blob: 4d9e71d43e7e2068eeef0859e20c8909d026ccc8 [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{
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{
J. Bruce Fields6dc0fe82006-03-26 01:37:24 -0800536 BUG_ON(!list_empty(&waiter->fl_block));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 list_add_tail(&waiter->fl_block, &blocker->fl_block);
538 waiter->fl_next = blocker;
539 if (IS_POSIX(blocker))
540 list_add(&waiter->fl_link, &blocked_list);
541}
542
543/* Wake up processes blocked waiting for blocker.
544 * If told to wait then schedule the processes until the block list
545 * is empty, otherwise empty the block list ourselves.
546 */
547static void locks_wake_up_blocks(struct file_lock *blocker)
548{
549 while (!list_empty(&blocker->fl_block)) {
550 struct file_lock *waiter = list_entry(blocker->fl_block.next,
551 struct file_lock, fl_block);
552 __locks_delete_block(waiter);
553 if (waiter->fl_lmops && waiter->fl_lmops->fl_notify)
554 waiter->fl_lmops->fl_notify(waiter);
555 else
556 wake_up(&waiter->fl_wait);
557 }
558}
559
560/* Insert file lock fl into an inode's lock list at the position indicated
561 * by pos. At the same time add the lock to the global file lock list.
562 */
563static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
564{
565 list_add(&fl->fl_link, &file_lock_list);
566
567 /* insert into file's list */
568 fl->fl_next = *pos;
569 *pos = fl;
570
571 if (fl->fl_ops && fl->fl_ops->fl_insert)
572 fl->fl_ops->fl_insert(fl);
573}
574
575/*
576 * Delete a lock and then free it.
577 * Wake up processes that are blocked waiting for this lock,
578 * notify the FS that the lock has been cleared and
579 * finally free the lock.
580 */
581static void locks_delete_lock(struct file_lock **thisfl_p)
582{
583 struct file_lock *fl = *thisfl_p;
584
585 *thisfl_p = fl->fl_next;
586 fl->fl_next = NULL;
587 list_del_init(&fl->fl_link);
588
589 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
590 if (fl->fl_fasync != NULL) {
591 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
592 fl->fl_fasync = NULL;
593 }
594
595 if (fl->fl_ops && fl->fl_ops->fl_remove)
596 fl->fl_ops->fl_remove(fl);
597
598 locks_wake_up_blocks(fl);
599 locks_free_lock(fl);
600}
601
602/* Determine if lock sys_fl blocks lock caller_fl. Common functionality
603 * checks for shared/exclusive status of overlapping locks.
604 */
605static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
606{
607 if (sys_fl->fl_type == F_WRLCK)
608 return 1;
609 if (caller_fl->fl_type == F_WRLCK)
610 return 1;
611 return 0;
612}
613
614/* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
615 * checking before calling the locks_conflict().
616 */
617static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
618{
619 /* POSIX locks owned by the same process do not conflict with
620 * each other.
621 */
622 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
623 return (0);
624
625 /* Check whether they overlap */
626 if (!locks_overlap(caller_fl, sys_fl))
627 return 0;
628
629 return (locks_conflict(caller_fl, sys_fl));
630}
631
632/* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
633 * checking before calling the locks_conflict().
634 */
635static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
636{
637 /* FLOCK locks referring to the same filp do not conflict with
638 * each other.
639 */
640 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
641 return (0);
642 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
643 return 0;
644
645 return (locks_conflict(caller_fl, sys_fl));
646}
647
648static int interruptible_sleep_on_locked(wait_queue_head_t *fl_wait, int timeout)
649{
650 int result = 0;
651 DECLARE_WAITQUEUE(wait, current);
652
653 __set_current_state(TASK_INTERRUPTIBLE);
654 add_wait_queue(fl_wait, &wait);
655 if (timeout == 0)
656 schedule();
657 else
658 result = schedule_timeout(timeout);
659 if (signal_pending(current))
660 result = -ERESTARTSYS;
661 remove_wait_queue(fl_wait, &wait);
662 __set_current_state(TASK_RUNNING);
663 return result;
664}
665
666static int locks_block_on_timeout(struct file_lock *blocker, struct file_lock *waiter, int time)
667{
668 int result;
669 locks_insert_block(blocker, waiter);
670 result = interruptible_sleep_on_locked(&waiter->fl_wait, time);
671 __locks_delete_block(waiter);
672 return result;
673}
674
Andy Adamson8dc7c312006-03-20 13:44:26 -0500675int
676posix_test_lock(struct file *filp, struct file_lock *fl,
677 struct file_lock *conflock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678{
679 struct file_lock *cfl;
680
681 lock_kernel();
682 for (cfl = filp->f_dentry->d_inode->i_flock; cfl; cfl = cfl->fl_next) {
683 if (!IS_POSIX(cfl))
684 continue;
685 if (posix_locks_conflict(cfl, fl))
686 break;
687 }
Andy Adamson8dc7c312006-03-20 13:44:26 -0500688 if (cfl) {
Trond Myklebust09969052006-03-20 13:44:38 -0500689 __locks_copy_lock(conflock, cfl);
Andy Adamson8dc7c312006-03-20 13:44:26 -0500690 unlock_kernel();
691 return 1;
692 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 unlock_kernel();
Andy Adamson8dc7c312006-03-20 13:44:26 -0500694 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695}
696
697EXPORT_SYMBOL(posix_test_lock);
698
699/* This function tests for deadlock condition before putting a process to
700 * sleep. The detection scheme is no longer recursive. Recursive was neat,
701 * but dangerous - we risked stack corruption if the lock data was bad, or
702 * if the recursion was too deep for any other reason.
703 *
704 * We rely on the fact that a task can only be on one lock's wait queue
705 * at a time. When we find blocked_task on a wait queue we can re-search
706 * with blocked_task equal to that queue's owner, until either blocked_task
707 * isn't found, or blocked_task is found on a queue owned by my_task.
708 *
709 * Note: the above assumption may not be true when handling lock requests
710 * from a broken NFS client. But broken NFS clients have a lot more to
711 * worry about than proper deadlock detection anyway... --okir
712 */
713int posix_locks_deadlock(struct file_lock *caller_fl,
714 struct file_lock *block_fl)
715{
716 struct list_head *tmp;
717
718next_task:
719 if (posix_same_owner(caller_fl, block_fl))
720 return 1;
721 list_for_each(tmp, &blocked_list) {
722 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
723 if (posix_same_owner(fl, block_fl)) {
724 fl = fl->fl_next;
725 block_fl = fl;
726 goto next_task;
727 }
728 }
729 return 0;
730}
731
732EXPORT_SYMBOL(posix_locks_deadlock);
733
734/* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
735 * at the head of the list, but that's secret knowledge known only to
736 * flock_lock_file and posix_lock_file.
737 */
738static int flock_lock_file(struct file *filp, struct file_lock *new_fl)
739{
740 struct file_lock **before;
741 struct inode * inode = filp->f_dentry->d_inode;
742 int error = 0;
743 int found = 0;
744
745 lock_kernel();
746 for_each_lock(inode, before) {
747 struct file_lock *fl = *before;
748 if (IS_POSIX(fl))
749 break;
750 if (IS_LEASE(fl))
751 continue;
752 if (filp != fl->fl_file)
753 continue;
754 if (new_fl->fl_type == fl->fl_type)
755 goto out;
756 found = 1;
757 locks_delete_lock(before);
758 break;
759 }
760 unlock_kernel();
761
762 if (new_fl->fl_type == F_UNLCK)
763 return 0;
764
765 /*
766 * If a higher-priority process was blocked on the old file lock,
767 * give it the opportunity to lock the file.
768 */
769 if (found)
770 cond_resched();
771
772 lock_kernel();
773 for_each_lock(inode, before) {
774 struct file_lock *fl = *before;
775 if (IS_POSIX(fl))
776 break;
777 if (IS_LEASE(fl))
778 continue;
779 if (!flock_locks_conflict(new_fl, fl))
780 continue;
781 error = -EAGAIN;
782 if (new_fl->fl_flags & FL_SLEEP) {
783 locks_insert_block(fl, new_fl);
784 }
785 goto out;
786 }
787 locks_insert_lock(&inode->i_flock, new_fl);
788 error = 0;
789
790out:
791 unlock_kernel();
792 return error;
793}
794
Andy Adamson5842add2006-03-26 01:37:26 -0800795static int __posix_lock_file_conf(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796{
797 struct file_lock *fl;
798 struct file_lock *new_fl, *new_fl2;
799 struct file_lock *left = NULL;
800 struct file_lock *right = NULL;
801 struct file_lock **before;
802 int error, added = 0;
803
804 /*
805 * We may need two file_lock structures for this operation,
806 * so we get them in advance to avoid races.
807 */
808 new_fl = locks_alloc_lock();
809 new_fl2 = locks_alloc_lock();
810
811 lock_kernel();
812 if (request->fl_type != F_UNLCK) {
813 for_each_lock(inode, before) {
814 struct file_lock *fl = *before;
815 if (!IS_POSIX(fl))
816 continue;
817 if (!posix_locks_conflict(request, fl))
818 continue;
Andy Adamson5842add2006-03-26 01:37:26 -0800819 if (conflock)
820 locks_copy_lock(conflock, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 error = -EAGAIN;
822 if (!(request->fl_flags & FL_SLEEP))
823 goto out;
824 error = -EDEADLK;
825 if (posix_locks_deadlock(request, fl))
826 goto out;
827 error = -EAGAIN;
828 locks_insert_block(fl, request);
829 goto out;
830 }
831 }
832
833 /* If we're just looking for a conflict, we're done. */
834 error = 0;
835 if (request->fl_flags & FL_ACCESS)
836 goto out;
837
838 error = -ENOLCK; /* "no luck" */
839 if (!(new_fl && new_fl2))
840 goto out;
841
842 /*
843 * We've allocated the new locks in advance, so there are no
844 * errors possible (and no blocking operations) from here on.
845 *
846 * Find the first old lock with the same owner as the new lock.
847 */
848
849 before = &inode->i_flock;
850
851 /* First skip locks owned by other processes. */
852 while ((fl = *before) && (!IS_POSIX(fl) ||
853 !posix_same_owner(request, fl))) {
854 before = &fl->fl_next;
855 }
856
857 /* Process locks with this owner. */
858 while ((fl = *before) && posix_same_owner(request, fl)) {
859 /* Detect adjacent or overlapping regions (if same lock type)
860 */
861 if (request->fl_type == fl->fl_type) {
Olaf Kirch449231d2005-08-25 16:25:35 -0700862 /* In all comparisons of start vs end, use
863 * "start - 1" rather than "end + 1". If end
864 * is OFFSET_MAX, end + 1 will become negative.
865 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 if (fl->fl_end < request->fl_start - 1)
867 goto next_lock;
868 /* If the next lock in the list has entirely bigger
869 * addresses than the new one, insert the lock here.
870 */
Olaf Kirch449231d2005-08-25 16:25:35 -0700871 if (fl->fl_start - 1 > request->fl_end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 break;
873
874 /* If we come here, the new and old lock are of the
875 * same type and adjacent or overlapping. Make one
876 * lock yielding from the lower start address of both
877 * locks to the higher end address.
878 */
879 if (fl->fl_start > request->fl_start)
880 fl->fl_start = request->fl_start;
881 else
882 request->fl_start = fl->fl_start;
883 if (fl->fl_end < request->fl_end)
884 fl->fl_end = request->fl_end;
885 else
886 request->fl_end = fl->fl_end;
887 if (added) {
888 locks_delete_lock(before);
889 continue;
890 }
891 request = fl;
892 added = 1;
893 }
894 else {
895 /* Processing for different lock types is a bit
896 * more complex.
897 */
898 if (fl->fl_end < request->fl_start)
899 goto next_lock;
900 if (fl->fl_start > request->fl_end)
901 break;
902 if (request->fl_type == F_UNLCK)
903 added = 1;
904 if (fl->fl_start < request->fl_start)
905 left = fl;
906 /* If the next lock in the list has a higher end
907 * address than the new one, insert the new one here.
908 */
909 if (fl->fl_end > request->fl_end) {
910 right = fl;
911 break;
912 }
913 if (fl->fl_start >= request->fl_start) {
914 /* The new lock completely replaces an old
915 * one (This may happen several times).
916 */
917 if (added) {
918 locks_delete_lock(before);
919 continue;
920 }
921 /* Replace the old lock with the new one.
922 * Wake up anybody waiting for the old one,
923 * as the change in lock type might satisfy
924 * their needs.
925 */
926 locks_wake_up_blocks(fl);
927 fl->fl_start = request->fl_start;
928 fl->fl_end = request->fl_end;
929 fl->fl_type = request->fl_type;
Trond Myklebust47831f32006-03-20 13:44:05 -0500930 locks_release_private(fl);
931 locks_copy_private(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 request = fl;
933 added = 1;
934 }
935 }
936 /* Go on to next lock.
937 */
938 next_lock:
939 before = &fl->fl_next;
940 }
941
942 error = 0;
943 if (!added) {
944 if (request->fl_type == F_UNLCK)
945 goto out;
946 locks_copy_lock(new_fl, request);
947 locks_insert_lock(before, new_fl);
948 new_fl = NULL;
949 }
950 if (right) {
951 if (left == right) {
952 /* The new lock breaks the old one in two pieces,
953 * so we have to use the second new lock.
954 */
955 left = new_fl2;
956 new_fl2 = NULL;
957 locks_copy_lock(left, right);
958 locks_insert_lock(before, left);
959 }
960 right->fl_start = request->fl_end + 1;
961 locks_wake_up_blocks(right);
962 }
963 if (left) {
964 left->fl_end = request->fl_start - 1;
965 locks_wake_up_blocks(left);
966 }
967 out:
968 unlock_kernel();
969 /*
970 * Free any unused locks.
971 */
972 if (new_fl)
973 locks_free_lock(new_fl);
974 if (new_fl2)
975 locks_free_lock(new_fl2);
976 return error;
977}
978
979/**
980 * posix_lock_file - Apply a POSIX-style lock to a file
981 * @filp: The file to apply the lock to
982 * @fl: The lock to be applied
983 *
984 * Add a POSIX style lock to a file.
985 * We merge adjacent & overlapping locks whenever possible.
986 * POSIX locks are sorted by owner task, then by starting address
987 */
988int posix_lock_file(struct file *filp, struct file_lock *fl)
989{
Andy Adamson5842add2006-03-26 01:37:26 -0800990 return __posix_lock_file_conf(filp->f_dentry->d_inode, fl, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991}
Andy Adamson5842add2006-03-26 01:37:26 -0800992EXPORT_SYMBOL(posix_lock_file);
993
994/**
995 * posix_lock_file_conf - Apply a POSIX-style lock to a file
996 * @filp: The file to apply the lock to
997 * @fl: The lock to be applied
998 * @conflock: Place to return a copy of the conflicting lock, if found.
999 *
1000 * Except for the conflock parameter, acts just like posix_lock_file.
1001 */
1002int posix_lock_file_conf(struct file *filp, struct file_lock *fl,
1003 struct file_lock *conflock)
1004{
1005 return __posix_lock_file_conf(filp->f_dentry->d_inode, fl, conflock);
1006}
1007EXPORT_SYMBOL(posix_lock_file_conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008
1009/**
1010 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1011 * @filp: The file to apply the lock to
1012 * @fl: The lock to be applied
1013 *
1014 * Add a POSIX style lock to a file.
1015 * We merge adjacent & overlapping locks whenever possible.
1016 * POSIX locks are sorted by owner task, then by starting address
1017 */
1018int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1019{
1020 int error;
1021 might_sleep ();
1022 for (;;) {
Andy Adamson5842add2006-03-26 01:37:26 -08001023 error = posix_lock_file(filp, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1025 break;
1026 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1027 if (!error)
1028 continue;
1029
1030 locks_delete_block(fl);
1031 break;
1032 }
1033 return error;
1034}
1035EXPORT_SYMBOL(posix_lock_file_wait);
1036
1037/**
1038 * locks_mandatory_locked - Check for an active lock
1039 * @inode: the file to check
1040 *
1041 * Searches the inode's list of locks to find any POSIX locks which conflict.
1042 * This function is called from locks_verify_locked() only.
1043 */
1044int locks_mandatory_locked(struct inode *inode)
1045{
1046 fl_owner_t owner = current->files;
1047 struct file_lock *fl;
1048
1049 /*
1050 * Search the lock list for this inode for any POSIX locks.
1051 */
1052 lock_kernel();
1053 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1054 if (!IS_POSIX(fl))
1055 continue;
1056 if (fl->fl_owner != owner)
1057 break;
1058 }
1059 unlock_kernel();
1060 return fl ? -EAGAIN : 0;
1061}
1062
1063/**
1064 * locks_mandatory_area - Check for a conflicting lock
1065 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1066 * for shared
1067 * @inode: the file to check
1068 * @filp: how the file was opened (if it was)
1069 * @offset: start of area to check
1070 * @count: length of area to check
1071 *
1072 * Searches the inode's list of locks to find any POSIX locks which conflict.
1073 * This function is called from rw_verify_area() and
1074 * locks_verify_truncate().
1075 */
1076int locks_mandatory_area(int read_write, struct inode *inode,
1077 struct file *filp, loff_t offset,
1078 size_t count)
1079{
1080 struct file_lock fl;
1081 int error;
1082
1083 locks_init_lock(&fl);
1084 fl.fl_owner = current->files;
1085 fl.fl_pid = current->tgid;
1086 fl.fl_file = filp;
1087 fl.fl_flags = FL_POSIX | FL_ACCESS;
1088 if (filp && !(filp->f_flags & O_NONBLOCK))
1089 fl.fl_flags |= FL_SLEEP;
1090 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1091 fl.fl_start = offset;
1092 fl.fl_end = offset + count - 1;
1093
1094 for (;;) {
Andy Adamson5842add2006-03-26 01:37:26 -08001095 error = __posix_lock_file_conf(inode, &fl, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 if (error != -EAGAIN)
1097 break;
1098 if (!(fl.fl_flags & FL_SLEEP))
1099 break;
1100 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1101 if (!error) {
1102 /*
1103 * If we've been sleeping someone might have
1104 * changed the permissions behind our back.
1105 */
1106 if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID)
1107 continue;
1108 }
1109
1110 locks_delete_block(&fl);
1111 break;
1112 }
1113
1114 return error;
1115}
1116
1117EXPORT_SYMBOL(locks_mandatory_area);
1118
1119/* We already had a lease on this file; just change its type */
1120int lease_modify(struct file_lock **before, int arg)
1121{
1122 struct file_lock *fl = *before;
1123 int error = assign_type(fl, arg);
1124
1125 if (error)
1126 return error;
1127 locks_wake_up_blocks(fl);
1128 if (arg == F_UNLCK)
1129 locks_delete_lock(before);
1130 return 0;
1131}
1132
1133EXPORT_SYMBOL(lease_modify);
1134
1135static void time_out_leases(struct inode *inode)
1136{
1137 struct file_lock **before;
1138 struct file_lock *fl;
1139
1140 before = &inode->i_flock;
1141 while ((fl = *before) && IS_LEASE(fl) && (fl->fl_type & F_INPROGRESS)) {
1142 if ((fl->fl_break_time == 0)
1143 || time_before(jiffies, fl->fl_break_time)) {
1144 before = &fl->fl_next;
1145 continue;
1146 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 lease_modify(before, fl->fl_type & ~F_INPROGRESS);
1148 if (fl == *before) /* lease_modify may have freed fl */
1149 before = &fl->fl_next;
1150 }
1151}
1152
1153/**
1154 * __break_lease - revoke all outstanding leases on file
1155 * @inode: the inode of the file to return
1156 * @mode: the open mode (read or write)
1157 *
1158 * break_lease (inlined for speed) has checked there already
1159 * is a lease on this file. Leases are broken on a call to open()
1160 * or truncate(). This function can sleep unless you
1161 * specified %O_NONBLOCK to your open().
1162 */
1163int __break_lease(struct inode *inode, unsigned int mode)
1164{
1165 int error = 0, future;
1166 struct file_lock *new_fl, *flock;
1167 struct file_lock *fl;
1168 int alloc_err;
1169 unsigned long break_time;
1170 int i_have_this_lease = 0;
1171
1172 alloc_err = lease_alloc(NULL, mode & FMODE_WRITE ? F_WRLCK : F_RDLCK,
1173 &new_fl);
1174
1175 lock_kernel();
1176
1177 time_out_leases(inode);
1178
1179 flock = inode->i_flock;
1180 if ((flock == NULL) || !IS_LEASE(flock))
1181 goto out;
1182
1183 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1184 if (fl->fl_owner == current->files)
1185 i_have_this_lease = 1;
1186
1187 if (mode & FMODE_WRITE) {
1188 /* If we want write access, we have to revoke any lease. */
1189 future = F_UNLCK | F_INPROGRESS;
1190 } else if (flock->fl_type & F_INPROGRESS) {
1191 /* If the lease is already being broken, we just leave it */
1192 future = flock->fl_type;
1193 } else if (flock->fl_type & F_WRLCK) {
1194 /* Downgrade the exclusive lease to a read-only lease. */
1195 future = F_RDLCK | F_INPROGRESS;
1196 } else {
1197 /* the existing lease was read-only, so we can read too. */
1198 goto out;
1199 }
1200
1201 if (alloc_err && !i_have_this_lease && ((mode & O_NONBLOCK) == 0)) {
1202 error = alloc_err;
1203 goto out;
1204 }
1205
1206 break_time = 0;
1207 if (lease_break_time > 0) {
1208 break_time = jiffies + lease_break_time * HZ;
1209 if (break_time == 0)
1210 break_time++; /* so that 0 means no break time */
1211 }
1212
1213 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1214 if (fl->fl_type != future) {
1215 fl->fl_type = future;
1216 fl->fl_break_time = break_time;
1217 /* lease must have lmops break callback */
1218 fl->fl_lmops->fl_break(fl);
1219 }
1220 }
1221
1222 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1223 error = -EWOULDBLOCK;
1224 goto out;
1225 }
1226
1227restart:
1228 break_time = flock->fl_break_time;
1229 if (break_time != 0) {
1230 break_time -= jiffies;
1231 if (break_time == 0)
1232 break_time++;
1233 }
1234 error = locks_block_on_timeout(flock, new_fl, break_time);
1235 if (error >= 0) {
1236 if (error == 0)
1237 time_out_leases(inode);
1238 /* Wait for the next lease that has not been broken yet */
1239 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1240 flock = flock->fl_next) {
1241 if (flock->fl_type & F_INPROGRESS)
1242 goto restart;
1243 }
1244 error = 0;
1245 }
1246
1247out:
1248 unlock_kernel();
1249 if (!alloc_err)
1250 locks_free_lock(new_fl);
1251 return error;
1252}
1253
1254EXPORT_SYMBOL(__break_lease);
1255
1256/**
1257 * lease_get_mtime
1258 * @inode: the inode
1259 * @time: pointer to a timespec which will contain the last modified time
1260 *
1261 * This is to force NFS clients to flush their caches for files with
1262 * exclusive leases. The justification is that if someone has an
1263 * exclusive lease, then they could be modifiying it.
1264 */
1265void lease_get_mtime(struct inode *inode, struct timespec *time)
1266{
1267 struct file_lock *flock = inode->i_flock;
1268 if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK))
1269 *time = current_fs_time(inode->i_sb);
1270 else
1271 *time = inode->i_mtime;
1272}
1273
1274EXPORT_SYMBOL(lease_get_mtime);
1275
1276/**
1277 * fcntl_getlease - Enquire what lease is currently active
1278 * @filp: the file
1279 *
1280 * The value returned by this function will be one of
1281 * (if no lease break is pending):
1282 *
1283 * %F_RDLCK to indicate a shared lease is held.
1284 *
1285 * %F_WRLCK to indicate an exclusive lease is held.
1286 *
1287 * %F_UNLCK to indicate no lease is held.
1288 *
1289 * (if a lease break is pending):
1290 *
1291 * %F_RDLCK to indicate an exclusive lease needs to be
1292 * changed to a shared lease (or removed).
1293 *
1294 * %F_UNLCK to indicate the lease needs to be removed.
1295 *
1296 * XXX: sfr & willy disagree over whether F_INPROGRESS
1297 * should be returned to userspace.
1298 */
1299int fcntl_getlease(struct file *filp)
1300{
1301 struct file_lock *fl;
1302 int type = F_UNLCK;
1303
1304 lock_kernel();
1305 time_out_leases(filp->f_dentry->d_inode);
1306 for (fl = filp->f_dentry->d_inode->i_flock; fl && IS_LEASE(fl);
1307 fl = fl->fl_next) {
1308 if (fl->fl_file == filp) {
1309 type = fl->fl_type & ~F_INPROGRESS;
1310 break;
1311 }
1312 }
1313 unlock_kernel();
1314 return type;
1315}
1316
1317/**
1318 * __setlease - sets a lease on an open file
1319 * @filp: file pointer
1320 * @arg: type of lease to obtain
1321 * @flp: input - file_lock to use, output - file_lock inserted
1322 *
1323 * The (input) flp->fl_lmops->fl_break function is required
1324 * by break_lease().
1325 *
1326 * Called with kernel lock held.
1327 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07001328static int __setlease(struct file *filp, long arg, struct file_lock **flp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329{
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001330 struct file_lock *fl, **before, **my_before = NULL, *lease;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331 struct dentry *dentry = filp->f_dentry;
1332 struct inode *inode = dentry->d_inode;
1333 int error, rdlease_count = 0, wrlease_count = 0;
1334
1335 time_out_leases(inode);
1336
1337 error = -EINVAL;
1338 if (!flp || !(*flp) || !(*flp)->fl_lmops || !(*flp)->fl_lmops->fl_break)
1339 goto out;
1340
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001341 lease = *flp;
1342
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 error = -EAGAIN;
1344 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1345 goto out;
1346 if ((arg == F_WRLCK)
1347 && ((atomic_read(&dentry->d_count) > 1)
1348 || (atomic_read(&inode->i_count) > 1)))
1349 goto out;
1350
1351 /*
1352 * At this point, we know that if there is an exclusive
1353 * lease on this file, then we hold it on this filp
1354 * (otherwise our open of this file would have blocked).
1355 * And if we are trying to acquire an exclusive lease,
1356 * then the file is not open by anyone (including us)
1357 * except for this filp.
1358 */
1359 for (before = &inode->i_flock;
1360 ((fl = *before) != NULL) && IS_LEASE(fl);
1361 before = &fl->fl_next) {
1362 if (lease->fl_lmops->fl_mylease(fl, lease))
1363 my_before = before;
1364 else if (fl->fl_type == (F_INPROGRESS | F_UNLCK))
1365 /*
1366 * Someone is in the process of opening this
1367 * file for writing so we may not take an
1368 * exclusive lease on it.
1369 */
1370 wrlease_count++;
1371 else
1372 rdlease_count++;
1373 }
1374
1375 if ((arg == F_RDLCK && (wrlease_count > 0)) ||
1376 (arg == F_WRLCK && ((rdlease_count + wrlease_count) > 0)))
1377 goto out;
1378
1379 if (my_before != NULL) {
1380 error = lease->fl_lmops->fl_change(my_before, arg);
1381 goto out;
1382 }
1383
1384 error = 0;
1385 if (arg == F_UNLCK)
1386 goto out;
1387
1388 error = -EINVAL;
1389 if (!leases_enable)
1390 goto out;
1391
1392 error = lease_alloc(filp, arg, &fl);
1393 if (error)
1394 goto out;
1395
1396 locks_copy_lock(fl, lease);
1397
1398 locks_insert_lock(before, fl);
1399
1400 *flp = fl;
1401out:
1402 return error;
1403}
1404
1405 /**
1406 * setlease - sets a lease on an open file
1407 * @filp: file pointer
1408 * @arg: type of lease to obtain
1409 * @lease: file_lock to use
1410 *
1411 * Call this to establish a lease on the file.
1412 * The fl_lmops fl_break function is required by break_lease
1413 */
1414
1415int setlease(struct file *filp, long arg, struct file_lock **lease)
1416{
1417 struct dentry *dentry = filp->f_dentry;
1418 struct inode *inode = dentry->d_inode;
1419 int error;
1420
1421 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1422 return -EACCES;
1423 if (!S_ISREG(inode->i_mode))
1424 return -EINVAL;
1425 error = security_file_lock(filp, arg);
1426 if (error)
1427 return error;
1428
1429 lock_kernel();
1430 error = __setlease(filp, arg, lease);
1431 unlock_kernel();
1432
1433 return error;
1434}
1435
1436EXPORT_SYMBOL(setlease);
1437
1438/**
1439 * fcntl_setlease - sets a lease on an open file
1440 * @fd: open file descriptor
1441 * @filp: file pointer
1442 * @arg: type of lease to obtain
1443 *
1444 * Call this fcntl to establish a lease on the file.
1445 * Note that you also need to call %F_SETSIG to
1446 * receive a signal when the lease is broken.
1447 */
1448int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1449{
1450 struct file_lock fl, *flp = &fl;
1451 struct dentry *dentry = filp->f_dentry;
1452 struct inode *inode = dentry->d_inode;
1453 int error;
1454
1455 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1456 return -EACCES;
1457 if (!S_ISREG(inode->i_mode))
1458 return -EINVAL;
1459 error = security_file_lock(filp, arg);
1460 if (error)
1461 return error;
1462
1463 locks_init_lock(&fl);
1464 error = lease_init(filp, arg, &fl);
1465 if (error)
1466 return error;
1467
1468 lock_kernel();
1469
1470 error = __setlease(filp, arg, &flp);
J. Bruce Fieldsdc15ae12005-11-10 19:08:00 -05001471 if (error || arg == F_UNLCK)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472 goto out_unlock;
1473
1474 error = fasync_helper(fd, filp, 1, &flp->fl_fasync);
1475 if (error < 0) {
1476 /* remove lease just inserted by __setlease */
1477 flp->fl_type = F_UNLCK | F_INPROGRESS;
1478 flp->fl_break_time = jiffies- 10;
1479 time_out_leases(inode);
1480 goto out_unlock;
1481 }
1482
1483 error = f_setown(filp, current->pid, 0);
1484out_unlock:
1485 unlock_kernel();
1486 return error;
1487}
1488
1489/**
1490 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1491 * @filp: The file to apply the lock to
1492 * @fl: The lock to be applied
1493 *
1494 * Add a FLOCK style lock to a file.
1495 */
1496int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1497{
1498 int error;
1499 might_sleep();
1500 for (;;) {
1501 error = flock_lock_file(filp, fl);
1502 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1503 break;
1504 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1505 if (!error)
1506 continue;
1507
1508 locks_delete_block(fl);
1509 break;
1510 }
1511 return error;
1512}
1513
1514EXPORT_SYMBOL(flock_lock_file_wait);
1515
1516/**
1517 * sys_flock: - flock() system call.
1518 * @fd: the file descriptor to lock.
1519 * @cmd: the type of lock to apply.
1520 *
1521 * Apply a %FL_FLOCK style lock to an open file descriptor.
1522 * The @cmd can be one of
1523 *
1524 * %LOCK_SH -- a shared lock.
1525 *
1526 * %LOCK_EX -- an exclusive lock.
1527 *
1528 * %LOCK_UN -- remove an existing lock.
1529 *
1530 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1531 *
1532 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1533 * processes read and write access respectively.
1534 */
1535asmlinkage long sys_flock(unsigned int fd, unsigned int cmd)
1536{
1537 struct file *filp;
1538 struct file_lock *lock;
1539 int can_sleep, unlock;
1540 int error;
1541
1542 error = -EBADF;
1543 filp = fget(fd);
1544 if (!filp)
1545 goto out;
1546
1547 can_sleep = !(cmd & LOCK_NB);
1548 cmd &= ~LOCK_NB;
1549 unlock = (cmd == LOCK_UN);
1550
1551 if (!unlock && !(cmd & LOCK_MAND) && !(filp->f_mode & 3))
1552 goto out_putf;
1553
1554 error = flock_make_lock(filp, &lock, cmd);
1555 if (error)
1556 goto out_putf;
1557 if (can_sleep)
1558 lock->fl_flags |= FL_SLEEP;
1559
1560 error = security_file_lock(filp, cmd);
1561 if (error)
1562 goto out_free;
1563
1564 if (filp->f_op && filp->f_op->flock)
1565 error = filp->f_op->flock(filp,
1566 (can_sleep) ? F_SETLKW : F_SETLK,
1567 lock);
1568 else
1569 error = flock_lock_file_wait(filp, lock);
1570
1571 out_free:
1572 if (list_empty(&lock->fl_link)) {
1573 locks_free_lock(lock);
1574 }
1575
1576 out_putf:
1577 fput(filp);
1578 out:
1579 return error;
1580}
1581
1582/* Report the first existing lock that would conflict with l.
1583 * This implements the F_GETLK command of fcntl().
1584 */
1585int fcntl_getlk(struct file *filp, struct flock __user *l)
1586{
Andy Adamson8dc7c312006-03-20 13:44:26 -05001587 struct file_lock *fl, cfl, file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588 struct flock flock;
1589 int error;
1590
1591 error = -EFAULT;
1592 if (copy_from_user(&flock, l, sizeof(flock)))
1593 goto out;
1594 error = -EINVAL;
1595 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1596 goto out;
1597
1598 error = flock_to_posix_lock(filp, &file_lock, &flock);
1599 if (error)
1600 goto out;
1601
1602 if (filp->f_op && filp->f_op->lock) {
1603 error = filp->f_op->lock(filp, F_GETLK, &file_lock);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00001604 if (file_lock.fl_ops && file_lock.fl_ops->fl_release_private)
1605 file_lock.fl_ops->fl_release_private(&file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606 if (error < 0)
1607 goto out;
1608 else
1609 fl = (file_lock.fl_type == F_UNLCK ? NULL : &file_lock);
1610 } else {
Andy Adamson8dc7c312006-03-20 13:44:26 -05001611 fl = (posix_test_lock(filp, &file_lock, &cfl) ? &cfl : NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612 }
1613
1614 flock.l_type = F_UNLCK;
1615 if (fl != NULL) {
1616 flock.l_pid = fl->fl_pid;
1617#if BITS_PER_LONG == 32
1618 /*
1619 * Make sure we can represent the posix lock via
1620 * legacy 32bit flock.
1621 */
1622 error = -EOVERFLOW;
1623 if (fl->fl_start > OFFT_OFFSET_MAX)
1624 goto out;
1625 if ((fl->fl_end != OFFSET_MAX)
1626 && (fl->fl_end > OFFT_OFFSET_MAX))
1627 goto out;
1628#endif
1629 flock.l_start = fl->fl_start;
1630 flock.l_len = fl->fl_end == OFFSET_MAX ? 0 :
1631 fl->fl_end - fl->fl_start + 1;
1632 flock.l_whence = 0;
1633 flock.l_type = fl->fl_type;
1634 }
1635 error = -EFAULT;
1636 if (!copy_to_user(l, &flock, sizeof(flock)))
1637 error = 0;
1638out:
1639 return error;
1640}
1641
1642/* Apply the lock described by l to an open file descriptor.
1643 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1644 */
Peter Staubachc2936212005-07-27 11:45:09 -07001645int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1646 struct flock __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647{
1648 struct file_lock *file_lock = locks_alloc_lock();
1649 struct flock flock;
1650 struct inode *inode;
1651 int error;
1652
1653 if (file_lock == NULL)
1654 return -ENOLCK;
1655
1656 /*
1657 * This might block, so we do it before checking the inode.
1658 */
1659 error = -EFAULT;
1660 if (copy_from_user(&flock, l, sizeof(flock)))
1661 goto out;
1662
1663 inode = filp->f_dentry->d_inode;
1664
1665 /* Don't allow mandatory locks on files that may be memory mapped
1666 * and shared.
1667 */
1668 if (IS_MANDLOCK(inode) &&
1669 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1670 mapping_writably_mapped(filp->f_mapping)) {
1671 error = -EAGAIN;
1672 goto out;
1673 }
1674
Peter Staubachc2936212005-07-27 11:45:09 -07001675again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676 error = flock_to_posix_lock(filp, file_lock, &flock);
1677 if (error)
1678 goto out;
1679 if (cmd == F_SETLKW) {
1680 file_lock->fl_flags |= FL_SLEEP;
1681 }
1682
1683 error = -EBADF;
1684 switch (flock.l_type) {
1685 case F_RDLCK:
1686 if (!(filp->f_mode & FMODE_READ))
1687 goto out;
1688 break;
1689 case F_WRLCK:
1690 if (!(filp->f_mode & FMODE_WRITE))
1691 goto out;
1692 break;
1693 case F_UNLCK:
1694 break;
1695 default:
1696 error = -EINVAL;
1697 goto out;
1698 }
1699
1700 error = security_file_lock(filp, file_lock->fl_type);
1701 if (error)
1702 goto out;
1703
Peter Staubachc2936212005-07-27 11:45:09 -07001704 if (filp->f_op && filp->f_op->lock != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705 error = filp->f_op->lock(filp, cmd, file_lock);
Peter Staubachc2936212005-07-27 11:45:09 -07001706 else {
1707 for (;;) {
Andy Adamson5842add2006-03-26 01:37:26 -08001708 error = posix_lock_file(filp, file_lock);
Peter Staubachc2936212005-07-27 11:45:09 -07001709 if ((error != -EAGAIN) || (cmd == F_SETLK))
1710 break;
1711 error = wait_event_interruptible(file_lock->fl_wait,
1712 !file_lock->fl_next);
1713 if (!error)
1714 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715
Peter Staubachc2936212005-07-27 11:45:09 -07001716 locks_delete_block(file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717 break;
Peter Staubachc2936212005-07-27 11:45:09 -07001718 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719 }
1720
Peter Staubachc2936212005-07-27 11:45:09 -07001721 /*
1722 * Attempt to detect a close/fcntl race and recover by
1723 * releasing the lock that was just acquired.
1724 */
1725 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1726 flock.l_type = F_UNLCK;
1727 goto again;
1728 }
1729
1730out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731 locks_free_lock(file_lock);
1732 return error;
1733}
1734
1735#if BITS_PER_LONG == 32
1736/* Report the first existing lock that would conflict with l.
1737 * This implements the F_GETLK command of fcntl().
1738 */
1739int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1740{
Andy Adamson8dc7c312006-03-20 13:44:26 -05001741 struct file_lock *fl, cfl, file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742 struct flock64 flock;
1743 int error;
1744
1745 error = -EFAULT;
1746 if (copy_from_user(&flock, l, sizeof(flock)))
1747 goto out;
1748 error = -EINVAL;
1749 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1750 goto out;
1751
1752 error = flock64_to_posix_lock(filp, &file_lock, &flock);
1753 if (error)
1754 goto out;
1755
1756 if (filp->f_op && filp->f_op->lock) {
1757 error = filp->f_op->lock(filp, F_GETLK, &file_lock);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00001758 if (file_lock.fl_ops && file_lock.fl_ops->fl_release_private)
1759 file_lock.fl_ops->fl_release_private(&file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760 if (error < 0)
1761 goto out;
1762 else
1763 fl = (file_lock.fl_type == F_UNLCK ? NULL : &file_lock);
1764 } else {
Andy Adamson8dc7c312006-03-20 13:44:26 -05001765 fl = (posix_test_lock(filp, &file_lock, &cfl) ? &cfl : NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766 }
1767
1768 flock.l_type = F_UNLCK;
1769 if (fl != NULL) {
1770 flock.l_pid = fl->fl_pid;
1771 flock.l_start = fl->fl_start;
1772 flock.l_len = fl->fl_end == OFFSET_MAX ? 0 :
1773 fl->fl_end - fl->fl_start + 1;
1774 flock.l_whence = 0;
1775 flock.l_type = fl->fl_type;
1776 }
1777 error = -EFAULT;
1778 if (!copy_to_user(l, &flock, sizeof(flock)))
1779 error = 0;
1780
1781out:
1782 return error;
1783}
1784
1785/* Apply the lock described by l to an open file descriptor.
1786 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1787 */
Peter Staubachc2936212005-07-27 11:45:09 -07001788int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
1789 struct flock64 __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790{
1791 struct file_lock *file_lock = locks_alloc_lock();
1792 struct flock64 flock;
1793 struct inode *inode;
1794 int error;
1795
1796 if (file_lock == NULL)
1797 return -ENOLCK;
1798
1799 /*
1800 * This might block, so we do it before checking the inode.
1801 */
1802 error = -EFAULT;
1803 if (copy_from_user(&flock, l, sizeof(flock)))
1804 goto out;
1805
1806 inode = filp->f_dentry->d_inode;
1807
1808 /* Don't allow mandatory locks on files that may be memory mapped
1809 * and shared.
1810 */
1811 if (IS_MANDLOCK(inode) &&
1812 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1813 mapping_writably_mapped(filp->f_mapping)) {
1814 error = -EAGAIN;
1815 goto out;
1816 }
1817
Peter Staubachc2936212005-07-27 11:45:09 -07001818again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819 error = flock64_to_posix_lock(filp, file_lock, &flock);
1820 if (error)
1821 goto out;
1822 if (cmd == F_SETLKW64) {
1823 file_lock->fl_flags |= FL_SLEEP;
1824 }
1825
1826 error = -EBADF;
1827 switch (flock.l_type) {
1828 case F_RDLCK:
1829 if (!(filp->f_mode & FMODE_READ))
1830 goto out;
1831 break;
1832 case F_WRLCK:
1833 if (!(filp->f_mode & FMODE_WRITE))
1834 goto out;
1835 break;
1836 case F_UNLCK:
1837 break;
1838 default:
1839 error = -EINVAL;
1840 goto out;
1841 }
1842
1843 error = security_file_lock(filp, file_lock->fl_type);
1844 if (error)
1845 goto out;
1846
Peter Staubachc2936212005-07-27 11:45:09 -07001847 if (filp->f_op && filp->f_op->lock != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001848 error = filp->f_op->lock(filp, cmd, file_lock);
Peter Staubachc2936212005-07-27 11:45:09 -07001849 else {
1850 for (;;) {
Andy Adamson5842add2006-03-26 01:37:26 -08001851 error = posix_lock_file(filp, file_lock);
Peter Staubachc2936212005-07-27 11:45:09 -07001852 if ((error != -EAGAIN) || (cmd == F_SETLK64))
1853 break;
1854 error = wait_event_interruptible(file_lock->fl_wait,
1855 !file_lock->fl_next);
1856 if (!error)
1857 continue;
1858
1859 locks_delete_block(file_lock);
1860 break;
1861 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862 }
1863
Peter Staubachc2936212005-07-27 11:45:09 -07001864 /*
1865 * Attempt to detect a close/fcntl race and recover by
1866 * releasing the lock that was just acquired.
1867 */
1868 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1869 flock.l_type = F_UNLCK;
1870 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871 }
1872
1873out:
1874 locks_free_lock(file_lock);
1875 return error;
1876}
1877#endif /* BITS_PER_LONG == 32 */
1878
1879/*
1880 * This function is called when the file is being removed
1881 * from the task's fd array. POSIX locks belonging to this task
1882 * are deleted at this time.
1883 */
1884void locks_remove_posix(struct file *filp, fl_owner_t owner)
1885{
1886 struct file_lock lock, **before;
1887
1888 /*
1889 * If there are no locks held on this file, we don't need to call
1890 * posix_lock_file(). Another process could be setting a lock on this
1891 * file at the same time, but we wouldn't remove that lock anyway.
1892 */
1893 before = &filp->f_dentry->d_inode->i_flock;
1894 if (*before == NULL)
1895 return;
1896
1897 lock.fl_type = F_UNLCK;
1898 lock.fl_flags = FL_POSIX;
1899 lock.fl_start = 0;
1900 lock.fl_end = OFFSET_MAX;
1901 lock.fl_owner = owner;
1902 lock.fl_pid = current->tgid;
1903 lock.fl_file = filp;
1904 lock.fl_ops = NULL;
1905 lock.fl_lmops = NULL;
1906
1907 if (filp->f_op && filp->f_op->lock != NULL) {
1908 filp->f_op->lock(filp, F_SETLK, &lock);
1909 goto out;
1910 }
1911
1912 /* Can't use posix_lock_file here; we need to remove it no matter
1913 * which pid we have.
1914 */
1915 lock_kernel();
1916 while (*before != NULL) {
1917 struct file_lock *fl = *before;
1918 if (IS_POSIX(fl) && posix_same_owner(fl, &lock)) {
1919 locks_delete_lock(before);
1920 continue;
1921 }
1922 before = &fl->fl_next;
1923 }
1924 unlock_kernel();
1925out:
1926 if (lock.fl_ops && lock.fl_ops->fl_release_private)
1927 lock.fl_ops->fl_release_private(&lock);
1928}
1929
1930EXPORT_SYMBOL(locks_remove_posix);
1931
1932/*
1933 * This function is called on the last close of an open file.
1934 */
1935void locks_remove_flock(struct file *filp)
1936{
1937 struct inode * inode = filp->f_dentry->d_inode;
1938 struct file_lock *fl;
1939 struct file_lock **before;
1940
1941 if (!inode->i_flock)
1942 return;
1943
1944 if (filp->f_op && filp->f_op->flock) {
1945 struct file_lock fl = {
1946 .fl_pid = current->tgid,
1947 .fl_file = filp,
1948 .fl_flags = FL_FLOCK,
1949 .fl_type = F_UNLCK,
1950 .fl_end = OFFSET_MAX,
1951 };
1952 filp->f_op->flock(filp, F_SETLKW, &fl);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00001953 if (fl.fl_ops && fl.fl_ops->fl_release_private)
1954 fl.fl_ops->fl_release_private(&fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955 }
1956
1957 lock_kernel();
1958 before = &inode->i_flock;
1959
1960 while ((fl = *before) != NULL) {
1961 if (fl->fl_file == filp) {
Peter Staubachc2936212005-07-27 11:45:09 -07001962 if (IS_FLOCK(fl)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963 locks_delete_lock(before);
1964 continue;
1965 }
1966 if (IS_LEASE(fl)) {
1967 lease_modify(before, F_UNLCK);
1968 continue;
1969 }
1970 /* What? */
1971 BUG();
1972 }
1973 before = &fl->fl_next;
1974 }
1975 unlock_kernel();
1976}
1977
1978/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979 * posix_unblock_lock - stop waiting for a file lock
1980 * @filp: how the file was opened
1981 * @waiter: the lock which was waiting
1982 *
1983 * lockd needs to block waiting for locks.
1984 */
J. Bruce Fields64a318e2006-01-03 09:55:46 +01001985int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986posix_unblock_lock(struct file *filp, struct file_lock *waiter)
1987{
J. Bruce Fields64a318e2006-01-03 09:55:46 +01001988 int status = 0;
1989
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990 lock_kernel();
J. Bruce Fields5996a292006-01-03 09:55:44 +01001991 if (waiter->fl_next)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992 __locks_delete_block(waiter);
J. Bruce Fields64a318e2006-01-03 09:55:46 +01001993 else
1994 status = -ENOENT;
J. Bruce Fields5996a292006-01-03 09:55:44 +01001995 unlock_kernel();
J. Bruce Fields64a318e2006-01-03 09:55:46 +01001996 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997}
1998
1999EXPORT_SYMBOL(posix_unblock_lock);
2000
2001static void lock_get_status(char* out, struct file_lock *fl, int id, char *pfx)
2002{
2003 struct inode *inode = NULL;
2004
2005 if (fl->fl_file != NULL)
2006 inode = fl->fl_file->f_dentry->d_inode;
2007
2008 out += sprintf(out, "%d:%s ", id, pfx);
2009 if (IS_POSIX(fl)) {
2010 out += sprintf(out, "%6s %s ",
2011 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2012 (inode == NULL) ? "*NOINODE*" :
2013 (IS_MANDLOCK(inode) &&
2014 (inode->i_mode & (S_IXGRP | S_ISGID)) == S_ISGID) ?
2015 "MANDATORY" : "ADVISORY ");
2016 } else if (IS_FLOCK(fl)) {
2017 if (fl->fl_type & LOCK_MAND) {
2018 out += sprintf(out, "FLOCK MSNFS ");
2019 } else {
2020 out += sprintf(out, "FLOCK ADVISORY ");
2021 }
2022 } else if (IS_LEASE(fl)) {
2023 out += sprintf(out, "LEASE ");
2024 if (fl->fl_type & F_INPROGRESS)
2025 out += sprintf(out, "BREAKING ");
2026 else if (fl->fl_file)
2027 out += sprintf(out, "ACTIVE ");
2028 else
2029 out += sprintf(out, "BREAKER ");
2030 } else {
2031 out += sprintf(out, "UNKNOWN UNKNOWN ");
2032 }
2033 if (fl->fl_type & LOCK_MAND) {
2034 out += sprintf(out, "%s ",
2035 (fl->fl_type & LOCK_READ)
2036 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2037 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2038 } else {
2039 out += sprintf(out, "%s ",
2040 (fl->fl_type & F_INPROGRESS)
2041 ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ "
2042 : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ ");
2043 }
2044 if (inode) {
2045#ifdef WE_CAN_BREAK_LSLK_NOW
2046 out += sprintf(out, "%d %s:%ld ", fl->fl_pid,
2047 inode->i_sb->s_id, inode->i_ino);
2048#else
2049 /* userspace relies on this representation of dev_t ;-( */
2050 out += sprintf(out, "%d %02x:%02x:%ld ", fl->fl_pid,
2051 MAJOR(inode->i_sb->s_dev),
2052 MINOR(inode->i_sb->s_dev), inode->i_ino);
2053#endif
2054 } else {
2055 out += sprintf(out, "%d <none>:0 ", fl->fl_pid);
2056 }
2057 if (IS_POSIX(fl)) {
2058 if (fl->fl_end == OFFSET_MAX)
2059 out += sprintf(out, "%Ld EOF\n", fl->fl_start);
2060 else
2061 out += sprintf(out, "%Ld %Ld\n", fl->fl_start,
2062 fl->fl_end);
2063 } else {
2064 out += sprintf(out, "0 EOF\n");
2065 }
2066}
2067
2068static void move_lock_status(char **p, off_t* pos, off_t offset)
2069{
2070 int len;
2071 len = strlen(*p);
2072 if(*pos >= offset) {
2073 /* the complete line is valid */
2074 *p += len;
2075 *pos += len;
2076 return;
2077 }
2078 if(*pos+len > offset) {
2079 /* use the second part of the line */
2080 int i = offset-*pos;
2081 memmove(*p,*p+i,len-i);
2082 *p += len-i;
2083 *pos += len;
2084 return;
2085 }
2086 /* discard the complete line */
2087 *pos += len;
2088}
2089
2090/**
2091 * get_locks_status - reports lock usage in /proc/locks
2092 * @buffer: address in userspace to write into
2093 * @start: ?
2094 * @offset: how far we are through the buffer
2095 * @length: how much to read
2096 */
2097
2098int get_locks_status(char *buffer, char **start, off_t offset, int length)
2099{
2100 struct list_head *tmp;
2101 char *q = buffer;
2102 off_t pos = 0;
2103 int i = 0;
2104
2105 lock_kernel();
2106 list_for_each(tmp, &file_lock_list) {
2107 struct list_head *btmp;
2108 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
2109 lock_get_status(q, fl, ++i, "");
2110 move_lock_status(&q, &pos, offset);
2111
2112 if(pos >= offset+length)
2113 goto done;
2114
2115 list_for_each(btmp, &fl->fl_block) {
2116 struct file_lock *bfl = list_entry(btmp,
2117 struct file_lock, fl_block);
2118 lock_get_status(q, bfl, i, " ->");
2119 move_lock_status(&q, &pos, offset);
2120
2121 if(pos >= offset+length)
2122 goto done;
2123 }
2124 }
2125done:
2126 unlock_kernel();
2127 *start = buffer;
2128 if(q-buffer < length)
2129 return (q-buffer);
2130 return length;
2131}
2132
2133/**
2134 * lock_may_read - checks that the region is free of locks
2135 * @inode: the inode that is being read
2136 * @start: the first byte to read
2137 * @len: the number of bytes to read
2138 *
2139 * Emulates Windows locking requirements. Whole-file
2140 * mandatory locks (share modes) can prohibit a read and
2141 * byte-range POSIX locks can prohibit a read if they overlap.
2142 *
2143 * N.B. this function is only ever called
2144 * from knfsd and ownership of locks is never checked.
2145 */
2146int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2147{
2148 struct file_lock *fl;
2149 int result = 1;
2150 lock_kernel();
2151 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2152 if (IS_POSIX(fl)) {
2153 if (fl->fl_type == F_RDLCK)
2154 continue;
2155 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2156 continue;
2157 } else if (IS_FLOCK(fl)) {
2158 if (!(fl->fl_type & LOCK_MAND))
2159 continue;
2160 if (fl->fl_type & LOCK_READ)
2161 continue;
2162 } else
2163 continue;
2164 result = 0;
2165 break;
2166 }
2167 unlock_kernel();
2168 return result;
2169}
2170
2171EXPORT_SYMBOL(lock_may_read);
2172
2173/**
2174 * lock_may_write - checks that the region is free of locks
2175 * @inode: the inode that is being written
2176 * @start: the first byte to write
2177 * @len: the number of bytes to write
2178 *
2179 * Emulates Windows locking requirements. Whole-file
2180 * mandatory locks (share modes) can prohibit a write and
2181 * byte-range POSIX locks can prohibit a write if they overlap.
2182 *
2183 * N.B. this function is only ever called
2184 * from knfsd and ownership of locks is never checked.
2185 */
2186int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2187{
2188 struct file_lock *fl;
2189 int result = 1;
2190 lock_kernel();
2191 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2192 if (IS_POSIX(fl)) {
2193 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2194 continue;
2195 } else if (IS_FLOCK(fl)) {
2196 if (!(fl->fl_type & LOCK_MAND))
2197 continue;
2198 if (fl->fl_type & LOCK_WRITE)
2199 continue;
2200 } else
2201 continue;
2202 result = 0;
2203 break;
2204 }
2205 unlock_kernel();
2206 return result;
2207}
2208
2209EXPORT_SYMBOL(lock_may_write);
2210
2211static inline void __steal_locks(struct file *file, fl_owner_t from)
2212{
2213 struct inode *inode = file->f_dentry->d_inode;
2214 struct file_lock *fl = inode->i_flock;
2215
2216 while (fl) {
2217 if (fl->fl_file == file && fl->fl_owner == from)
2218 fl->fl_owner = current->files;
2219 fl = fl->fl_next;
2220 }
2221}
2222
2223/* When getting ready for executing a binary, we make sure that current
2224 * has a files_struct on its own. Before dropping the old files_struct,
2225 * we take over ownership of all locks for all file descriptors we own.
2226 * Note that we may accidentally steal a lock for a file that a sibling
2227 * has created since the unshare_files() call.
2228 */
2229void steal_locks(fl_owner_t from)
2230{
2231 struct files_struct *files = current->files;
2232 int i, j;
Dipankar Sarmabadf1662005-09-09 13:04:10 -07002233 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234
2235 if (from == files)
2236 return;
2237
2238 lock_kernel();
2239 j = 0;
Dipankar Sarma4fb3a532005-09-16 19:28:13 -07002240 rcu_read_lock();
Dipankar Sarmabadf1662005-09-09 13:04:10 -07002241 fdt = files_fdtable(files);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242 for (;;) {
2243 unsigned long set;
2244 i = j * __NFDBITS;
Dipankar Sarmabadf1662005-09-09 13:04:10 -07002245 if (i >= fdt->max_fdset || i >= fdt->max_fds)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246 break;
Dipankar Sarmabadf1662005-09-09 13:04:10 -07002247 set = fdt->open_fds->fds_bits[j++];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248 while (set) {
2249 if (set & 1) {
Dipankar Sarmabadf1662005-09-09 13:04:10 -07002250 struct file *file = fdt->fd[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002251 if (file)
2252 __steal_locks(file, from);
2253 }
2254 i++;
2255 set >>= 1;
2256 }
2257 }
Dipankar Sarma4fb3a532005-09-16 19:28:13 -07002258 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002259 unlock_kernel();
2260}
2261EXPORT_SYMBOL(steal_locks);
2262
2263static int __init filelock_init(void)
2264{
2265 filelock_cache = kmem_cache_create("file_lock_cache",
2266 sizeof(struct file_lock), 0, SLAB_PANIC,
2267 init_once, NULL);
2268 return 0;
2269}
2270
2271core_initcall(filelock_init);