blob: f7daa5f48949c41ea02b687e1521a440387b0592 [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
142LIST_HEAD(file_lock_list);
143
144EXPORT_SYMBOL(file_lock_list);
145
146static LIST_HEAD(blocked_list);
147
148static kmem_cache_t *filelock_cache;
149
150/* Allocate an empty lock structure. */
151static struct file_lock *locks_alloc_lock(void)
152{
153 return kmem_cache_alloc(filelock_cache, SLAB_KERNEL);
154}
155
156/* Free a lock which is not in use. */
157static inline void locks_free_lock(struct file_lock *fl)
158{
159 if (fl == NULL) {
160 BUG();
161 return;
162 }
163 if (waitqueue_active(&fl->fl_wait))
164 panic("Attempting to free lock with active wait queue");
165
166 if (!list_empty(&fl->fl_block))
167 panic("Attempting to free lock with active block list");
168
169 if (!list_empty(&fl->fl_link))
170 panic("Attempting to free lock on active lock list");
171
172 if (fl->fl_ops) {
173 if (fl->fl_ops->fl_release_private)
174 fl->fl_ops->fl_release_private(fl);
175 fl->fl_ops = NULL;
176 }
177
178 if (fl->fl_lmops) {
179 if (fl->fl_lmops->fl_release_private)
180 fl->fl_lmops->fl_release_private(fl);
181 fl->fl_lmops = NULL;
182 }
183
184 kmem_cache_free(filelock_cache, fl);
185}
186
187void locks_init_lock(struct file_lock *fl)
188{
189 INIT_LIST_HEAD(&fl->fl_link);
190 INIT_LIST_HEAD(&fl->fl_block);
191 init_waitqueue_head(&fl->fl_wait);
192 fl->fl_next = NULL;
193 fl->fl_fasync = NULL;
194 fl->fl_owner = NULL;
195 fl->fl_pid = 0;
196 fl->fl_file = NULL;
197 fl->fl_flags = 0;
198 fl->fl_type = 0;
199 fl->fl_start = fl->fl_end = 0;
200 fl->fl_ops = NULL;
201 fl->fl_lmops = NULL;
202}
203
204EXPORT_SYMBOL(locks_init_lock);
205
206/*
207 * Initialises the fields of the file lock which are invariant for
208 * free file_locks.
209 */
210static void init_once(void *foo, kmem_cache_t *cache, unsigned long flags)
211{
212 struct file_lock *lock = (struct file_lock *) foo;
213
214 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) !=
215 SLAB_CTOR_CONSTRUCTOR)
216 return;
217
218 locks_init_lock(lock);
219}
220
221/*
222 * Initialize a new lock from an existing file_lock structure.
223 */
224void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
225{
226 new->fl_owner = fl->fl_owner;
227 new->fl_pid = fl->fl_pid;
228 new->fl_file = fl->fl_file;
229 new->fl_flags = fl->fl_flags;
230 new->fl_type = fl->fl_type;
231 new->fl_start = fl->fl_start;
232 new->fl_end = fl->fl_end;
233 new->fl_ops = fl->fl_ops;
234 new->fl_lmops = fl->fl_lmops;
235 if (fl->fl_ops && fl->fl_ops->fl_copy_lock)
236 fl->fl_ops->fl_copy_lock(new, fl);
237 if (fl->fl_lmops && fl->fl_lmops->fl_copy_lock)
238 fl->fl_lmops->fl_copy_lock(new, fl);
239}
240
241EXPORT_SYMBOL(locks_copy_lock);
242
243static inline int flock_translate_cmd(int cmd) {
244 if (cmd & LOCK_MAND)
245 return cmd & (LOCK_MAND | LOCK_RW);
246 switch (cmd) {
247 case LOCK_SH:
248 return F_RDLCK;
249 case LOCK_EX:
250 return F_WRLCK;
251 case LOCK_UN:
252 return F_UNLCK;
253 }
254 return -EINVAL;
255}
256
257/* Fill in a file_lock structure with an appropriate FLOCK lock. */
258static int flock_make_lock(struct file *filp, struct file_lock **lock,
259 unsigned int cmd)
260{
261 struct file_lock *fl;
262 int type = flock_translate_cmd(cmd);
263 if (type < 0)
264 return type;
265
266 fl = locks_alloc_lock();
267 if (fl == NULL)
268 return -ENOMEM;
269
270 fl->fl_file = filp;
271 fl->fl_pid = current->tgid;
272 fl->fl_flags = FL_FLOCK;
273 fl->fl_type = type;
274 fl->fl_end = OFFSET_MAX;
275
276 *lock = fl;
277 return 0;
278}
279
280static int assign_type(struct file_lock *fl, int type)
281{
282 switch (type) {
283 case F_RDLCK:
284 case F_WRLCK:
285 case F_UNLCK:
286 fl->fl_type = type;
287 break;
288 default:
289 return -EINVAL;
290 }
291 return 0;
292}
293
294/* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
295 * style lock.
296 */
297static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
298 struct flock *l)
299{
300 off_t start, end;
301
302 switch (l->l_whence) {
303 case 0: /*SEEK_SET*/
304 start = 0;
305 break;
306 case 1: /*SEEK_CUR*/
307 start = filp->f_pos;
308 break;
309 case 2: /*SEEK_END*/
310 start = i_size_read(filp->f_dentry->d_inode);
311 break;
312 default:
313 return -EINVAL;
314 }
315
316 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
317 POSIX-2001 defines it. */
318 start += l->l_start;
319 end = start + l->l_len - 1;
320 if (l->l_len < 0) {
321 end = start - 1;
322 start += l->l_len;
323 }
324
325 if (start < 0)
326 return -EINVAL;
327 if (l->l_len > 0 && end < 0)
328 return -EOVERFLOW;
329
330 fl->fl_start = start; /* we record the absolute position */
331 fl->fl_end = end;
332 if (l->l_len == 0)
333 fl->fl_end = OFFSET_MAX;
334
335 fl->fl_owner = current->files;
336 fl->fl_pid = current->tgid;
337 fl->fl_file = filp;
338 fl->fl_flags = FL_POSIX;
339 fl->fl_ops = NULL;
340 fl->fl_lmops = NULL;
341
342 return assign_type(fl, l->l_type);
343}
344
345#if BITS_PER_LONG == 32
346static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
347 struct flock64 *l)
348{
349 loff_t start;
350
351 switch (l->l_whence) {
352 case 0: /*SEEK_SET*/
353 start = 0;
354 break;
355 case 1: /*SEEK_CUR*/
356 start = filp->f_pos;
357 break;
358 case 2: /*SEEK_END*/
359 start = i_size_read(filp->f_dentry->d_inode);
360 break;
361 default:
362 return -EINVAL;
363 }
364
365 if (((start += l->l_start) < 0) || (l->l_len < 0))
366 return -EINVAL;
367 fl->fl_end = start + l->l_len - 1;
368 if (l->l_len > 0 && fl->fl_end < 0)
369 return -EOVERFLOW;
370 fl->fl_start = start; /* we record the absolute position */
371 if (l->l_len == 0)
372 fl->fl_end = OFFSET_MAX;
373
374 fl->fl_owner = current->files;
375 fl->fl_pid = current->tgid;
376 fl->fl_file = filp;
377 fl->fl_flags = FL_POSIX;
378 fl->fl_ops = NULL;
379 fl->fl_lmops = NULL;
380
381 switch (l->l_type) {
382 case F_RDLCK:
383 case F_WRLCK:
384 case F_UNLCK:
385 fl->fl_type = l->l_type;
386 break;
387 default:
388 return -EINVAL;
389 }
390
391 return (0);
392}
393#endif
394
395/* default lease lock manager operations */
396static void lease_break_callback(struct file_lock *fl)
397{
398 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
399}
400
401static void lease_release_private_callback(struct file_lock *fl)
402{
403 if (!fl->fl_file)
404 return;
405
406 f_delown(fl->fl_file);
407 fl->fl_file->f_owner.signum = 0;
408}
409
Adrian Bunk75c96f82005-05-05 16:16:09 -0700410static int lease_mylease_callback(struct file_lock *fl, struct file_lock *try)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411{
412 return fl->fl_file == try->fl_file;
413}
414
Adrian Bunk75c96f82005-05-05 16:16:09 -0700415static struct lock_manager_operations lease_manager_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 .fl_break = lease_break_callback,
417 .fl_release_private = lease_release_private_callback,
418 .fl_mylease = lease_mylease_callback,
419 .fl_change = lease_modify,
420};
421
422/*
423 * Initialize a lease, use the default lock manager operations
424 */
425static int lease_init(struct file *filp, int type, struct file_lock *fl)
426 {
427 fl->fl_owner = current->files;
428 fl->fl_pid = current->tgid;
429
430 fl->fl_file = filp;
431 fl->fl_flags = FL_LEASE;
432 if (assign_type(fl, type) != 0) {
433 locks_free_lock(fl);
434 return -EINVAL;
435 }
436 fl->fl_start = 0;
437 fl->fl_end = OFFSET_MAX;
438 fl->fl_ops = NULL;
439 fl->fl_lmops = &lease_manager_ops;
440 return 0;
441}
442
443/* Allocate a file_lock initialised to this type of lease */
444static int lease_alloc(struct file *filp, int type, struct file_lock **flp)
445{
446 struct file_lock *fl = locks_alloc_lock();
447 int error;
448
449 if (fl == NULL)
450 return -ENOMEM;
451
452 error = lease_init(filp, type, fl);
453 if (error)
454 return error;
455 *flp = fl;
456 return 0;
457}
458
459/* Check if two locks overlap each other.
460 */
461static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
462{
463 return ((fl1->fl_end >= fl2->fl_start) &&
464 (fl2->fl_end >= fl1->fl_start));
465}
466
467/*
468 * Check whether two locks have the same owner.
469 */
470static inline int
471posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
472{
473 if (fl1->fl_lmops && fl1->fl_lmops->fl_compare_owner)
474 return fl2->fl_lmops == fl1->fl_lmops &&
475 fl1->fl_lmops->fl_compare_owner(fl1, fl2);
476 return fl1->fl_owner == fl2->fl_owner;
477}
478
479/* Remove waiter from blocker's block list.
480 * When blocker ends up pointing to itself then the list is empty.
481 */
482static inline void __locks_delete_block(struct file_lock *waiter)
483{
484 list_del_init(&waiter->fl_block);
485 list_del_init(&waiter->fl_link);
486 waiter->fl_next = NULL;
487}
488
489/*
490 */
491static void locks_delete_block(struct file_lock *waiter)
492{
493 lock_kernel();
494 __locks_delete_block(waiter);
495 unlock_kernel();
496}
497
498/* Insert waiter into blocker's block list.
499 * We use a circular list so that processes can be easily woken up in
500 * the order they blocked. The documentation doesn't require this but
501 * it seems like the reasonable thing to do.
502 */
503static void locks_insert_block(struct file_lock *blocker,
504 struct file_lock *waiter)
505{
506 if (!list_empty(&waiter->fl_block)) {
507 printk(KERN_ERR "locks_insert_block: removing duplicated lock "
508 "(pid=%d %Ld-%Ld type=%d)\n", waiter->fl_pid,
509 waiter->fl_start, waiter->fl_end, waiter->fl_type);
510 __locks_delete_block(waiter);
511 }
512 list_add_tail(&waiter->fl_block, &blocker->fl_block);
513 waiter->fl_next = blocker;
514 if (IS_POSIX(blocker))
515 list_add(&waiter->fl_link, &blocked_list);
516}
517
518/* Wake up processes blocked waiting for blocker.
519 * If told to wait then schedule the processes until the block list
520 * is empty, otherwise empty the block list ourselves.
521 */
522static void locks_wake_up_blocks(struct file_lock *blocker)
523{
524 while (!list_empty(&blocker->fl_block)) {
525 struct file_lock *waiter = list_entry(blocker->fl_block.next,
526 struct file_lock, fl_block);
527 __locks_delete_block(waiter);
528 if (waiter->fl_lmops && waiter->fl_lmops->fl_notify)
529 waiter->fl_lmops->fl_notify(waiter);
530 else
531 wake_up(&waiter->fl_wait);
532 }
533}
534
535/* Insert file lock fl into an inode's lock list at the position indicated
536 * by pos. At the same time add the lock to the global file lock list.
537 */
538static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
539{
540 list_add(&fl->fl_link, &file_lock_list);
541
542 /* insert into file's list */
543 fl->fl_next = *pos;
544 *pos = fl;
545
546 if (fl->fl_ops && fl->fl_ops->fl_insert)
547 fl->fl_ops->fl_insert(fl);
548}
549
550/*
551 * Delete a lock and then free it.
552 * Wake up processes that are blocked waiting for this lock,
553 * notify the FS that the lock has been cleared and
554 * finally free the lock.
555 */
556static void locks_delete_lock(struct file_lock **thisfl_p)
557{
558 struct file_lock *fl = *thisfl_p;
559
560 *thisfl_p = fl->fl_next;
561 fl->fl_next = NULL;
562 list_del_init(&fl->fl_link);
563
564 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
565 if (fl->fl_fasync != NULL) {
566 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
567 fl->fl_fasync = NULL;
568 }
569
570 if (fl->fl_ops && fl->fl_ops->fl_remove)
571 fl->fl_ops->fl_remove(fl);
572
573 locks_wake_up_blocks(fl);
574 locks_free_lock(fl);
575}
576
577/* Determine if lock sys_fl blocks lock caller_fl. Common functionality
578 * checks for shared/exclusive status of overlapping locks.
579 */
580static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
581{
582 if (sys_fl->fl_type == F_WRLCK)
583 return 1;
584 if (caller_fl->fl_type == F_WRLCK)
585 return 1;
586 return 0;
587}
588
589/* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
590 * checking before calling the locks_conflict().
591 */
592static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
593{
594 /* POSIX locks owned by the same process do not conflict with
595 * each other.
596 */
597 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
598 return (0);
599
600 /* Check whether they overlap */
601 if (!locks_overlap(caller_fl, sys_fl))
602 return 0;
603
604 return (locks_conflict(caller_fl, sys_fl));
605}
606
607/* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
608 * checking before calling the locks_conflict().
609 */
610static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
611{
612 /* FLOCK locks referring to the same filp do not conflict with
613 * each other.
614 */
615 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
616 return (0);
617 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
618 return 0;
619
620 return (locks_conflict(caller_fl, sys_fl));
621}
622
623static int interruptible_sleep_on_locked(wait_queue_head_t *fl_wait, int timeout)
624{
625 int result = 0;
626 DECLARE_WAITQUEUE(wait, current);
627
628 __set_current_state(TASK_INTERRUPTIBLE);
629 add_wait_queue(fl_wait, &wait);
630 if (timeout == 0)
631 schedule();
632 else
633 result = schedule_timeout(timeout);
634 if (signal_pending(current))
635 result = -ERESTARTSYS;
636 remove_wait_queue(fl_wait, &wait);
637 __set_current_state(TASK_RUNNING);
638 return result;
639}
640
641static int locks_block_on_timeout(struct file_lock *blocker, struct file_lock *waiter, int time)
642{
643 int result;
644 locks_insert_block(blocker, waiter);
645 result = interruptible_sleep_on_locked(&waiter->fl_wait, time);
646 __locks_delete_block(waiter);
647 return result;
648}
649
650struct file_lock *
651posix_test_lock(struct file *filp, struct file_lock *fl)
652{
653 struct file_lock *cfl;
654
655 lock_kernel();
656 for (cfl = filp->f_dentry->d_inode->i_flock; cfl; cfl = cfl->fl_next) {
657 if (!IS_POSIX(cfl))
658 continue;
659 if (posix_locks_conflict(cfl, fl))
660 break;
661 }
662 unlock_kernel();
663
664 return (cfl);
665}
666
667EXPORT_SYMBOL(posix_test_lock);
668
669/* This function tests for deadlock condition before putting a process to
670 * sleep. The detection scheme is no longer recursive. Recursive was neat,
671 * but dangerous - we risked stack corruption if the lock data was bad, or
672 * if the recursion was too deep for any other reason.
673 *
674 * We rely on the fact that a task can only be on one lock's wait queue
675 * at a time. When we find blocked_task on a wait queue we can re-search
676 * with blocked_task equal to that queue's owner, until either blocked_task
677 * isn't found, or blocked_task is found on a queue owned by my_task.
678 *
679 * Note: the above assumption may not be true when handling lock requests
680 * from a broken NFS client. But broken NFS clients have a lot more to
681 * worry about than proper deadlock detection anyway... --okir
682 */
683int posix_locks_deadlock(struct file_lock *caller_fl,
684 struct file_lock *block_fl)
685{
686 struct list_head *tmp;
687
688next_task:
689 if (posix_same_owner(caller_fl, block_fl))
690 return 1;
691 list_for_each(tmp, &blocked_list) {
692 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
693 if (posix_same_owner(fl, block_fl)) {
694 fl = fl->fl_next;
695 block_fl = fl;
696 goto next_task;
697 }
698 }
699 return 0;
700}
701
702EXPORT_SYMBOL(posix_locks_deadlock);
703
704/* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
705 * at the head of the list, but that's secret knowledge known only to
706 * flock_lock_file and posix_lock_file.
707 */
708static int flock_lock_file(struct file *filp, struct file_lock *new_fl)
709{
710 struct file_lock **before;
711 struct inode * inode = filp->f_dentry->d_inode;
712 int error = 0;
713 int found = 0;
714
715 lock_kernel();
716 for_each_lock(inode, before) {
717 struct file_lock *fl = *before;
718 if (IS_POSIX(fl))
719 break;
720 if (IS_LEASE(fl))
721 continue;
722 if (filp != fl->fl_file)
723 continue;
724 if (new_fl->fl_type == fl->fl_type)
725 goto out;
726 found = 1;
727 locks_delete_lock(before);
728 break;
729 }
730 unlock_kernel();
731
732 if (new_fl->fl_type == F_UNLCK)
733 return 0;
734
735 /*
736 * If a higher-priority process was blocked on the old file lock,
737 * give it the opportunity to lock the file.
738 */
739 if (found)
740 cond_resched();
741
742 lock_kernel();
743 for_each_lock(inode, before) {
744 struct file_lock *fl = *before;
745 if (IS_POSIX(fl))
746 break;
747 if (IS_LEASE(fl))
748 continue;
749 if (!flock_locks_conflict(new_fl, fl))
750 continue;
751 error = -EAGAIN;
752 if (new_fl->fl_flags & FL_SLEEP) {
753 locks_insert_block(fl, new_fl);
754 }
755 goto out;
756 }
757 locks_insert_lock(&inode->i_flock, new_fl);
758 error = 0;
759
760out:
761 unlock_kernel();
762 return error;
763}
764
765EXPORT_SYMBOL(posix_lock_file);
766
767static int __posix_lock_file(struct inode *inode, struct file_lock *request)
768{
769 struct file_lock *fl;
770 struct file_lock *new_fl, *new_fl2;
771 struct file_lock *left = NULL;
772 struct file_lock *right = NULL;
773 struct file_lock **before;
774 int error, added = 0;
775
776 /*
777 * We may need two file_lock structures for this operation,
778 * so we get them in advance to avoid races.
779 */
780 new_fl = locks_alloc_lock();
781 new_fl2 = locks_alloc_lock();
782
783 lock_kernel();
784 if (request->fl_type != F_UNLCK) {
785 for_each_lock(inode, before) {
786 struct file_lock *fl = *before;
787 if (!IS_POSIX(fl))
788 continue;
789 if (!posix_locks_conflict(request, fl))
790 continue;
791 error = -EAGAIN;
792 if (!(request->fl_flags & FL_SLEEP))
793 goto out;
794 error = -EDEADLK;
795 if (posix_locks_deadlock(request, fl))
796 goto out;
797 error = -EAGAIN;
798 locks_insert_block(fl, request);
799 goto out;
800 }
801 }
802
803 /* If we're just looking for a conflict, we're done. */
804 error = 0;
805 if (request->fl_flags & FL_ACCESS)
806 goto out;
807
808 error = -ENOLCK; /* "no luck" */
809 if (!(new_fl && new_fl2))
810 goto out;
811
812 /*
813 * We've allocated the new locks in advance, so there are no
814 * errors possible (and no blocking operations) from here on.
815 *
816 * Find the first old lock with the same owner as the new lock.
817 */
818
819 before = &inode->i_flock;
820
821 /* First skip locks owned by other processes. */
822 while ((fl = *before) && (!IS_POSIX(fl) ||
823 !posix_same_owner(request, fl))) {
824 before = &fl->fl_next;
825 }
826
827 /* Process locks with this owner. */
828 while ((fl = *before) && posix_same_owner(request, fl)) {
829 /* Detect adjacent or overlapping regions (if same lock type)
830 */
831 if (request->fl_type == fl->fl_type) {
832 if (fl->fl_end < request->fl_start - 1)
833 goto next_lock;
834 /* If the next lock in the list has entirely bigger
835 * addresses than the new one, insert the lock here.
836 */
837 if (fl->fl_start > request->fl_end + 1)
838 break;
839
840 /* If we come here, the new and old lock are of the
841 * same type and adjacent or overlapping. Make one
842 * lock yielding from the lower start address of both
843 * locks to the higher end address.
844 */
845 if (fl->fl_start > request->fl_start)
846 fl->fl_start = request->fl_start;
847 else
848 request->fl_start = fl->fl_start;
849 if (fl->fl_end < request->fl_end)
850 fl->fl_end = request->fl_end;
851 else
852 request->fl_end = fl->fl_end;
853 if (added) {
854 locks_delete_lock(before);
855 continue;
856 }
857 request = fl;
858 added = 1;
859 }
860 else {
861 /* Processing for different lock types is a bit
862 * more complex.
863 */
864 if (fl->fl_end < request->fl_start)
865 goto next_lock;
866 if (fl->fl_start > request->fl_end)
867 break;
868 if (request->fl_type == F_UNLCK)
869 added = 1;
870 if (fl->fl_start < request->fl_start)
871 left = fl;
872 /* If the next lock in the list has a higher end
873 * address than the new one, insert the new one here.
874 */
875 if (fl->fl_end > request->fl_end) {
876 right = fl;
877 break;
878 }
879 if (fl->fl_start >= request->fl_start) {
880 /* The new lock completely replaces an old
881 * one (This may happen several times).
882 */
883 if (added) {
884 locks_delete_lock(before);
885 continue;
886 }
887 /* Replace the old lock with the new one.
888 * Wake up anybody waiting for the old one,
889 * as the change in lock type might satisfy
890 * their needs.
891 */
892 locks_wake_up_blocks(fl);
893 fl->fl_start = request->fl_start;
894 fl->fl_end = request->fl_end;
895 fl->fl_type = request->fl_type;
896 fl->fl_u = request->fl_u;
897 request = fl;
898 added = 1;
899 }
900 }
901 /* Go on to next lock.
902 */
903 next_lock:
904 before = &fl->fl_next;
905 }
906
907 error = 0;
908 if (!added) {
909 if (request->fl_type == F_UNLCK)
910 goto out;
911 locks_copy_lock(new_fl, request);
912 locks_insert_lock(before, new_fl);
913 new_fl = NULL;
914 }
915 if (right) {
916 if (left == right) {
917 /* The new lock breaks the old one in two pieces,
918 * so we have to use the second new lock.
919 */
920 left = new_fl2;
921 new_fl2 = NULL;
922 locks_copy_lock(left, right);
923 locks_insert_lock(before, left);
924 }
925 right->fl_start = request->fl_end + 1;
926 locks_wake_up_blocks(right);
927 }
928 if (left) {
929 left->fl_end = request->fl_start - 1;
930 locks_wake_up_blocks(left);
931 }
932 out:
933 unlock_kernel();
934 /*
935 * Free any unused locks.
936 */
937 if (new_fl)
938 locks_free_lock(new_fl);
939 if (new_fl2)
940 locks_free_lock(new_fl2);
941 return error;
942}
943
944/**
945 * posix_lock_file - Apply a POSIX-style lock to a file
946 * @filp: The file to apply the lock to
947 * @fl: The lock to be applied
948 *
949 * Add a POSIX style lock to a file.
950 * We merge adjacent & overlapping locks whenever possible.
951 * POSIX locks are sorted by owner task, then by starting address
952 */
953int posix_lock_file(struct file *filp, struct file_lock *fl)
954{
955 return __posix_lock_file(filp->f_dentry->d_inode, fl);
956}
957
958/**
959 * posix_lock_file_wait - Apply a POSIX-style lock to a file
960 * @filp: The file to apply the lock to
961 * @fl: The lock to be applied
962 *
963 * Add a POSIX style lock to a file.
964 * We merge adjacent & overlapping locks whenever possible.
965 * POSIX locks are sorted by owner task, then by starting address
966 */
967int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
968{
969 int error;
970 might_sleep ();
971 for (;;) {
972 error = __posix_lock_file(filp->f_dentry->d_inode, fl);
973 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
974 break;
975 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
976 if (!error)
977 continue;
978
979 locks_delete_block(fl);
980 break;
981 }
982 return error;
983}
984EXPORT_SYMBOL(posix_lock_file_wait);
985
986/**
987 * locks_mandatory_locked - Check for an active lock
988 * @inode: the file to check
989 *
990 * Searches the inode's list of locks to find any POSIX locks which conflict.
991 * This function is called from locks_verify_locked() only.
992 */
993int locks_mandatory_locked(struct inode *inode)
994{
995 fl_owner_t owner = current->files;
996 struct file_lock *fl;
997
998 /*
999 * Search the lock list for this inode for any POSIX locks.
1000 */
1001 lock_kernel();
1002 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1003 if (!IS_POSIX(fl))
1004 continue;
1005 if (fl->fl_owner != owner)
1006 break;
1007 }
1008 unlock_kernel();
1009 return fl ? -EAGAIN : 0;
1010}
1011
1012/**
1013 * locks_mandatory_area - Check for a conflicting lock
1014 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1015 * for shared
1016 * @inode: the file to check
1017 * @filp: how the file was opened (if it was)
1018 * @offset: start of area to check
1019 * @count: length of area to check
1020 *
1021 * Searches the inode's list of locks to find any POSIX locks which conflict.
1022 * This function is called from rw_verify_area() and
1023 * locks_verify_truncate().
1024 */
1025int locks_mandatory_area(int read_write, struct inode *inode,
1026 struct file *filp, loff_t offset,
1027 size_t count)
1028{
1029 struct file_lock fl;
1030 int error;
1031
1032 locks_init_lock(&fl);
1033 fl.fl_owner = current->files;
1034 fl.fl_pid = current->tgid;
1035 fl.fl_file = filp;
1036 fl.fl_flags = FL_POSIX | FL_ACCESS;
1037 if (filp && !(filp->f_flags & O_NONBLOCK))
1038 fl.fl_flags |= FL_SLEEP;
1039 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1040 fl.fl_start = offset;
1041 fl.fl_end = offset + count - 1;
1042
1043 for (;;) {
1044 error = __posix_lock_file(inode, &fl);
1045 if (error != -EAGAIN)
1046 break;
1047 if (!(fl.fl_flags & FL_SLEEP))
1048 break;
1049 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1050 if (!error) {
1051 /*
1052 * If we've been sleeping someone might have
1053 * changed the permissions behind our back.
1054 */
1055 if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID)
1056 continue;
1057 }
1058
1059 locks_delete_block(&fl);
1060 break;
1061 }
1062
1063 return error;
1064}
1065
1066EXPORT_SYMBOL(locks_mandatory_area);
1067
1068/* We already had a lease on this file; just change its type */
1069int lease_modify(struct file_lock **before, int arg)
1070{
1071 struct file_lock *fl = *before;
1072 int error = assign_type(fl, arg);
1073
1074 if (error)
1075 return error;
1076 locks_wake_up_blocks(fl);
1077 if (arg == F_UNLCK)
1078 locks_delete_lock(before);
1079 return 0;
1080}
1081
1082EXPORT_SYMBOL(lease_modify);
1083
1084static void time_out_leases(struct inode *inode)
1085{
1086 struct file_lock **before;
1087 struct file_lock *fl;
1088
1089 before = &inode->i_flock;
1090 while ((fl = *before) && IS_LEASE(fl) && (fl->fl_type & F_INPROGRESS)) {
1091 if ((fl->fl_break_time == 0)
1092 || time_before(jiffies, fl->fl_break_time)) {
1093 before = &fl->fl_next;
1094 continue;
1095 }
1096 printk(KERN_INFO "lease broken - owner pid = %d\n", fl->fl_pid);
1097 lease_modify(before, fl->fl_type & ~F_INPROGRESS);
1098 if (fl == *before) /* lease_modify may have freed fl */
1099 before = &fl->fl_next;
1100 }
1101}
1102
1103/**
1104 * __break_lease - revoke all outstanding leases on file
1105 * @inode: the inode of the file to return
1106 * @mode: the open mode (read or write)
1107 *
1108 * break_lease (inlined for speed) has checked there already
1109 * is a lease on this file. Leases are broken on a call to open()
1110 * or truncate(). This function can sleep unless you
1111 * specified %O_NONBLOCK to your open().
1112 */
1113int __break_lease(struct inode *inode, unsigned int mode)
1114{
1115 int error = 0, future;
1116 struct file_lock *new_fl, *flock;
1117 struct file_lock *fl;
1118 int alloc_err;
1119 unsigned long break_time;
1120 int i_have_this_lease = 0;
1121
1122 alloc_err = lease_alloc(NULL, mode & FMODE_WRITE ? F_WRLCK : F_RDLCK,
1123 &new_fl);
1124
1125 lock_kernel();
1126
1127 time_out_leases(inode);
1128
1129 flock = inode->i_flock;
1130 if ((flock == NULL) || !IS_LEASE(flock))
1131 goto out;
1132
1133 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1134 if (fl->fl_owner == current->files)
1135 i_have_this_lease = 1;
1136
1137 if (mode & FMODE_WRITE) {
1138 /* If we want write access, we have to revoke any lease. */
1139 future = F_UNLCK | F_INPROGRESS;
1140 } else if (flock->fl_type & F_INPROGRESS) {
1141 /* If the lease is already being broken, we just leave it */
1142 future = flock->fl_type;
1143 } else if (flock->fl_type & F_WRLCK) {
1144 /* Downgrade the exclusive lease to a read-only lease. */
1145 future = F_RDLCK | F_INPROGRESS;
1146 } else {
1147 /* the existing lease was read-only, so we can read too. */
1148 goto out;
1149 }
1150
1151 if (alloc_err && !i_have_this_lease && ((mode & O_NONBLOCK) == 0)) {
1152 error = alloc_err;
1153 goto out;
1154 }
1155
1156 break_time = 0;
1157 if (lease_break_time > 0) {
1158 break_time = jiffies + lease_break_time * HZ;
1159 if (break_time == 0)
1160 break_time++; /* so that 0 means no break time */
1161 }
1162
1163 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1164 if (fl->fl_type != future) {
1165 fl->fl_type = future;
1166 fl->fl_break_time = break_time;
1167 /* lease must have lmops break callback */
1168 fl->fl_lmops->fl_break(fl);
1169 }
1170 }
1171
1172 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1173 error = -EWOULDBLOCK;
1174 goto out;
1175 }
1176
1177restart:
1178 break_time = flock->fl_break_time;
1179 if (break_time != 0) {
1180 break_time -= jiffies;
1181 if (break_time == 0)
1182 break_time++;
1183 }
1184 error = locks_block_on_timeout(flock, new_fl, break_time);
1185 if (error >= 0) {
1186 if (error == 0)
1187 time_out_leases(inode);
1188 /* Wait for the next lease that has not been broken yet */
1189 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1190 flock = flock->fl_next) {
1191 if (flock->fl_type & F_INPROGRESS)
1192 goto restart;
1193 }
1194 error = 0;
1195 }
1196
1197out:
1198 unlock_kernel();
1199 if (!alloc_err)
1200 locks_free_lock(new_fl);
1201 return error;
1202}
1203
1204EXPORT_SYMBOL(__break_lease);
1205
1206/**
1207 * lease_get_mtime
1208 * @inode: the inode
1209 * @time: pointer to a timespec which will contain the last modified time
1210 *
1211 * This is to force NFS clients to flush their caches for files with
1212 * exclusive leases. The justification is that if someone has an
1213 * exclusive lease, then they could be modifiying it.
1214 */
1215void lease_get_mtime(struct inode *inode, struct timespec *time)
1216{
1217 struct file_lock *flock = inode->i_flock;
1218 if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK))
1219 *time = current_fs_time(inode->i_sb);
1220 else
1221 *time = inode->i_mtime;
1222}
1223
1224EXPORT_SYMBOL(lease_get_mtime);
1225
1226/**
1227 * fcntl_getlease - Enquire what lease is currently active
1228 * @filp: the file
1229 *
1230 * The value returned by this function will be one of
1231 * (if no lease break is pending):
1232 *
1233 * %F_RDLCK to indicate a shared lease is held.
1234 *
1235 * %F_WRLCK to indicate an exclusive lease is held.
1236 *
1237 * %F_UNLCK to indicate no lease is held.
1238 *
1239 * (if a lease break is pending):
1240 *
1241 * %F_RDLCK to indicate an exclusive lease needs to be
1242 * changed to a shared lease (or removed).
1243 *
1244 * %F_UNLCK to indicate the lease needs to be removed.
1245 *
1246 * XXX: sfr & willy disagree over whether F_INPROGRESS
1247 * should be returned to userspace.
1248 */
1249int fcntl_getlease(struct file *filp)
1250{
1251 struct file_lock *fl;
1252 int type = F_UNLCK;
1253
1254 lock_kernel();
1255 time_out_leases(filp->f_dentry->d_inode);
1256 for (fl = filp->f_dentry->d_inode->i_flock; fl && IS_LEASE(fl);
1257 fl = fl->fl_next) {
1258 if (fl->fl_file == filp) {
1259 type = fl->fl_type & ~F_INPROGRESS;
1260 break;
1261 }
1262 }
1263 unlock_kernel();
1264 return type;
1265}
1266
1267/**
1268 * __setlease - sets a lease on an open file
1269 * @filp: file pointer
1270 * @arg: type of lease to obtain
1271 * @flp: input - file_lock to use, output - file_lock inserted
1272 *
1273 * The (input) flp->fl_lmops->fl_break function is required
1274 * by break_lease().
1275 *
1276 * Called with kernel lock held.
1277 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07001278static int __setlease(struct file *filp, long arg, struct file_lock **flp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279{
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001280 struct file_lock *fl, **before, **my_before = NULL, *lease;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281 struct dentry *dentry = filp->f_dentry;
1282 struct inode *inode = dentry->d_inode;
1283 int error, rdlease_count = 0, wrlease_count = 0;
1284
1285 time_out_leases(inode);
1286
1287 error = -EINVAL;
1288 if (!flp || !(*flp) || !(*flp)->fl_lmops || !(*flp)->fl_lmops->fl_break)
1289 goto out;
1290
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001291 lease = *flp;
1292
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 error = -EAGAIN;
1294 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1295 goto out;
1296 if ((arg == F_WRLCK)
1297 && ((atomic_read(&dentry->d_count) > 1)
1298 || (atomic_read(&inode->i_count) > 1)))
1299 goto out;
1300
1301 /*
1302 * At this point, we know that if there is an exclusive
1303 * lease on this file, then we hold it on this filp
1304 * (otherwise our open of this file would have blocked).
1305 * And if we are trying to acquire an exclusive lease,
1306 * then the file is not open by anyone (including us)
1307 * except for this filp.
1308 */
1309 for (before = &inode->i_flock;
1310 ((fl = *before) != NULL) && IS_LEASE(fl);
1311 before = &fl->fl_next) {
1312 if (lease->fl_lmops->fl_mylease(fl, lease))
1313 my_before = before;
1314 else if (fl->fl_type == (F_INPROGRESS | F_UNLCK))
1315 /*
1316 * Someone is in the process of opening this
1317 * file for writing so we may not take an
1318 * exclusive lease on it.
1319 */
1320 wrlease_count++;
1321 else
1322 rdlease_count++;
1323 }
1324
1325 if ((arg == F_RDLCK && (wrlease_count > 0)) ||
1326 (arg == F_WRLCK && ((rdlease_count + wrlease_count) > 0)))
1327 goto out;
1328
1329 if (my_before != NULL) {
1330 error = lease->fl_lmops->fl_change(my_before, arg);
1331 goto out;
1332 }
1333
1334 error = 0;
1335 if (arg == F_UNLCK)
1336 goto out;
1337
1338 error = -EINVAL;
1339 if (!leases_enable)
1340 goto out;
1341
1342 error = lease_alloc(filp, arg, &fl);
1343 if (error)
1344 goto out;
1345
1346 locks_copy_lock(fl, lease);
1347
1348 locks_insert_lock(before, fl);
1349
1350 *flp = fl;
1351out:
1352 return error;
1353}
1354
1355 /**
1356 * setlease - sets a lease on an open file
1357 * @filp: file pointer
1358 * @arg: type of lease to obtain
1359 * @lease: file_lock to use
1360 *
1361 * Call this to establish a lease on the file.
1362 * The fl_lmops fl_break function is required by break_lease
1363 */
1364
1365int setlease(struct file *filp, long arg, struct file_lock **lease)
1366{
1367 struct dentry *dentry = filp->f_dentry;
1368 struct inode *inode = dentry->d_inode;
1369 int error;
1370
1371 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1372 return -EACCES;
1373 if (!S_ISREG(inode->i_mode))
1374 return -EINVAL;
1375 error = security_file_lock(filp, arg);
1376 if (error)
1377 return error;
1378
1379 lock_kernel();
1380 error = __setlease(filp, arg, lease);
1381 unlock_kernel();
1382
1383 return error;
1384}
1385
1386EXPORT_SYMBOL(setlease);
1387
1388/**
1389 * fcntl_setlease - sets a lease on an open file
1390 * @fd: open file descriptor
1391 * @filp: file pointer
1392 * @arg: type of lease to obtain
1393 *
1394 * Call this fcntl to establish a lease on the file.
1395 * Note that you also need to call %F_SETSIG to
1396 * receive a signal when the lease is broken.
1397 */
1398int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1399{
1400 struct file_lock fl, *flp = &fl;
1401 struct dentry *dentry = filp->f_dentry;
1402 struct inode *inode = dentry->d_inode;
1403 int error;
1404
1405 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1406 return -EACCES;
1407 if (!S_ISREG(inode->i_mode))
1408 return -EINVAL;
1409 error = security_file_lock(filp, arg);
1410 if (error)
1411 return error;
1412
1413 locks_init_lock(&fl);
1414 error = lease_init(filp, arg, &fl);
1415 if (error)
1416 return error;
1417
1418 lock_kernel();
1419
1420 error = __setlease(filp, arg, &flp);
1421 if (error)
1422 goto out_unlock;
1423
1424 error = fasync_helper(fd, filp, 1, &flp->fl_fasync);
1425 if (error < 0) {
1426 /* remove lease just inserted by __setlease */
1427 flp->fl_type = F_UNLCK | F_INPROGRESS;
1428 flp->fl_break_time = jiffies- 10;
1429 time_out_leases(inode);
1430 goto out_unlock;
1431 }
1432
1433 error = f_setown(filp, current->pid, 0);
1434out_unlock:
1435 unlock_kernel();
1436 return error;
1437}
1438
1439/**
1440 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1441 * @filp: The file to apply the lock to
1442 * @fl: The lock to be applied
1443 *
1444 * Add a FLOCK style lock to a file.
1445 */
1446int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1447{
1448 int error;
1449 might_sleep();
1450 for (;;) {
1451 error = flock_lock_file(filp, fl);
1452 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1453 break;
1454 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1455 if (!error)
1456 continue;
1457
1458 locks_delete_block(fl);
1459 break;
1460 }
1461 return error;
1462}
1463
1464EXPORT_SYMBOL(flock_lock_file_wait);
1465
1466/**
1467 * sys_flock: - flock() system call.
1468 * @fd: the file descriptor to lock.
1469 * @cmd: the type of lock to apply.
1470 *
1471 * Apply a %FL_FLOCK style lock to an open file descriptor.
1472 * The @cmd can be one of
1473 *
1474 * %LOCK_SH -- a shared lock.
1475 *
1476 * %LOCK_EX -- an exclusive lock.
1477 *
1478 * %LOCK_UN -- remove an existing lock.
1479 *
1480 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1481 *
1482 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1483 * processes read and write access respectively.
1484 */
1485asmlinkage long sys_flock(unsigned int fd, unsigned int cmd)
1486{
1487 struct file *filp;
1488 struct file_lock *lock;
1489 int can_sleep, unlock;
1490 int error;
1491
1492 error = -EBADF;
1493 filp = fget(fd);
1494 if (!filp)
1495 goto out;
1496
1497 can_sleep = !(cmd & LOCK_NB);
1498 cmd &= ~LOCK_NB;
1499 unlock = (cmd == LOCK_UN);
1500
1501 if (!unlock && !(cmd & LOCK_MAND) && !(filp->f_mode & 3))
1502 goto out_putf;
1503
1504 error = flock_make_lock(filp, &lock, cmd);
1505 if (error)
1506 goto out_putf;
1507 if (can_sleep)
1508 lock->fl_flags |= FL_SLEEP;
1509
1510 error = security_file_lock(filp, cmd);
1511 if (error)
1512 goto out_free;
1513
1514 if (filp->f_op && filp->f_op->flock)
1515 error = filp->f_op->flock(filp,
1516 (can_sleep) ? F_SETLKW : F_SETLK,
1517 lock);
1518 else
1519 error = flock_lock_file_wait(filp, lock);
1520
1521 out_free:
1522 if (list_empty(&lock->fl_link)) {
1523 locks_free_lock(lock);
1524 }
1525
1526 out_putf:
1527 fput(filp);
1528 out:
1529 return error;
1530}
1531
1532/* Report the first existing lock that would conflict with l.
1533 * This implements the F_GETLK command of fcntl().
1534 */
1535int fcntl_getlk(struct file *filp, struct flock __user *l)
1536{
1537 struct file_lock *fl, file_lock;
1538 struct flock flock;
1539 int error;
1540
1541 error = -EFAULT;
1542 if (copy_from_user(&flock, l, sizeof(flock)))
1543 goto out;
1544 error = -EINVAL;
1545 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1546 goto out;
1547
1548 error = flock_to_posix_lock(filp, &file_lock, &flock);
1549 if (error)
1550 goto out;
1551
1552 if (filp->f_op && filp->f_op->lock) {
1553 error = filp->f_op->lock(filp, F_GETLK, &file_lock);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00001554 if (file_lock.fl_ops && file_lock.fl_ops->fl_release_private)
1555 file_lock.fl_ops->fl_release_private(&file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556 if (error < 0)
1557 goto out;
1558 else
1559 fl = (file_lock.fl_type == F_UNLCK ? NULL : &file_lock);
1560 } else {
1561 fl = posix_test_lock(filp, &file_lock);
1562 }
1563
1564 flock.l_type = F_UNLCK;
1565 if (fl != NULL) {
1566 flock.l_pid = fl->fl_pid;
1567#if BITS_PER_LONG == 32
1568 /*
1569 * Make sure we can represent the posix lock via
1570 * legacy 32bit flock.
1571 */
1572 error = -EOVERFLOW;
1573 if (fl->fl_start > OFFT_OFFSET_MAX)
1574 goto out;
1575 if ((fl->fl_end != OFFSET_MAX)
1576 && (fl->fl_end > OFFT_OFFSET_MAX))
1577 goto out;
1578#endif
1579 flock.l_start = fl->fl_start;
1580 flock.l_len = fl->fl_end == OFFSET_MAX ? 0 :
1581 fl->fl_end - fl->fl_start + 1;
1582 flock.l_whence = 0;
1583 flock.l_type = fl->fl_type;
1584 }
1585 error = -EFAULT;
1586 if (!copy_to_user(l, &flock, sizeof(flock)))
1587 error = 0;
1588out:
1589 return error;
1590}
1591
1592/* Apply the lock described by l to an open file descriptor.
1593 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1594 */
Peter Staubachc2936212005-07-27 11:45:09 -07001595int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1596 struct flock __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597{
1598 struct file_lock *file_lock = locks_alloc_lock();
1599 struct flock flock;
1600 struct inode *inode;
1601 int error;
1602
1603 if (file_lock == NULL)
1604 return -ENOLCK;
1605
1606 /*
1607 * This might block, so we do it before checking the inode.
1608 */
1609 error = -EFAULT;
1610 if (copy_from_user(&flock, l, sizeof(flock)))
1611 goto out;
1612
1613 inode = filp->f_dentry->d_inode;
1614
1615 /* Don't allow mandatory locks on files that may be memory mapped
1616 * and shared.
1617 */
1618 if (IS_MANDLOCK(inode) &&
1619 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1620 mapping_writably_mapped(filp->f_mapping)) {
1621 error = -EAGAIN;
1622 goto out;
1623 }
1624
Peter Staubachc2936212005-07-27 11:45:09 -07001625again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626 error = flock_to_posix_lock(filp, file_lock, &flock);
1627 if (error)
1628 goto out;
1629 if (cmd == F_SETLKW) {
1630 file_lock->fl_flags |= FL_SLEEP;
1631 }
1632
1633 error = -EBADF;
1634 switch (flock.l_type) {
1635 case F_RDLCK:
1636 if (!(filp->f_mode & FMODE_READ))
1637 goto out;
1638 break;
1639 case F_WRLCK:
1640 if (!(filp->f_mode & FMODE_WRITE))
1641 goto out;
1642 break;
1643 case F_UNLCK:
1644 break;
1645 default:
1646 error = -EINVAL;
1647 goto out;
1648 }
1649
1650 error = security_file_lock(filp, file_lock->fl_type);
1651 if (error)
1652 goto out;
1653
Peter Staubachc2936212005-07-27 11:45:09 -07001654 if (filp->f_op && filp->f_op->lock != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655 error = filp->f_op->lock(filp, cmd, file_lock);
Peter Staubachc2936212005-07-27 11:45:09 -07001656 else {
1657 for (;;) {
1658 error = __posix_lock_file(inode, file_lock);
1659 if ((error != -EAGAIN) || (cmd == F_SETLK))
1660 break;
1661 error = wait_event_interruptible(file_lock->fl_wait,
1662 !file_lock->fl_next);
1663 if (!error)
1664 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665
Peter Staubachc2936212005-07-27 11:45:09 -07001666 locks_delete_block(file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 break;
Peter Staubachc2936212005-07-27 11:45:09 -07001668 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669 }
1670
Peter Staubachc2936212005-07-27 11:45:09 -07001671 /*
1672 * Attempt to detect a close/fcntl race and recover by
1673 * releasing the lock that was just acquired.
1674 */
1675 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1676 flock.l_type = F_UNLCK;
1677 goto again;
1678 }
1679
1680out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681 locks_free_lock(file_lock);
1682 return error;
1683}
1684
1685#if BITS_PER_LONG == 32
1686/* Report the first existing lock that would conflict with l.
1687 * This implements the F_GETLK command of fcntl().
1688 */
1689int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1690{
1691 struct file_lock *fl, file_lock;
1692 struct flock64 flock;
1693 int error;
1694
1695 error = -EFAULT;
1696 if (copy_from_user(&flock, l, sizeof(flock)))
1697 goto out;
1698 error = -EINVAL;
1699 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1700 goto out;
1701
1702 error = flock64_to_posix_lock(filp, &file_lock, &flock);
1703 if (error)
1704 goto out;
1705
1706 if (filp->f_op && filp->f_op->lock) {
1707 error = filp->f_op->lock(filp, F_GETLK, &file_lock);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00001708 if (file_lock.fl_ops && file_lock.fl_ops->fl_release_private)
1709 file_lock.fl_ops->fl_release_private(&file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 if (error < 0)
1711 goto out;
1712 else
1713 fl = (file_lock.fl_type == F_UNLCK ? NULL : &file_lock);
1714 } else {
1715 fl = posix_test_lock(filp, &file_lock);
1716 }
1717
1718 flock.l_type = F_UNLCK;
1719 if (fl != NULL) {
1720 flock.l_pid = fl->fl_pid;
1721 flock.l_start = fl->fl_start;
1722 flock.l_len = fl->fl_end == OFFSET_MAX ? 0 :
1723 fl->fl_end - fl->fl_start + 1;
1724 flock.l_whence = 0;
1725 flock.l_type = fl->fl_type;
1726 }
1727 error = -EFAULT;
1728 if (!copy_to_user(l, &flock, sizeof(flock)))
1729 error = 0;
1730
1731out:
1732 return error;
1733}
1734
1735/* Apply the lock described by l to an open file descriptor.
1736 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1737 */
Peter Staubachc2936212005-07-27 11:45:09 -07001738int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
1739 struct flock64 __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740{
1741 struct file_lock *file_lock = locks_alloc_lock();
1742 struct flock64 flock;
1743 struct inode *inode;
1744 int error;
1745
1746 if (file_lock == NULL)
1747 return -ENOLCK;
1748
1749 /*
1750 * This might block, so we do it before checking the inode.
1751 */
1752 error = -EFAULT;
1753 if (copy_from_user(&flock, l, sizeof(flock)))
1754 goto out;
1755
1756 inode = filp->f_dentry->d_inode;
1757
1758 /* Don't allow mandatory locks on files that may be memory mapped
1759 * and shared.
1760 */
1761 if (IS_MANDLOCK(inode) &&
1762 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1763 mapping_writably_mapped(filp->f_mapping)) {
1764 error = -EAGAIN;
1765 goto out;
1766 }
1767
Peter Staubachc2936212005-07-27 11:45:09 -07001768again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769 error = flock64_to_posix_lock(filp, file_lock, &flock);
1770 if (error)
1771 goto out;
1772 if (cmd == F_SETLKW64) {
1773 file_lock->fl_flags |= FL_SLEEP;
1774 }
1775
1776 error = -EBADF;
1777 switch (flock.l_type) {
1778 case F_RDLCK:
1779 if (!(filp->f_mode & FMODE_READ))
1780 goto out;
1781 break;
1782 case F_WRLCK:
1783 if (!(filp->f_mode & FMODE_WRITE))
1784 goto out;
1785 break;
1786 case F_UNLCK:
1787 break;
1788 default:
1789 error = -EINVAL;
1790 goto out;
1791 }
1792
1793 error = security_file_lock(filp, file_lock->fl_type);
1794 if (error)
1795 goto out;
1796
Peter Staubachc2936212005-07-27 11:45:09 -07001797 if (filp->f_op && filp->f_op->lock != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798 error = filp->f_op->lock(filp, cmd, file_lock);
Peter Staubachc2936212005-07-27 11:45:09 -07001799 else {
1800 for (;;) {
1801 error = __posix_lock_file(inode, file_lock);
1802 if ((error != -EAGAIN) || (cmd == F_SETLK64))
1803 break;
1804 error = wait_event_interruptible(file_lock->fl_wait,
1805 !file_lock->fl_next);
1806 if (!error)
1807 continue;
1808
1809 locks_delete_block(file_lock);
1810 break;
1811 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812 }
1813
Peter Staubachc2936212005-07-27 11:45:09 -07001814 /*
1815 * Attempt to detect a close/fcntl race and recover by
1816 * releasing the lock that was just acquired.
1817 */
1818 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1819 flock.l_type = F_UNLCK;
1820 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821 }
1822
1823out:
1824 locks_free_lock(file_lock);
1825 return error;
1826}
1827#endif /* BITS_PER_LONG == 32 */
1828
1829/*
1830 * This function is called when the file is being removed
1831 * from the task's fd array. POSIX locks belonging to this task
1832 * are deleted at this time.
1833 */
1834void locks_remove_posix(struct file *filp, fl_owner_t owner)
1835{
1836 struct file_lock lock, **before;
1837
1838 /*
1839 * If there are no locks held on this file, we don't need to call
1840 * posix_lock_file(). Another process could be setting a lock on this
1841 * file at the same time, but we wouldn't remove that lock anyway.
1842 */
1843 before = &filp->f_dentry->d_inode->i_flock;
1844 if (*before == NULL)
1845 return;
1846
1847 lock.fl_type = F_UNLCK;
1848 lock.fl_flags = FL_POSIX;
1849 lock.fl_start = 0;
1850 lock.fl_end = OFFSET_MAX;
1851 lock.fl_owner = owner;
1852 lock.fl_pid = current->tgid;
1853 lock.fl_file = filp;
1854 lock.fl_ops = NULL;
1855 lock.fl_lmops = NULL;
1856
1857 if (filp->f_op && filp->f_op->lock != NULL) {
1858 filp->f_op->lock(filp, F_SETLK, &lock);
1859 goto out;
1860 }
1861
1862 /* Can't use posix_lock_file here; we need to remove it no matter
1863 * which pid we have.
1864 */
1865 lock_kernel();
1866 while (*before != NULL) {
1867 struct file_lock *fl = *before;
1868 if (IS_POSIX(fl) && posix_same_owner(fl, &lock)) {
1869 locks_delete_lock(before);
1870 continue;
1871 }
1872 before = &fl->fl_next;
1873 }
1874 unlock_kernel();
1875out:
1876 if (lock.fl_ops && lock.fl_ops->fl_release_private)
1877 lock.fl_ops->fl_release_private(&lock);
1878}
1879
1880EXPORT_SYMBOL(locks_remove_posix);
1881
1882/*
1883 * This function is called on the last close of an open file.
1884 */
1885void locks_remove_flock(struct file *filp)
1886{
1887 struct inode * inode = filp->f_dentry->d_inode;
1888 struct file_lock *fl;
1889 struct file_lock **before;
1890
1891 if (!inode->i_flock)
1892 return;
1893
1894 if (filp->f_op && filp->f_op->flock) {
1895 struct file_lock fl = {
1896 .fl_pid = current->tgid,
1897 .fl_file = filp,
1898 .fl_flags = FL_FLOCK,
1899 .fl_type = F_UNLCK,
1900 .fl_end = OFFSET_MAX,
1901 };
1902 filp->f_op->flock(filp, F_SETLKW, &fl);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00001903 if (fl.fl_ops && fl.fl_ops->fl_release_private)
1904 fl.fl_ops->fl_release_private(&fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905 }
1906
1907 lock_kernel();
1908 before = &inode->i_flock;
1909
1910 while ((fl = *before) != NULL) {
1911 if (fl->fl_file == filp) {
Peter Staubachc2936212005-07-27 11:45:09 -07001912 if (IS_FLOCK(fl)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913 locks_delete_lock(before);
1914 continue;
1915 }
1916 if (IS_LEASE(fl)) {
1917 lease_modify(before, F_UNLCK);
1918 continue;
1919 }
1920 /* What? */
1921 BUG();
1922 }
1923 before = &fl->fl_next;
1924 }
1925 unlock_kernel();
1926}
1927
1928/**
1929 * posix_block_lock - blocks waiting for a file lock
1930 * @blocker: the lock which is blocking
1931 * @waiter: the lock which conflicts and has to wait
1932 *
1933 * lockd needs to block waiting for locks.
1934 */
1935void
1936posix_block_lock(struct file_lock *blocker, struct file_lock *waiter)
1937{
1938 locks_insert_block(blocker, waiter);
1939}
1940
1941EXPORT_SYMBOL(posix_block_lock);
1942
1943/**
1944 * posix_unblock_lock - stop waiting for a file lock
1945 * @filp: how the file was opened
1946 * @waiter: the lock which was waiting
1947 *
1948 * lockd needs to block waiting for locks.
1949 */
1950void
1951posix_unblock_lock(struct file *filp, struct file_lock *waiter)
1952{
1953 /*
1954 * A remote machine may cancel the lock request after it's been
1955 * granted locally. If that happens, we need to delete the lock.
1956 */
1957 lock_kernel();
1958 if (waiter->fl_next) {
1959 __locks_delete_block(waiter);
1960 unlock_kernel();
1961 } else {
1962 unlock_kernel();
1963 waiter->fl_type = F_UNLCK;
1964 posix_lock_file(filp, waiter);
1965 }
1966}
1967
1968EXPORT_SYMBOL(posix_unblock_lock);
1969
1970static void lock_get_status(char* out, struct file_lock *fl, int id, char *pfx)
1971{
1972 struct inode *inode = NULL;
1973
1974 if (fl->fl_file != NULL)
1975 inode = fl->fl_file->f_dentry->d_inode;
1976
1977 out += sprintf(out, "%d:%s ", id, pfx);
1978 if (IS_POSIX(fl)) {
1979 out += sprintf(out, "%6s %s ",
1980 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
1981 (inode == NULL) ? "*NOINODE*" :
1982 (IS_MANDLOCK(inode) &&
1983 (inode->i_mode & (S_IXGRP | S_ISGID)) == S_ISGID) ?
1984 "MANDATORY" : "ADVISORY ");
1985 } else if (IS_FLOCK(fl)) {
1986 if (fl->fl_type & LOCK_MAND) {
1987 out += sprintf(out, "FLOCK MSNFS ");
1988 } else {
1989 out += sprintf(out, "FLOCK ADVISORY ");
1990 }
1991 } else if (IS_LEASE(fl)) {
1992 out += sprintf(out, "LEASE ");
1993 if (fl->fl_type & F_INPROGRESS)
1994 out += sprintf(out, "BREAKING ");
1995 else if (fl->fl_file)
1996 out += sprintf(out, "ACTIVE ");
1997 else
1998 out += sprintf(out, "BREAKER ");
1999 } else {
2000 out += sprintf(out, "UNKNOWN UNKNOWN ");
2001 }
2002 if (fl->fl_type & LOCK_MAND) {
2003 out += sprintf(out, "%s ",
2004 (fl->fl_type & LOCK_READ)
2005 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2006 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2007 } else {
2008 out += sprintf(out, "%s ",
2009 (fl->fl_type & F_INPROGRESS)
2010 ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ "
2011 : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ ");
2012 }
2013 if (inode) {
2014#ifdef WE_CAN_BREAK_LSLK_NOW
2015 out += sprintf(out, "%d %s:%ld ", fl->fl_pid,
2016 inode->i_sb->s_id, inode->i_ino);
2017#else
2018 /* userspace relies on this representation of dev_t ;-( */
2019 out += sprintf(out, "%d %02x:%02x:%ld ", fl->fl_pid,
2020 MAJOR(inode->i_sb->s_dev),
2021 MINOR(inode->i_sb->s_dev), inode->i_ino);
2022#endif
2023 } else {
2024 out += sprintf(out, "%d <none>:0 ", fl->fl_pid);
2025 }
2026 if (IS_POSIX(fl)) {
2027 if (fl->fl_end == OFFSET_MAX)
2028 out += sprintf(out, "%Ld EOF\n", fl->fl_start);
2029 else
2030 out += sprintf(out, "%Ld %Ld\n", fl->fl_start,
2031 fl->fl_end);
2032 } else {
2033 out += sprintf(out, "0 EOF\n");
2034 }
2035}
2036
2037static void move_lock_status(char **p, off_t* pos, off_t offset)
2038{
2039 int len;
2040 len = strlen(*p);
2041 if(*pos >= offset) {
2042 /* the complete line is valid */
2043 *p += len;
2044 *pos += len;
2045 return;
2046 }
2047 if(*pos+len > offset) {
2048 /* use the second part of the line */
2049 int i = offset-*pos;
2050 memmove(*p,*p+i,len-i);
2051 *p += len-i;
2052 *pos += len;
2053 return;
2054 }
2055 /* discard the complete line */
2056 *pos += len;
2057}
2058
2059/**
2060 * get_locks_status - reports lock usage in /proc/locks
2061 * @buffer: address in userspace to write into
2062 * @start: ?
2063 * @offset: how far we are through the buffer
2064 * @length: how much to read
2065 */
2066
2067int get_locks_status(char *buffer, char **start, off_t offset, int length)
2068{
2069 struct list_head *tmp;
2070 char *q = buffer;
2071 off_t pos = 0;
2072 int i = 0;
2073
2074 lock_kernel();
2075 list_for_each(tmp, &file_lock_list) {
2076 struct list_head *btmp;
2077 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
2078 lock_get_status(q, fl, ++i, "");
2079 move_lock_status(&q, &pos, offset);
2080
2081 if(pos >= offset+length)
2082 goto done;
2083
2084 list_for_each(btmp, &fl->fl_block) {
2085 struct file_lock *bfl = list_entry(btmp,
2086 struct file_lock, fl_block);
2087 lock_get_status(q, bfl, i, " ->");
2088 move_lock_status(&q, &pos, offset);
2089
2090 if(pos >= offset+length)
2091 goto done;
2092 }
2093 }
2094done:
2095 unlock_kernel();
2096 *start = buffer;
2097 if(q-buffer < length)
2098 return (q-buffer);
2099 return length;
2100}
2101
2102/**
2103 * lock_may_read - checks that the region is free of locks
2104 * @inode: the inode that is being read
2105 * @start: the first byte to read
2106 * @len: the number of bytes to read
2107 *
2108 * Emulates Windows locking requirements. Whole-file
2109 * mandatory locks (share modes) can prohibit a read and
2110 * byte-range POSIX locks can prohibit a read if they overlap.
2111 *
2112 * N.B. this function is only ever called
2113 * from knfsd and ownership of locks is never checked.
2114 */
2115int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2116{
2117 struct file_lock *fl;
2118 int result = 1;
2119 lock_kernel();
2120 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2121 if (IS_POSIX(fl)) {
2122 if (fl->fl_type == F_RDLCK)
2123 continue;
2124 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2125 continue;
2126 } else if (IS_FLOCK(fl)) {
2127 if (!(fl->fl_type & LOCK_MAND))
2128 continue;
2129 if (fl->fl_type & LOCK_READ)
2130 continue;
2131 } else
2132 continue;
2133 result = 0;
2134 break;
2135 }
2136 unlock_kernel();
2137 return result;
2138}
2139
2140EXPORT_SYMBOL(lock_may_read);
2141
2142/**
2143 * lock_may_write - checks that the region is free of locks
2144 * @inode: the inode that is being written
2145 * @start: the first byte to write
2146 * @len: the number of bytes to write
2147 *
2148 * Emulates Windows locking requirements. Whole-file
2149 * mandatory locks (share modes) can prohibit a write and
2150 * byte-range POSIX locks can prohibit a write if they overlap.
2151 *
2152 * N.B. this function is only ever called
2153 * from knfsd and ownership of locks is never checked.
2154 */
2155int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2156{
2157 struct file_lock *fl;
2158 int result = 1;
2159 lock_kernel();
2160 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2161 if (IS_POSIX(fl)) {
2162 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2163 continue;
2164 } else if (IS_FLOCK(fl)) {
2165 if (!(fl->fl_type & LOCK_MAND))
2166 continue;
2167 if (fl->fl_type & LOCK_WRITE)
2168 continue;
2169 } else
2170 continue;
2171 result = 0;
2172 break;
2173 }
2174 unlock_kernel();
2175 return result;
2176}
2177
2178EXPORT_SYMBOL(lock_may_write);
2179
2180static inline void __steal_locks(struct file *file, fl_owner_t from)
2181{
2182 struct inode *inode = file->f_dentry->d_inode;
2183 struct file_lock *fl = inode->i_flock;
2184
2185 while (fl) {
2186 if (fl->fl_file == file && fl->fl_owner == from)
2187 fl->fl_owner = current->files;
2188 fl = fl->fl_next;
2189 }
2190}
2191
2192/* When getting ready for executing a binary, we make sure that current
2193 * has a files_struct on its own. Before dropping the old files_struct,
2194 * we take over ownership of all locks for all file descriptors we own.
2195 * Note that we may accidentally steal a lock for a file that a sibling
2196 * has created since the unshare_files() call.
2197 */
2198void steal_locks(fl_owner_t from)
2199{
2200 struct files_struct *files = current->files;
2201 int i, j;
Dipankar Sarmabadf1662005-09-09 13:04:10 -07002202 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203
2204 if (from == files)
2205 return;
2206
2207 lock_kernel();
2208 j = 0;
Dipankar Sarma4fb3a532005-09-16 19:28:13 -07002209 rcu_read_lock();
Dipankar Sarmabadf1662005-09-09 13:04:10 -07002210 fdt = files_fdtable(files);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002211 for (;;) {
2212 unsigned long set;
2213 i = j * __NFDBITS;
Dipankar Sarmabadf1662005-09-09 13:04:10 -07002214 if (i >= fdt->max_fdset || i >= fdt->max_fds)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215 break;
Dipankar Sarmabadf1662005-09-09 13:04:10 -07002216 set = fdt->open_fds->fds_bits[j++];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217 while (set) {
2218 if (set & 1) {
Dipankar Sarmabadf1662005-09-09 13:04:10 -07002219 struct file *file = fdt->fd[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220 if (file)
2221 __steal_locks(file, from);
2222 }
2223 i++;
2224 set >>= 1;
2225 }
2226 }
Dipankar Sarma4fb3a532005-09-16 19:28:13 -07002227 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228 unlock_kernel();
2229}
2230EXPORT_SYMBOL(steal_locks);
2231
2232static int __init filelock_init(void)
2233{
2234 filelock_cache = kmem_cache_create("file_lock_cache",
2235 sizeof(struct file_lock), 0, SLAB_PANIC,
2236 init_once, NULL);
2237 return 0;
2238}
2239
2240core_initcall(filelock_init);