blob: 56f996e98bbcd03459480ce9ea505e2663b46129 [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
145static kmem_cache_t *filelock_cache;
146
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{
171 if (fl == NULL) {
172 BUG();
173 return;
174 }
175 if (waitqueue_active(&fl->fl_wait))
176 panic("Attempting to free lock with active wait queue");
177
178 if (!list_empty(&fl->fl_block))
179 panic("Attempting to free lock with active block list");
180
181 if (!list_empty(&fl->fl_link))
182 panic("Attempting to free lock on active lock list");
183
Trond Myklebust47831f32006-03-20 13:44:05 -0500184 locks_release_private(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 kmem_cache_free(filelock_cache, fl);
186}
187
188void locks_init_lock(struct file_lock *fl)
189{
190 INIT_LIST_HEAD(&fl->fl_link);
191 INIT_LIST_HEAD(&fl->fl_block);
192 init_waitqueue_head(&fl->fl_wait);
193 fl->fl_next = NULL;
194 fl->fl_fasync = NULL;
195 fl->fl_owner = NULL;
196 fl->fl_pid = 0;
197 fl->fl_file = NULL;
198 fl->fl_flags = 0;
199 fl->fl_type = 0;
200 fl->fl_start = fl->fl_end = 0;
201 fl->fl_ops = NULL;
202 fl->fl_lmops = NULL;
203}
204
205EXPORT_SYMBOL(locks_init_lock);
206
207/*
208 * Initialises the fields of the file lock which are invariant for
209 * free file_locks.
210 */
211static void init_once(void *foo, kmem_cache_t *cache, unsigned long flags)
212{
213 struct file_lock *lock = (struct file_lock *) foo;
214
215 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) !=
216 SLAB_CTOR_CONSTRUCTOR)
217 return;
218
219 locks_init_lock(lock);
220}
221
Trond Myklebust47831f32006-03-20 13:44:05 -0500222static void locks_copy_private(struct file_lock *new, struct file_lock *fl)
223{
224 if (fl->fl_ops) {
225 if (fl->fl_ops->fl_copy_lock)
226 fl->fl_ops->fl_copy_lock(new, fl);
227 new->fl_ops = fl->fl_ops;
228 }
229 if (fl->fl_lmops) {
230 if (fl->fl_lmops->fl_copy_lock)
231 fl->fl_lmops->fl_copy_lock(new, fl);
232 new->fl_lmops = fl->fl_lmops;
233 }
234}
235
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236/*
237 * Initialize a new lock from an existing file_lock structure.
238 */
Trond Myklebust09969052006-03-20 13:44:38 -0500239static void __locks_copy_lock(struct file_lock *new, const struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240{
241 new->fl_owner = fl->fl_owner;
242 new->fl_pid = fl->fl_pid;
Trond Myklebust09969052006-03-20 13:44:38 -0500243 new->fl_file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244 new->fl_flags = fl->fl_flags;
245 new->fl_type = fl->fl_type;
246 new->fl_start = fl->fl_start;
247 new->fl_end = fl->fl_end;
Trond Myklebust09969052006-03-20 13:44:38 -0500248 new->fl_ops = NULL;
249 new->fl_lmops = NULL;
250}
251
252void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
253{
254 locks_release_private(new);
255
256 __locks_copy_lock(new, fl);
257 new->fl_file = fl->fl_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 new->fl_ops = fl->fl_ops;
259 new->fl_lmops = fl->fl_lmops;
Trond Myklebust47831f32006-03-20 13:44:05 -0500260
261 locks_copy_private(new, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262}
263
264EXPORT_SYMBOL(locks_copy_lock);
265
266static inline int flock_translate_cmd(int cmd) {
267 if (cmd & LOCK_MAND)
268 return cmd & (LOCK_MAND | LOCK_RW);
269 switch (cmd) {
270 case LOCK_SH:
271 return F_RDLCK;
272 case LOCK_EX:
273 return F_WRLCK;
274 case LOCK_UN:
275 return F_UNLCK;
276 }
277 return -EINVAL;
278}
279
280/* Fill in a file_lock structure with an appropriate FLOCK lock. */
281static int flock_make_lock(struct file *filp, struct file_lock **lock,
282 unsigned int cmd)
283{
284 struct file_lock *fl;
285 int type = flock_translate_cmd(cmd);
286 if (type < 0)
287 return type;
288
289 fl = locks_alloc_lock();
290 if (fl == NULL)
291 return -ENOMEM;
292
293 fl->fl_file = filp;
294 fl->fl_pid = current->tgid;
295 fl->fl_flags = FL_FLOCK;
296 fl->fl_type = type;
297 fl->fl_end = OFFSET_MAX;
298
299 *lock = fl;
300 return 0;
301}
302
303static int assign_type(struct file_lock *fl, int type)
304{
305 switch (type) {
306 case F_RDLCK:
307 case F_WRLCK:
308 case F_UNLCK:
309 fl->fl_type = type;
310 break;
311 default:
312 return -EINVAL;
313 }
314 return 0;
315}
316
317/* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
318 * style lock.
319 */
320static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
321 struct flock *l)
322{
323 off_t start, end;
324
325 switch (l->l_whence) {
326 case 0: /*SEEK_SET*/
327 start = 0;
328 break;
329 case 1: /*SEEK_CUR*/
330 start = filp->f_pos;
331 break;
332 case 2: /*SEEK_END*/
333 start = i_size_read(filp->f_dentry->d_inode);
334 break;
335 default:
336 return -EINVAL;
337 }
338
339 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
340 POSIX-2001 defines it. */
341 start += l->l_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 if (start < 0)
343 return -EINVAL;
Trond Myklebust4c780a42005-10-18 14:20:21 -0700344 fl->fl_end = OFFSET_MAX;
345 if (l->l_len > 0) {
346 end = start + l->l_len - 1;
347 fl->fl_end = end;
348 } else if (l->l_len < 0) {
349 end = start - 1;
350 fl->fl_end = end;
351 start += l->l_len;
352 if (start < 0)
353 return -EINVAL;
354 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 fl->fl_start = start; /* we record the absolute position */
Trond Myklebust4c780a42005-10-18 14:20:21 -0700356 if (fl->fl_end < fl->fl_start)
357 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358
359 fl->fl_owner = current->files;
360 fl->fl_pid = current->tgid;
361 fl->fl_file = filp;
362 fl->fl_flags = FL_POSIX;
363 fl->fl_ops = NULL;
364 fl->fl_lmops = NULL;
365
366 return assign_type(fl, l->l_type);
367}
368
369#if BITS_PER_LONG == 32
370static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
371 struct flock64 *l)
372{
373 loff_t start;
374
375 switch (l->l_whence) {
376 case 0: /*SEEK_SET*/
377 start = 0;
378 break;
379 case 1: /*SEEK_CUR*/
380 start = filp->f_pos;
381 break;
382 case 2: /*SEEK_END*/
383 start = i_size_read(filp->f_dentry->d_inode);
384 break;
385 default:
386 return -EINVAL;
387 }
388
Trond Myklebust4c780a42005-10-18 14:20:21 -0700389 start += l->l_start;
390 if (start < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 return -EINVAL;
Trond Myklebust4c780a42005-10-18 14:20:21 -0700392 fl->fl_end = OFFSET_MAX;
393 if (l->l_len > 0) {
394 fl->fl_end = start + l->l_len - 1;
395 } else if (l->l_len < 0) {
396 fl->fl_end = start - 1;
397 start += l->l_len;
398 if (start < 0)
399 return -EINVAL;
400 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401 fl->fl_start = start; /* we record the absolute position */
Trond Myklebust4c780a42005-10-18 14:20:21 -0700402 if (fl->fl_end < fl->fl_start)
403 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404
405 fl->fl_owner = current->files;
406 fl->fl_pid = current->tgid;
407 fl->fl_file = filp;
408 fl->fl_flags = FL_POSIX;
409 fl->fl_ops = NULL;
410 fl->fl_lmops = NULL;
411
412 switch (l->l_type) {
413 case F_RDLCK:
414 case F_WRLCK:
415 case F_UNLCK:
416 fl->fl_type = l->l_type;
417 break;
418 default:
419 return -EINVAL;
420 }
421
422 return (0);
423}
424#endif
425
426/* default lease lock manager operations */
427static void lease_break_callback(struct file_lock *fl)
428{
429 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
430}
431
432static void lease_release_private_callback(struct file_lock *fl)
433{
434 if (!fl->fl_file)
435 return;
436
437 f_delown(fl->fl_file);
438 fl->fl_file->f_owner.signum = 0;
439}
440
Adrian Bunk75c96f82005-05-05 16:16:09 -0700441static int lease_mylease_callback(struct file_lock *fl, struct file_lock *try)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442{
443 return fl->fl_file == try->fl_file;
444}
445
Adrian Bunk75c96f82005-05-05 16:16:09 -0700446static struct lock_manager_operations lease_manager_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 .fl_break = lease_break_callback,
448 .fl_release_private = lease_release_private_callback,
449 .fl_mylease = lease_mylease_callback,
450 .fl_change = lease_modify,
451};
452
453/*
454 * Initialize a lease, use the default lock manager operations
455 */
456static int lease_init(struct file *filp, int type, struct file_lock *fl)
457 {
458 fl->fl_owner = current->files;
459 fl->fl_pid = current->tgid;
460
461 fl->fl_file = filp;
462 fl->fl_flags = FL_LEASE;
463 if (assign_type(fl, type) != 0) {
464 locks_free_lock(fl);
465 return -EINVAL;
466 }
467 fl->fl_start = 0;
468 fl->fl_end = OFFSET_MAX;
469 fl->fl_ops = NULL;
470 fl->fl_lmops = &lease_manager_ops;
471 return 0;
472}
473
474/* Allocate a file_lock initialised to this type of lease */
475static int lease_alloc(struct file *filp, int type, struct file_lock **flp)
476{
477 struct file_lock *fl = locks_alloc_lock();
478 int error;
479
480 if (fl == NULL)
481 return -ENOMEM;
482
483 error = lease_init(filp, type, fl);
484 if (error)
485 return error;
486 *flp = fl;
487 return 0;
488}
489
490/* Check if two locks overlap each other.
491 */
492static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
493{
494 return ((fl1->fl_end >= fl2->fl_start) &&
495 (fl2->fl_end >= fl1->fl_start));
496}
497
498/*
499 * Check whether two locks have the same owner.
500 */
Matt Mackall33443c42006-01-08 01:05:22 -0800501static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502{
503 if (fl1->fl_lmops && fl1->fl_lmops->fl_compare_owner)
504 return fl2->fl_lmops == fl1->fl_lmops &&
505 fl1->fl_lmops->fl_compare_owner(fl1, fl2);
506 return fl1->fl_owner == fl2->fl_owner;
507}
508
509/* Remove waiter from blocker's block list.
510 * When blocker ends up pointing to itself then the list is empty.
511 */
Matt Mackall33443c42006-01-08 01:05:22 -0800512static void __locks_delete_block(struct file_lock *waiter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513{
514 list_del_init(&waiter->fl_block);
515 list_del_init(&waiter->fl_link);
516 waiter->fl_next = NULL;
517}
518
519/*
520 */
521static void locks_delete_block(struct file_lock *waiter)
522{
523 lock_kernel();
524 __locks_delete_block(waiter);
525 unlock_kernel();
526}
527
528/* Insert waiter into blocker's block list.
529 * We use a circular list so that processes can be easily woken up in
530 * the order they blocked. The documentation doesn't require this but
531 * it seems like the reasonable thing to do.
532 */
533static void locks_insert_block(struct file_lock *blocker,
534 struct file_lock *waiter)
535{
536 if (!list_empty(&waiter->fl_block)) {
537 printk(KERN_ERR "locks_insert_block: removing duplicated lock "
538 "(pid=%d %Ld-%Ld type=%d)\n", waiter->fl_pid,
539 waiter->fl_start, waiter->fl_end, waiter->fl_type);
540 __locks_delete_block(waiter);
541 }
542 list_add_tail(&waiter->fl_block, &blocker->fl_block);
543 waiter->fl_next = blocker;
544 if (IS_POSIX(blocker))
545 list_add(&waiter->fl_link, &blocked_list);
546}
547
548/* Wake up processes blocked waiting for blocker.
549 * If told to wait then schedule the processes until the block list
550 * is empty, otherwise empty the block list ourselves.
551 */
552static void locks_wake_up_blocks(struct file_lock *blocker)
553{
554 while (!list_empty(&blocker->fl_block)) {
555 struct file_lock *waiter = list_entry(blocker->fl_block.next,
556 struct file_lock, fl_block);
557 __locks_delete_block(waiter);
558 if (waiter->fl_lmops && waiter->fl_lmops->fl_notify)
559 waiter->fl_lmops->fl_notify(waiter);
560 else
561 wake_up(&waiter->fl_wait);
562 }
563}
564
565/* Insert file lock fl into an inode's lock list at the position indicated
566 * by pos. At the same time add the lock to the global file lock list.
567 */
568static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
569{
570 list_add(&fl->fl_link, &file_lock_list);
571
572 /* insert into file's list */
573 fl->fl_next = *pos;
574 *pos = fl;
575
576 if (fl->fl_ops && fl->fl_ops->fl_insert)
577 fl->fl_ops->fl_insert(fl);
578}
579
580/*
581 * Delete a lock and then free it.
582 * Wake up processes that are blocked waiting for this lock,
583 * notify the FS that the lock has been cleared and
584 * finally free the lock.
585 */
586static void locks_delete_lock(struct file_lock **thisfl_p)
587{
588 struct file_lock *fl = *thisfl_p;
589
590 *thisfl_p = fl->fl_next;
591 fl->fl_next = NULL;
592 list_del_init(&fl->fl_link);
593
594 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
595 if (fl->fl_fasync != NULL) {
596 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
597 fl->fl_fasync = NULL;
598 }
599
600 if (fl->fl_ops && fl->fl_ops->fl_remove)
601 fl->fl_ops->fl_remove(fl);
602
603 locks_wake_up_blocks(fl);
604 locks_free_lock(fl);
605}
606
607/* Determine if lock sys_fl blocks lock caller_fl. Common functionality
608 * checks for shared/exclusive status of overlapping locks.
609 */
610static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
611{
612 if (sys_fl->fl_type == F_WRLCK)
613 return 1;
614 if (caller_fl->fl_type == F_WRLCK)
615 return 1;
616 return 0;
617}
618
619/* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
620 * checking before calling the locks_conflict().
621 */
622static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
623{
624 /* POSIX locks owned by the same process do not conflict with
625 * each other.
626 */
627 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
628 return (0);
629
630 /* Check whether they overlap */
631 if (!locks_overlap(caller_fl, sys_fl))
632 return 0;
633
634 return (locks_conflict(caller_fl, sys_fl));
635}
636
637/* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
638 * checking before calling the locks_conflict().
639 */
640static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
641{
642 /* FLOCK locks referring to the same filp do not conflict with
643 * each other.
644 */
645 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
646 return (0);
647 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
648 return 0;
649
650 return (locks_conflict(caller_fl, sys_fl));
651}
652
653static int interruptible_sleep_on_locked(wait_queue_head_t *fl_wait, int timeout)
654{
655 int result = 0;
656 DECLARE_WAITQUEUE(wait, current);
657
658 __set_current_state(TASK_INTERRUPTIBLE);
659 add_wait_queue(fl_wait, &wait);
660 if (timeout == 0)
661 schedule();
662 else
663 result = schedule_timeout(timeout);
664 if (signal_pending(current))
665 result = -ERESTARTSYS;
666 remove_wait_queue(fl_wait, &wait);
667 __set_current_state(TASK_RUNNING);
668 return result;
669}
670
671static int locks_block_on_timeout(struct file_lock *blocker, struct file_lock *waiter, int time)
672{
673 int result;
674 locks_insert_block(blocker, waiter);
675 result = interruptible_sleep_on_locked(&waiter->fl_wait, time);
676 __locks_delete_block(waiter);
677 return result;
678}
679
Andy Adamson8dc7c312006-03-20 13:44:26 -0500680int
681posix_test_lock(struct file *filp, struct file_lock *fl,
682 struct file_lock *conflock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683{
684 struct file_lock *cfl;
685
686 lock_kernel();
687 for (cfl = filp->f_dentry->d_inode->i_flock; cfl; cfl = cfl->fl_next) {
688 if (!IS_POSIX(cfl))
689 continue;
690 if (posix_locks_conflict(cfl, fl))
691 break;
692 }
Andy Adamson8dc7c312006-03-20 13:44:26 -0500693 if (cfl) {
Trond Myklebust09969052006-03-20 13:44:38 -0500694 __locks_copy_lock(conflock, cfl);
Andy Adamson8dc7c312006-03-20 13:44:26 -0500695 unlock_kernel();
696 return 1;
697 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698 unlock_kernel();
Andy Adamson8dc7c312006-03-20 13:44:26 -0500699 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700}
701
702EXPORT_SYMBOL(posix_test_lock);
703
704/* This function tests for deadlock condition before putting a process to
705 * sleep. The detection scheme is no longer recursive. Recursive was neat,
706 * but dangerous - we risked stack corruption if the lock data was bad, or
707 * if the recursion was too deep for any other reason.
708 *
709 * We rely on the fact that a task can only be on one lock's wait queue
710 * at a time. When we find blocked_task on a wait queue we can re-search
711 * with blocked_task equal to that queue's owner, until either blocked_task
712 * isn't found, or blocked_task is found on a queue owned by my_task.
713 *
714 * Note: the above assumption may not be true when handling lock requests
715 * from a broken NFS client. But broken NFS clients have a lot more to
716 * worry about than proper deadlock detection anyway... --okir
717 */
718int posix_locks_deadlock(struct file_lock *caller_fl,
719 struct file_lock *block_fl)
720{
721 struct list_head *tmp;
722
723next_task:
724 if (posix_same_owner(caller_fl, block_fl))
725 return 1;
726 list_for_each(tmp, &blocked_list) {
727 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
728 if (posix_same_owner(fl, block_fl)) {
729 fl = fl->fl_next;
730 block_fl = fl;
731 goto next_task;
732 }
733 }
734 return 0;
735}
736
737EXPORT_SYMBOL(posix_locks_deadlock);
738
739/* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
740 * at the head of the list, but that's secret knowledge known only to
741 * flock_lock_file and posix_lock_file.
742 */
743static int flock_lock_file(struct file *filp, struct file_lock *new_fl)
744{
745 struct file_lock **before;
746 struct inode * inode = filp->f_dentry->d_inode;
747 int error = 0;
748 int found = 0;
749
750 lock_kernel();
751 for_each_lock(inode, before) {
752 struct file_lock *fl = *before;
753 if (IS_POSIX(fl))
754 break;
755 if (IS_LEASE(fl))
756 continue;
757 if (filp != fl->fl_file)
758 continue;
759 if (new_fl->fl_type == fl->fl_type)
760 goto out;
761 found = 1;
762 locks_delete_lock(before);
763 break;
764 }
765 unlock_kernel();
766
767 if (new_fl->fl_type == F_UNLCK)
768 return 0;
769
770 /*
771 * If a higher-priority process was blocked on the old file lock,
772 * give it the opportunity to lock the file.
773 */
774 if (found)
775 cond_resched();
776
777 lock_kernel();
778 for_each_lock(inode, before) {
779 struct file_lock *fl = *before;
780 if (IS_POSIX(fl))
781 break;
782 if (IS_LEASE(fl))
783 continue;
784 if (!flock_locks_conflict(new_fl, fl))
785 continue;
786 error = -EAGAIN;
787 if (new_fl->fl_flags & FL_SLEEP) {
788 locks_insert_block(fl, new_fl);
789 }
790 goto out;
791 }
792 locks_insert_lock(&inode->i_flock, new_fl);
793 error = 0;
794
795out:
796 unlock_kernel();
797 return error;
798}
799
800EXPORT_SYMBOL(posix_lock_file);
801
802static int __posix_lock_file(struct inode *inode, struct file_lock *request)
803{
804 struct file_lock *fl;
805 struct file_lock *new_fl, *new_fl2;
806 struct file_lock *left = NULL;
807 struct file_lock *right = NULL;
808 struct file_lock **before;
809 int error, added = 0;
810
811 /*
812 * We may need two file_lock structures for this operation,
813 * so we get them in advance to avoid races.
814 */
815 new_fl = locks_alloc_lock();
816 new_fl2 = locks_alloc_lock();
817
818 lock_kernel();
819 if (request->fl_type != F_UNLCK) {
820 for_each_lock(inode, before) {
821 struct file_lock *fl = *before;
822 if (!IS_POSIX(fl))
823 continue;
824 if (!posix_locks_conflict(request, fl))
825 continue;
826 error = -EAGAIN;
827 if (!(request->fl_flags & FL_SLEEP))
828 goto out;
829 error = -EDEADLK;
830 if (posix_locks_deadlock(request, fl))
831 goto out;
832 error = -EAGAIN;
833 locks_insert_block(fl, request);
834 goto out;
835 }
836 }
837
838 /* If we're just looking for a conflict, we're done. */
839 error = 0;
840 if (request->fl_flags & FL_ACCESS)
841 goto out;
842
843 error = -ENOLCK; /* "no luck" */
844 if (!(new_fl && new_fl2))
845 goto out;
846
847 /*
848 * We've allocated the new locks in advance, so there are no
849 * errors possible (and no blocking operations) from here on.
850 *
851 * Find the first old lock with the same owner as the new lock.
852 */
853
854 before = &inode->i_flock;
855
856 /* First skip locks owned by other processes. */
857 while ((fl = *before) && (!IS_POSIX(fl) ||
858 !posix_same_owner(request, fl))) {
859 before = &fl->fl_next;
860 }
861
862 /* Process locks with this owner. */
863 while ((fl = *before) && posix_same_owner(request, fl)) {
864 /* Detect adjacent or overlapping regions (if same lock type)
865 */
866 if (request->fl_type == fl->fl_type) {
Olaf Kirch449231d2005-08-25 16:25:35 -0700867 /* In all comparisons of start vs end, use
868 * "start - 1" rather than "end + 1". If end
869 * is OFFSET_MAX, end + 1 will become negative.
870 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 if (fl->fl_end < request->fl_start - 1)
872 goto next_lock;
873 /* If the next lock in the list has entirely bigger
874 * addresses than the new one, insert the lock here.
875 */
Olaf Kirch449231d2005-08-25 16:25:35 -0700876 if (fl->fl_start - 1 > request->fl_end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 break;
878
879 /* If we come here, the new and old lock are of the
880 * same type and adjacent or overlapping. Make one
881 * lock yielding from the lower start address of both
882 * locks to the higher end address.
883 */
884 if (fl->fl_start > request->fl_start)
885 fl->fl_start = request->fl_start;
886 else
887 request->fl_start = fl->fl_start;
888 if (fl->fl_end < request->fl_end)
889 fl->fl_end = request->fl_end;
890 else
891 request->fl_end = fl->fl_end;
892 if (added) {
893 locks_delete_lock(before);
894 continue;
895 }
896 request = fl;
897 added = 1;
898 }
899 else {
900 /* Processing for different lock types is a bit
901 * more complex.
902 */
903 if (fl->fl_end < request->fl_start)
904 goto next_lock;
905 if (fl->fl_start > request->fl_end)
906 break;
907 if (request->fl_type == F_UNLCK)
908 added = 1;
909 if (fl->fl_start < request->fl_start)
910 left = fl;
911 /* If the next lock in the list has a higher end
912 * address than the new one, insert the new one here.
913 */
914 if (fl->fl_end > request->fl_end) {
915 right = fl;
916 break;
917 }
918 if (fl->fl_start >= request->fl_start) {
919 /* The new lock completely replaces an old
920 * one (This may happen several times).
921 */
922 if (added) {
923 locks_delete_lock(before);
924 continue;
925 }
926 /* Replace the old lock with the new one.
927 * Wake up anybody waiting for the old one,
928 * as the change in lock type might satisfy
929 * their needs.
930 */
931 locks_wake_up_blocks(fl);
932 fl->fl_start = request->fl_start;
933 fl->fl_end = request->fl_end;
934 fl->fl_type = request->fl_type;
Trond Myklebust47831f32006-03-20 13:44:05 -0500935 locks_release_private(fl);
936 locks_copy_private(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 request = fl;
938 added = 1;
939 }
940 }
941 /* Go on to next lock.
942 */
943 next_lock:
944 before = &fl->fl_next;
945 }
946
947 error = 0;
948 if (!added) {
949 if (request->fl_type == F_UNLCK)
950 goto out;
951 locks_copy_lock(new_fl, request);
952 locks_insert_lock(before, new_fl);
953 new_fl = NULL;
954 }
955 if (right) {
956 if (left == right) {
957 /* The new lock breaks the old one in two pieces,
958 * so we have to use the second new lock.
959 */
960 left = new_fl2;
961 new_fl2 = NULL;
962 locks_copy_lock(left, right);
963 locks_insert_lock(before, left);
964 }
965 right->fl_start = request->fl_end + 1;
966 locks_wake_up_blocks(right);
967 }
968 if (left) {
969 left->fl_end = request->fl_start - 1;
970 locks_wake_up_blocks(left);
971 }
972 out:
973 unlock_kernel();
974 /*
975 * Free any unused locks.
976 */
977 if (new_fl)
978 locks_free_lock(new_fl);
979 if (new_fl2)
980 locks_free_lock(new_fl2);
981 return error;
982}
983
984/**
985 * posix_lock_file - Apply a POSIX-style lock to a file
986 * @filp: The file to apply the lock to
987 * @fl: The lock to be applied
988 *
989 * Add a POSIX style lock to a file.
990 * We merge adjacent & overlapping locks whenever possible.
991 * POSIX locks are sorted by owner task, then by starting address
992 */
993int posix_lock_file(struct file *filp, struct file_lock *fl)
994{
995 return __posix_lock_file(filp->f_dentry->d_inode, fl);
996}
997
998/**
999 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1000 * @filp: The file to apply the lock to
1001 * @fl: The lock to be applied
1002 *
1003 * Add a POSIX style lock to a file.
1004 * We merge adjacent & overlapping locks whenever possible.
1005 * POSIX locks are sorted by owner task, then by starting address
1006 */
1007int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1008{
1009 int error;
1010 might_sleep ();
1011 for (;;) {
1012 error = __posix_lock_file(filp->f_dentry->d_inode, fl);
1013 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1014 break;
1015 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1016 if (!error)
1017 continue;
1018
1019 locks_delete_block(fl);
1020 break;
1021 }
1022 return error;
1023}
1024EXPORT_SYMBOL(posix_lock_file_wait);
1025
1026/**
1027 * locks_mandatory_locked - Check for an active lock
1028 * @inode: the file to check
1029 *
1030 * Searches the inode's list of locks to find any POSIX locks which conflict.
1031 * This function is called from locks_verify_locked() only.
1032 */
1033int locks_mandatory_locked(struct inode *inode)
1034{
1035 fl_owner_t owner = current->files;
1036 struct file_lock *fl;
1037
1038 /*
1039 * Search the lock list for this inode for any POSIX locks.
1040 */
1041 lock_kernel();
1042 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1043 if (!IS_POSIX(fl))
1044 continue;
1045 if (fl->fl_owner != owner)
1046 break;
1047 }
1048 unlock_kernel();
1049 return fl ? -EAGAIN : 0;
1050}
1051
1052/**
1053 * locks_mandatory_area - Check for a conflicting lock
1054 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1055 * for shared
1056 * @inode: the file to check
1057 * @filp: how the file was opened (if it was)
1058 * @offset: start of area to check
1059 * @count: length of area to check
1060 *
1061 * Searches the inode's list of locks to find any POSIX locks which conflict.
1062 * This function is called from rw_verify_area() and
1063 * locks_verify_truncate().
1064 */
1065int locks_mandatory_area(int read_write, struct inode *inode,
1066 struct file *filp, loff_t offset,
1067 size_t count)
1068{
1069 struct file_lock fl;
1070 int error;
1071
1072 locks_init_lock(&fl);
1073 fl.fl_owner = current->files;
1074 fl.fl_pid = current->tgid;
1075 fl.fl_file = filp;
1076 fl.fl_flags = FL_POSIX | FL_ACCESS;
1077 if (filp && !(filp->f_flags & O_NONBLOCK))
1078 fl.fl_flags |= FL_SLEEP;
1079 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1080 fl.fl_start = offset;
1081 fl.fl_end = offset + count - 1;
1082
1083 for (;;) {
1084 error = __posix_lock_file(inode, &fl);
1085 if (error != -EAGAIN)
1086 break;
1087 if (!(fl.fl_flags & FL_SLEEP))
1088 break;
1089 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1090 if (!error) {
1091 /*
1092 * If we've been sleeping someone might have
1093 * changed the permissions behind our back.
1094 */
1095 if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID)
1096 continue;
1097 }
1098
1099 locks_delete_block(&fl);
1100 break;
1101 }
1102
1103 return error;
1104}
1105
1106EXPORT_SYMBOL(locks_mandatory_area);
1107
1108/* We already had a lease on this file; just change its type */
1109int lease_modify(struct file_lock **before, int arg)
1110{
1111 struct file_lock *fl = *before;
1112 int error = assign_type(fl, arg);
1113
1114 if (error)
1115 return error;
1116 locks_wake_up_blocks(fl);
1117 if (arg == F_UNLCK)
1118 locks_delete_lock(before);
1119 return 0;
1120}
1121
1122EXPORT_SYMBOL(lease_modify);
1123
1124static void time_out_leases(struct inode *inode)
1125{
1126 struct file_lock **before;
1127 struct file_lock *fl;
1128
1129 before = &inode->i_flock;
1130 while ((fl = *before) && IS_LEASE(fl) && (fl->fl_type & F_INPROGRESS)) {
1131 if ((fl->fl_break_time == 0)
1132 || time_before(jiffies, fl->fl_break_time)) {
1133 before = &fl->fl_next;
1134 continue;
1135 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 lease_modify(before, fl->fl_type & ~F_INPROGRESS);
1137 if (fl == *before) /* lease_modify may have freed fl */
1138 before = &fl->fl_next;
1139 }
1140}
1141
1142/**
1143 * __break_lease - revoke all outstanding leases on file
1144 * @inode: the inode of the file to return
1145 * @mode: the open mode (read or write)
1146 *
1147 * break_lease (inlined for speed) has checked there already
1148 * is a lease on this file. Leases are broken on a call to open()
1149 * or truncate(). This function can sleep unless you
1150 * specified %O_NONBLOCK to your open().
1151 */
1152int __break_lease(struct inode *inode, unsigned int mode)
1153{
1154 int error = 0, future;
1155 struct file_lock *new_fl, *flock;
1156 struct file_lock *fl;
1157 int alloc_err;
1158 unsigned long break_time;
1159 int i_have_this_lease = 0;
1160
1161 alloc_err = lease_alloc(NULL, mode & FMODE_WRITE ? F_WRLCK : F_RDLCK,
1162 &new_fl);
1163
1164 lock_kernel();
1165
1166 time_out_leases(inode);
1167
1168 flock = inode->i_flock;
1169 if ((flock == NULL) || !IS_LEASE(flock))
1170 goto out;
1171
1172 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1173 if (fl->fl_owner == current->files)
1174 i_have_this_lease = 1;
1175
1176 if (mode & FMODE_WRITE) {
1177 /* If we want write access, we have to revoke any lease. */
1178 future = F_UNLCK | F_INPROGRESS;
1179 } else if (flock->fl_type & F_INPROGRESS) {
1180 /* If the lease is already being broken, we just leave it */
1181 future = flock->fl_type;
1182 } else if (flock->fl_type & F_WRLCK) {
1183 /* Downgrade the exclusive lease to a read-only lease. */
1184 future = F_RDLCK | F_INPROGRESS;
1185 } else {
1186 /* the existing lease was read-only, so we can read too. */
1187 goto out;
1188 }
1189
1190 if (alloc_err && !i_have_this_lease && ((mode & O_NONBLOCK) == 0)) {
1191 error = alloc_err;
1192 goto out;
1193 }
1194
1195 break_time = 0;
1196 if (lease_break_time > 0) {
1197 break_time = jiffies + lease_break_time * HZ;
1198 if (break_time == 0)
1199 break_time++; /* so that 0 means no break time */
1200 }
1201
1202 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1203 if (fl->fl_type != future) {
1204 fl->fl_type = future;
1205 fl->fl_break_time = break_time;
1206 /* lease must have lmops break callback */
1207 fl->fl_lmops->fl_break(fl);
1208 }
1209 }
1210
1211 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1212 error = -EWOULDBLOCK;
1213 goto out;
1214 }
1215
1216restart:
1217 break_time = flock->fl_break_time;
1218 if (break_time != 0) {
1219 break_time -= jiffies;
1220 if (break_time == 0)
1221 break_time++;
1222 }
1223 error = locks_block_on_timeout(flock, new_fl, break_time);
1224 if (error >= 0) {
1225 if (error == 0)
1226 time_out_leases(inode);
1227 /* Wait for the next lease that has not been broken yet */
1228 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1229 flock = flock->fl_next) {
1230 if (flock->fl_type & F_INPROGRESS)
1231 goto restart;
1232 }
1233 error = 0;
1234 }
1235
1236out:
1237 unlock_kernel();
1238 if (!alloc_err)
1239 locks_free_lock(new_fl);
1240 return error;
1241}
1242
1243EXPORT_SYMBOL(__break_lease);
1244
1245/**
1246 * lease_get_mtime
1247 * @inode: the inode
1248 * @time: pointer to a timespec which will contain the last modified time
1249 *
1250 * This is to force NFS clients to flush their caches for files with
1251 * exclusive leases. The justification is that if someone has an
1252 * exclusive lease, then they could be modifiying it.
1253 */
1254void lease_get_mtime(struct inode *inode, struct timespec *time)
1255{
1256 struct file_lock *flock = inode->i_flock;
1257 if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK))
1258 *time = current_fs_time(inode->i_sb);
1259 else
1260 *time = inode->i_mtime;
1261}
1262
1263EXPORT_SYMBOL(lease_get_mtime);
1264
1265/**
1266 * fcntl_getlease - Enquire what lease is currently active
1267 * @filp: the file
1268 *
1269 * The value returned by this function will be one of
1270 * (if no lease break is pending):
1271 *
1272 * %F_RDLCK to indicate a shared lease is held.
1273 *
1274 * %F_WRLCK to indicate an exclusive lease is held.
1275 *
1276 * %F_UNLCK to indicate no lease is held.
1277 *
1278 * (if a lease break is pending):
1279 *
1280 * %F_RDLCK to indicate an exclusive lease needs to be
1281 * changed to a shared lease (or removed).
1282 *
1283 * %F_UNLCK to indicate the lease needs to be removed.
1284 *
1285 * XXX: sfr & willy disagree over whether F_INPROGRESS
1286 * should be returned to userspace.
1287 */
1288int fcntl_getlease(struct file *filp)
1289{
1290 struct file_lock *fl;
1291 int type = F_UNLCK;
1292
1293 lock_kernel();
1294 time_out_leases(filp->f_dentry->d_inode);
1295 for (fl = filp->f_dentry->d_inode->i_flock; fl && IS_LEASE(fl);
1296 fl = fl->fl_next) {
1297 if (fl->fl_file == filp) {
1298 type = fl->fl_type & ~F_INPROGRESS;
1299 break;
1300 }
1301 }
1302 unlock_kernel();
1303 return type;
1304}
1305
1306/**
1307 * __setlease - sets a lease on an open file
1308 * @filp: file pointer
1309 * @arg: type of lease to obtain
1310 * @flp: input - file_lock to use, output - file_lock inserted
1311 *
1312 * The (input) flp->fl_lmops->fl_break function is required
1313 * by break_lease().
1314 *
1315 * Called with kernel lock held.
1316 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07001317static int __setlease(struct file *filp, long arg, struct file_lock **flp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318{
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001319 struct file_lock *fl, **before, **my_before = NULL, *lease;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320 struct dentry *dentry = filp->f_dentry;
1321 struct inode *inode = dentry->d_inode;
1322 int error, rdlease_count = 0, wrlease_count = 0;
1323
1324 time_out_leases(inode);
1325
1326 error = -EINVAL;
1327 if (!flp || !(*flp) || !(*flp)->fl_lmops || !(*flp)->fl_lmops->fl_break)
1328 goto out;
1329
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001330 lease = *flp;
1331
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 error = -EAGAIN;
1333 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1334 goto out;
1335 if ((arg == F_WRLCK)
1336 && ((atomic_read(&dentry->d_count) > 1)
1337 || (atomic_read(&inode->i_count) > 1)))
1338 goto out;
1339
1340 /*
1341 * At this point, we know that if there is an exclusive
1342 * lease on this file, then we hold it on this filp
1343 * (otherwise our open of this file would have blocked).
1344 * And if we are trying to acquire an exclusive lease,
1345 * then the file is not open by anyone (including us)
1346 * except for this filp.
1347 */
1348 for (before = &inode->i_flock;
1349 ((fl = *before) != NULL) && IS_LEASE(fl);
1350 before = &fl->fl_next) {
1351 if (lease->fl_lmops->fl_mylease(fl, lease))
1352 my_before = before;
1353 else if (fl->fl_type == (F_INPROGRESS | F_UNLCK))
1354 /*
1355 * Someone is in the process of opening this
1356 * file for writing so we may not take an
1357 * exclusive lease on it.
1358 */
1359 wrlease_count++;
1360 else
1361 rdlease_count++;
1362 }
1363
1364 if ((arg == F_RDLCK && (wrlease_count > 0)) ||
1365 (arg == F_WRLCK && ((rdlease_count + wrlease_count) > 0)))
1366 goto out;
1367
1368 if (my_before != NULL) {
1369 error = lease->fl_lmops->fl_change(my_before, arg);
1370 goto out;
1371 }
1372
1373 error = 0;
1374 if (arg == F_UNLCK)
1375 goto out;
1376
1377 error = -EINVAL;
1378 if (!leases_enable)
1379 goto out;
1380
1381 error = lease_alloc(filp, arg, &fl);
1382 if (error)
1383 goto out;
1384
1385 locks_copy_lock(fl, lease);
1386
1387 locks_insert_lock(before, fl);
1388
1389 *flp = fl;
1390out:
1391 return error;
1392}
1393
1394 /**
1395 * setlease - sets a lease on an open file
1396 * @filp: file pointer
1397 * @arg: type of lease to obtain
1398 * @lease: file_lock to use
1399 *
1400 * Call this to establish a lease on the file.
1401 * The fl_lmops fl_break function is required by break_lease
1402 */
1403
1404int setlease(struct file *filp, long arg, struct file_lock **lease)
1405{
1406 struct dentry *dentry = filp->f_dentry;
1407 struct inode *inode = dentry->d_inode;
1408 int error;
1409
1410 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1411 return -EACCES;
1412 if (!S_ISREG(inode->i_mode))
1413 return -EINVAL;
1414 error = security_file_lock(filp, arg);
1415 if (error)
1416 return error;
1417
1418 lock_kernel();
1419 error = __setlease(filp, arg, lease);
1420 unlock_kernel();
1421
1422 return error;
1423}
1424
1425EXPORT_SYMBOL(setlease);
1426
1427/**
1428 * fcntl_setlease - sets a lease on an open file
1429 * @fd: open file descriptor
1430 * @filp: file pointer
1431 * @arg: type of lease to obtain
1432 *
1433 * Call this fcntl to establish a lease on the file.
1434 * Note that you also need to call %F_SETSIG to
1435 * receive a signal when the lease is broken.
1436 */
1437int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1438{
1439 struct file_lock fl, *flp = &fl;
1440 struct dentry *dentry = filp->f_dentry;
1441 struct inode *inode = dentry->d_inode;
1442 int error;
1443
1444 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1445 return -EACCES;
1446 if (!S_ISREG(inode->i_mode))
1447 return -EINVAL;
1448 error = security_file_lock(filp, arg);
1449 if (error)
1450 return error;
1451
1452 locks_init_lock(&fl);
1453 error = lease_init(filp, arg, &fl);
1454 if (error)
1455 return error;
1456
1457 lock_kernel();
1458
1459 error = __setlease(filp, arg, &flp);
J. Bruce Fieldsdc15ae12005-11-10 19:08:00 -05001460 if (error || arg == F_UNLCK)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 goto out_unlock;
1462
1463 error = fasync_helper(fd, filp, 1, &flp->fl_fasync);
1464 if (error < 0) {
1465 /* remove lease just inserted by __setlease */
1466 flp->fl_type = F_UNLCK | F_INPROGRESS;
1467 flp->fl_break_time = jiffies- 10;
1468 time_out_leases(inode);
1469 goto out_unlock;
1470 }
1471
1472 error = f_setown(filp, current->pid, 0);
1473out_unlock:
1474 unlock_kernel();
1475 return error;
1476}
1477
1478/**
1479 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1480 * @filp: The file to apply the lock to
1481 * @fl: The lock to be applied
1482 *
1483 * Add a FLOCK style lock to a file.
1484 */
1485int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1486{
1487 int error;
1488 might_sleep();
1489 for (;;) {
1490 error = flock_lock_file(filp, fl);
1491 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1492 break;
1493 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1494 if (!error)
1495 continue;
1496
1497 locks_delete_block(fl);
1498 break;
1499 }
1500 return error;
1501}
1502
1503EXPORT_SYMBOL(flock_lock_file_wait);
1504
1505/**
1506 * sys_flock: - flock() system call.
1507 * @fd: the file descriptor to lock.
1508 * @cmd: the type of lock to apply.
1509 *
1510 * Apply a %FL_FLOCK style lock to an open file descriptor.
1511 * The @cmd can be one of
1512 *
1513 * %LOCK_SH -- a shared lock.
1514 *
1515 * %LOCK_EX -- an exclusive lock.
1516 *
1517 * %LOCK_UN -- remove an existing lock.
1518 *
1519 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1520 *
1521 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1522 * processes read and write access respectively.
1523 */
1524asmlinkage long sys_flock(unsigned int fd, unsigned int cmd)
1525{
1526 struct file *filp;
1527 struct file_lock *lock;
1528 int can_sleep, unlock;
1529 int error;
1530
1531 error = -EBADF;
1532 filp = fget(fd);
1533 if (!filp)
1534 goto out;
1535
1536 can_sleep = !(cmd & LOCK_NB);
1537 cmd &= ~LOCK_NB;
1538 unlock = (cmd == LOCK_UN);
1539
1540 if (!unlock && !(cmd & LOCK_MAND) && !(filp->f_mode & 3))
1541 goto out_putf;
1542
1543 error = flock_make_lock(filp, &lock, cmd);
1544 if (error)
1545 goto out_putf;
1546 if (can_sleep)
1547 lock->fl_flags |= FL_SLEEP;
1548
1549 error = security_file_lock(filp, cmd);
1550 if (error)
1551 goto out_free;
1552
1553 if (filp->f_op && filp->f_op->flock)
1554 error = filp->f_op->flock(filp,
1555 (can_sleep) ? F_SETLKW : F_SETLK,
1556 lock);
1557 else
1558 error = flock_lock_file_wait(filp, lock);
1559
1560 out_free:
1561 if (list_empty(&lock->fl_link)) {
1562 locks_free_lock(lock);
1563 }
1564
1565 out_putf:
1566 fput(filp);
1567 out:
1568 return error;
1569}
1570
1571/* Report the first existing lock that would conflict with l.
1572 * This implements the F_GETLK command of fcntl().
1573 */
1574int fcntl_getlk(struct file *filp, struct flock __user *l)
1575{
Andy Adamson8dc7c312006-03-20 13:44:26 -05001576 struct file_lock *fl, cfl, file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577 struct flock flock;
1578 int error;
1579
1580 error = -EFAULT;
1581 if (copy_from_user(&flock, l, sizeof(flock)))
1582 goto out;
1583 error = -EINVAL;
1584 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1585 goto out;
1586
1587 error = flock_to_posix_lock(filp, &file_lock, &flock);
1588 if (error)
1589 goto out;
1590
1591 if (filp->f_op && filp->f_op->lock) {
1592 error = filp->f_op->lock(filp, F_GETLK, &file_lock);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00001593 if (file_lock.fl_ops && file_lock.fl_ops->fl_release_private)
1594 file_lock.fl_ops->fl_release_private(&file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595 if (error < 0)
1596 goto out;
1597 else
1598 fl = (file_lock.fl_type == F_UNLCK ? NULL : &file_lock);
1599 } else {
Andy Adamson8dc7c312006-03-20 13:44:26 -05001600 fl = (posix_test_lock(filp, &file_lock, &cfl) ? &cfl : NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601 }
1602
1603 flock.l_type = F_UNLCK;
1604 if (fl != NULL) {
1605 flock.l_pid = fl->fl_pid;
1606#if BITS_PER_LONG == 32
1607 /*
1608 * Make sure we can represent the posix lock via
1609 * legacy 32bit flock.
1610 */
1611 error = -EOVERFLOW;
1612 if (fl->fl_start > OFFT_OFFSET_MAX)
1613 goto out;
1614 if ((fl->fl_end != OFFSET_MAX)
1615 && (fl->fl_end > OFFT_OFFSET_MAX))
1616 goto out;
1617#endif
1618 flock.l_start = fl->fl_start;
1619 flock.l_len = fl->fl_end == OFFSET_MAX ? 0 :
1620 fl->fl_end - fl->fl_start + 1;
1621 flock.l_whence = 0;
1622 flock.l_type = fl->fl_type;
1623 }
1624 error = -EFAULT;
1625 if (!copy_to_user(l, &flock, sizeof(flock)))
1626 error = 0;
1627out:
1628 return error;
1629}
1630
1631/* Apply the lock described by l to an open file descriptor.
1632 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1633 */
Peter Staubachc2936212005-07-27 11:45:09 -07001634int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1635 struct flock __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636{
1637 struct file_lock *file_lock = locks_alloc_lock();
1638 struct flock flock;
1639 struct inode *inode;
1640 int error;
1641
1642 if (file_lock == NULL)
1643 return -ENOLCK;
1644
1645 /*
1646 * This might block, so we do it before checking the inode.
1647 */
1648 error = -EFAULT;
1649 if (copy_from_user(&flock, l, sizeof(flock)))
1650 goto out;
1651
1652 inode = filp->f_dentry->d_inode;
1653
1654 /* Don't allow mandatory locks on files that may be memory mapped
1655 * and shared.
1656 */
1657 if (IS_MANDLOCK(inode) &&
1658 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1659 mapping_writably_mapped(filp->f_mapping)) {
1660 error = -EAGAIN;
1661 goto out;
1662 }
1663
Peter Staubachc2936212005-07-27 11:45:09 -07001664again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 error = flock_to_posix_lock(filp, file_lock, &flock);
1666 if (error)
1667 goto out;
1668 if (cmd == F_SETLKW) {
1669 file_lock->fl_flags |= FL_SLEEP;
1670 }
1671
1672 error = -EBADF;
1673 switch (flock.l_type) {
1674 case F_RDLCK:
1675 if (!(filp->f_mode & FMODE_READ))
1676 goto out;
1677 break;
1678 case F_WRLCK:
1679 if (!(filp->f_mode & FMODE_WRITE))
1680 goto out;
1681 break;
1682 case F_UNLCK:
1683 break;
1684 default:
1685 error = -EINVAL;
1686 goto out;
1687 }
1688
1689 error = security_file_lock(filp, file_lock->fl_type);
1690 if (error)
1691 goto out;
1692
Peter Staubachc2936212005-07-27 11:45:09 -07001693 if (filp->f_op && filp->f_op->lock != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694 error = filp->f_op->lock(filp, cmd, file_lock);
Peter Staubachc2936212005-07-27 11:45:09 -07001695 else {
1696 for (;;) {
1697 error = __posix_lock_file(inode, file_lock);
1698 if ((error != -EAGAIN) || (cmd == F_SETLK))
1699 break;
1700 error = wait_event_interruptible(file_lock->fl_wait,
1701 !file_lock->fl_next);
1702 if (!error)
1703 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704
Peter Staubachc2936212005-07-27 11:45:09 -07001705 locks_delete_block(file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706 break;
Peter Staubachc2936212005-07-27 11:45:09 -07001707 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708 }
1709
Peter Staubachc2936212005-07-27 11:45:09 -07001710 /*
1711 * Attempt to detect a close/fcntl race and recover by
1712 * releasing the lock that was just acquired.
1713 */
1714 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1715 flock.l_type = F_UNLCK;
1716 goto again;
1717 }
1718
1719out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720 locks_free_lock(file_lock);
1721 return error;
1722}
1723
1724#if BITS_PER_LONG == 32
1725/* Report the first existing lock that would conflict with l.
1726 * This implements the F_GETLK command of fcntl().
1727 */
1728int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1729{
Andy Adamson8dc7c312006-03-20 13:44:26 -05001730 struct file_lock *fl, cfl, file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731 struct flock64 flock;
1732 int error;
1733
1734 error = -EFAULT;
1735 if (copy_from_user(&flock, l, sizeof(flock)))
1736 goto out;
1737 error = -EINVAL;
1738 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1739 goto out;
1740
1741 error = flock64_to_posix_lock(filp, &file_lock, &flock);
1742 if (error)
1743 goto out;
1744
1745 if (filp->f_op && filp->f_op->lock) {
1746 error = filp->f_op->lock(filp, F_GETLK, &file_lock);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00001747 if (file_lock.fl_ops && file_lock.fl_ops->fl_release_private)
1748 file_lock.fl_ops->fl_release_private(&file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749 if (error < 0)
1750 goto out;
1751 else
1752 fl = (file_lock.fl_type == F_UNLCK ? NULL : &file_lock);
1753 } else {
Andy Adamson8dc7c312006-03-20 13:44:26 -05001754 fl = (posix_test_lock(filp, &file_lock, &cfl) ? &cfl : NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 }
1756
1757 flock.l_type = F_UNLCK;
1758 if (fl != NULL) {
1759 flock.l_pid = fl->fl_pid;
1760 flock.l_start = fl->fl_start;
1761 flock.l_len = fl->fl_end == OFFSET_MAX ? 0 :
1762 fl->fl_end - fl->fl_start + 1;
1763 flock.l_whence = 0;
1764 flock.l_type = fl->fl_type;
1765 }
1766 error = -EFAULT;
1767 if (!copy_to_user(l, &flock, sizeof(flock)))
1768 error = 0;
1769
1770out:
1771 return error;
1772}
1773
1774/* Apply the lock described by l to an open file descriptor.
1775 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1776 */
Peter Staubachc2936212005-07-27 11:45:09 -07001777int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
1778 struct flock64 __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779{
1780 struct file_lock *file_lock = locks_alloc_lock();
1781 struct flock64 flock;
1782 struct inode *inode;
1783 int error;
1784
1785 if (file_lock == NULL)
1786 return -ENOLCK;
1787
1788 /*
1789 * This might block, so we do it before checking the inode.
1790 */
1791 error = -EFAULT;
1792 if (copy_from_user(&flock, l, sizeof(flock)))
1793 goto out;
1794
1795 inode = filp->f_dentry->d_inode;
1796
1797 /* Don't allow mandatory locks on files that may be memory mapped
1798 * and shared.
1799 */
1800 if (IS_MANDLOCK(inode) &&
1801 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1802 mapping_writably_mapped(filp->f_mapping)) {
1803 error = -EAGAIN;
1804 goto out;
1805 }
1806
Peter Staubachc2936212005-07-27 11:45:09 -07001807again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808 error = flock64_to_posix_lock(filp, file_lock, &flock);
1809 if (error)
1810 goto out;
1811 if (cmd == F_SETLKW64) {
1812 file_lock->fl_flags |= FL_SLEEP;
1813 }
1814
1815 error = -EBADF;
1816 switch (flock.l_type) {
1817 case F_RDLCK:
1818 if (!(filp->f_mode & FMODE_READ))
1819 goto out;
1820 break;
1821 case F_WRLCK:
1822 if (!(filp->f_mode & FMODE_WRITE))
1823 goto out;
1824 break;
1825 case F_UNLCK:
1826 break;
1827 default:
1828 error = -EINVAL;
1829 goto out;
1830 }
1831
1832 error = security_file_lock(filp, file_lock->fl_type);
1833 if (error)
1834 goto out;
1835
Peter Staubachc2936212005-07-27 11:45:09 -07001836 if (filp->f_op && filp->f_op->lock != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837 error = filp->f_op->lock(filp, cmd, file_lock);
Peter Staubachc2936212005-07-27 11:45:09 -07001838 else {
1839 for (;;) {
1840 error = __posix_lock_file(inode, file_lock);
1841 if ((error != -EAGAIN) || (cmd == F_SETLK64))
1842 break;
1843 error = wait_event_interruptible(file_lock->fl_wait,
1844 !file_lock->fl_next);
1845 if (!error)
1846 continue;
1847
1848 locks_delete_block(file_lock);
1849 break;
1850 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851 }
1852
Peter Staubachc2936212005-07-27 11:45:09 -07001853 /*
1854 * Attempt to detect a close/fcntl race and recover by
1855 * releasing the lock that was just acquired.
1856 */
1857 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1858 flock.l_type = F_UNLCK;
1859 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860 }
1861
1862out:
1863 locks_free_lock(file_lock);
1864 return error;
1865}
1866#endif /* BITS_PER_LONG == 32 */
1867
1868/*
1869 * This function is called when the file is being removed
1870 * from the task's fd array. POSIX locks belonging to this task
1871 * are deleted at this time.
1872 */
1873void locks_remove_posix(struct file *filp, fl_owner_t owner)
1874{
1875 struct file_lock lock, **before;
1876
1877 /*
1878 * If there are no locks held on this file, we don't need to call
1879 * posix_lock_file(). Another process could be setting a lock on this
1880 * file at the same time, but we wouldn't remove that lock anyway.
1881 */
1882 before = &filp->f_dentry->d_inode->i_flock;
1883 if (*before == NULL)
1884 return;
1885
1886 lock.fl_type = F_UNLCK;
1887 lock.fl_flags = FL_POSIX;
1888 lock.fl_start = 0;
1889 lock.fl_end = OFFSET_MAX;
1890 lock.fl_owner = owner;
1891 lock.fl_pid = current->tgid;
1892 lock.fl_file = filp;
1893 lock.fl_ops = NULL;
1894 lock.fl_lmops = NULL;
1895
1896 if (filp->f_op && filp->f_op->lock != NULL) {
1897 filp->f_op->lock(filp, F_SETLK, &lock);
1898 goto out;
1899 }
1900
1901 /* Can't use posix_lock_file here; we need to remove it no matter
1902 * which pid we have.
1903 */
1904 lock_kernel();
1905 while (*before != NULL) {
1906 struct file_lock *fl = *before;
1907 if (IS_POSIX(fl) && posix_same_owner(fl, &lock)) {
1908 locks_delete_lock(before);
1909 continue;
1910 }
1911 before = &fl->fl_next;
1912 }
1913 unlock_kernel();
1914out:
1915 if (lock.fl_ops && lock.fl_ops->fl_release_private)
1916 lock.fl_ops->fl_release_private(&lock);
1917}
1918
1919EXPORT_SYMBOL(locks_remove_posix);
1920
1921/*
1922 * This function is called on the last close of an open file.
1923 */
1924void locks_remove_flock(struct file *filp)
1925{
1926 struct inode * inode = filp->f_dentry->d_inode;
1927 struct file_lock *fl;
1928 struct file_lock **before;
1929
1930 if (!inode->i_flock)
1931 return;
1932
1933 if (filp->f_op && filp->f_op->flock) {
1934 struct file_lock fl = {
1935 .fl_pid = current->tgid,
1936 .fl_file = filp,
1937 .fl_flags = FL_FLOCK,
1938 .fl_type = F_UNLCK,
1939 .fl_end = OFFSET_MAX,
1940 };
1941 filp->f_op->flock(filp, F_SETLKW, &fl);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00001942 if (fl.fl_ops && fl.fl_ops->fl_release_private)
1943 fl.fl_ops->fl_release_private(&fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944 }
1945
1946 lock_kernel();
1947 before = &inode->i_flock;
1948
1949 while ((fl = *before) != NULL) {
1950 if (fl->fl_file == filp) {
Peter Staubachc2936212005-07-27 11:45:09 -07001951 if (IS_FLOCK(fl)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952 locks_delete_lock(before);
1953 continue;
1954 }
1955 if (IS_LEASE(fl)) {
1956 lease_modify(before, F_UNLCK);
1957 continue;
1958 }
1959 /* What? */
1960 BUG();
1961 }
1962 before = &fl->fl_next;
1963 }
1964 unlock_kernel();
1965}
1966
1967/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968 * posix_unblock_lock - stop waiting for a file lock
1969 * @filp: how the file was opened
1970 * @waiter: the lock which was waiting
1971 *
1972 * lockd needs to block waiting for locks.
1973 */
J. Bruce Fields64a318e2006-01-03 09:55:46 +01001974int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975posix_unblock_lock(struct file *filp, struct file_lock *waiter)
1976{
J. Bruce Fields64a318e2006-01-03 09:55:46 +01001977 int status = 0;
1978
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979 lock_kernel();
J. Bruce Fields5996a292006-01-03 09:55:44 +01001980 if (waiter->fl_next)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981 __locks_delete_block(waiter);
J. Bruce Fields64a318e2006-01-03 09:55:46 +01001982 else
1983 status = -ENOENT;
J. Bruce Fields5996a292006-01-03 09:55:44 +01001984 unlock_kernel();
J. Bruce Fields64a318e2006-01-03 09:55:46 +01001985 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986}
1987
1988EXPORT_SYMBOL(posix_unblock_lock);
1989
1990static void lock_get_status(char* out, struct file_lock *fl, int id, char *pfx)
1991{
1992 struct inode *inode = NULL;
1993
1994 if (fl->fl_file != NULL)
1995 inode = fl->fl_file->f_dentry->d_inode;
1996
1997 out += sprintf(out, "%d:%s ", id, pfx);
1998 if (IS_POSIX(fl)) {
1999 out += sprintf(out, "%6s %s ",
2000 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2001 (inode == NULL) ? "*NOINODE*" :
2002 (IS_MANDLOCK(inode) &&
2003 (inode->i_mode & (S_IXGRP | S_ISGID)) == S_ISGID) ?
2004 "MANDATORY" : "ADVISORY ");
2005 } else if (IS_FLOCK(fl)) {
2006 if (fl->fl_type & LOCK_MAND) {
2007 out += sprintf(out, "FLOCK MSNFS ");
2008 } else {
2009 out += sprintf(out, "FLOCK ADVISORY ");
2010 }
2011 } else if (IS_LEASE(fl)) {
2012 out += sprintf(out, "LEASE ");
2013 if (fl->fl_type & F_INPROGRESS)
2014 out += sprintf(out, "BREAKING ");
2015 else if (fl->fl_file)
2016 out += sprintf(out, "ACTIVE ");
2017 else
2018 out += sprintf(out, "BREAKER ");
2019 } else {
2020 out += sprintf(out, "UNKNOWN UNKNOWN ");
2021 }
2022 if (fl->fl_type & LOCK_MAND) {
2023 out += sprintf(out, "%s ",
2024 (fl->fl_type & LOCK_READ)
2025 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2026 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2027 } else {
2028 out += sprintf(out, "%s ",
2029 (fl->fl_type & F_INPROGRESS)
2030 ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ "
2031 : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ ");
2032 }
2033 if (inode) {
2034#ifdef WE_CAN_BREAK_LSLK_NOW
2035 out += sprintf(out, "%d %s:%ld ", fl->fl_pid,
2036 inode->i_sb->s_id, inode->i_ino);
2037#else
2038 /* userspace relies on this representation of dev_t ;-( */
2039 out += sprintf(out, "%d %02x:%02x:%ld ", fl->fl_pid,
2040 MAJOR(inode->i_sb->s_dev),
2041 MINOR(inode->i_sb->s_dev), inode->i_ino);
2042#endif
2043 } else {
2044 out += sprintf(out, "%d <none>:0 ", fl->fl_pid);
2045 }
2046 if (IS_POSIX(fl)) {
2047 if (fl->fl_end == OFFSET_MAX)
2048 out += sprintf(out, "%Ld EOF\n", fl->fl_start);
2049 else
2050 out += sprintf(out, "%Ld %Ld\n", fl->fl_start,
2051 fl->fl_end);
2052 } else {
2053 out += sprintf(out, "0 EOF\n");
2054 }
2055}
2056
2057static void move_lock_status(char **p, off_t* pos, off_t offset)
2058{
2059 int len;
2060 len = strlen(*p);
2061 if(*pos >= offset) {
2062 /* the complete line is valid */
2063 *p += len;
2064 *pos += len;
2065 return;
2066 }
2067 if(*pos+len > offset) {
2068 /* use the second part of the line */
2069 int i = offset-*pos;
2070 memmove(*p,*p+i,len-i);
2071 *p += len-i;
2072 *pos += len;
2073 return;
2074 }
2075 /* discard the complete line */
2076 *pos += len;
2077}
2078
2079/**
2080 * get_locks_status - reports lock usage in /proc/locks
2081 * @buffer: address in userspace to write into
2082 * @start: ?
2083 * @offset: how far we are through the buffer
2084 * @length: how much to read
2085 */
2086
2087int get_locks_status(char *buffer, char **start, off_t offset, int length)
2088{
2089 struct list_head *tmp;
2090 char *q = buffer;
2091 off_t pos = 0;
2092 int i = 0;
2093
2094 lock_kernel();
2095 list_for_each(tmp, &file_lock_list) {
2096 struct list_head *btmp;
2097 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
2098 lock_get_status(q, fl, ++i, "");
2099 move_lock_status(&q, &pos, offset);
2100
2101 if(pos >= offset+length)
2102 goto done;
2103
2104 list_for_each(btmp, &fl->fl_block) {
2105 struct file_lock *bfl = list_entry(btmp,
2106 struct file_lock, fl_block);
2107 lock_get_status(q, bfl, i, " ->");
2108 move_lock_status(&q, &pos, offset);
2109
2110 if(pos >= offset+length)
2111 goto done;
2112 }
2113 }
2114done:
2115 unlock_kernel();
2116 *start = buffer;
2117 if(q-buffer < length)
2118 return (q-buffer);
2119 return length;
2120}
2121
2122/**
2123 * lock_may_read - checks that the region is free of locks
2124 * @inode: the inode that is being read
2125 * @start: the first byte to read
2126 * @len: the number of bytes to read
2127 *
2128 * Emulates Windows locking requirements. Whole-file
2129 * mandatory locks (share modes) can prohibit a read and
2130 * byte-range POSIX locks can prohibit a read if they overlap.
2131 *
2132 * N.B. this function is only ever called
2133 * from knfsd and ownership of locks is never checked.
2134 */
2135int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2136{
2137 struct file_lock *fl;
2138 int result = 1;
2139 lock_kernel();
2140 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2141 if (IS_POSIX(fl)) {
2142 if (fl->fl_type == F_RDLCK)
2143 continue;
2144 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2145 continue;
2146 } else if (IS_FLOCK(fl)) {
2147 if (!(fl->fl_type & LOCK_MAND))
2148 continue;
2149 if (fl->fl_type & LOCK_READ)
2150 continue;
2151 } else
2152 continue;
2153 result = 0;
2154 break;
2155 }
2156 unlock_kernel();
2157 return result;
2158}
2159
2160EXPORT_SYMBOL(lock_may_read);
2161
2162/**
2163 * lock_may_write - checks that the region is free of locks
2164 * @inode: the inode that is being written
2165 * @start: the first byte to write
2166 * @len: the number of bytes to write
2167 *
2168 * Emulates Windows locking requirements. Whole-file
2169 * mandatory locks (share modes) can prohibit a write and
2170 * byte-range POSIX locks can prohibit a write if they overlap.
2171 *
2172 * N.B. this function is only ever called
2173 * from knfsd and ownership of locks is never checked.
2174 */
2175int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2176{
2177 struct file_lock *fl;
2178 int result = 1;
2179 lock_kernel();
2180 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2181 if (IS_POSIX(fl)) {
2182 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2183 continue;
2184 } else if (IS_FLOCK(fl)) {
2185 if (!(fl->fl_type & LOCK_MAND))
2186 continue;
2187 if (fl->fl_type & LOCK_WRITE)
2188 continue;
2189 } else
2190 continue;
2191 result = 0;
2192 break;
2193 }
2194 unlock_kernel();
2195 return result;
2196}
2197
2198EXPORT_SYMBOL(lock_may_write);
2199
2200static inline void __steal_locks(struct file *file, fl_owner_t from)
2201{
2202 struct inode *inode = file->f_dentry->d_inode;
2203 struct file_lock *fl = inode->i_flock;
2204
2205 while (fl) {
2206 if (fl->fl_file == file && fl->fl_owner == from)
2207 fl->fl_owner = current->files;
2208 fl = fl->fl_next;
2209 }
2210}
2211
2212/* When getting ready for executing a binary, we make sure that current
2213 * has a files_struct on its own. Before dropping the old files_struct,
2214 * we take over ownership of all locks for all file descriptors we own.
2215 * Note that we may accidentally steal a lock for a file that a sibling
2216 * has created since the unshare_files() call.
2217 */
2218void steal_locks(fl_owner_t from)
2219{
2220 struct files_struct *files = current->files;
2221 int i, j;
Dipankar Sarmabadf1662005-09-09 13:04:10 -07002222 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223
2224 if (from == files)
2225 return;
2226
2227 lock_kernel();
2228 j = 0;
Dipankar Sarma4fb3a532005-09-16 19:28:13 -07002229 rcu_read_lock();
Dipankar Sarmabadf1662005-09-09 13:04:10 -07002230 fdt = files_fdtable(files);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002231 for (;;) {
2232 unsigned long set;
2233 i = j * __NFDBITS;
Dipankar Sarmabadf1662005-09-09 13:04:10 -07002234 if (i >= fdt->max_fdset || i >= fdt->max_fds)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002235 break;
Dipankar Sarmabadf1662005-09-09 13:04:10 -07002236 set = fdt->open_fds->fds_bits[j++];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237 while (set) {
2238 if (set & 1) {
Dipankar Sarmabadf1662005-09-09 13:04:10 -07002239 struct file *file = fdt->fd[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240 if (file)
2241 __steal_locks(file, from);
2242 }
2243 i++;
2244 set >>= 1;
2245 }
2246 }
Dipankar Sarma4fb3a532005-09-16 19:28:13 -07002247 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248 unlock_kernel();
2249}
2250EXPORT_SYMBOL(steal_locks);
2251
2252static int __init filelock_init(void)
2253{
2254 filelock_cache = kmem_cache_create("file_lock_cache",
2255 sizeof(struct file_lock), 0, SLAB_PANIC,
2256 init_once, NULL);
2257 return 0;
2258}
2259
2260core_initcall(filelock_init);