blob: 11956b6179ff81a9924ab099896a07ba3892bc0b [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>
127
128#include <asm/semaphore.h>
129#include <asm/uaccess.h>
130
131#define IS_POSIX(fl) (fl->fl_flags & FL_POSIX)
132#define IS_FLOCK(fl) (fl->fl_flags & FL_FLOCK)
133#define IS_LEASE(fl) (fl->fl_flags & FL_LEASE)
134
135int leases_enable = 1;
136int lease_break_time = 45;
137
138#define for_each_lock(inode, lockp) \
139 for (lockp = &inode->i_flock; *lockp != NULL; lockp = &(*lockp)->fl_next)
140
141LIST_HEAD(file_lock_list);
142
143EXPORT_SYMBOL(file_lock_list);
144
145static LIST_HEAD(blocked_list);
146
147static kmem_cache_t *filelock_cache;
148
149/* Allocate an empty lock structure. */
150static struct file_lock *locks_alloc_lock(void)
151{
152 return kmem_cache_alloc(filelock_cache, SLAB_KERNEL);
153}
154
155/* Free a lock which is not in use. */
156static inline void locks_free_lock(struct file_lock *fl)
157{
158 if (fl == NULL) {
159 BUG();
160 return;
161 }
162 if (waitqueue_active(&fl->fl_wait))
163 panic("Attempting to free lock with active wait queue");
164
165 if (!list_empty(&fl->fl_block))
166 panic("Attempting to free lock with active block list");
167
168 if (!list_empty(&fl->fl_link))
169 panic("Attempting to free lock on active lock list");
170
171 if (fl->fl_ops) {
172 if (fl->fl_ops->fl_release_private)
173 fl->fl_ops->fl_release_private(fl);
174 fl->fl_ops = NULL;
175 }
176
177 if (fl->fl_lmops) {
178 if (fl->fl_lmops->fl_release_private)
179 fl->fl_lmops->fl_release_private(fl);
180 fl->fl_lmops = NULL;
181 }
182
183 kmem_cache_free(filelock_cache, fl);
184}
185
186void locks_init_lock(struct file_lock *fl)
187{
188 INIT_LIST_HEAD(&fl->fl_link);
189 INIT_LIST_HEAD(&fl->fl_block);
190 init_waitqueue_head(&fl->fl_wait);
191 fl->fl_next = NULL;
192 fl->fl_fasync = NULL;
193 fl->fl_owner = NULL;
194 fl->fl_pid = 0;
195 fl->fl_file = NULL;
196 fl->fl_flags = 0;
197 fl->fl_type = 0;
198 fl->fl_start = fl->fl_end = 0;
199 fl->fl_ops = NULL;
200 fl->fl_lmops = NULL;
201}
202
203EXPORT_SYMBOL(locks_init_lock);
204
205/*
206 * Initialises the fields of the file lock which are invariant for
207 * free file_locks.
208 */
209static void init_once(void *foo, kmem_cache_t *cache, unsigned long flags)
210{
211 struct file_lock *lock = (struct file_lock *) foo;
212
213 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) !=
214 SLAB_CTOR_CONSTRUCTOR)
215 return;
216
217 locks_init_lock(lock);
218}
219
220/*
221 * Initialize a new lock from an existing file_lock structure.
222 */
223void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
224{
225 new->fl_owner = fl->fl_owner;
226 new->fl_pid = fl->fl_pid;
227 new->fl_file = fl->fl_file;
228 new->fl_flags = fl->fl_flags;
229 new->fl_type = fl->fl_type;
230 new->fl_start = fl->fl_start;
231 new->fl_end = fl->fl_end;
232 new->fl_ops = fl->fl_ops;
233 new->fl_lmops = fl->fl_lmops;
234 if (fl->fl_ops && fl->fl_ops->fl_copy_lock)
235 fl->fl_ops->fl_copy_lock(new, fl);
236 if (fl->fl_lmops && fl->fl_lmops->fl_copy_lock)
237 fl->fl_lmops->fl_copy_lock(new, fl);
238}
239
240EXPORT_SYMBOL(locks_copy_lock);
241
242static inline int flock_translate_cmd(int cmd) {
243 if (cmd & LOCK_MAND)
244 return cmd & (LOCK_MAND | LOCK_RW);
245 switch (cmd) {
246 case LOCK_SH:
247 return F_RDLCK;
248 case LOCK_EX:
249 return F_WRLCK;
250 case LOCK_UN:
251 return F_UNLCK;
252 }
253 return -EINVAL;
254}
255
256/* Fill in a file_lock structure with an appropriate FLOCK lock. */
257static int flock_make_lock(struct file *filp, struct file_lock **lock,
258 unsigned int cmd)
259{
260 struct file_lock *fl;
261 int type = flock_translate_cmd(cmd);
262 if (type < 0)
263 return type;
264
265 fl = locks_alloc_lock();
266 if (fl == NULL)
267 return -ENOMEM;
268
269 fl->fl_file = filp;
270 fl->fl_pid = current->tgid;
271 fl->fl_flags = FL_FLOCK;
272 fl->fl_type = type;
273 fl->fl_end = OFFSET_MAX;
274
275 *lock = fl;
276 return 0;
277}
278
279static int assign_type(struct file_lock *fl, int type)
280{
281 switch (type) {
282 case F_RDLCK:
283 case F_WRLCK:
284 case F_UNLCK:
285 fl->fl_type = type;
286 break;
287 default:
288 return -EINVAL;
289 }
290 return 0;
291}
292
293/* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
294 * style lock.
295 */
296static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
297 struct flock *l)
298{
299 off_t start, end;
300
301 switch (l->l_whence) {
302 case 0: /*SEEK_SET*/
303 start = 0;
304 break;
305 case 1: /*SEEK_CUR*/
306 start = filp->f_pos;
307 break;
308 case 2: /*SEEK_END*/
309 start = i_size_read(filp->f_dentry->d_inode);
310 break;
311 default:
312 return -EINVAL;
313 }
314
315 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
316 POSIX-2001 defines it. */
317 start += l->l_start;
318 end = start + l->l_len - 1;
319 if (l->l_len < 0) {
320 end = start - 1;
321 start += l->l_len;
322 }
323
324 if (start < 0)
325 return -EINVAL;
326 if (l->l_len > 0 && end < 0)
327 return -EOVERFLOW;
328
329 fl->fl_start = start; /* we record the absolute position */
330 fl->fl_end = end;
331 if (l->l_len == 0)
332 fl->fl_end = OFFSET_MAX;
333
334 fl->fl_owner = current->files;
335 fl->fl_pid = current->tgid;
336 fl->fl_file = filp;
337 fl->fl_flags = FL_POSIX;
338 fl->fl_ops = NULL;
339 fl->fl_lmops = NULL;
340
341 return assign_type(fl, l->l_type);
342}
343
344#if BITS_PER_LONG == 32
345static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
346 struct flock64 *l)
347{
348 loff_t start;
349
350 switch (l->l_whence) {
351 case 0: /*SEEK_SET*/
352 start = 0;
353 break;
354 case 1: /*SEEK_CUR*/
355 start = filp->f_pos;
356 break;
357 case 2: /*SEEK_END*/
358 start = i_size_read(filp->f_dentry->d_inode);
359 break;
360 default:
361 return -EINVAL;
362 }
363
364 if (((start += l->l_start) < 0) || (l->l_len < 0))
365 return -EINVAL;
366 fl->fl_end = start + l->l_len - 1;
367 if (l->l_len > 0 && fl->fl_end < 0)
368 return -EOVERFLOW;
369 fl->fl_start = start; /* we record the absolute position */
370 if (l->l_len == 0)
371 fl->fl_end = OFFSET_MAX;
372
373 fl->fl_owner = current->files;
374 fl->fl_pid = current->tgid;
375 fl->fl_file = filp;
376 fl->fl_flags = FL_POSIX;
377 fl->fl_ops = NULL;
378 fl->fl_lmops = NULL;
379
380 switch (l->l_type) {
381 case F_RDLCK:
382 case F_WRLCK:
383 case F_UNLCK:
384 fl->fl_type = l->l_type;
385 break;
386 default:
387 return -EINVAL;
388 }
389
390 return (0);
391}
392#endif
393
394/* default lease lock manager operations */
395static void lease_break_callback(struct file_lock *fl)
396{
397 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
398}
399
400static void lease_release_private_callback(struct file_lock *fl)
401{
402 if (!fl->fl_file)
403 return;
404
405 f_delown(fl->fl_file);
406 fl->fl_file->f_owner.signum = 0;
407}
408
Adrian Bunk75c96f82005-05-05 16:16:09 -0700409static int lease_mylease_callback(struct file_lock *fl, struct file_lock *try)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410{
411 return fl->fl_file == try->fl_file;
412}
413
Adrian Bunk75c96f82005-05-05 16:16:09 -0700414static struct lock_manager_operations lease_manager_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 .fl_break = lease_break_callback,
416 .fl_release_private = lease_release_private_callback,
417 .fl_mylease = lease_mylease_callback,
418 .fl_change = lease_modify,
419};
420
421/*
422 * Initialize a lease, use the default lock manager operations
423 */
424static int lease_init(struct file *filp, int type, struct file_lock *fl)
425 {
426 fl->fl_owner = current->files;
427 fl->fl_pid = current->tgid;
428
429 fl->fl_file = filp;
430 fl->fl_flags = FL_LEASE;
431 if (assign_type(fl, type) != 0) {
432 locks_free_lock(fl);
433 return -EINVAL;
434 }
435 fl->fl_start = 0;
436 fl->fl_end = OFFSET_MAX;
437 fl->fl_ops = NULL;
438 fl->fl_lmops = &lease_manager_ops;
439 return 0;
440}
441
442/* Allocate a file_lock initialised to this type of lease */
443static int lease_alloc(struct file *filp, int type, struct file_lock **flp)
444{
445 struct file_lock *fl = locks_alloc_lock();
446 int error;
447
448 if (fl == NULL)
449 return -ENOMEM;
450
451 error = lease_init(filp, type, fl);
452 if (error)
453 return error;
454 *flp = fl;
455 return 0;
456}
457
458/* Check if two locks overlap each other.
459 */
460static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
461{
462 return ((fl1->fl_end >= fl2->fl_start) &&
463 (fl2->fl_end >= fl1->fl_start));
464}
465
466/*
467 * Check whether two locks have the same owner.
468 */
469static inline int
470posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
471{
472 if (fl1->fl_lmops && fl1->fl_lmops->fl_compare_owner)
473 return fl2->fl_lmops == fl1->fl_lmops &&
474 fl1->fl_lmops->fl_compare_owner(fl1, fl2);
475 return fl1->fl_owner == fl2->fl_owner;
476}
477
478/* Remove waiter from blocker's block list.
479 * When blocker ends up pointing to itself then the list is empty.
480 */
481static inline void __locks_delete_block(struct file_lock *waiter)
482{
483 list_del_init(&waiter->fl_block);
484 list_del_init(&waiter->fl_link);
485 waiter->fl_next = NULL;
486}
487
488/*
489 */
490static void locks_delete_block(struct file_lock *waiter)
491{
492 lock_kernel();
493 __locks_delete_block(waiter);
494 unlock_kernel();
495}
496
497/* Insert waiter into blocker's block list.
498 * We use a circular list so that processes can be easily woken up in
499 * the order they blocked. The documentation doesn't require this but
500 * it seems like the reasonable thing to do.
501 */
502static void locks_insert_block(struct file_lock *blocker,
503 struct file_lock *waiter)
504{
505 if (!list_empty(&waiter->fl_block)) {
506 printk(KERN_ERR "locks_insert_block: removing duplicated lock "
507 "(pid=%d %Ld-%Ld type=%d)\n", waiter->fl_pid,
508 waiter->fl_start, waiter->fl_end, waiter->fl_type);
509 __locks_delete_block(waiter);
510 }
511 list_add_tail(&waiter->fl_block, &blocker->fl_block);
512 waiter->fl_next = blocker;
513 if (IS_POSIX(blocker))
514 list_add(&waiter->fl_link, &blocked_list);
515}
516
517/* Wake up processes blocked waiting for blocker.
518 * If told to wait then schedule the processes until the block list
519 * is empty, otherwise empty the block list ourselves.
520 */
521static void locks_wake_up_blocks(struct file_lock *blocker)
522{
523 while (!list_empty(&blocker->fl_block)) {
524 struct file_lock *waiter = list_entry(blocker->fl_block.next,
525 struct file_lock, fl_block);
526 __locks_delete_block(waiter);
527 if (waiter->fl_lmops && waiter->fl_lmops->fl_notify)
528 waiter->fl_lmops->fl_notify(waiter);
529 else
530 wake_up(&waiter->fl_wait);
531 }
532}
533
534/* Insert file lock fl into an inode's lock list at the position indicated
535 * by pos. At the same time add the lock to the global file lock list.
536 */
537static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
538{
539 list_add(&fl->fl_link, &file_lock_list);
540
541 /* insert into file's list */
542 fl->fl_next = *pos;
543 *pos = fl;
544
545 if (fl->fl_ops && fl->fl_ops->fl_insert)
546 fl->fl_ops->fl_insert(fl);
547}
548
549/*
550 * Delete a lock and then free it.
551 * Wake up processes that are blocked waiting for this lock,
552 * notify the FS that the lock has been cleared and
553 * finally free the lock.
554 */
555static void locks_delete_lock(struct file_lock **thisfl_p)
556{
557 struct file_lock *fl = *thisfl_p;
558
559 *thisfl_p = fl->fl_next;
560 fl->fl_next = NULL;
561 list_del_init(&fl->fl_link);
562
563 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
564 if (fl->fl_fasync != NULL) {
565 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
566 fl->fl_fasync = NULL;
567 }
568
569 if (fl->fl_ops && fl->fl_ops->fl_remove)
570 fl->fl_ops->fl_remove(fl);
571
572 locks_wake_up_blocks(fl);
573 locks_free_lock(fl);
574}
575
576/* Determine if lock sys_fl blocks lock caller_fl. Common functionality
577 * checks for shared/exclusive status of overlapping locks.
578 */
579static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
580{
581 if (sys_fl->fl_type == F_WRLCK)
582 return 1;
583 if (caller_fl->fl_type == F_WRLCK)
584 return 1;
585 return 0;
586}
587
588/* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
589 * checking before calling the locks_conflict().
590 */
591static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
592{
593 /* POSIX locks owned by the same process do not conflict with
594 * each other.
595 */
596 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
597 return (0);
598
599 /* Check whether they overlap */
600 if (!locks_overlap(caller_fl, sys_fl))
601 return 0;
602
603 return (locks_conflict(caller_fl, sys_fl));
604}
605
606/* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
607 * checking before calling the locks_conflict().
608 */
609static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
610{
611 /* FLOCK locks referring to the same filp do not conflict with
612 * each other.
613 */
614 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
615 return (0);
616 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
617 return 0;
618
619 return (locks_conflict(caller_fl, sys_fl));
620}
621
622static int interruptible_sleep_on_locked(wait_queue_head_t *fl_wait, int timeout)
623{
624 int result = 0;
625 DECLARE_WAITQUEUE(wait, current);
626
627 __set_current_state(TASK_INTERRUPTIBLE);
628 add_wait_queue(fl_wait, &wait);
629 if (timeout == 0)
630 schedule();
631 else
632 result = schedule_timeout(timeout);
633 if (signal_pending(current))
634 result = -ERESTARTSYS;
635 remove_wait_queue(fl_wait, &wait);
636 __set_current_state(TASK_RUNNING);
637 return result;
638}
639
640static int locks_block_on_timeout(struct file_lock *blocker, struct file_lock *waiter, int time)
641{
642 int result;
643 locks_insert_block(blocker, waiter);
644 result = interruptible_sleep_on_locked(&waiter->fl_wait, time);
645 __locks_delete_block(waiter);
646 return result;
647}
648
649struct file_lock *
650posix_test_lock(struct file *filp, struct file_lock *fl)
651{
652 struct file_lock *cfl;
653
654 lock_kernel();
655 for (cfl = filp->f_dentry->d_inode->i_flock; cfl; cfl = cfl->fl_next) {
656 if (!IS_POSIX(cfl))
657 continue;
658 if (posix_locks_conflict(cfl, fl))
659 break;
660 }
661 unlock_kernel();
662
663 return (cfl);
664}
665
666EXPORT_SYMBOL(posix_test_lock);
667
668/* This function tests for deadlock condition before putting a process to
669 * sleep. The detection scheme is no longer recursive. Recursive was neat,
670 * but dangerous - we risked stack corruption if the lock data was bad, or
671 * if the recursion was too deep for any other reason.
672 *
673 * We rely on the fact that a task can only be on one lock's wait queue
674 * at a time. When we find blocked_task on a wait queue we can re-search
675 * with blocked_task equal to that queue's owner, until either blocked_task
676 * isn't found, or blocked_task is found on a queue owned by my_task.
677 *
678 * Note: the above assumption may not be true when handling lock requests
679 * from a broken NFS client. But broken NFS clients have a lot more to
680 * worry about than proper deadlock detection anyway... --okir
681 */
682int posix_locks_deadlock(struct file_lock *caller_fl,
683 struct file_lock *block_fl)
684{
685 struct list_head *tmp;
686
687next_task:
688 if (posix_same_owner(caller_fl, block_fl))
689 return 1;
690 list_for_each(tmp, &blocked_list) {
691 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
692 if (posix_same_owner(fl, block_fl)) {
693 fl = fl->fl_next;
694 block_fl = fl;
695 goto next_task;
696 }
697 }
698 return 0;
699}
700
701EXPORT_SYMBOL(posix_locks_deadlock);
702
703/* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
704 * at the head of the list, but that's secret knowledge known only to
705 * flock_lock_file and posix_lock_file.
706 */
707static int flock_lock_file(struct file *filp, struct file_lock *new_fl)
708{
709 struct file_lock **before;
710 struct inode * inode = filp->f_dentry->d_inode;
711 int error = 0;
712 int found = 0;
713
714 lock_kernel();
715 for_each_lock(inode, before) {
716 struct file_lock *fl = *before;
717 if (IS_POSIX(fl))
718 break;
719 if (IS_LEASE(fl))
720 continue;
721 if (filp != fl->fl_file)
722 continue;
723 if (new_fl->fl_type == fl->fl_type)
724 goto out;
725 found = 1;
726 locks_delete_lock(before);
727 break;
728 }
729 unlock_kernel();
730
731 if (new_fl->fl_type == F_UNLCK)
732 return 0;
733
734 /*
735 * If a higher-priority process was blocked on the old file lock,
736 * give it the opportunity to lock the file.
737 */
738 if (found)
739 cond_resched();
740
741 lock_kernel();
742 for_each_lock(inode, before) {
743 struct file_lock *fl = *before;
744 if (IS_POSIX(fl))
745 break;
746 if (IS_LEASE(fl))
747 continue;
748 if (!flock_locks_conflict(new_fl, fl))
749 continue;
750 error = -EAGAIN;
751 if (new_fl->fl_flags & FL_SLEEP) {
752 locks_insert_block(fl, new_fl);
753 }
754 goto out;
755 }
756 locks_insert_lock(&inode->i_flock, new_fl);
757 error = 0;
758
759out:
760 unlock_kernel();
761 return error;
762}
763
764EXPORT_SYMBOL(posix_lock_file);
765
766static int __posix_lock_file(struct inode *inode, struct file_lock *request)
767{
768 struct file_lock *fl;
769 struct file_lock *new_fl, *new_fl2;
770 struct file_lock *left = NULL;
771 struct file_lock *right = NULL;
772 struct file_lock **before;
773 int error, added = 0;
774
775 /*
776 * We may need two file_lock structures for this operation,
777 * so we get them in advance to avoid races.
778 */
779 new_fl = locks_alloc_lock();
780 new_fl2 = locks_alloc_lock();
781
782 lock_kernel();
783 if (request->fl_type != F_UNLCK) {
784 for_each_lock(inode, before) {
785 struct file_lock *fl = *before;
786 if (!IS_POSIX(fl))
787 continue;
788 if (!posix_locks_conflict(request, fl))
789 continue;
790 error = -EAGAIN;
791 if (!(request->fl_flags & FL_SLEEP))
792 goto out;
793 error = -EDEADLK;
794 if (posix_locks_deadlock(request, fl))
795 goto out;
796 error = -EAGAIN;
797 locks_insert_block(fl, request);
798 goto out;
799 }
800 }
801
802 /* If we're just looking for a conflict, we're done. */
803 error = 0;
804 if (request->fl_flags & FL_ACCESS)
805 goto out;
806
807 error = -ENOLCK; /* "no luck" */
808 if (!(new_fl && new_fl2))
809 goto out;
810
811 /*
812 * We've allocated the new locks in advance, so there are no
813 * errors possible (and no blocking operations) from here on.
814 *
815 * Find the first old lock with the same owner as the new lock.
816 */
817
818 before = &inode->i_flock;
819
820 /* First skip locks owned by other processes. */
821 while ((fl = *before) && (!IS_POSIX(fl) ||
822 !posix_same_owner(request, fl))) {
823 before = &fl->fl_next;
824 }
825
826 /* Process locks with this owner. */
827 while ((fl = *before) && posix_same_owner(request, fl)) {
828 /* Detect adjacent or overlapping regions (if same lock type)
829 */
830 if (request->fl_type == fl->fl_type) {
831 if (fl->fl_end < request->fl_start - 1)
832 goto next_lock;
833 /* If the next lock in the list has entirely bigger
834 * addresses than the new one, insert the lock here.
835 */
836 if (fl->fl_start > request->fl_end + 1)
837 break;
838
839 /* If we come here, the new and old lock are of the
840 * same type and adjacent or overlapping. Make one
841 * lock yielding from the lower start address of both
842 * locks to the higher end address.
843 */
844 if (fl->fl_start > request->fl_start)
845 fl->fl_start = request->fl_start;
846 else
847 request->fl_start = fl->fl_start;
848 if (fl->fl_end < request->fl_end)
849 fl->fl_end = request->fl_end;
850 else
851 request->fl_end = fl->fl_end;
852 if (added) {
853 locks_delete_lock(before);
854 continue;
855 }
856 request = fl;
857 added = 1;
858 }
859 else {
860 /* Processing for different lock types is a bit
861 * more complex.
862 */
863 if (fl->fl_end < request->fl_start)
864 goto next_lock;
865 if (fl->fl_start > request->fl_end)
866 break;
867 if (request->fl_type == F_UNLCK)
868 added = 1;
869 if (fl->fl_start < request->fl_start)
870 left = fl;
871 /* If the next lock in the list has a higher end
872 * address than the new one, insert the new one here.
873 */
874 if (fl->fl_end > request->fl_end) {
875 right = fl;
876 break;
877 }
878 if (fl->fl_start >= request->fl_start) {
879 /* The new lock completely replaces an old
880 * one (This may happen several times).
881 */
882 if (added) {
883 locks_delete_lock(before);
884 continue;
885 }
886 /* Replace the old lock with the new one.
887 * Wake up anybody waiting for the old one,
888 * as the change in lock type might satisfy
889 * their needs.
890 */
891 locks_wake_up_blocks(fl);
892 fl->fl_start = request->fl_start;
893 fl->fl_end = request->fl_end;
894 fl->fl_type = request->fl_type;
895 fl->fl_u = request->fl_u;
896 request = fl;
897 added = 1;
898 }
899 }
900 /* Go on to next lock.
901 */
902 next_lock:
903 before = &fl->fl_next;
904 }
905
906 error = 0;
907 if (!added) {
908 if (request->fl_type == F_UNLCK)
909 goto out;
910 locks_copy_lock(new_fl, request);
911 locks_insert_lock(before, new_fl);
912 new_fl = NULL;
913 }
914 if (right) {
915 if (left == right) {
916 /* The new lock breaks the old one in two pieces,
917 * so we have to use the second new lock.
918 */
919 left = new_fl2;
920 new_fl2 = NULL;
921 locks_copy_lock(left, right);
922 locks_insert_lock(before, left);
923 }
924 right->fl_start = request->fl_end + 1;
925 locks_wake_up_blocks(right);
926 }
927 if (left) {
928 left->fl_end = request->fl_start - 1;
929 locks_wake_up_blocks(left);
930 }
931 out:
932 unlock_kernel();
933 /*
934 * Free any unused locks.
935 */
936 if (new_fl)
937 locks_free_lock(new_fl);
938 if (new_fl2)
939 locks_free_lock(new_fl2);
940 return error;
941}
942
943/**
944 * posix_lock_file - Apply a POSIX-style lock to a file
945 * @filp: The file to apply the lock to
946 * @fl: The lock to be applied
947 *
948 * Add a POSIX style lock to a file.
949 * We merge adjacent & overlapping locks whenever possible.
950 * POSIX locks are sorted by owner task, then by starting address
951 */
952int posix_lock_file(struct file *filp, struct file_lock *fl)
953{
954 return __posix_lock_file(filp->f_dentry->d_inode, fl);
955}
956
957/**
958 * posix_lock_file_wait - Apply a POSIX-style lock to a file
959 * @filp: The file to apply the lock to
960 * @fl: The lock to be applied
961 *
962 * Add a POSIX style lock to a file.
963 * We merge adjacent & overlapping locks whenever possible.
964 * POSIX locks are sorted by owner task, then by starting address
965 */
966int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
967{
968 int error;
969 might_sleep ();
970 for (;;) {
971 error = __posix_lock_file(filp->f_dentry->d_inode, fl);
972 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
973 break;
974 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
975 if (!error)
976 continue;
977
978 locks_delete_block(fl);
979 break;
980 }
981 return error;
982}
983EXPORT_SYMBOL(posix_lock_file_wait);
984
985/**
986 * locks_mandatory_locked - Check for an active lock
987 * @inode: the file to check
988 *
989 * Searches the inode's list of locks to find any POSIX locks which conflict.
990 * This function is called from locks_verify_locked() only.
991 */
992int locks_mandatory_locked(struct inode *inode)
993{
994 fl_owner_t owner = current->files;
995 struct file_lock *fl;
996
997 /*
998 * Search the lock list for this inode for any POSIX locks.
999 */
1000 lock_kernel();
1001 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1002 if (!IS_POSIX(fl))
1003 continue;
1004 if (fl->fl_owner != owner)
1005 break;
1006 }
1007 unlock_kernel();
1008 return fl ? -EAGAIN : 0;
1009}
1010
1011/**
1012 * locks_mandatory_area - Check for a conflicting lock
1013 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1014 * for shared
1015 * @inode: the file to check
1016 * @filp: how the file was opened (if it was)
1017 * @offset: start of area to check
1018 * @count: length of area to check
1019 *
1020 * Searches the inode's list of locks to find any POSIX locks which conflict.
1021 * This function is called from rw_verify_area() and
1022 * locks_verify_truncate().
1023 */
1024int locks_mandatory_area(int read_write, struct inode *inode,
1025 struct file *filp, loff_t offset,
1026 size_t count)
1027{
1028 struct file_lock fl;
1029 int error;
1030
1031 locks_init_lock(&fl);
1032 fl.fl_owner = current->files;
1033 fl.fl_pid = current->tgid;
1034 fl.fl_file = filp;
1035 fl.fl_flags = FL_POSIX | FL_ACCESS;
1036 if (filp && !(filp->f_flags & O_NONBLOCK))
1037 fl.fl_flags |= FL_SLEEP;
1038 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1039 fl.fl_start = offset;
1040 fl.fl_end = offset + count - 1;
1041
1042 for (;;) {
1043 error = __posix_lock_file(inode, &fl);
1044 if (error != -EAGAIN)
1045 break;
1046 if (!(fl.fl_flags & FL_SLEEP))
1047 break;
1048 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1049 if (!error) {
1050 /*
1051 * If we've been sleeping someone might have
1052 * changed the permissions behind our back.
1053 */
1054 if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID)
1055 continue;
1056 }
1057
1058 locks_delete_block(&fl);
1059 break;
1060 }
1061
1062 return error;
1063}
1064
1065EXPORT_SYMBOL(locks_mandatory_area);
1066
1067/* We already had a lease on this file; just change its type */
1068int lease_modify(struct file_lock **before, int arg)
1069{
1070 struct file_lock *fl = *before;
1071 int error = assign_type(fl, arg);
1072
1073 if (error)
1074 return error;
1075 locks_wake_up_blocks(fl);
1076 if (arg == F_UNLCK)
1077 locks_delete_lock(before);
1078 return 0;
1079}
1080
1081EXPORT_SYMBOL(lease_modify);
1082
1083static void time_out_leases(struct inode *inode)
1084{
1085 struct file_lock **before;
1086 struct file_lock *fl;
1087
1088 before = &inode->i_flock;
1089 while ((fl = *before) && IS_LEASE(fl) && (fl->fl_type & F_INPROGRESS)) {
1090 if ((fl->fl_break_time == 0)
1091 || time_before(jiffies, fl->fl_break_time)) {
1092 before = &fl->fl_next;
1093 continue;
1094 }
1095 printk(KERN_INFO "lease broken - owner pid = %d\n", fl->fl_pid);
1096 lease_modify(before, fl->fl_type & ~F_INPROGRESS);
1097 if (fl == *before) /* lease_modify may have freed fl */
1098 before = &fl->fl_next;
1099 }
1100}
1101
1102/**
1103 * __break_lease - revoke all outstanding leases on file
1104 * @inode: the inode of the file to return
1105 * @mode: the open mode (read or write)
1106 *
1107 * break_lease (inlined for speed) has checked there already
1108 * is a lease on this file. Leases are broken on a call to open()
1109 * or truncate(). This function can sleep unless you
1110 * specified %O_NONBLOCK to your open().
1111 */
1112int __break_lease(struct inode *inode, unsigned int mode)
1113{
1114 int error = 0, future;
1115 struct file_lock *new_fl, *flock;
1116 struct file_lock *fl;
1117 int alloc_err;
1118 unsigned long break_time;
1119 int i_have_this_lease = 0;
1120
1121 alloc_err = lease_alloc(NULL, mode & FMODE_WRITE ? F_WRLCK : F_RDLCK,
1122 &new_fl);
1123
1124 lock_kernel();
1125
1126 time_out_leases(inode);
1127
1128 flock = inode->i_flock;
1129 if ((flock == NULL) || !IS_LEASE(flock))
1130 goto out;
1131
1132 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1133 if (fl->fl_owner == current->files)
1134 i_have_this_lease = 1;
1135
1136 if (mode & FMODE_WRITE) {
1137 /* If we want write access, we have to revoke any lease. */
1138 future = F_UNLCK | F_INPROGRESS;
1139 } else if (flock->fl_type & F_INPROGRESS) {
1140 /* If the lease is already being broken, we just leave it */
1141 future = flock->fl_type;
1142 } else if (flock->fl_type & F_WRLCK) {
1143 /* Downgrade the exclusive lease to a read-only lease. */
1144 future = F_RDLCK | F_INPROGRESS;
1145 } else {
1146 /* the existing lease was read-only, so we can read too. */
1147 goto out;
1148 }
1149
1150 if (alloc_err && !i_have_this_lease && ((mode & O_NONBLOCK) == 0)) {
1151 error = alloc_err;
1152 goto out;
1153 }
1154
1155 break_time = 0;
1156 if (lease_break_time > 0) {
1157 break_time = jiffies + lease_break_time * HZ;
1158 if (break_time == 0)
1159 break_time++; /* so that 0 means no break time */
1160 }
1161
1162 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1163 if (fl->fl_type != future) {
1164 fl->fl_type = future;
1165 fl->fl_break_time = break_time;
1166 /* lease must have lmops break callback */
1167 fl->fl_lmops->fl_break(fl);
1168 }
1169 }
1170
1171 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1172 error = -EWOULDBLOCK;
1173 goto out;
1174 }
1175
1176restart:
1177 break_time = flock->fl_break_time;
1178 if (break_time != 0) {
1179 break_time -= jiffies;
1180 if (break_time == 0)
1181 break_time++;
1182 }
1183 error = locks_block_on_timeout(flock, new_fl, break_time);
1184 if (error >= 0) {
1185 if (error == 0)
1186 time_out_leases(inode);
1187 /* Wait for the next lease that has not been broken yet */
1188 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1189 flock = flock->fl_next) {
1190 if (flock->fl_type & F_INPROGRESS)
1191 goto restart;
1192 }
1193 error = 0;
1194 }
1195
1196out:
1197 unlock_kernel();
1198 if (!alloc_err)
1199 locks_free_lock(new_fl);
1200 return error;
1201}
1202
1203EXPORT_SYMBOL(__break_lease);
1204
1205/**
1206 * lease_get_mtime
1207 * @inode: the inode
1208 * @time: pointer to a timespec which will contain the last modified time
1209 *
1210 * This is to force NFS clients to flush their caches for files with
1211 * exclusive leases. The justification is that if someone has an
1212 * exclusive lease, then they could be modifiying it.
1213 */
1214void lease_get_mtime(struct inode *inode, struct timespec *time)
1215{
1216 struct file_lock *flock = inode->i_flock;
1217 if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK))
1218 *time = current_fs_time(inode->i_sb);
1219 else
1220 *time = inode->i_mtime;
1221}
1222
1223EXPORT_SYMBOL(lease_get_mtime);
1224
1225/**
1226 * fcntl_getlease - Enquire what lease is currently active
1227 * @filp: the file
1228 *
1229 * The value returned by this function will be one of
1230 * (if no lease break is pending):
1231 *
1232 * %F_RDLCK to indicate a shared lease is held.
1233 *
1234 * %F_WRLCK to indicate an exclusive lease is held.
1235 *
1236 * %F_UNLCK to indicate no lease is held.
1237 *
1238 * (if a lease break is pending):
1239 *
1240 * %F_RDLCK to indicate an exclusive lease needs to be
1241 * changed to a shared lease (or removed).
1242 *
1243 * %F_UNLCK to indicate the lease needs to be removed.
1244 *
1245 * XXX: sfr & willy disagree over whether F_INPROGRESS
1246 * should be returned to userspace.
1247 */
1248int fcntl_getlease(struct file *filp)
1249{
1250 struct file_lock *fl;
1251 int type = F_UNLCK;
1252
1253 lock_kernel();
1254 time_out_leases(filp->f_dentry->d_inode);
1255 for (fl = filp->f_dentry->d_inode->i_flock; fl && IS_LEASE(fl);
1256 fl = fl->fl_next) {
1257 if (fl->fl_file == filp) {
1258 type = fl->fl_type & ~F_INPROGRESS;
1259 break;
1260 }
1261 }
1262 unlock_kernel();
1263 return type;
1264}
1265
1266/**
1267 * __setlease - sets a lease on an open file
1268 * @filp: file pointer
1269 * @arg: type of lease to obtain
1270 * @flp: input - file_lock to use, output - file_lock inserted
1271 *
1272 * The (input) flp->fl_lmops->fl_break function is required
1273 * by break_lease().
1274 *
1275 * Called with kernel lock held.
1276 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07001277static int __setlease(struct file *filp, long arg, struct file_lock **flp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278{
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001279 struct file_lock *fl, **before, **my_before = NULL, *lease;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 struct dentry *dentry = filp->f_dentry;
1281 struct inode *inode = dentry->d_inode;
1282 int error, rdlease_count = 0, wrlease_count = 0;
1283
1284 time_out_leases(inode);
1285
1286 error = -EINVAL;
1287 if (!flp || !(*flp) || !(*flp)->fl_lmops || !(*flp)->fl_lmops->fl_break)
1288 goto out;
1289
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001290 lease = *flp;
1291
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292 error = -EAGAIN;
1293 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1294 goto out;
1295 if ((arg == F_WRLCK)
1296 && ((atomic_read(&dentry->d_count) > 1)
1297 || (atomic_read(&inode->i_count) > 1)))
1298 goto out;
1299
1300 /*
1301 * At this point, we know that if there is an exclusive
1302 * lease on this file, then we hold it on this filp
1303 * (otherwise our open of this file would have blocked).
1304 * And if we are trying to acquire an exclusive lease,
1305 * then the file is not open by anyone (including us)
1306 * except for this filp.
1307 */
1308 for (before = &inode->i_flock;
1309 ((fl = *before) != NULL) && IS_LEASE(fl);
1310 before = &fl->fl_next) {
1311 if (lease->fl_lmops->fl_mylease(fl, lease))
1312 my_before = before;
1313 else if (fl->fl_type == (F_INPROGRESS | F_UNLCK))
1314 /*
1315 * Someone is in the process of opening this
1316 * file for writing so we may not take an
1317 * exclusive lease on it.
1318 */
1319 wrlease_count++;
1320 else
1321 rdlease_count++;
1322 }
1323
1324 if ((arg == F_RDLCK && (wrlease_count > 0)) ||
1325 (arg == F_WRLCK && ((rdlease_count + wrlease_count) > 0)))
1326 goto out;
1327
1328 if (my_before != NULL) {
1329 error = lease->fl_lmops->fl_change(my_before, arg);
1330 goto out;
1331 }
1332
1333 error = 0;
1334 if (arg == F_UNLCK)
1335 goto out;
1336
1337 error = -EINVAL;
1338 if (!leases_enable)
1339 goto out;
1340
1341 error = lease_alloc(filp, arg, &fl);
1342 if (error)
1343 goto out;
1344
1345 locks_copy_lock(fl, lease);
1346
1347 locks_insert_lock(before, fl);
1348
1349 *flp = fl;
1350out:
1351 return error;
1352}
1353
1354 /**
1355 * setlease - sets a lease on an open file
1356 * @filp: file pointer
1357 * @arg: type of lease to obtain
1358 * @lease: file_lock to use
1359 *
1360 * Call this to establish a lease on the file.
1361 * The fl_lmops fl_break function is required by break_lease
1362 */
1363
1364int setlease(struct file *filp, long arg, struct file_lock **lease)
1365{
1366 struct dentry *dentry = filp->f_dentry;
1367 struct inode *inode = dentry->d_inode;
1368 int error;
1369
1370 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1371 return -EACCES;
1372 if (!S_ISREG(inode->i_mode))
1373 return -EINVAL;
1374 error = security_file_lock(filp, arg);
1375 if (error)
1376 return error;
1377
1378 lock_kernel();
1379 error = __setlease(filp, arg, lease);
1380 unlock_kernel();
1381
1382 return error;
1383}
1384
1385EXPORT_SYMBOL(setlease);
1386
1387/**
1388 * fcntl_setlease - sets a lease on an open file
1389 * @fd: open file descriptor
1390 * @filp: file pointer
1391 * @arg: type of lease to obtain
1392 *
1393 * Call this fcntl to establish a lease on the file.
1394 * Note that you also need to call %F_SETSIG to
1395 * receive a signal when the lease is broken.
1396 */
1397int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1398{
1399 struct file_lock fl, *flp = &fl;
1400 struct dentry *dentry = filp->f_dentry;
1401 struct inode *inode = dentry->d_inode;
1402 int error;
1403
1404 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1405 return -EACCES;
1406 if (!S_ISREG(inode->i_mode))
1407 return -EINVAL;
1408 error = security_file_lock(filp, arg);
1409 if (error)
1410 return error;
1411
1412 locks_init_lock(&fl);
1413 error = lease_init(filp, arg, &fl);
1414 if (error)
1415 return error;
1416
1417 lock_kernel();
1418
1419 error = __setlease(filp, arg, &flp);
1420 if (error)
1421 goto out_unlock;
1422
1423 error = fasync_helper(fd, filp, 1, &flp->fl_fasync);
1424 if (error < 0) {
1425 /* remove lease just inserted by __setlease */
1426 flp->fl_type = F_UNLCK | F_INPROGRESS;
1427 flp->fl_break_time = jiffies- 10;
1428 time_out_leases(inode);
1429 goto out_unlock;
1430 }
1431
1432 error = f_setown(filp, current->pid, 0);
1433out_unlock:
1434 unlock_kernel();
1435 return error;
1436}
1437
1438/**
1439 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1440 * @filp: The file to apply the lock to
1441 * @fl: The lock to be applied
1442 *
1443 * Add a FLOCK style lock to a file.
1444 */
1445int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1446{
1447 int error;
1448 might_sleep();
1449 for (;;) {
1450 error = flock_lock_file(filp, fl);
1451 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1452 break;
1453 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1454 if (!error)
1455 continue;
1456
1457 locks_delete_block(fl);
1458 break;
1459 }
1460 return error;
1461}
1462
1463EXPORT_SYMBOL(flock_lock_file_wait);
1464
1465/**
1466 * sys_flock: - flock() system call.
1467 * @fd: the file descriptor to lock.
1468 * @cmd: the type of lock to apply.
1469 *
1470 * Apply a %FL_FLOCK style lock to an open file descriptor.
1471 * The @cmd can be one of
1472 *
1473 * %LOCK_SH -- a shared lock.
1474 *
1475 * %LOCK_EX -- an exclusive lock.
1476 *
1477 * %LOCK_UN -- remove an existing lock.
1478 *
1479 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1480 *
1481 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1482 * processes read and write access respectively.
1483 */
1484asmlinkage long sys_flock(unsigned int fd, unsigned int cmd)
1485{
1486 struct file *filp;
1487 struct file_lock *lock;
1488 int can_sleep, unlock;
1489 int error;
1490
1491 error = -EBADF;
1492 filp = fget(fd);
1493 if (!filp)
1494 goto out;
1495
1496 can_sleep = !(cmd & LOCK_NB);
1497 cmd &= ~LOCK_NB;
1498 unlock = (cmd == LOCK_UN);
1499
1500 if (!unlock && !(cmd & LOCK_MAND) && !(filp->f_mode & 3))
1501 goto out_putf;
1502
1503 error = flock_make_lock(filp, &lock, cmd);
1504 if (error)
1505 goto out_putf;
1506 if (can_sleep)
1507 lock->fl_flags |= FL_SLEEP;
1508
1509 error = security_file_lock(filp, cmd);
1510 if (error)
1511 goto out_free;
1512
1513 if (filp->f_op && filp->f_op->flock)
1514 error = filp->f_op->flock(filp,
1515 (can_sleep) ? F_SETLKW : F_SETLK,
1516 lock);
1517 else
1518 error = flock_lock_file_wait(filp, lock);
1519
1520 out_free:
1521 if (list_empty(&lock->fl_link)) {
1522 locks_free_lock(lock);
1523 }
1524
1525 out_putf:
1526 fput(filp);
1527 out:
1528 return error;
1529}
1530
1531/* Report the first existing lock that would conflict with l.
1532 * This implements the F_GETLK command of fcntl().
1533 */
1534int fcntl_getlk(struct file *filp, struct flock __user *l)
1535{
1536 struct file_lock *fl, file_lock;
1537 struct flock flock;
1538 int error;
1539
1540 error = -EFAULT;
1541 if (copy_from_user(&flock, l, sizeof(flock)))
1542 goto out;
1543 error = -EINVAL;
1544 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1545 goto out;
1546
1547 error = flock_to_posix_lock(filp, &file_lock, &flock);
1548 if (error)
1549 goto out;
1550
1551 if (filp->f_op && filp->f_op->lock) {
1552 error = filp->f_op->lock(filp, F_GETLK, &file_lock);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00001553 if (file_lock.fl_ops && file_lock.fl_ops->fl_release_private)
1554 file_lock.fl_ops->fl_release_private(&file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555 if (error < 0)
1556 goto out;
1557 else
1558 fl = (file_lock.fl_type == F_UNLCK ? NULL : &file_lock);
1559 } else {
1560 fl = posix_test_lock(filp, &file_lock);
1561 }
1562
1563 flock.l_type = F_UNLCK;
1564 if (fl != NULL) {
1565 flock.l_pid = fl->fl_pid;
1566#if BITS_PER_LONG == 32
1567 /*
1568 * Make sure we can represent the posix lock via
1569 * legacy 32bit flock.
1570 */
1571 error = -EOVERFLOW;
1572 if (fl->fl_start > OFFT_OFFSET_MAX)
1573 goto out;
1574 if ((fl->fl_end != OFFSET_MAX)
1575 && (fl->fl_end > OFFT_OFFSET_MAX))
1576 goto out;
1577#endif
1578 flock.l_start = fl->fl_start;
1579 flock.l_len = fl->fl_end == OFFSET_MAX ? 0 :
1580 fl->fl_end - fl->fl_start + 1;
1581 flock.l_whence = 0;
1582 flock.l_type = fl->fl_type;
1583 }
1584 error = -EFAULT;
1585 if (!copy_to_user(l, &flock, sizeof(flock)))
1586 error = 0;
1587out:
1588 return error;
1589}
1590
1591/* Apply the lock described by l to an open file descriptor.
1592 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1593 */
Peter Staubachc2936212005-07-27 11:45:09 -07001594int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1595 struct flock __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596{
1597 struct file_lock *file_lock = locks_alloc_lock();
1598 struct flock flock;
1599 struct inode *inode;
1600 int error;
1601
1602 if (file_lock == NULL)
1603 return -ENOLCK;
1604
1605 /*
1606 * This might block, so we do it before checking the inode.
1607 */
1608 error = -EFAULT;
1609 if (copy_from_user(&flock, l, sizeof(flock)))
1610 goto out;
1611
1612 inode = filp->f_dentry->d_inode;
1613
1614 /* Don't allow mandatory locks on files that may be memory mapped
1615 * and shared.
1616 */
1617 if (IS_MANDLOCK(inode) &&
1618 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1619 mapping_writably_mapped(filp->f_mapping)) {
1620 error = -EAGAIN;
1621 goto out;
1622 }
1623
Peter Staubachc2936212005-07-27 11:45:09 -07001624again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625 error = flock_to_posix_lock(filp, file_lock, &flock);
1626 if (error)
1627 goto out;
1628 if (cmd == F_SETLKW) {
1629 file_lock->fl_flags |= FL_SLEEP;
1630 }
1631
1632 error = -EBADF;
1633 switch (flock.l_type) {
1634 case F_RDLCK:
1635 if (!(filp->f_mode & FMODE_READ))
1636 goto out;
1637 break;
1638 case F_WRLCK:
1639 if (!(filp->f_mode & FMODE_WRITE))
1640 goto out;
1641 break;
1642 case F_UNLCK:
1643 break;
1644 default:
1645 error = -EINVAL;
1646 goto out;
1647 }
1648
1649 error = security_file_lock(filp, file_lock->fl_type);
1650 if (error)
1651 goto out;
1652
Peter Staubachc2936212005-07-27 11:45:09 -07001653 if (filp->f_op && filp->f_op->lock != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 error = filp->f_op->lock(filp, cmd, file_lock);
Peter Staubachc2936212005-07-27 11:45:09 -07001655 else {
1656 for (;;) {
1657 error = __posix_lock_file(inode, file_lock);
1658 if ((error != -EAGAIN) || (cmd == F_SETLK))
1659 break;
1660 error = wait_event_interruptible(file_lock->fl_wait,
1661 !file_lock->fl_next);
1662 if (!error)
1663 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664
Peter Staubachc2936212005-07-27 11:45:09 -07001665 locks_delete_block(file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666 break;
Peter Staubachc2936212005-07-27 11:45:09 -07001667 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668 }
1669
Peter Staubachc2936212005-07-27 11:45:09 -07001670 /*
1671 * Attempt to detect a close/fcntl race and recover by
1672 * releasing the lock that was just acquired.
1673 */
1674 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1675 flock.l_type = F_UNLCK;
1676 goto again;
1677 }
1678
1679out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 locks_free_lock(file_lock);
1681 return error;
1682}
1683
1684#if BITS_PER_LONG == 32
1685/* Report the first existing lock that would conflict with l.
1686 * This implements the F_GETLK command of fcntl().
1687 */
1688int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1689{
1690 struct file_lock *fl, file_lock;
1691 struct flock64 flock;
1692 int error;
1693
1694 error = -EFAULT;
1695 if (copy_from_user(&flock, l, sizeof(flock)))
1696 goto out;
1697 error = -EINVAL;
1698 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1699 goto out;
1700
1701 error = flock64_to_posix_lock(filp, &file_lock, &flock);
1702 if (error)
1703 goto out;
1704
1705 if (filp->f_op && filp->f_op->lock) {
1706 error = filp->f_op->lock(filp, F_GETLK, &file_lock);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00001707 if (file_lock.fl_ops && file_lock.fl_ops->fl_release_private)
1708 file_lock.fl_ops->fl_release_private(&file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709 if (error < 0)
1710 goto out;
1711 else
1712 fl = (file_lock.fl_type == F_UNLCK ? NULL : &file_lock);
1713 } else {
1714 fl = posix_test_lock(filp, &file_lock);
1715 }
1716
1717 flock.l_type = F_UNLCK;
1718 if (fl != NULL) {
1719 flock.l_pid = fl->fl_pid;
1720 flock.l_start = fl->fl_start;
1721 flock.l_len = fl->fl_end == OFFSET_MAX ? 0 :
1722 fl->fl_end - fl->fl_start + 1;
1723 flock.l_whence = 0;
1724 flock.l_type = fl->fl_type;
1725 }
1726 error = -EFAULT;
1727 if (!copy_to_user(l, &flock, sizeof(flock)))
1728 error = 0;
1729
1730out:
1731 return error;
1732}
1733
1734/* Apply the lock described by l to an open file descriptor.
1735 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1736 */
Peter Staubachc2936212005-07-27 11:45:09 -07001737int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
1738 struct flock64 __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739{
1740 struct file_lock *file_lock = locks_alloc_lock();
1741 struct flock64 flock;
1742 struct inode *inode;
1743 int error;
1744
1745 if (file_lock == NULL)
1746 return -ENOLCK;
1747
1748 /*
1749 * This might block, so we do it before checking the inode.
1750 */
1751 error = -EFAULT;
1752 if (copy_from_user(&flock, l, sizeof(flock)))
1753 goto out;
1754
1755 inode = filp->f_dentry->d_inode;
1756
1757 /* Don't allow mandatory locks on files that may be memory mapped
1758 * and shared.
1759 */
1760 if (IS_MANDLOCK(inode) &&
1761 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1762 mapping_writably_mapped(filp->f_mapping)) {
1763 error = -EAGAIN;
1764 goto out;
1765 }
1766
Peter Staubachc2936212005-07-27 11:45:09 -07001767again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768 error = flock64_to_posix_lock(filp, file_lock, &flock);
1769 if (error)
1770 goto out;
1771 if (cmd == F_SETLKW64) {
1772 file_lock->fl_flags |= FL_SLEEP;
1773 }
1774
1775 error = -EBADF;
1776 switch (flock.l_type) {
1777 case F_RDLCK:
1778 if (!(filp->f_mode & FMODE_READ))
1779 goto out;
1780 break;
1781 case F_WRLCK:
1782 if (!(filp->f_mode & FMODE_WRITE))
1783 goto out;
1784 break;
1785 case F_UNLCK:
1786 break;
1787 default:
1788 error = -EINVAL;
1789 goto out;
1790 }
1791
1792 error = security_file_lock(filp, file_lock->fl_type);
1793 if (error)
1794 goto out;
1795
Peter Staubachc2936212005-07-27 11:45:09 -07001796 if (filp->f_op && filp->f_op->lock != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797 error = filp->f_op->lock(filp, cmd, file_lock);
Peter Staubachc2936212005-07-27 11:45:09 -07001798 else {
1799 for (;;) {
1800 error = __posix_lock_file(inode, file_lock);
1801 if ((error != -EAGAIN) || (cmd == F_SETLK64))
1802 break;
1803 error = wait_event_interruptible(file_lock->fl_wait,
1804 !file_lock->fl_next);
1805 if (!error)
1806 continue;
1807
1808 locks_delete_block(file_lock);
1809 break;
1810 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811 }
1812
Peter Staubachc2936212005-07-27 11:45:09 -07001813 /*
1814 * Attempt to detect a close/fcntl race and recover by
1815 * releasing the lock that was just acquired.
1816 */
1817 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1818 flock.l_type = F_UNLCK;
1819 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820 }
1821
1822out:
1823 locks_free_lock(file_lock);
1824 return error;
1825}
1826#endif /* BITS_PER_LONG == 32 */
1827
1828/*
1829 * This function is called when the file is being removed
1830 * from the task's fd array. POSIX locks belonging to this task
1831 * are deleted at this time.
1832 */
1833void locks_remove_posix(struct file *filp, fl_owner_t owner)
1834{
1835 struct file_lock lock, **before;
1836
1837 /*
1838 * If there are no locks held on this file, we don't need to call
1839 * posix_lock_file(). Another process could be setting a lock on this
1840 * file at the same time, but we wouldn't remove that lock anyway.
1841 */
1842 before = &filp->f_dentry->d_inode->i_flock;
1843 if (*before == NULL)
1844 return;
1845
1846 lock.fl_type = F_UNLCK;
1847 lock.fl_flags = FL_POSIX;
1848 lock.fl_start = 0;
1849 lock.fl_end = OFFSET_MAX;
1850 lock.fl_owner = owner;
1851 lock.fl_pid = current->tgid;
1852 lock.fl_file = filp;
1853 lock.fl_ops = NULL;
1854 lock.fl_lmops = NULL;
1855
1856 if (filp->f_op && filp->f_op->lock != NULL) {
1857 filp->f_op->lock(filp, F_SETLK, &lock);
1858 goto out;
1859 }
1860
1861 /* Can't use posix_lock_file here; we need to remove it no matter
1862 * which pid we have.
1863 */
1864 lock_kernel();
1865 while (*before != NULL) {
1866 struct file_lock *fl = *before;
1867 if (IS_POSIX(fl) && posix_same_owner(fl, &lock)) {
1868 locks_delete_lock(before);
1869 continue;
1870 }
1871 before = &fl->fl_next;
1872 }
1873 unlock_kernel();
1874out:
1875 if (lock.fl_ops && lock.fl_ops->fl_release_private)
1876 lock.fl_ops->fl_release_private(&lock);
1877}
1878
1879EXPORT_SYMBOL(locks_remove_posix);
1880
1881/*
1882 * This function is called on the last close of an open file.
1883 */
1884void locks_remove_flock(struct file *filp)
1885{
1886 struct inode * inode = filp->f_dentry->d_inode;
1887 struct file_lock *fl;
1888 struct file_lock **before;
1889
1890 if (!inode->i_flock)
1891 return;
1892
1893 if (filp->f_op && filp->f_op->flock) {
1894 struct file_lock fl = {
1895 .fl_pid = current->tgid,
1896 .fl_file = filp,
1897 .fl_flags = FL_FLOCK,
1898 .fl_type = F_UNLCK,
1899 .fl_end = OFFSET_MAX,
1900 };
1901 filp->f_op->flock(filp, F_SETLKW, &fl);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00001902 if (fl.fl_ops && fl.fl_ops->fl_release_private)
1903 fl.fl_ops->fl_release_private(&fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904 }
1905
1906 lock_kernel();
1907 before = &inode->i_flock;
1908
1909 while ((fl = *before) != NULL) {
1910 if (fl->fl_file == filp) {
Peter Staubachc2936212005-07-27 11:45:09 -07001911 if (IS_FLOCK(fl)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912 locks_delete_lock(before);
1913 continue;
1914 }
1915 if (IS_LEASE(fl)) {
1916 lease_modify(before, F_UNLCK);
1917 continue;
1918 }
1919 /* What? */
1920 BUG();
1921 }
1922 before = &fl->fl_next;
1923 }
1924 unlock_kernel();
1925}
1926
1927/**
1928 * posix_block_lock - blocks waiting for a file lock
1929 * @blocker: the lock which is blocking
1930 * @waiter: the lock which conflicts and has to wait
1931 *
1932 * lockd needs to block waiting for locks.
1933 */
1934void
1935posix_block_lock(struct file_lock *blocker, struct file_lock *waiter)
1936{
1937 locks_insert_block(blocker, waiter);
1938}
1939
1940EXPORT_SYMBOL(posix_block_lock);
1941
1942/**
1943 * posix_unblock_lock - stop waiting for a file lock
1944 * @filp: how the file was opened
1945 * @waiter: the lock which was waiting
1946 *
1947 * lockd needs to block waiting for locks.
1948 */
1949void
1950posix_unblock_lock(struct file *filp, struct file_lock *waiter)
1951{
1952 /*
1953 * A remote machine may cancel the lock request after it's been
1954 * granted locally. If that happens, we need to delete the lock.
1955 */
1956 lock_kernel();
1957 if (waiter->fl_next) {
1958 __locks_delete_block(waiter);
1959 unlock_kernel();
1960 } else {
1961 unlock_kernel();
1962 waiter->fl_type = F_UNLCK;
1963 posix_lock_file(filp, waiter);
1964 }
1965}
1966
1967EXPORT_SYMBOL(posix_unblock_lock);
1968
1969static void lock_get_status(char* out, struct file_lock *fl, int id, char *pfx)
1970{
1971 struct inode *inode = NULL;
1972
1973 if (fl->fl_file != NULL)
1974 inode = fl->fl_file->f_dentry->d_inode;
1975
1976 out += sprintf(out, "%d:%s ", id, pfx);
1977 if (IS_POSIX(fl)) {
1978 out += sprintf(out, "%6s %s ",
1979 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
1980 (inode == NULL) ? "*NOINODE*" :
1981 (IS_MANDLOCK(inode) &&
1982 (inode->i_mode & (S_IXGRP | S_ISGID)) == S_ISGID) ?
1983 "MANDATORY" : "ADVISORY ");
1984 } else if (IS_FLOCK(fl)) {
1985 if (fl->fl_type & LOCK_MAND) {
1986 out += sprintf(out, "FLOCK MSNFS ");
1987 } else {
1988 out += sprintf(out, "FLOCK ADVISORY ");
1989 }
1990 } else if (IS_LEASE(fl)) {
1991 out += sprintf(out, "LEASE ");
1992 if (fl->fl_type & F_INPROGRESS)
1993 out += sprintf(out, "BREAKING ");
1994 else if (fl->fl_file)
1995 out += sprintf(out, "ACTIVE ");
1996 else
1997 out += sprintf(out, "BREAKER ");
1998 } else {
1999 out += sprintf(out, "UNKNOWN UNKNOWN ");
2000 }
2001 if (fl->fl_type & LOCK_MAND) {
2002 out += sprintf(out, "%s ",
2003 (fl->fl_type & LOCK_READ)
2004 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2005 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2006 } else {
2007 out += sprintf(out, "%s ",
2008 (fl->fl_type & F_INPROGRESS)
2009 ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ "
2010 : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ ");
2011 }
2012 if (inode) {
2013#ifdef WE_CAN_BREAK_LSLK_NOW
2014 out += sprintf(out, "%d %s:%ld ", fl->fl_pid,
2015 inode->i_sb->s_id, inode->i_ino);
2016#else
2017 /* userspace relies on this representation of dev_t ;-( */
2018 out += sprintf(out, "%d %02x:%02x:%ld ", fl->fl_pid,
2019 MAJOR(inode->i_sb->s_dev),
2020 MINOR(inode->i_sb->s_dev), inode->i_ino);
2021#endif
2022 } else {
2023 out += sprintf(out, "%d <none>:0 ", fl->fl_pid);
2024 }
2025 if (IS_POSIX(fl)) {
2026 if (fl->fl_end == OFFSET_MAX)
2027 out += sprintf(out, "%Ld EOF\n", fl->fl_start);
2028 else
2029 out += sprintf(out, "%Ld %Ld\n", fl->fl_start,
2030 fl->fl_end);
2031 } else {
2032 out += sprintf(out, "0 EOF\n");
2033 }
2034}
2035
2036static void move_lock_status(char **p, off_t* pos, off_t offset)
2037{
2038 int len;
2039 len = strlen(*p);
2040 if(*pos >= offset) {
2041 /* the complete line is valid */
2042 *p += len;
2043 *pos += len;
2044 return;
2045 }
2046 if(*pos+len > offset) {
2047 /* use the second part of the line */
2048 int i = offset-*pos;
2049 memmove(*p,*p+i,len-i);
2050 *p += len-i;
2051 *pos += len;
2052 return;
2053 }
2054 /* discard the complete line */
2055 *pos += len;
2056}
2057
2058/**
2059 * get_locks_status - reports lock usage in /proc/locks
2060 * @buffer: address in userspace to write into
2061 * @start: ?
2062 * @offset: how far we are through the buffer
2063 * @length: how much to read
2064 */
2065
2066int get_locks_status(char *buffer, char **start, off_t offset, int length)
2067{
2068 struct list_head *tmp;
2069 char *q = buffer;
2070 off_t pos = 0;
2071 int i = 0;
2072
2073 lock_kernel();
2074 list_for_each(tmp, &file_lock_list) {
2075 struct list_head *btmp;
2076 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
2077 lock_get_status(q, fl, ++i, "");
2078 move_lock_status(&q, &pos, offset);
2079
2080 if(pos >= offset+length)
2081 goto done;
2082
2083 list_for_each(btmp, &fl->fl_block) {
2084 struct file_lock *bfl = list_entry(btmp,
2085 struct file_lock, fl_block);
2086 lock_get_status(q, bfl, i, " ->");
2087 move_lock_status(&q, &pos, offset);
2088
2089 if(pos >= offset+length)
2090 goto done;
2091 }
2092 }
2093done:
2094 unlock_kernel();
2095 *start = buffer;
2096 if(q-buffer < length)
2097 return (q-buffer);
2098 return length;
2099}
2100
2101/**
2102 * lock_may_read - checks that the region is free of locks
2103 * @inode: the inode that is being read
2104 * @start: the first byte to read
2105 * @len: the number of bytes to read
2106 *
2107 * Emulates Windows locking requirements. Whole-file
2108 * mandatory locks (share modes) can prohibit a read and
2109 * byte-range POSIX locks can prohibit a read if they overlap.
2110 *
2111 * N.B. this function is only ever called
2112 * from knfsd and ownership of locks is never checked.
2113 */
2114int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2115{
2116 struct file_lock *fl;
2117 int result = 1;
2118 lock_kernel();
2119 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2120 if (IS_POSIX(fl)) {
2121 if (fl->fl_type == F_RDLCK)
2122 continue;
2123 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2124 continue;
2125 } else if (IS_FLOCK(fl)) {
2126 if (!(fl->fl_type & LOCK_MAND))
2127 continue;
2128 if (fl->fl_type & LOCK_READ)
2129 continue;
2130 } else
2131 continue;
2132 result = 0;
2133 break;
2134 }
2135 unlock_kernel();
2136 return result;
2137}
2138
2139EXPORT_SYMBOL(lock_may_read);
2140
2141/**
2142 * lock_may_write - checks that the region is free of locks
2143 * @inode: the inode that is being written
2144 * @start: the first byte to write
2145 * @len: the number of bytes to write
2146 *
2147 * Emulates Windows locking requirements. Whole-file
2148 * mandatory locks (share modes) can prohibit a write and
2149 * byte-range POSIX locks can prohibit a write if they overlap.
2150 *
2151 * N.B. this function is only ever called
2152 * from knfsd and ownership of locks is never checked.
2153 */
2154int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2155{
2156 struct file_lock *fl;
2157 int result = 1;
2158 lock_kernel();
2159 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2160 if (IS_POSIX(fl)) {
2161 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2162 continue;
2163 } else if (IS_FLOCK(fl)) {
2164 if (!(fl->fl_type & LOCK_MAND))
2165 continue;
2166 if (fl->fl_type & LOCK_WRITE)
2167 continue;
2168 } else
2169 continue;
2170 result = 0;
2171 break;
2172 }
2173 unlock_kernel();
2174 return result;
2175}
2176
2177EXPORT_SYMBOL(lock_may_write);
2178
2179static inline void __steal_locks(struct file *file, fl_owner_t from)
2180{
2181 struct inode *inode = file->f_dentry->d_inode;
2182 struct file_lock *fl = inode->i_flock;
2183
2184 while (fl) {
2185 if (fl->fl_file == file && fl->fl_owner == from)
2186 fl->fl_owner = current->files;
2187 fl = fl->fl_next;
2188 }
2189}
2190
2191/* When getting ready for executing a binary, we make sure that current
2192 * has a files_struct on its own. Before dropping the old files_struct,
2193 * we take over ownership of all locks for all file descriptors we own.
2194 * Note that we may accidentally steal a lock for a file that a sibling
2195 * has created since the unshare_files() call.
2196 */
2197void steal_locks(fl_owner_t from)
2198{
2199 struct files_struct *files = current->files;
2200 int i, j;
2201
2202 if (from == files)
2203 return;
2204
2205 lock_kernel();
2206 j = 0;
2207 for (;;) {
2208 unsigned long set;
2209 i = j * __NFDBITS;
2210 if (i >= files->max_fdset || i >= files->max_fds)
2211 break;
2212 set = files->open_fds->fds_bits[j++];
2213 while (set) {
2214 if (set & 1) {
2215 struct file *file = files->fd[i];
2216 if (file)
2217 __steal_locks(file, from);
2218 }
2219 i++;
2220 set >>= 1;
2221 }
2222 }
2223 unlock_kernel();
2224}
2225EXPORT_SYMBOL(steal_locks);
2226
2227static int __init filelock_init(void)
2228{
2229 filelock_cache = kmem_cache_create("file_lock_cache",
2230 sizeof(struct file_lock), 0, SLAB_PANIC,
2231 init_once, NULL);
2232 return 0;
2233}
2234
2235core_initcall(filelock_init);