blob: 231b23c12c3d685b861697bb2cd1720ec631bc51 [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
Trond Myklebust47831f32006-03-20 13:44:05 -0500156static void locks_release_private(struct file_lock *fl)
157{
158 if (fl->fl_ops) {
159 if (fl->fl_ops->fl_release_private)
160 fl->fl_ops->fl_release_private(fl);
161 fl->fl_ops = NULL;
162 }
163 if (fl->fl_lmops) {
164 if (fl->fl_lmops->fl_release_private)
165 fl->fl_lmops->fl_release_private(fl);
166 fl->fl_lmops = NULL;
167 }
168
169}
170
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171/* Free a lock which is not in use. */
Matt Mackall33443c42006-01-08 01:05:22 -0800172static void locks_free_lock(struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173{
174 if (fl == NULL) {
175 BUG();
176 return;
177 }
178 if (waitqueue_active(&fl->fl_wait))
179 panic("Attempting to free lock with active wait queue");
180
181 if (!list_empty(&fl->fl_block))
182 panic("Attempting to free lock with active block list");
183
184 if (!list_empty(&fl->fl_link))
185 panic("Attempting to free lock on active lock list");
186
Trond Myklebust47831f32006-03-20 13:44:05 -0500187 locks_release_private(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188 kmem_cache_free(filelock_cache, fl);
189}
190
191void locks_init_lock(struct file_lock *fl)
192{
193 INIT_LIST_HEAD(&fl->fl_link);
194 INIT_LIST_HEAD(&fl->fl_block);
195 init_waitqueue_head(&fl->fl_wait);
196 fl->fl_next = NULL;
197 fl->fl_fasync = NULL;
198 fl->fl_owner = NULL;
199 fl->fl_pid = 0;
200 fl->fl_file = NULL;
201 fl->fl_flags = 0;
202 fl->fl_type = 0;
203 fl->fl_start = fl->fl_end = 0;
204 fl->fl_ops = NULL;
205 fl->fl_lmops = NULL;
206}
207
208EXPORT_SYMBOL(locks_init_lock);
209
210/*
211 * Initialises the fields of the file lock which are invariant for
212 * free file_locks.
213 */
214static void init_once(void *foo, kmem_cache_t *cache, unsigned long flags)
215{
216 struct file_lock *lock = (struct file_lock *) foo;
217
218 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) !=
219 SLAB_CTOR_CONSTRUCTOR)
220 return;
221
222 locks_init_lock(lock);
223}
224
Trond Myklebust47831f32006-03-20 13:44:05 -0500225static void locks_copy_private(struct file_lock *new, struct file_lock *fl)
226{
227 if (fl->fl_ops) {
228 if (fl->fl_ops->fl_copy_lock)
229 fl->fl_ops->fl_copy_lock(new, fl);
230 new->fl_ops = fl->fl_ops;
231 }
232 if (fl->fl_lmops) {
233 if (fl->fl_lmops->fl_copy_lock)
234 fl->fl_lmops->fl_copy_lock(new, fl);
235 new->fl_lmops = fl->fl_lmops;
236 }
237}
238
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239/*
240 * Initialize a new lock from an existing file_lock structure.
241 */
242void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
243{
Trond Myklebust47831f32006-03-20 13:44:05 -0500244 locks_release_private(new);
245
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 new->fl_owner = fl->fl_owner;
247 new->fl_pid = fl->fl_pid;
248 new->fl_file = fl->fl_file;
249 new->fl_flags = fl->fl_flags;
250 new->fl_type = fl->fl_type;
251 new->fl_start = fl->fl_start;
252 new->fl_end = fl->fl_end;
253 new->fl_ops = fl->fl_ops;
254 new->fl_lmops = fl->fl_lmops;
Trond Myklebust47831f32006-03-20 13:44:05 -0500255
256 locks_copy_private(new, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257}
258
259EXPORT_SYMBOL(locks_copy_lock);
260
261static inline int flock_translate_cmd(int cmd) {
262 if (cmd & LOCK_MAND)
263 return cmd & (LOCK_MAND | LOCK_RW);
264 switch (cmd) {
265 case LOCK_SH:
266 return F_RDLCK;
267 case LOCK_EX:
268 return F_WRLCK;
269 case LOCK_UN:
270 return F_UNLCK;
271 }
272 return -EINVAL;
273}
274
275/* Fill in a file_lock structure with an appropriate FLOCK lock. */
276static int flock_make_lock(struct file *filp, struct file_lock **lock,
277 unsigned int cmd)
278{
279 struct file_lock *fl;
280 int type = flock_translate_cmd(cmd);
281 if (type < 0)
282 return type;
283
284 fl = locks_alloc_lock();
285 if (fl == NULL)
286 return -ENOMEM;
287
288 fl->fl_file = filp;
289 fl->fl_pid = current->tgid;
290 fl->fl_flags = FL_FLOCK;
291 fl->fl_type = type;
292 fl->fl_end = OFFSET_MAX;
293
294 *lock = fl;
295 return 0;
296}
297
298static int assign_type(struct file_lock *fl, int type)
299{
300 switch (type) {
301 case F_RDLCK:
302 case F_WRLCK:
303 case F_UNLCK:
304 fl->fl_type = type;
305 break;
306 default:
307 return -EINVAL;
308 }
309 return 0;
310}
311
312/* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
313 * style lock.
314 */
315static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
316 struct flock *l)
317{
318 off_t start, end;
319
320 switch (l->l_whence) {
321 case 0: /*SEEK_SET*/
322 start = 0;
323 break;
324 case 1: /*SEEK_CUR*/
325 start = filp->f_pos;
326 break;
327 case 2: /*SEEK_END*/
328 start = i_size_read(filp->f_dentry->d_inode);
329 break;
330 default:
331 return -EINVAL;
332 }
333
334 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
335 POSIX-2001 defines it. */
336 start += l->l_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 if (start < 0)
338 return -EINVAL;
Trond Myklebust4c780a42005-10-18 14:20:21 -0700339 fl->fl_end = OFFSET_MAX;
340 if (l->l_len > 0) {
341 end = start + l->l_len - 1;
342 fl->fl_end = end;
343 } else if (l->l_len < 0) {
344 end = start - 1;
345 fl->fl_end = end;
346 start += l->l_len;
347 if (start < 0)
348 return -EINVAL;
349 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 fl->fl_start = start; /* we record the absolute position */
Trond Myklebust4c780a42005-10-18 14:20:21 -0700351 if (fl->fl_end < fl->fl_start)
352 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353
354 fl->fl_owner = current->files;
355 fl->fl_pid = current->tgid;
356 fl->fl_file = filp;
357 fl->fl_flags = FL_POSIX;
358 fl->fl_ops = NULL;
359 fl->fl_lmops = NULL;
360
361 return assign_type(fl, l->l_type);
362}
363
364#if BITS_PER_LONG == 32
365static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
366 struct flock64 *l)
367{
368 loff_t start;
369
370 switch (l->l_whence) {
371 case 0: /*SEEK_SET*/
372 start = 0;
373 break;
374 case 1: /*SEEK_CUR*/
375 start = filp->f_pos;
376 break;
377 case 2: /*SEEK_END*/
378 start = i_size_read(filp->f_dentry->d_inode);
379 break;
380 default:
381 return -EINVAL;
382 }
383
Trond Myklebust4c780a42005-10-18 14:20:21 -0700384 start += l->l_start;
385 if (start < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386 return -EINVAL;
Trond Myklebust4c780a42005-10-18 14:20:21 -0700387 fl->fl_end = OFFSET_MAX;
388 if (l->l_len > 0) {
389 fl->fl_end = start + l->l_len - 1;
390 } else if (l->l_len < 0) {
391 fl->fl_end = start - 1;
392 start += l->l_len;
393 if (start < 0)
394 return -EINVAL;
395 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396 fl->fl_start = start; /* we record the absolute position */
Trond Myklebust4c780a42005-10-18 14:20:21 -0700397 if (fl->fl_end < fl->fl_start)
398 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399
400 fl->fl_owner = current->files;
401 fl->fl_pid = current->tgid;
402 fl->fl_file = filp;
403 fl->fl_flags = FL_POSIX;
404 fl->fl_ops = NULL;
405 fl->fl_lmops = NULL;
406
407 switch (l->l_type) {
408 case F_RDLCK:
409 case F_WRLCK:
410 case F_UNLCK:
411 fl->fl_type = l->l_type;
412 break;
413 default:
414 return -EINVAL;
415 }
416
417 return (0);
418}
419#endif
420
421/* default lease lock manager operations */
422static void lease_break_callback(struct file_lock *fl)
423{
424 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
425}
426
427static void lease_release_private_callback(struct file_lock *fl)
428{
429 if (!fl->fl_file)
430 return;
431
432 f_delown(fl->fl_file);
433 fl->fl_file->f_owner.signum = 0;
434}
435
Adrian Bunk75c96f82005-05-05 16:16:09 -0700436static int lease_mylease_callback(struct file_lock *fl, struct file_lock *try)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437{
438 return fl->fl_file == try->fl_file;
439}
440
Adrian Bunk75c96f82005-05-05 16:16:09 -0700441static struct lock_manager_operations lease_manager_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 .fl_break = lease_break_callback,
443 .fl_release_private = lease_release_private_callback,
444 .fl_mylease = lease_mylease_callback,
445 .fl_change = lease_modify,
446};
447
448/*
449 * Initialize a lease, use the default lock manager operations
450 */
451static int lease_init(struct file *filp, int type, struct file_lock *fl)
452 {
453 fl->fl_owner = current->files;
454 fl->fl_pid = current->tgid;
455
456 fl->fl_file = filp;
457 fl->fl_flags = FL_LEASE;
458 if (assign_type(fl, type) != 0) {
459 locks_free_lock(fl);
460 return -EINVAL;
461 }
462 fl->fl_start = 0;
463 fl->fl_end = OFFSET_MAX;
464 fl->fl_ops = NULL;
465 fl->fl_lmops = &lease_manager_ops;
466 return 0;
467}
468
469/* Allocate a file_lock initialised to this type of lease */
470static int lease_alloc(struct file *filp, int type, struct file_lock **flp)
471{
472 struct file_lock *fl = locks_alloc_lock();
473 int error;
474
475 if (fl == NULL)
476 return -ENOMEM;
477
478 error = lease_init(filp, type, fl);
479 if (error)
480 return error;
481 *flp = fl;
482 return 0;
483}
484
485/* Check if two locks overlap each other.
486 */
487static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
488{
489 return ((fl1->fl_end >= fl2->fl_start) &&
490 (fl2->fl_end >= fl1->fl_start));
491}
492
493/*
494 * Check whether two locks have the same owner.
495 */
Matt Mackall33443c42006-01-08 01:05:22 -0800496static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497{
498 if (fl1->fl_lmops && fl1->fl_lmops->fl_compare_owner)
499 return fl2->fl_lmops == fl1->fl_lmops &&
500 fl1->fl_lmops->fl_compare_owner(fl1, fl2);
501 return fl1->fl_owner == fl2->fl_owner;
502}
503
504/* Remove waiter from blocker's block list.
505 * When blocker ends up pointing to itself then the list is empty.
506 */
Matt Mackall33443c42006-01-08 01:05:22 -0800507static void __locks_delete_block(struct file_lock *waiter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508{
509 list_del_init(&waiter->fl_block);
510 list_del_init(&waiter->fl_link);
511 waiter->fl_next = NULL;
512}
513
514/*
515 */
516static void locks_delete_block(struct file_lock *waiter)
517{
518 lock_kernel();
519 __locks_delete_block(waiter);
520 unlock_kernel();
521}
522
523/* Insert waiter into blocker's block list.
524 * We use a circular list so that processes can be easily woken up in
525 * the order they blocked. The documentation doesn't require this but
526 * it seems like the reasonable thing to do.
527 */
528static void locks_insert_block(struct file_lock *blocker,
529 struct file_lock *waiter)
530{
531 if (!list_empty(&waiter->fl_block)) {
532 printk(KERN_ERR "locks_insert_block: removing duplicated lock "
533 "(pid=%d %Ld-%Ld type=%d)\n", waiter->fl_pid,
534 waiter->fl_start, waiter->fl_end, waiter->fl_type);
535 __locks_delete_block(waiter);
536 }
537 list_add_tail(&waiter->fl_block, &blocker->fl_block);
538 waiter->fl_next = blocker;
539 if (IS_POSIX(blocker))
540 list_add(&waiter->fl_link, &blocked_list);
541}
542
543/* Wake up processes blocked waiting for blocker.
544 * If told to wait then schedule the processes until the block list
545 * is empty, otherwise empty the block list ourselves.
546 */
547static void locks_wake_up_blocks(struct file_lock *blocker)
548{
549 while (!list_empty(&blocker->fl_block)) {
550 struct file_lock *waiter = list_entry(blocker->fl_block.next,
551 struct file_lock, fl_block);
552 __locks_delete_block(waiter);
553 if (waiter->fl_lmops && waiter->fl_lmops->fl_notify)
554 waiter->fl_lmops->fl_notify(waiter);
555 else
556 wake_up(&waiter->fl_wait);
557 }
558}
559
560/* Insert file lock fl into an inode's lock list at the position indicated
561 * by pos. At the same time add the lock to the global file lock list.
562 */
563static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
564{
565 list_add(&fl->fl_link, &file_lock_list);
566
567 /* insert into file's list */
568 fl->fl_next = *pos;
569 *pos = fl;
570
571 if (fl->fl_ops && fl->fl_ops->fl_insert)
572 fl->fl_ops->fl_insert(fl);
573}
574
575/*
576 * Delete a lock and then free it.
577 * Wake up processes that are blocked waiting for this lock,
578 * notify the FS that the lock has been cleared and
579 * finally free the lock.
580 */
581static void locks_delete_lock(struct file_lock **thisfl_p)
582{
583 struct file_lock *fl = *thisfl_p;
584
585 *thisfl_p = fl->fl_next;
586 fl->fl_next = NULL;
587 list_del_init(&fl->fl_link);
588
589 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
590 if (fl->fl_fasync != NULL) {
591 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
592 fl->fl_fasync = NULL;
593 }
594
595 if (fl->fl_ops && fl->fl_ops->fl_remove)
596 fl->fl_ops->fl_remove(fl);
597
598 locks_wake_up_blocks(fl);
599 locks_free_lock(fl);
600}
601
602/* Determine if lock sys_fl blocks lock caller_fl. Common functionality
603 * checks for shared/exclusive status of overlapping locks.
604 */
605static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
606{
607 if (sys_fl->fl_type == F_WRLCK)
608 return 1;
609 if (caller_fl->fl_type == F_WRLCK)
610 return 1;
611 return 0;
612}
613
614/* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
615 * checking before calling the locks_conflict().
616 */
617static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
618{
619 /* POSIX locks owned by the same process do not conflict with
620 * each other.
621 */
622 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
623 return (0);
624
625 /* Check whether they overlap */
626 if (!locks_overlap(caller_fl, sys_fl))
627 return 0;
628
629 return (locks_conflict(caller_fl, sys_fl));
630}
631
632/* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
633 * checking before calling the locks_conflict().
634 */
635static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
636{
637 /* FLOCK locks referring to the same filp do not conflict with
638 * each other.
639 */
640 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
641 return (0);
642 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
643 return 0;
644
645 return (locks_conflict(caller_fl, sys_fl));
646}
647
648static int interruptible_sleep_on_locked(wait_queue_head_t *fl_wait, int timeout)
649{
650 int result = 0;
651 DECLARE_WAITQUEUE(wait, current);
652
653 __set_current_state(TASK_INTERRUPTIBLE);
654 add_wait_queue(fl_wait, &wait);
655 if (timeout == 0)
656 schedule();
657 else
658 result = schedule_timeout(timeout);
659 if (signal_pending(current))
660 result = -ERESTARTSYS;
661 remove_wait_queue(fl_wait, &wait);
662 __set_current_state(TASK_RUNNING);
663 return result;
664}
665
666static int locks_block_on_timeout(struct file_lock *blocker, struct file_lock *waiter, int time)
667{
668 int result;
669 locks_insert_block(blocker, waiter);
670 result = interruptible_sleep_on_locked(&waiter->fl_wait, time);
671 __locks_delete_block(waiter);
672 return result;
673}
674
Andy Adamson8dc7c312006-03-20 13:44:26 -0500675int
676posix_test_lock(struct file *filp, struct file_lock *fl,
677 struct file_lock *conflock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678{
679 struct file_lock *cfl;
680
681 lock_kernel();
682 for (cfl = filp->f_dentry->d_inode->i_flock; cfl; cfl = cfl->fl_next) {
683 if (!IS_POSIX(cfl))
684 continue;
685 if (posix_locks_conflict(cfl, fl))
686 break;
687 }
Andy Adamson8dc7c312006-03-20 13:44:26 -0500688 if (cfl) {
689 locks_copy_lock(conflock, cfl);
690 unlock_kernel();
691 return 1;
692 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 unlock_kernel();
Andy Adamson8dc7c312006-03-20 13:44:26 -0500694 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695}
696
697EXPORT_SYMBOL(posix_test_lock);
698
699/* This function tests for deadlock condition before putting a process to
700 * sleep. The detection scheme is no longer recursive. Recursive was neat,
701 * but dangerous - we risked stack corruption if the lock data was bad, or
702 * if the recursion was too deep for any other reason.
703 *
704 * We rely on the fact that a task can only be on one lock's wait queue
705 * at a time. When we find blocked_task on a wait queue we can re-search
706 * with blocked_task equal to that queue's owner, until either blocked_task
707 * isn't found, or blocked_task is found on a queue owned by my_task.
708 *
709 * Note: the above assumption may not be true when handling lock requests
710 * from a broken NFS client. But broken NFS clients have a lot more to
711 * worry about than proper deadlock detection anyway... --okir
712 */
713int posix_locks_deadlock(struct file_lock *caller_fl,
714 struct file_lock *block_fl)
715{
716 struct list_head *tmp;
717
718next_task:
719 if (posix_same_owner(caller_fl, block_fl))
720 return 1;
721 list_for_each(tmp, &blocked_list) {
722 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
723 if (posix_same_owner(fl, block_fl)) {
724 fl = fl->fl_next;
725 block_fl = fl;
726 goto next_task;
727 }
728 }
729 return 0;
730}
731
732EXPORT_SYMBOL(posix_locks_deadlock);
733
734/* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
735 * at the head of the list, but that's secret knowledge known only to
736 * flock_lock_file and posix_lock_file.
737 */
738static int flock_lock_file(struct file *filp, struct file_lock *new_fl)
739{
740 struct file_lock **before;
741 struct inode * inode = filp->f_dentry->d_inode;
742 int error = 0;
743 int found = 0;
744
745 lock_kernel();
746 for_each_lock(inode, before) {
747 struct file_lock *fl = *before;
748 if (IS_POSIX(fl))
749 break;
750 if (IS_LEASE(fl))
751 continue;
752 if (filp != fl->fl_file)
753 continue;
754 if (new_fl->fl_type == fl->fl_type)
755 goto out;
756 found = 1;
757 locks_delete_lock(before);
758 break;
759 }
760 unlock_kernel();
761
762 if (new_fl->fl_type == F_UNLCK)
763 return 0;
764
765 /*
766 * If a higher-priority process was blocked on the old file lock,
767 * give it the opportunity to lock the file.
768 */
769 if (found)
770 cond_resched();
771
772 lock_kernel();
773 for_each_lock(inode, before) {
774 struct file_lock *fl = *before;
775 if (IS_POSIX(fl))
776 break;
777 if (IS_LEASE(fl))
778 continue;
779 if (!flock_locks_conflict(new_fl, fl))
780 continue;
781 error = -EAGAIN;
782 if (new_fl->fl_flags & FL_SLEEP) {
783 locks_insert_block(fl, new_fl);
784 }
785 goto out;
786 }
787 locks_insert_lock(&inode->i_flock, new_fl);
788 error = 0;
789
790out:
791 unlock_kernel();
792 return error;
793}
794
795EXPORT_SYMBOL(posix_lock_file);
796
797static int __posix_lock_file(struct inode *inode, struct file_lock *request)
798{
799 struct file_lock *fl;
800 struct file_lock *new_fl, *new_fl2;
801 struct file_lock *left = NULL;
802 struct file_lock *right = NULL;
803 struct file_lock **before;
804 int error, added = 0;
805
806 /*
807 * We may need two file_lock structures for this operation,
808 * so we get them in advance to avoid races.
809 */
810 new_fl = locks_alloc_lock();
811 new_fl2 = locks_alloc_lock();
812
813 lock_kernel();
814 if (request->fl_type != F_UNLCK) {
815 for_each_lock(inode, before) {
816 struct file_lock *fl = *before;
817 if (!IS_POSIX(fl))
818 continue;
819 if (!posix_locks_conflict(request, fl))
820 continue;
821 error = -EAGAIN;
822 if (!(request->fl_flags & FL_SLEEP))
823 goto out;
824 error = -EDEADLK;
825 if (posix_locks_deadlock(request, fl))
826 goto out;
827 error = -EAGAIN;
828 locks_insert_block(fl, request);
829 goto out;
830 }
831 }
832
833 /* If we're just looking for a conflict, we're done. */
834 error = 0;
835 if (request->fl_flags & FL_ACCESS)
836 goto out;
837
838 error = -ENOLCK; /* "no luck" */
839 if (!(new_fl && new_fl2))
840 goto out;
841
842 /*
843 * We've allocated the new locks in advance, so there are no
844 * errors possible (and no blocking operations) from here on.
845 *
846 * Find the first old lock with the same owner as the new lock.
847 */
848
849 before = &inode->i_flock;
850
851 /* First skip locks owned by other processes. */
852 while ((fl = *before) && (!IS_POSIX(fl) ||
853 !posix_same_owner(request, fl))) {
854 before = &fl->fl_next;
855 }
856
857 /* Process locks with this owner. */
858 while ((fl = *before) && posix_same_owner(request, fl)) {
859 /* Detect adjacent or overlapping regions (if same lock type)
860 */
861 if (request->fl_type == fl->fl_type) {
Olaf Kirch449231d2005-08-25 16:25:35 -0700862 /* In all comparisons of start vs end, use
863 * "start - 1" rather than "end + 1". If end
864 * is OFFSET_MAX, end + 1 will become negative.
865 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 if (fl->fl_end < request->fl_start - 1)
867 goto next_lock;
868 /* If the next lock in the list has entirely bigger
869 * addresses than the new one, insert the lock here.
870 */
Olaf Kirch449231d2005-08-25 16:25:35 -0700871 if (fl->fl_start - 1 > request->fl_end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 break;
873
874 /* If we come here, the new and old lock are of the
875 * same type and adjacent or overlapping. Make one
876 * lock yielding from the lower start address of both
877 * locks to the higher end address.
878 */
879 if (fl->fl_start > request->fl_start)
880 fl->fl_start = request->fl_start;
881 else
882 request->fl_start = fl->fl_start;
883 if (fl->fl_end < request->fl_end)
884 fl->fl_end = request->fl_end;
885 else
886 request->fl_end = fl->fl_end;
887 if (added) {
888 locks_delete_lock(before);
889 continue;
890 }
891 request = fl;
892 added = 1;
893 }
894 else {
895 /* Processing for different lock types is a bit
896 * more complex.
897 */
898 if (fl->fl_end < request->fl_start)
899 goto next_lock;
900 if (fl->fl_start > request->fl_end)
901 break;
902 if (request->fl_type == F_UNLCK)
903 added = 1;
904 if (fl->fl_start < request->fl_start)
905 left = fl;
906 /* If the next lock in the list has a higher end
907 * address than the new one, insert the new one here.
908 */
909 if (fl->fl_end > request->fl_end) {
910 right = fl;
911 break;
912 }
913 if (fl->fl_start >= request->fl_start) {
914 /* The new lock completely replaces an old
915 * one (This may happen several times).
916 */
917 if (added) {
918 locks_delete_lock(before);
919 continue;
920 }
921 /* Replace the old lock with the new one.
922 * Wake up anybody waiting for the old one,
923 * as the change in lock type might satisfy
924 * their needs.
925 */
926 locks_wake_up_blocks(fl);
927 fl->fl_start = request->fl_start;
928 fl->fl_end = request->fl_end;
929 fl->fl_type = request->fl_type;
Trond Myklebust47831f32006-03-20 13:44:05 -0500930 locks_release_private(fl);
931 locks_copy_private(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 request = fl;
933 added = 1;
934 }
935 }
936 /* Go on to next lock.
937 */
938 next_lock:
939 before = &fl->fl_next;
940 }
941
942 error = 0;
943 if (!added) {
944 if (request->fl_type == F_UNLCK)
945 goto out;
946 locks_copy_lock(new_fl, request);
947 locks_insert_lock(before, new_fl);
948 new_fl = NULL;
949 }
950 if (right) {
951 if (left == right) {
952 /* The new lock breaks the old one in two pieces,
953 * so we have to use the second new lock.
954 */
955 left = new_fl2;
956 new_fl2 = NULL;
957 locks_copy_lock(left, right);
958 locks_insert_lock(before, left);
959 }
960 right->fl_start = request->fl_end + 1;
961 locks_wake_up_blocks(right);
962 }
963 if (left) {
964 left->fl_end = request->fl_start - 1;
965 locks_wake_up_blocks(left);
966 }
967 out:
968 unlock_kernel();
969 /*
970 * Free any unused locks.
971 */
972 if (new_fl)
973 locks_free_lock(new_fl);
974 if (new_fl2)
975 locks_free_lock(new_fl2);
976 return error;
977}
978
979/**
980 * posix_lock_file - Apply a POSIX-style lock to a file
981 * @filp: The file to apply the lock to
982 * @fl: The lock to be applied
983 *
984 * Add a POSIX style lock to a file.
985 * We merge adjacent & overlapping locks whenever possible.
986 * POSIX locks are sorted by owner task, then by starting address
987 */
988int posix_lock_file(struct file *filp, struct file_lock *fl)
989{
990 return __posix_lock_file(filp->f_dentry->d_inode, fl);
991}
992
993/**
994 * posix_lock_file_wait - Apply a POSIX-style lock to a file
995 * @filp: The file to apply the lock to
996 * @fl: The lock to be applied
997 *
998 * Add a POSIX style lock to a file.
999 * We merge adjacent & overlapping locks whenever possible.
1000 * POSIX locks are sorted by owner task, then by starting address
1001 */
1002int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1003{
1004 int error;
1005 might_sleep ();
1006 for (;;) {
1007 error = __posix_lock_file(filp->f_dentry->d_inode, fl);
1008 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1009 break;
1010 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1011 if (!error)
1012 continue;
1013
1014 locks_delete_block(fl);
1015 break;
1016 }
1017 return error;
1018}
1019EXPORT_SYMBOL(posix_lock_file_wait);
1020
1021/**
1022 * locks_mandatory_locked - Check for an active lock
1023 * @inode: the file to check
1024 *
1025 * Searches the inode's list of locks to find any POSIX locks which conflict.
1026 * This function is called from locks_verify_locked() only.
1027 */
1028int locks_mandatory_locked(struct inode *inode)
1029{
1030 fl_owner_t owner = current->files;
1031 struct file_lock *fl;
1032
1033 /*
1034 * Search the lock list for this inode for any POSIX locks.
1035 */
1036 lock_kernel();
1037 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1038 if (!IS_POSIX(fl))
1039 continue;
1040 if (fl->fl_owner != owner)
1041 break;
1042 }
1043 unlock_kernel();
1044 return fl ? -EAGAIN : 0;
1045}
1046
1047/**
1048 * locks_mandatory_area - Check for a conflicting lock
1049 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1050 * for shared
1051 * @inode: the file to check
1052 * @filp: how the file was opened (if it was)
1053 * @offset: start of area to check
1054 * @count: length of area to check
1055 *
1056 * Searches the inode's list of locks to find any POSIX locks which conflict.
1057 * This function is called from rw_verify_area() and
1058 * locks_verify_truncate().
1059 */
1060int locks_mandatory_area(int read_write, struct inode *inode,
1061 struct file *filp, loff_t offset,
1062 size_t count)
1063{
1064 struct file_lock fl;
1065 int error;
1066
1067 locks_init_lock(&fl);
1068 fl.fl_owner = current->files;
1069 fl.fl_pid = current->tgid;
1070 fl.fl_file = filp;
1071 fl.fl_flags = FL_POSIX | FL_ACCESS;
1072 if (filp && !(filp->f_flags & O_NONBLOCK))
1073 fl.fl_flags |= FL_SLEEP;
1074 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1075 fl.fl_start = offset;
1076 fl.fl_end = offset + count - 1;
1077
1078 for (;;) {
1079 error = __posix_lock_file(inode, &fl);
1080 if (error != -EAGAIN)
1081 break;
1082 if (!(fl.fl_flags & FL_SLEEP))
1083 break;
1084 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1085 if (!error) {
1086 /*
1087 * If we've been sleeping someone might have
1088 * changed the permissions behind our back.
1089 */
1090 if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID)
1091 continue;
1092 }
1093
1094 locks_delete_block(&fl);
1095 break;
1096 }
1097
1098 return error;
1099}
1100
1101EXPORT_SYMBOL(locks_mandatory_area);
1102
1103/* We already had a lease on this file; just change its type */
1104int lease_modify(struct file_lock **before, int arg)
1105{
1106 struct file_lock *fl = *before;
1107 int error = assign_type(fl, arg);
1108
1109 if (error)
1110 return error;
1111 locks_wake_up_blocks(fl);
1112 if (arg == F_UNLCK)
1113 locks_delete_lock(before);
1114 return 0;
1115}
1116
1117EXPORT_SYMBOL(lease_modify);
1118
1119static void time_out_leases(struct inode *inode)
1120{
1121 struct file_lock **before;
1122 struct file_lock *fl;
1123
1124 before = &inode->i_flock;
1125 while ((fl = *before) && IS_LEASE(fl) && (fl->fl_type & F_INPROGRESS)) {
1126 if ((fl->fl_break_time == 0)
1127 || time_before(jiffies, fl->fl_break_time)) {
1128 before = &fl->fl_next;
1129 continue;
1130 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131 lease_modify(before, fl->fl_type & ~F_INPROGRESS);
1132 if (fl == *before) /* lease_modify may have freed fl */
1133 before = &fl->fl_next;
1134 }
1135}
1136
1137/**
1138 * __break_lease - revoke all outstanding leases on file
1139 * @inode: the inode of the file to return
1140 * @mode: the open mode (read or write)
1141 *
1142 * break_lease (inlined for speed) has checked there already
1143 * is a lease on this file. Leases are broken on a call to open()
1144 * or truncate(). This function can sleep unless you
1145 * specified %O_NONBLOCK to your open().
1146 */
1147int __break_lease(struct inode *inode, unsigned int mode)
1148{
1149 int error = 0, future;
1150 struct file_lock *new_fl, *flock;
1151 struct file_lock *fl;
1152 int alloc_err;
1153 unsigned long break_time;
1154 int i_have_this_lease = 0;
1155
1156 alloc_err = lease_alloc(NULL, mode & FMODE_WRITE ? F_WRLCK : F_RDLCK,
1157 &new_fl);
1158
1159 lock_kernel();
1160
1161 time_out_leases(inode);
1162
1163 flock = inode->i_flock;
1164 if ((flock == NULL) || !IS_LEASE(flock))
1165 goto out;
1166
1167 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1168 if (fl->fl_owner == current->files)
1169 i_have_this_lease = 1;
1170
1171 if (mode & FMODE_WRITE) {
1172 /* If we want write access, we have to revoke any lease. */
1173 future = F_UNLCK | F_INPROGRESS;
1174 } else if (flock->fl_type & F_INPROGRESS) {
1175 /* If the lease is already being broken, we just leave it */
1176 future = flock->fl_type;
1177 } else if (flock->fl_type & F_WRLCK) {
1178 /* Downgrade the exclusive lease to a read-only lease. */
1179 future = F_RDLCK | F_INPROGRESS;
1180 } else {
1181 /* the existing lease was read-only, so we can read too. */
1182 goto out;
1183 }
1184
1185 if (alloc_err && !i_have_this_lease && ((mode & O_NONBLOCK) == 0)) {
1186 error = alloc_err;
1187 goto out;
1188 }
1189
1190 break_time = 0;
1191 if (lease_break_time > 0) {
1192 break_time = jiffies + lease_break_time * HZ;
1193 if (break_time == 0)
1194 break_time++; /* so that 0 means no break time */
1195 }
1196
1197 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1198 if (fl->fl_type != future) {
1199 fl->fl_type = future;
1200 fl->fl_break_time = break_time;
1201 /* lease must have lmops break callback */
1202 fl->fl_lmops->fl_break(fl);
1203 }
1204 }
1205
1206 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1207 error = -EWOULDBLOCK;
1208 goto out;
1209 }
1210
1211restart:
1212 break_time = flock->fl_break_time;
1213 if (break_time != 0) {
1214 break_time -= jiffies;
1215 if (break_time == 0)
1216 break_time++;
1217 }
1218 error = locks_block_on_timeout(flock, new_fl, break_time);
1219 if (error >= 0) {
1220 if (error == 0)
1221 time_out_leases(inode);
1222 /* Wait for the next lease that has not been broken yet */
1223 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1224 flock = flock->fl_next) {
1225 if (flock->fl_type & F_INPROGRESS)
1226 goto restart;
1227 }
1228 error = 0;
1229 }
1230
1231out:
1232 unlock_kernel();
1233 if (!alloc_err)
1234 locks_free_lock(new_fl);
1235 return error;
1236}
1237
1238EXPORT_SYMBOL(__break_lease);
1239
1240/**
1241 * lease_get_mtime
1242 * @inode: the inode
1243 * @time: pointer to a timespec which will contain the last modified time
1244 *
1245 * This is to force NFS clients to flush their caches for files with
1246 * exclusive leases. The justification is that if someone has an
1247 * exclusive lease, then they could be modifiying it.
1248 */
1249void lease_get_mtime(struct inode *inode, struct timespec *time)
1250{
1251 struct file_lock *flock = inode->i_flock;
1252 if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK))
1253 *time = current_fs_time(inode->i_sb);
1254 else
1255 *time = inode->i_mtime;
1256}
1257
1258EXPORT_SYMBOL(lease_get_mtime);
1259
1260/**
1261 * fcntl_getlease - Enquire what lease is currently active
1262 * @filp: the file
1263 *
1264 * The value returned by this function will be one of
1265 * (if no lease break is pending):
1266 *
1267 * %F_RDLCK to indicate a shared lease is held.
1268 *
1269 * %F_WRLCK to indicate an exclusive lease is held.
1270 *
1271 * %F_UNLCK to indicate no lease is held.
1272 *
1273 * (if a lease break is pending):
1274 *
1275 * %F_RDLCK to indicate an exclusive lease needs to be
1276 * changed to a shared lease (or removed).
1277 *
1278 * %F_UNLCK to indicate the lease needs to be removed.
1279 *
1280 * XXX: sfr & willy disagree over whether F_INPROGRESS
1281 * should be returned to userspace.
1282 */
1283int fcntl_getlease(struct file *filp)
1284{
1285 struct file_lock *fl;
1286 int type = F_UNLCK;
1287
1288 lock_kernel();
1289 time_out_leases(filp->f_dentry->d_inode);
1290 for (fl = filp->f_dentry->d_inode->i_flock; fl && IS_LEASE(fl);
1291 fl = fl->fl_next) {
1292 if (fl->fl_file == filp) {
1293 type = fl->fl_type & ~F_INPROGRESS;
1294 break;
1295 }
1296 }
1297 unlock_kernel();
1298 return type;
1299}
1300
1301/**
1302 * __setlease - sets a lease on an open file
1303 * @filp: file pointer
1304 * @arg: type of lease to obtain
1305 * @flp: input - file_lock to use, output - file_lock inserted
1306 *
1307 * The (input) flp->fl_lmops->fl_break function is required
1308 * by break_lease().
1309 *
1310 * Called with kernel lock held.
1311 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07001312static int __setlease(struct file *filp, long arg, struct file_lock **flp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313{
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001314 struct file_lock *fl, **before, **my_before = NULL, *lease;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315 struct dentry *dentry = filp->f_dentry;
1316 struct inode *inode = dentry->d_inode;
1317 int error, rdlease_count = 0, wrlease_count = 0;
1318
1319 time_out_leases(inode);
1320
1321 error = -EINVAL;
1322 if (!flp || !(*flp) || !(*flp)->fl_lmops || !(*flp)->fl_lmops->fl_break)
1323 goto out;
1324
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001325 lease = *flp;
1326
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327 error = -EAGAIN;
1328 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1329 goto out;
1330 if ((arg == F_WRLCK)
1331 && ((atomic_read(&dentry->d_count) > 1)
1332 || (atomic_read(&inode->i_count) > 1)))
1333 goto out;
1334
1335 /*
1336 * At this point, we know that if there is an exclusive
1337 * lease on this file, then we hold it on this filp
1338 * (otherwise our open of this file would have blocked).
1339 * And if we are trying to acquire an exclusive lease,
1340 * then the file is not open by anyone (including us)
1341 * except for this filp.
1342 */
1343 for (before = &inode->i_flock;
1344 ((fl = *before) != NULL) && IS_LEASE(fl);
1345 before = &fl->fl_next) {
1346 if (lease->fl_lmops->fl_mylease(fl, lease))
1347 my_before = before;
1348 else if (fl->fl_type == (F_INPROGRESS | F_UNLCK))
1349 /*
1350 * Someone is in the process of opening this
1351 * file for writing so we may not take an
1352 * exclusive lease on it.
1353 */
1354 wrlease_count++;
1355 else
1356 rdlease_count++;
1357 }
1358
1359 if ((arg == F_RDLCK && (wrlease_count > 0)) ||
1360 (arg == F_WRLCK && ((rdlease_count + wrlease_count) > 0)))
1361 goto out;
1362
1363 if (my_before != NULL) {
1364 error = lease->fl_lmops->fl_change(my_before, arg);
1365 goto out;
1366 }
1367
1368 error = 0;
1369 if (arg == F_UNLCK)
1370 goto out;
1371
1372 error = -EINVAL;
1373 if (!leases_enable)
1374 goto out;
1375
1376 error = lease_alloc(filp, arg, &fl);
1377 if (error)
1378 goto out;
1379
1380 locks_copy_lock(fl, lease);
1381
1382 locks_insert_lock(before, fl);
1383
1384 *flp = fl;
1385out:
1386 return error;
1387}
1388
1389 /**
1390 * setlease - sets a lease on an open file
1391 * @filp: file pointer
1392 * @arg: type of lease to obtain
1393 * @lease: file_lock to use
1394 *
1395 * Call this to establish a lease on the file.
1396 * The fl_lmops fl_break function is required by break_lease
1397 */
1398
1399int setlease(struct file *filp, long arg, struct file_lock **lease)
1400{
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 lock_kernel();
1414 error = __setlease(filp, arg, lease);
1415 unlock_kernel();
1416
1417 return error;
1418}
1419
1420EXPORT_SYMBOL(setlease);
1421
1422/**
1423 * fcntl_setlease - sets a lease on an open file
1424 * @fd: open file descriptor
1425 * @filp: file pointer
1426 * @arg: type of lease to obtain
1427 *
1428 * Call this fcntl to establish a lease on the file.
1429 * Note that you also need to call %F_SETSIG to
1430 * receive a signal when the lease is broken.
1431 */
1432int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1433{
1434 struct file_lock fl, *flp = &fl;
1435 struct dentry *dentry = filp->f_dentry;
1436 struct inode *inode = dentry->d_inode;
1437 int error;
1438
1439 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1440 return -EACCES;
1441 if (!S_ISREG(inode->i_mode))
1442 return -EINVAL;
1443 error = security_file_lock(filp, arg);
1444 if (error)
1445 return error;
1446
1447 locks_init_lock(&fl);
1448 error = lease_init(filp, arg, &fl);
1449 if (error)
1450 return error;
1451
1452 lock_kernel();
1453
1454 error = __setlease(filp, arg, &flp);
J. Bruce Fieldsdc15ae12005-11-10 19:08:00 -05001455 if (error || arg == F_UNLCK)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456 goto out_unlock;
1457
1458 error = fasync_helper(fd, filp, 1, &flp->fl_fasync);
1459 if (error < 0) {
1460 /* remove lease just inserted by __setlease */
1461 flp->fl_type = F_UNLCK | F_INPROGRESS;
1462 flp->fl_break_time = jiffies- 10;
1463 time_out_leases(inode);
1464 goto out_unlock;
1465 }
1466
1467 error = f_setown(filp, current->pid, 0);
1468out_unlock:
1469 unlock_kernel();
1470 return error;
1471}
1472
1473/**
1474 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1475 * @filp: The file to apply the lock to
1476 * @fl: The lock to be applied
1477 *
1478 * Add a FLOCK style lock to a file.
1479 */
1480int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1481{
1482 int error;
1483 might_sleep();
1484 for (;;) {
1485 error = flock_lock_file(filp, fl);
1486 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1487 break;
1488 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1489 if (!error)
1490 continue;
1491
1492 locks_delete_block(fl);
1493 break;
1494 }
1495 return error;
1496}
1497
1498EXPORT_SYMBOL(flock_lock_file_wait);
1499
1500/**
1501 * sys_flock: - flock() system call.
1502 * @fd: the file descriptor to lock.
1503 * @cmd: the type of lock to apply.
1504 *
1505 * Apply a %FL_FLOCK style lock to an open file descriptor.
1506 * The @cmd can be one of
1507 *
1508 * %LOCK_SH -- a shared lock.
1509 *
1510 * %LOCK_EX -- an exclusive lock.
1511 *
1512 * %LOCK_UN -- remove an existing lock.
1513 *
1514 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1515 *
1516 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1517 * processes read and write access respectively.
1518 */
1519asmlinkage long sys_flock(unsigned int fd, unsigned int cmd)
1520{
1521 struct file *filp;
1522 struct file_lock *lock;
1523 int can_sleep, unlock;
1524 int error;
1525
1526 error = -EBADF;
1527 filp = fget(fd);
1528 if (!filp)
1529 goto out;
1530
1531 can_sleep = !(cmd & LOCK_NB);
1532 cmd &= ~LOCK_NB;
1533 unlock = (cmd == LOCK_UN);
1534
1535 if (!unlock && !(cmd & LOCK_MAND) && !(filp->f_mode & 3))
1536 goto out_putf;
1537
1538 error = flock_make_lock(filp, &lock, cmd);
1539 if (error)
1540 goto out_putf;
1541 if (can_sleep)
1542 lock->fl_flags |= FL_SLEEP;
1543
1544 error = security_file_lock(filp, cmd);
1545 if (error)
1546 goto out_free;
1547
1548 if (filp->f_op && filp->f_op->flock)
1549 error = filp->f_op->flock(filp,
1550 (can_sleep) ? F_SETLKW : F_SETLK,
1551 lock);
1552 else
1553 error = flock_lock_file_wait(filp, lock);
1554
1555 out_free:
1556 if (list_empty(&lock->fl_link)) {
1557 locks_free_lock(lock);
1558 }
1559
1560 out_putf:
1561 fput(filp);
1562 out:
1563 return error;
1564}
1565
1566/* Report the first existing lock that would conflict with l.
1567 * This implements the F_GETLK command of fcntl().
1568 */
1569int fcntl_getlk(struct file *filp, struct flock __user *l)
1570{
Andy Adamson8dc7c312006-03-20 13:44:26 -05001571 struct file_lock *fl, cfl, file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572 struct flock flock;
1573 int error;
1574
1575 error = -EFAULT;
1576 if (copy_from_user(&flock, l, sizeof(flock)))
1577 goto out;
1578 error = -EINVAL;
1579 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1580 goto out;
1581
1582 error = flock_to_posix_lock(filp, &file_lock, &flock);
1583 if (error)
1584 goto out;
1585
1586 if (filp->f_op && filp->f_op->lock) {
1587 error = filp->f_op->lock(filp, F_GETLK, &file_lock);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00001588 if (file_lock.fl_ops && file_lock.fl_ops->fl_release_private)
1589 file_lock.fl_ops->fl_release_private(&file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590 if (error < 0)
1591 goto out;
1592 else
1593 fl = (file_lock.fl_type == F_UNLCK ? NULL : &file_lock);
1594 } else {
Andy Adamson8dc7c312006-03-20 13:44:26 -05001595 fl = (posix_test_lock(filp, &file_lock, &cfl) ? &cfl : NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 }
1597
1598 flock.l_type = F_UNLCK;
1599 if (fl != NULL) {
1600 flock.l_pid = fl->fl_pid;
1601#if BITS_PER_LONG == 32
1602 /*
1603 * Make sure we can represent the posix lock via
1604 * legacy 32bit flock.
1605 */
1606 error = -EOVERFLOW;
1607 if (fl->fl_start > OFFT_OFFSET_MAX)
1608 goto out;
1609 if ((fl->fl_end != OFFSET_MAX)
1610 && (fl->fl_end > OFFT_OFFSET_MAX))
1611 goto out;
1612#endif
1613 flock.l_start = fl->fl_start;
1614 flock.l_len = fl->fl_end == OFFSET_MAX ? 0 :
1615 fl->fl_end - fl->fl_start + 1;
1616 flock.l_whence = 0;
1617 flock.l_type = fl->fl_type;
1618 }
1619 error = -EFAULT;
1620 if (!copy_to_user(l, &flock, sizeof(flock)))
1621 error = 0;
1622out:
1623 return error;
1624}
1625
1626/* Apply the lock described by l to an open file descriptor.
1627 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1628 */
Peter Staubachc2936212005-07-27 11:45:09 -07001629int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1630 struct flock __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631{
1632 struct file_lock *file_lock = locks_alloc_lock();
1633 struct flock flock;
1634 struct inode *inode;
1635 int error;
1636
1637 if (file_lock == NULL)
1638 return -ENOLCK;
1639
1640 /*
1641 * This might block, so we do it before checking the inode.
1642 */
1643 error = -EFAULT;
1644 if (copy_from_user(&flock, l, sizeof(flock)))
1645 goto out;
1646
1647 inode = filp->f_dentry->d_inode;
1648
1649 /* Don't allow mandatory locks on files that may be memory mapped
1650 * and shared.
1651 */
1652 if (IS_MANDLOCK(inode) &&
1653 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1654 mapping_writably_mapped(filp->f_mapping)) {
1655 error = -EAGAIN;
1656 goto out;
1657 }
1658
Peter Staubachc2936212005-07-27 11:45:09 -07001659again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660 error = flock_to_posix_lock(filp, file_lock, &flock);
1661 if (error)
1662 goto out;
1663 if (cmd == F_SETLKW) {
1664 file_lock->fl_flags |= FL_SLEEP;
1665 }
1666
1667 error = -EBADF;
1668 switch (flock.l_type) {
1669 case F_RDLCK:
1670 if (!(filp->f_mode & FMODE_READ))
1671 goto out;
1672 break;
1673 case F_WRLCK:
1674 if (!(filp->f_mode & FMODE_WRITE))
1675 goto out;
1676 break;
1677 case F_UNLCK:
1678 break;
1679 default:
1680 error = -EINVAL;
1681 goto out;
1682 }
1683
1684 error = security_file_lock(filp, file_lock->fl_type);
1685 if (error)
1686 goto out;
1687
Peter Staubachc2936212005-07-27 11:45:09 -07001688 if (filp->f_op && filp->f_op->lock != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689 error = filp->f_op->lock(filp, cmd, file_lock);
Peter Staubachc2936212005-07-27 11:45:09 -07001690 else {
1691 for (;;) {
1692 error = __posix_lock_file(inode, file_lock);
1693 if ((error != -EAGAIN) || (cmd == F_SETLK))
1694 break;
1695 error = wait_event_interruptible(file_lock->fl_wait,
1696 !file_lock->fl_next);
1697 if (!error)
1698 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699
Peter Staubachc2936212005-07-27 11:45:09 -07001700 locks_delete_block(file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701 break;
Peter Staubachc2936212005-07-27 11:45:09 -07001702 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703 }
1704
Peter Staubachc2936212005-07-27 11:45:09 -07001705 /*
1706 * Attempt to detect a close/fcntl race and recover by
1707 * releasing the lock that was just acquired.
1708 */
1709 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1710 flock.l_type = F_UNLCK;
1711 goto again;
1712 }
1713
1714out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 locks_free_lock(file_lock);
1716 return error;
1717}
1718
1719#if BITS_PER_LONG == 32
1720/* Report the first existing lock that would conflict with l.
1721 * This implements the F_GETLK command of fcntl().
1722 */
1723int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1724{
Andy Adamson8dc7c312006-03-20 13:44:26 -05001725 struct file_lock *fl, cfl, file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726 struct flock64 flock;
1727 int error;
1728
1729 error = -EFAULT;
1730 if (copy_from_user(&flock, l, sizeof(flock)))
1731 goto out;
1732 error = -EINVAL;
1733 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1734 goto out;
1735
1736 error = flock64_to_posix_lock(filp, &file_lock, &flock);
1737 if (error)
1738 goto out;
1739
1740 if (filp->f_op && filp->f_op->lock) {
1741 error = filp->f_op->lock(filp, F_GETLK, &file_lock);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00001742 if (file_lock.fl_ops && file_lock.fl_ops->fl_release_private)
1743 file_lock.fl_ops->fl_release_private(&file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744 if (error < 0)
1745 goto out;
1746 else
1747 fl = (file_lock.fl_type == F_UNLCK ? NULL : &file_lock);
1748 } else {
Andy Adamson8dc7c312006-03-20 13:44:26 -05001749 fl = (posix_test_lock(filp, &file_lock, &cfl) ? &cfl : NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750 }
1751
1752 flock.l_type = F_UNLCK;
1753 if (fl != NULL) {
1754 flock.l_pid = fl->fl_pid;
1755 flock.l_start = fl->fl_start;
1756 flock.l_len = fl->fl_end == OFFSET_MAX ? 0 :
1757 fl->fl_end - fl->fl_start + 1;
1758 flock.l_whence = 0;
1759 flock.l_type = fl->fl_type;
1760 }
1761 error = -EFAULT;
1762 if (!copy_to_user(l, &flock, sizeof(flock)))
1763 error = 0;
1764
1765out:
1766 return error;
1767}
1768
1769/* Apply the lock described by l to an open file descriptor.
1770 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1771 */
Peter Staubachc2936212005-07-27 11:45:09 -07001772int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
1773 struct flock64 __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774{
1775 struct file_lock *file_lock = locks_alloc_lock();
1776 struct flock64 flock;
1777 struct inode *inode;
1778 int error;
1779
1780 if (file_lock == NULL)
1781 return -ENOLCK;
1782
1783 /*
1784 * This might block, so we do it before checking the inode.
1785 */
1786 error = -EFAULT;
1787 if (copy_from_user(&flock, l, sizeof(flock)))
1788 goto out;
1789
1790 inode = filp->f_dentry->d_inode;
1791
1792 /* Don't allow mandatory locks on files that may be memory mapped
1793 * and shared.
1794 */
1795 if (IS_MANDLOCK(inode) &&
1796 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1797 mapping_writably_mapped(filp->f_mapping)) {
1798 error = -EAGAIN;
1799 goto out;
1800 }
1801
Peter Staubachc2936212005-07-27 11:45:09 -07001802again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803 error = flock64_to_posix_lock(filp, file_lock, &flock);
1804 if (error)
1805 goto out;
1806 if (cmd == F_SETLKW64) {
1807 file_lock->fl_flags |= FL_SLEEP;
1808 }
1809
1810 error = -EBADF;
1811 switch (flock.l_type) {
1812 case F_RDLCK:
1813 if (!(filp->f_mode & FMODE_READ))
1814 goto out;
1815 break;
1816 case F_WRLCK:
1817 if (!(filp->f_mode & FMODE_WRITE))
1818 goto out;
1819 break;
1820 case F_UNLCK:
1821 break;
1822 default:
1823 error = -EINVAL;
1824 goto out;
1825 }
1826
1827 error = security_file_lock(filp, file_lock->fl_type);
1828 if (error)
1829 goto out;
1830
Peter Staubachc2936212005-07-27 11:45:09 -07001831 if (filp->f_op && filp->f_op->lock != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832 error = filp->f_op->lock(filp, cmd, file_lock);
Peter Staubachc2936212005-07-27 11:45:09 -07001833 else {
1834 for (;;) {
1835 error = __posix_lock_file(inode, file_lock);
1836 if ((error != -EAGAIN) || (cmd == F_SETLK64))
1837 break;
1838 error = wait_event_interruptible(file_lock->fl_wait,
1839 !file_lock->fl_next);
1840 if (!error)
1841 continue;
1842
1843 locks_delete_block(file_lock);
1844 break;
1845 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846 }
1847
Peter Staubachc2936212005-07-27 11:45:09 -07001848 /*
1849 * Attempt to detect a close/fcntl race and recover by
1850 * releasing the lock that was just acquired.
1851 */
1852 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1853 flock.l_type = F_UNLCK;
1854 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855 }
1856
1857out:
1858 locks_free_lock(file_lock);
1859 return error;
1860}
1861#endif /* BITS_PER_LONG == 32 */
1862
1863/*
1864 * This function is called when the file is being removed
1865 * from the task's fd array. POSIX locks belonging to this task
1866 * are deleted at this time.
1867 */
1868void locks_remove_posix(struct file *filp, fl_owner_t owner)
1869{
1870 struct file_lock lock, **before;
1871
1872 /*
1873 * If there are no locks held on this file, we don't need to call
1874 * posix_lock_file(). Another process could be setting a lock on this
1875 * file at the same time, but we wouldn't remove that lock anyway.
1876 */
1877 before = &filp->f_dentry->d_inode->i_flock;
1878 if (*before == NULL)
1879 return;
1880
1881 lock.fl_type = F_UNLCK;
1882 lock.fl_flags = FL_POSIX;
1883 lock.fl_start = 0;
1884 lock.fl_end = OFFSET_MAX;
1885 lock.fl_owner = owner;
1886 lock.fl_pid = current->tgid;
1887 lock.fl_file = filp;
1888 lock.fl_ops = NULL;
1889 lock.fl_lmops = NULL;
1890
1891 if (filp->f_op && filp->f_op->lock != NULL) {
1892 filp->f_op->lock(filp, F_SETLK, &lock);
1893 goto out;
1894 }
1895
1896 /* Can't use posix_lock_file here; we need to remove it no matter
1897 * which pid we have.
1898 */
1899 lock_kernel();
1900 while (*before != NULL) {
1901 struct file_lock *fl = *before;
1902 if (IS_POSIX(fl) && posix_same_owner(fl, &lock)) {
1903 locks_delete_lock(before);
1904 continue;
1905 }
1906 before = &fl->fl_next;
1907 }
1908 unlock_kernel();
1909out:
1910 if (lock.fl_ops && lock.fl_ops->fl_release_private)
1911 lock.fl_ops->fl_release_private(&lock);
1912}
1913
1914EXPORT_SYMBOL(locks_remove_posix);
1915
1916/*
1917 * This function is called on the last close of an open file.
1918 */
1919void locks_remove_flock(struct file *filp)
1920{
1921 struct inode * inode = filp->f_dentry->d_inode;
1922 struct file_lock *fl;
1923 struct file_lock **before;
1924
1925 if (!inode->i_flock)
1926 return;
1927
1928 if (filp->f_op && filp->f_op->flock) {
1929 struct file_lock fl = {
1930 .fl_pid = current->tgid,
1931 .fl_file = filp,
1932 .fl_flags = FL_FLOCK,
1933 .fl_type = F_UNLCK,
1934 .fl_end = OFFSET_MAX,
1935 };
1936 filp->f_op->flock(filp, F_SETLKW, &fl);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00001937 if (fl.fl_ops && fl.fl_ops->fl_release_private)
1938 fl.fl_ops->fl_release_private(&fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939 }
1940
1941 lock_kernel();
1942 before = &inode->i_flock;
1943
1944 while ((fl = *before) != NULL) {
1945 if (fl->fl_file == filp) {
Peter Staubachc2936212005-07-27 11:45:09 -07001946 if (IS_FLOCK(fl)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947 locks_delete_lock(before);
1948 continue;
1949 }
1950 if (IS_LEASE(fl)) {
1951 lease_modify(before, F_UNLCK);
1952 continue;
1953 }
1954 /* What? */
1955 BUG();
1956 }
1957 before = &fl->fl_next;
1958 }
1959 unlock_kernel();
1960}
1961
1962/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963 * posix_unblock_lock - stop waiting for a file lock
1964 * @filp: how the file was opened
1965 * @waiter: the lock which was waiting
1966 *
1967 * lockd needs to block waiting for locks.
1968 */
J. Bruce Fields64a318e2006-01-03 09:55:46 +01001969int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970posix_unblock_lock(struct file *filp, struct file_lock *waiter)
1971{
J. Bruce Fields64a318e2006-01-03 09:55:46 +01001972 int status = 0;
1973
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974 lock_kernel();
J. Bruce Fields5996a292006-01-03 09:55:44 +01001975 if (waiter->fl_next)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976 __locks_delete_block(waiter);
J. Bruce Fields64a318e2006-01-03 09:55:46 +01001977 else
1978 status = -ENOENT;
J. Bruce Fields5996a292006-01-03 09:55:44 +01001979 unlock_kernel();
J. Bruce Fields64a318e2006-01-03 09:55:46 +01001980 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981}
1982
1983EXPORT_SYMBOL(posix_unblock_lock);
1984
1985static void lock_get_status(char* out, struct file_lock *fl, int id, char *pfx)
1986{
1987 struct inode *inode = NULL;
1988
1989 if (fl->fl_file != NULL)
1990 inode = fl->fl_file->f_dentry->d_inode;
1991
1992 out += sprintf(out, "%d:%s ", id, pfx);
1993 if (IS_POSIX(fl)) {
1994 out += sprintf(out, "%6s %s ",
1995 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
1996 (inode == NULL) ? "*NOINODE*" :
1997 (IS_MANDLOCK(inode) &&
1998 (inode->i_mode & (S_IXGRP | S_ISGID)) == S_ISGID) ?
1999 "MANDATORY" : "ADVISORY ");
2000 } else if (IS_FLOCK(fl)) {
2001 if (fl->fl_type & LOCK_MAND) {
2002 out += sprintf(out, "FLOCK MSNFS ");
2003 } else {
2004 out += sprintf(out, "FLOCK ADVISORY ");
2005 }
2006 } else if (IS_LEASE(fl)) {
2007 out += sprintf(out, "LEASE ");
2008 if (fl->fl_type & F_INPROGRESS)
2009 out += sprintf(out, "BREAKING ");
2010 else if (fl->fl_file)
2011 out += sprintf(out, "ACTIVE ");
2012 else
2013 out += sprintf(out, "BREAKER ");
2014 } else {
2015 out += sprintf(out, "UNKNOWN UNKNOWN ");
2016 }
2017 if (fl->fl_type & LOCK_MAND) {
2018 out += sprintf(out, "%s ",
2019 (fl->fl_type & LOCK_READ)
2020 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2021 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2022 } else {
2023 out += sprintf(out, "%s ",
2024 (fl->fl_type & F_INPROGRESS)
2025 ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ "
2026 : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ ");
2027 }
2028 if (inode) {
2029#ifdef WE_CAN_BREAK_LSLK_NOW
2030 out += sprintf(out, "%d %s:%ld ", fl->fl_pid,
2031 inode->i_sb->s_id, inode->i_ino);
2032#else
2033 /* userspace relies on this representation of dev_t ;-( */
2034 out += sprintf(out, "%d %02x:%02x:%ld ", fl->fl_pid,
2035 MAJOR(inode->i_sb->s_dev),
2036 MINOR(inode->i_sb->s_dev), inode->i_ino);
2037#endif
2038 } else {
2039 out += sprintf(out, "%d <none>:0 ", fl->fl_pid);
2040 }
2041 if (IS_POSIX(fl)) {
2042 if (fl->fl_end == OFFSET_MAX)
2043 out += sprintf(out, "%Ld EOF\n", fl->fl_start);
2044 else
2045 out += sprintf(out, "%Ld %Ld\n", fl->fl_start,
2046 fl->fl_end);
2047 } else {
2048 out += sprintf(out, "0 EOF\n");
2049 }
2050}
2051
2052static void move_lock_status(char **p, off_t* pos, off_t offset)
2053{
2054 int len;
2055 len = strlen(*p);
2056 if(*pos >= offset) {
2057 /* the complete line is valid */
2058 *p += len;
2059 *pos += len;
2060 return;
2061 }
2062 if(*pos+len > offset) {
2063 /* use the second part of the line */
2064 int i = offset-*pos;
2065 memmove(*p,*p+i,len-i);
2066 *p += len-i;
2067 *pos += len;
2068 return;
2069 }
2070 /* discard the complete line */
2071 *pos += len;
2072}
2073
2074/**
2075 * get_locks_status - reports lock usage in /proc/locks
2076 * @buffer: address in userspace to write into
2077 * @start: ?
2078 * @offset: how far we are through the buffer
2079 * @length: how much to read
2080 */
2081
2082int get_locks_status(char *buffer, char **start, off_t offset, int length)
2083{
2084 struct list_head *tmp;
2085 char *q = buffer;
2086 off_t pos = 0;
2087 int i = 0;
2088
2089 lock_kernel();
2090 list_for_each(tmp, &file_lock_list) {
2091 struct list_head *btmp;
2092 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
2093 lock_get_status(q, fl, ++i, "");
2094 move_lock_status(&q, &pos, offset);
2095
2096 if(pos >= offset+length)
2097 goto done;
2098
2099 list_for_each(btmp, &fl->fl_block) {
2100 struct file_lock *bfl = list_entry(btmp,
2101 struct file_lock, fl_block);
2102 lock_get_status(q, bfl, i, " ->");
2103 move_lock_status(&q, &pos, offset);
2104
2105 if(pos >= offset+length)
2106 goto done;
2107 }
2108 }
2109done:
2110 unlock_kernel();
2111 *start = buffer;
2112 if(q-buffer < length)
2113 return (q-buffer);
2114 return length;
2115}
2116
2117/**
2118 * lock_may_read - checks that the region is free of locks
2119 * @inode: the inode that is being read
2120 * @start: the first byte to read
2121 * @len: the number of bytes to read
2122 *
2123 * Emulates Windows locking requirements. Whole-file
2124 * mandatory locks (share modes) can prohibit a read and
2125 * byte-range POSIX locks can prohibit a read if they overlap.
2126 *
2127 * N.B. this function is only ever called
2128 * from knfsd and ownership of locks is never checked.
2129 */
2130int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2131{
2132 struct file_lock *fl;
2133 int result = 1;
2134 lock_kernel();
2135 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2136 if (IS_POSIX(fl)) {
2137 if (fl->fl_type == F_RDLCK)
2138 continue;
2139 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2140 continue;
2141 } else if (IS_FLOCK(fl)) {
2142 if (!(fl->fl_type & LOCK_MAND))
2143 continue;
2144 if (fl->fl_type & LOCK_READ)
2145 continue;
2146 } else
2147 continue;
2148 result = 0;
2149 break;
2150 }
2151 unlock_kernel();
2152 return result;
2153}
2154
2155EXPORT_SYMBOL(lock_may_read);
2156
2157/**
2158 * lock_may_write - checks that the region is free of locks
2159 * @inode: the inode that is being written
2160 * @start: the first byte to write
2161 * @len: the number of bytes to write
2162 *
2163 * Emulates Windows locking requirements. Whole-file
2164 * mandatory locks (share modes) can prohibit a write and
2165 * byte-range POSIX locks can prohibit a write if they overlap.
2166 *
2167 * N.B. this function is only ever called
2168 * from knfsd and ownership of locks is never checked.
2169 */
2170int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2171{
2172 struct file_lock *fl;
2173 int result = 1;
2174 lock_kernel();
2175 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2176 if (IS_POSIX(fl)) {
2177 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2178 continue;
2179 } else if (IS_FLOCK(fl)) {
2180 if (!(fl->fl_type & LOCK_MAND))
2181 continue;
2182 if (fl->fl_type & LOCK_WRITE)
2183 continue;
2184 } else
2185 continue;
2186 result = 0;
2187 break;
2188 }
2189 unlock_kernel();
2190 return result;
2191}
2192
2193EXPORT_SYMBOL(lock_may_write);
2194
2195static inline void __steal_locks(struct file *file, fl_owner_t from)
2196{
2197 struct inode *inode = file->f_dentry->d_inode;
2198 struct file_lock *fl = inode->i_flock;
2199
2200 while (fl) {
2201 if (fl->fl_file == file && fl->fl_owner == from)
2202 fl->fl_owner = current->files;
2203 fl = fl->fl_next;
2204 }
2205}
2206
2207/* When getting ready for executing a binary, we make sure that current
2208 * has a files_struct on its own. Before dropping the old files_struct,
2209 * we take over ownership of all locks for all file descriptors we own.
2210 * Note that we may accidentally steal a lock for a file that a sibling
2211 * has created since the unshare_files() call.
2212 */
2213void steal_locks(fl_owner_t from)
2214{
2215 struct files_struct *files = current->files;
2216 int i, j;
Dipankar Sarmabadf1662005-09-09 13:04:10 -07002217 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218
2219 if (from == files)
2220 return;
2221
2222 lock_kernel();
2223 j = 0;
Dipankar Sarma4fb3a532005-09-16 19:28:13 -07002224 rcu_read_lock();
Dipankar Sarmabadf1662005-09-09 13:04:10 -07002225 fdt = files_fdtable(files);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226 for (;;) {
2227 unsigned long set;
2228 i = j * __NFDBITS;
Dipankar Sarmabadf1662005-09-09 13:04:10 -07002229 if (i >= fdt->max_fdset || i >= fdt->max_fds)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002230 break;
Dipankar Sarmabadf1662005-09-09 13:04:10 -07002231 set = fdt->open_fds->fds_bits[j++];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002232 while (set) {
2233 if (set & 1) {
Dipankar Sarmabadf1662005-09-09 13:04:10 -07002234 struct file *file = fdt->fd[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002235 if (file)
2236 __steal_locks(file, from);
2237 }
2238 i++;
2239 set >>= 1;
2240 }
2241 }
Dipankar Sarma4fb3a532005-09-16 19:28:13 -07002242 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002243 unlock_kernel();
2244}
2245EXPORT_SYMBOL(steal_locks);
2246
2247static int __init filelock_init(void)
2248{
2249 filelock_cache = kmem_cache_create("file_lock_cache",
2250 sizeof(struct file_lock), 0, SLAB_PANIC,
2251 init_once, NULL);
2252 return 0;
2253}
2254
2255core_initcall(filelock_init);