &cfl) ? &cfl : NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605 }
1606
1607 flock.l_type = F_UNLCK;
1608 if (fl != NULL) {
1609 flock.l_pid = fl->fl_pid;
1610#if BITS_PER_LONG == 32
1611 /*
1612 * Make sure we can represent the posix lock via
1613 * legacy 32bit flock.
1614 */
1615 error = -EOVERFLOW;
1616 if (fl->fl_start > OFFT_OFFSET_MAX)
1617 goto out;
1618 if ((fl->fl_end != OFFSET_MAX)
1619 && (fl->fl_end > OFFT_OFFSET_MAX))
1620 goto out;
1621#endif
1622 flock.l_start = fl->fl_start;
1623 flock.l_len = fl->fl_end == OFFSET_MAX ? 0 :
1624 fl->fl_end - fl->fl_start + 1;
1625 flock.l_whence = 0;
1626 flock.l_type = fl->fl_type;
1627 }
1628 error = -EFAULT;
1629 if (!copy_to_user(l, &flock, sizeof(flock)))
1630 error = 0;
1631out:
1632 return error;
1633}
1634
1635/* Apply the lock described by l to an open file descriptor.
1636 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1637 */
Peter Staubachc2936212005-07-27 11:45:09 -07001638int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1639 struct flock __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640{
1641 struct file_lock *file_lock = locks_alloc_lock();
1642 struct flock flock;
1643 struct inode *inode;
1644 int error;
1645
1646 if (file_lock == NULL)
1647 return -ENOLCK;
1648
1649 /*
1650 * This might block, so we do it before checking the inode.
1651 */
1652 error = -EFAULT;
1653 if (copy_from_user(&flock, l, sizeof(flock)))
1654 goto out;
1655
1656 inode = filp->f_dentry->d_inode;
1657
1658 /* Don't allow mandatory locks on files that may be memory mapped
1659 * and shared.
1660 */
1661 if (IS_MANDLOCK(inode) &&
1662 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1663 mapping_writably_mapped(filp->f_mapping)) {
1664 error = -EAGAIN;
1665 goto out;
1666 }
1667
Peter Staubachc2936212005-07-27 11:45:09 -07001668again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669 error = flock_to_posix_lock(filp, file_lock, &flock);
1670 if (error)
1671 goto out;
1672 if (cmd == F_SETLKW) {
1673 file_lock->fl_flags |= FL_SLEEP;
1674 }
1675
1676 error = -EBADF;
1677 switch (flock.l_type) {
1678 case F_RDLCK:
1679 if (!(filp->f_mode & FMODE_READ))
1680 goto out;
1681 break;
1682 case F_WRLCK:
1683 if (!(filp->f_mode & FMODE_WRITE))
1684 goto out;
1685 break;
1686 case F_UNLCK:
1687 break;
1688 default:
1689 error = -EINVAL;
1690 goto out;
1691 }
1692
1693 error = security_file_lock(filp, file_lock->fl_type);
1694 if (error)
1695 goto out;
1696
Peter Staubachc2936212005-07-27 11:45:09 -07001697 if (filp->f_op && filp->f_op->lock != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698 error = filp->f_op->lock(filp, cmd, file_lock);
Peter Staubachc2936212005-07-27 11:45:09 -07001699 else {
1700 for (;;) {
Andy Adamson5842add2006-03-26 01:37:26 -08001701 error = posix_lock_file(filp, file_lock);
Peter Staubachc2936212005-07-27 11:45:09 -07001702 if ((error != -EAGAIN) || (cmd == F_SETLK))
1703 break;
1704 error = wait_event_interruptible(file_lock->fl_wait,
1705 !file_lock->fl_next);
1706 if (!error)
1707 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708
Peter Staubachc2936212005-07-27 11:45:09 -07001709 locks_delete_block(file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 break;
Peter Staubachc2936212005-07-27 11:45:09 -07001711 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712 }
1713
Peter Staubachc2936212005-07-27 11:45:09 -07001714 /*
1715 * Attempt to detect a close/fcntl race and recover by
1716 * releasing the lock that was just acquired.
1717 */
1718 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1719 flock.l_type = F_UNLCK;
1720 goto again;
1721 }
1722
1723out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724 locks_free_lock(file_lock);
1725 return error;
1726}
1727
1728#if BITS_PER_LONG == 32
1729/* Report the first existing lock that would conflict with l.
1730 * This implements the F_GETLK command of fcntl().
1731 */
1732int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1733{
Andy Adamson8dc7c312006-03-20 13:44:26 -05001734 struct file_lock *fl, cfl, file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735 struct flock64 flock;
1736 int error;
1737
1738 error = -EFAULT;
1739 if (copy_from_user(&flock, l, sizeof(flock)))
1740 goto out;
1741 error = -EINVAL;
1742 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1743 goto out;
1744
1745 error = flock64_to_posix_lock(filp, &file_lock, &flock);
1746 if (error)
1747 goto out;
1748
1749 if (filp->f_op && filp->f_op->lock) {
1750 error = filp->f_op->lock(filp, F_GETLK, &file_lock);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00001751 if (file_lock.fl_ops && file_lock.fl_ops->fl_release_private)
1752 file_lock.fl_ops->fl_release_private(&file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753 if (error < 0)
1754 goto out;
1755 else
1756 fl = (file_lock.fl_type == F_UNLCK ? NULL : &file_lock);
1757 } else {
Andy Adamson8dc7c312006-03-20 13:44:26 -05001758 fl = (posix_test_lock(filp, &file_lock, &cfl) ? &cfl : NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759 }
1760
1761 flock.l_type = F_UNLCK;
1762 if (fl != NULL) {
1763 flock.l_pid = fl->fl_pid;
1764 flock.l_start = fl->fl_start;
1765 flock.l_len = fl->fl_end == OFFSET_MAX ? 0 :
1766 fl->fl_end - fl->fl_start + 1;
1767 flock.l_whence = 0;
1768 flock.l_type = fl->fl_type;
1769 }
1770 error = -EFAULT;
1771 if (!copy_to_user(l, &flock, sizeof(flock)))
1772 error = 0;
1773
1774out:
1775 return error;
1776}
1777
1778/* Apply the lock described by l to an open file descriptor.
1779 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1780 */
Peter Staubachc2936212005-07-27 11:45:09 -07001781int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
1782 struct flock64 __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783{
1784 struct file_lock *file_lock = locks_alloc_lock();
1785 struct flock64 flock;
1786 struct inode *inode;
1787 int error;
1788
1789 if (file_lock == NULL)
1790 return -ENOLCK;
1791
1792 /*
1793 * This might block, so we do it before checking the inode.
1794 */
1795 error = -EFAULT;
1796 if (copy_from_user(&flock, l, sizeof(flock)))
1797 goto out;
1798
1799 inode = filp->f_dentry->d_inode;
1800
1801 /* Don't allow mandatory locks on files that may be memory mapped
1802 * and shared.
1803 */
1804 if (IS_MANDLOCK(inode) &&
1805 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1806 mapping_writably_mapped(filp->f_mapping)) {
1807 error = -EAGAIN;
1808 goto out;
1809 }
1810
Peter Staubachc2936212005-07-27 11:45:09 -07001811again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812 error = flock64_to_posix_lock(filp, file_lock, &flock);
1813 if (error)
1814 goto out;
1815 if (cmd == F_SETLKW64) {
1816 file_lock->fl_flags |= FL_SLEEP;
1817 }
1818
1819 error = -EBADF;
1820 switch (flock.l_type) {
1821 case F_RDLCK:
1822 if (!(filp->f_mode & FMODE_READ))
1823 goto out;
1824 break;
1825 case F_WRLCK:
1826 if (!(filp->f_mode & FMODE_WRITE))
1827 goto out;
1828 break;
1829 case F_UNLCK:
1830 break;
1831 default:
1832 error = -EINVAL;
1833 goto out;
1834 }
1835
1836 error = security_file_lock(filp, file_lock->fl_type);
1837 if (error)
1838 goto out;
1839
Peter Staubachc2936212005-07-27 11:45:09 -07001840 if (filp->f_op && filp->f_op->lock != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841 error = filp->f_op->lock(filp, cmd, file_lock);
Peter Staubachc2936212005-07-27 11:45:09 -07001842 else {
1843 for (;;) {
Andy Adamson5842add2006-03-26 01:37:26 -08001844 error = posix_lock_file(filp, file_lock);
Peter Staubachc2936212005-07-27 11:45:09 -07001845 if ((error != -EAGAIN) || (cmd == F_SETLK64))
1846 break;
1847 error = wait_event_interruptible(file_lock->fl_wait,
1848 !file_lock->fl_next);
1849 if (!error)
1850 continue;
1851
1852 locks_delete_block(file_lock);
1853 break;
1854 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855 }
1856
Peter Staubachc2936212005-07-27 11:45:09 -07001857 /*
1858 * Attempt to detect a close/fcntl race and recover by
1859 * releasing the lock that was just acquired.
1860 */
1861 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1862 flock.l_type = F_UNLCK;
1863 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864 }
1865
1866out:
1867 locks_free_lock(file_lock);
1868 return error;
1869}
1870#endif /* BITS_PER_LONG == 32 */
1871
1872/*
1873 * This function is called when the file is being removed
1874 * from the task's fd array. POSIX locks belonging to this task
1875 * are deleted at this time.
1876 */
1877void locks_remove_posix(struct file *filp, fl_owner_t owner)
1878{
1879 struct file_lock lock, **before;
1880
1881 /*
1882 * If there are no locks held on this file, we don't need to call
1883 * posix_lock_file(). Another process could be setting a lock on this
1884 * file at the same time, but we wouldn't remove that lock anyway.
1885 */
1886 before = &filp->f_dentry->d_inode->i_flock;
1887 if (*before == NULL)
1888 return;
1889
1890 lock.fl_type = F_UNLCK;
1891 lock.fl_flags = FL_POSIX;
1892 lock.fl_start = 0;
1893 lock.fl_end = OFFSET_MAX;
1894 lock.fl_owner = owner;
1895 lock.fl_pid = current->tgid;
1896 lock.fl_file = filp;
1897 lock.fl_ops = NULL;
1898 lock.fl_lmops = NULL;
1899
1900 if (filp->f_op && filp->f_op->lock != NULL) {
1901 filp->f_op->lock(filp, F_SETLK, &lock);
1902 goto out;
1903 }
1904
1905 /* Can't use posix_lock_file here; we need to remove it no matter
1906 * which pid we have.
1907 */
1908 lock_kernel();
1909 while (*before != NULL) {
1910 struct file_lock *fl = *before;
1911 if (IS_POSIX(fl) && posix_same_owner(fl, &lock)) {
1912 locks_delete_lock(before);
1913 continue;
1914 }
1915 before = &fl->fl_next;
1916 }
1917 unlock_kernel();
1918out:
1919 if (lock.fl_ops && lock.fl_ops->fl_release_private)
1920 lock.fl_ops->fl_release_private(&lock);
1921}
1922
1923EXPORT_SYMBOL(locks_remove_posix);
1924
1925/*
1926 * This function is called on the last close of an open file.
1927 */
1928void locks_remove_flock(struct file *filp)
1929{
1930 struct inode * inode = filp->f_dentry->d_inode;
1931 struct file_lock *fl;
1932 struct file_lock **before;
1933
1934 if (!inode->i_flock)
1935 return;
1936
1937 if (filp->f_op && filp->f_op->flock) {
1938 struct file_lock fl = {
1939 .fl_pid = current->tgid,
1940 .fl_file = filp,
1941 .fl_flags = FL_FLOCK,
1942 .fl_type = F_UNLCK,
1943 .fl_end = OFFSET_MAX,
1944 };
1945 filp->f_op->flock(filp, F_SETLKW, &fl);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00001946 if (fl.fl_ops && fl.fl_ops->fl_release_private)
1947 fl.fl_ops->fl_release_private(&fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001948 }
1949
1950 lock_kernel();
1951 before = &inode->i_flock;
1952
1953 while ((fl = *before) != NULL) {
1954 if (fl->fl_file == filp) {
Peter Staubachc2936212005-07-27 11:45:09 -07001955 if (IS_FLOCK(fl)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956 locks_delete_lock(before);
1957 continue;
1958 }
1959 if (IS_LEASE(fl)) {
1960 lease_modify(before, F_UNLCK);
1961 continue;
1962 }
1963 /* What? */
1964 BUG();
1965 }
1966 before = &fl->fl_next;
1967 }
1968 unlock_kernel();
1969}
1970
1971/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972 * posix_unblock_lock - stop waiting for a file lock
1973 * @filp: how the file was opened
1974 * @waiter: the lock which was waiting
1975 *
1976 * lockd needs to block waiting for locks.
1977 */
J. Bruce Fields64a318e2006-01-03 09:55:46 +01001978int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979posix_unblock_lock(struct file *filp, struct file_lock *waiter)
1980{
J. Bruce Fields64a318e2006-01-03 09:55:46 +01001981 int status = 0;
1982
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983 lock_kernel();
J. Bruce Fields5996a292006-01-03 09:55:44 +01001984 if (waiter->fl_next)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985 __locks_delete_block(waiter);
J. Bruce Fields64a318e2006-01-03 09:55:46 +01001986 else
1987 status = -ENOENT;
J. Bruce Fields5996a292006-01-03 09:55:44 +01001988 unlock_kernel();
J. Bruce Fields64a318e2006-01-03 09:55:46 +01001989 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990}
1991
1992EXPORT_SYMBOL(posix_unblock_lock);
1993
1994static void lock_get_status(char* out, struct file_lock *fl, int id, char *pfx)
1995{
1996 struct inode *inode = NULL;
1997
1998 if (fl->fl_file != NULL)
1999 inode = fl->fl_file->f_dentry->d_inode;
2000
2001 out += sprintf(out, "%d:%s ", id, pfx);
2002 if (IS_POSIX(fl)) {
2003 out += sprintf(out, "%6s %s ",
2004 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2005 (inode == NULL) ? "*NOINODE*" :
2006 (IS_MANDLOCK(inode) &&
2007 (inode->i_mode & (S_IXGRP | S_ISGID)) == S_ISGID) ?
2008 "MANDATORY" : "ADVISORY ");
2009 } else if (IS_FLOCK(fl)) {
2010 if (fl->fl_type & LOCK_MAND) {
2011 out += sprintf(out, "FLOCK MSNFS ");
2012 } else {
2013 out += sprintf(out, "FLOCK ADVISORY ");
2014 }
2015 } else if (IS_LEASE(fl)) {
2016 out += sprintf(out, "LEASE ");
2017 if (fl->fl_type & F_INPROGRESS)
2018 out += sprintf(out, "BREAKING ");
2019 else if (fl->fl_file)
2020 out += sprintf(out, "ACTIVE ");
2021 else
2022 out += sprintf(out, "BREAKER ");
2023 } else {
2024 out += sprintf(out, "UNKNOWN UNKNOWN ");
2025 }
2026 if (fl->fl_type & LOCK_MAND) {
2027 out += sprintf(out, "%s ",
2028 (fl->fl_type & LOCK_READ)
2029 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2030 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2031 } else {
2032 out += sprintf(out, "%s ",
2033 (fl->fl_type & F_INPROGRESS)
2034 ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ "
2035 : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ ");
2036 }
2037 if (inode) {
2038#ifdef WE_CAN_BREAK_LSLK_NOW
2039 out += sprintf(out, "%d %s:%ld ", fl->fl_pid,
2040 inode->i_sb->s_id, inode->i_ino);
2041#else
2042 /* userspace relies on this representation of dev_t ;-( */
2043 out += sprintf(out, "%d %02x:%02x:%ld ", fl->fl_pid,
2044 MAJOR(inode->i_sb->s_dev),
2045 MINOR(inode->i_sb->s_dev), inode->i_ino);
2046#endif
2047 } else {
2048 out += sprintf(out, "%d <none>:0 ", fl->fl_pid);
2049 }
2050 if (IS_POSIX(fl)) {
2051 if (fl->fl_end == OFFSET_MAX)
2052 out += sprintf(out, "%Ld EOF\n", fl->fl_start);
2053 else
2054 out += sprintf(out, "%Ld %Ld\n", fl->fl_start,
2055 fl->fl_end);
2056 } else {
2057 out += sprintf(out, "0 EOF\n");
2058 }
2059}
2060
2061static void move_lock_status(char **p, off_t* pos, off_t offset)
2062{
2063 int len;
2064 len = strlen(*p);
2065 if(*pos >= offset) {
2066 /* the complete line is valid */
2067 *p += len;
2068 *pos += len;
2069 return;
2070 }
2071 if(*pos+len > offset) {
2072 /* use the second part of the line */
2073 int i = offset-*pos;
2074 memmove(*p,*p+i,len-i);
2075 *p += len-i;
2076 *pos += len;
2077 return;
2078 }
2079 /* discard the complete line */
2080 *pos += len;
2081}
2082
2083/**
2084 * get_locks_status - reports lock usage in /proc/locks
2085 * @buffer: address in userspace to write into
2086 * @start: ?
2087 * @offset: how far we are through the buffer
2088 * @length: how much to read
2089 */
2090
2091int get_locks_status(char *buffer, char **start, off_t offset, int length)
2092{
2093 struct list_head *tmp;
2094 char *q = buffer;
2095 off_t pos = 0;
2096 int i = 0;
2097
2098 lock_kernel();
2099 list_for_each(tmp, &file_lock_list) {
2100 struct list_head *btmp;
2101 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
2102 lock_get_status(q, fl, ++i, "");
2103 move_lock_status(&q, &pos, offset);
2104
2105 if(pos >= offset+length)
2106 goto done;
2107
2108 list_for_each(btmp, &fl->fl_block) {
2109 struct file_lock *bfl = list_entry(btmp,
2110 struct file_lock, fl_block);
2111 lock_get_status(q, bfl, i, " ->");
2112 move_lock_status(&q, &pos, offset);
2113
2114 if(pos >= offset+length)
2115 goto done;
2116 }
2117 }
2118done:
2119 unlock_kernel();
2120 *start = buffer;
2121 if(q-buffer < length)
2122 return (q-buffer);
2123 return length;
2124}
2125
2126/**
2127 * lock_may_read - checks that the region is free of locks
2128 * @inode: the inode that is being read
2129 * @start: the first byte to read
2130 * @len: the number of bytes to read
2131 *
2132 * Emulates Windows locking requirements. Whole-file
2133 * mandatory locks (share modes) can prohibit a read and
2134 * byte-range POSIX locks can prohibit a read if they overlap.
2135 *
2136 * N.B. this function is only ever called
2137 * from knfsd and ownership of locks is never checked.
2138 */
2139int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2140{
2141 struct file_lock *fl;
2142 int result = 1;
2143 lock_kernel();
2144 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2145 if (IS_POSIX(fl)) {
2146 if (fl->fl_type == F_RDLCK)
2147 continue;
2148 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2149 continue;
2150 } else if (IS_FLOCK(fl)) {
2151 if (!(fl->fl_type & LOCK_MAND))
2152 continue;
2153 if (fl->fl_type & LOCK_READ)
2154 continue;
2155 } else
2156 continue;
2157 result = 0;
2158 break;
2159 }
2160 unlock_kernel();
2161 return result;
2162}
2163
2164EXPORT_SYMBOL(lock_may_read);
2165
2166/**
2167 * lock_may_write - checks that the region is free of locks
2168 * @inode: the inode that is being written
2169 * @start: the first byte to write
2170 * @len: the number of bytes to write
2171 *
2172 * Emulates Windows locking requirements. Whole-file
2173 * mandatory locks (share modes) can prohibit a write and
2174 * byte-range POSIX locks can prohibit a write if they overlap.
2175 *
2176 * N.B. this function is only ever called
2177 * from knfsd and ownership of locks is never checked.
2178 */
2179int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2180{
2181 struct file_lock *fl;
2182 int result = 1;
2183 lock_kernel();
2184 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2185 if (IS_POSIX(fl)) {
2186 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2187 continue;
2188 } else if (IS_FLOCK(fl)) {
2189 if (!(fl->fl_type & LOCK_MAND))
2190 continue;
2191 if (fl->fl_type & LOCK_WRITE)
2192 continue;
2193 } else
2194 continue;
2195 result = 0;
2196 break;
2197 }
2198 unlock_kernel();
2199 return result;
2200}
2201
2202EXPORT_SYMBOL(lock_may_write);
2203
2204static inline void __steal_locks(struct file *file, fl_owner_t from)
2205{
2206 struct inode *inode = file->f_dentry->d_inode;
2207 struct file_lock *fl = inode->i_flock;
2208
2209 while (fl) {
2210 if (fl->fl_file == file && fl->fl_owner == from)
2211 fl->fl_owner = current->files;
2212 fl = fl->fl_next;
2213 }
2214}
2215
2216/* When getting ready for executing a binary, we make sure that current
2217 * has a files_struct on its own. Before dropping the old files_struct,
2218 * we take over ownership of all locks for all file descriptors we own.
2219 * Note that we may accidentally steal a lock for a file that a sibling
2220 * has created since the unshare_files() call.
2221 */
2222void steal_locks(fl_owner_t from)
2223{
2224 struct files_struct *files = current->files;
2225 int i, j;
Dipankar Sarmabadf1662005-09-09 13:04:10 -07002226 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227
2228 if (from == files)
2229 return;
2230
2231 lock_kernel();
2232 j = 0;
Dipankar Sarmaca99c1d2006-04-18 22:21:46 -07002233
2234 /*
2235 * We are not taking a ref to the file structures, so
2236 * we need to acquire ->file_lock.
2237 */
2238 spin_lock(&files->file_lock);
Dipankar Sarmabadf1662005-09-09 13:04:10 -07002239 fdt = files_fdtable(files);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240 for (;;) {
2241 unsigned long set;
2242 i = j * __NFDBITS;
Dipankar Sarmabadf1662005-09-09 13:04:10 -07002243 if (i >= fdt->max_fdset || i >= fdt->max_fds)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244 break;
Dipankar Sarmabadf1662005-09-09 13:04:10 -07002245 set = fdt->open_fds->fds_bits[j++];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246 while (set) {
2247 if (set & 1) {
Dipankar Sarmabadf1662005-09-09 13:04:10 -07002248 struct file *file = fdt->fd[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002249 if (file)
2250 __steal_locks(file, from);
2251 }
2252 i++;
2253 set >>= 1;
2254 }
2255 }
Dipankar Sarmaca99c1d2006-04-18 22:21:46 -07002256 spin_unlock(&files->file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002257 unlock_kernel();
2258}
2259EXPORT_SYMBOL(steal_locks);
2260
2261static int __init filelock_init(void)
2262{
2263 filelock_cache = kmem_cache_create("file_lock_cache",
2264 sizeof(struct file_lock), 0, SLAB_PANIC,
2265 init_once, NULL);
2266 return 0;
2267}
2268
2269core_initcall(filelock_init);