blob: d2c5306e3db0fec60a4f366d561cd30083d4a076 [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
675struct file_lock *
676posix_test_lock(struct file *filp, struct file_lock *fl)
677{
678 struct file_lock *cfl;
679
680 lock_kernel();
681 for (cfl = filp->f_dentry->d_inode->i_flock; cfl; cfl = cfl->fl_next) {
682 if (!IS_POSIX(cfl))
683 continue;
684 if (posix_locks_conflict(cfl, fl))
685 break;
686 }
687 unlock_kernel();
688
689 return (cfl);
690}
691
692EXPORT_SYMBOL(posix_test_lock);
693
694/* This function tests for deadlock condition before putting a process to
695 * sleep. The detection scheme is no longer recursive. Recursive was neat,
696 * but dangerous - we risked stack corruption if the lock data was bad, or
697 * if the recursion was too deep for any other reason.
698 *
699 * We rely on the fact that a task can only be on one lock's wait queue
700 * at a time. When we find blocked_task on a wait queue we can re-search
701 * with blocked_task equal to that queue's owner, until either blocked_task
702 * isn't found, or blocked_task is found on a queue owned by my_task.
703 *
704 * Note: the above assumption may not be true when handling lock requests
705 * from a broken NFS client. But broken NFS clients have a lot more to
706 * worry about than proper deadlock detection anyway... --okir
707 */
708int posix_locks_deadlock(struct file_lock *caller_fl,
709 struct file_lock *block_fl)
710{
711 struct list_head *tmp;
712
713next_task:
714 if (posix_same_owner(caller_fl, block_fl))
715 return 1;
716 list_for_each(tmp, &blocked_list) {
717 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
718 if (posix_same_owner(fl, block_fl)) {
719 fl = fl->fl_next;
720 block_fl = fl;
721 goto next_task;
722 }
723 }
724 return 0;
725}
726
727EXPORT_SYMBOL(posix_locks_deadlock);
728
729/* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
730 * at the head of the list, but that's secret knowledge known only to
731 * flock_lock_file and posix_lock_file.
732 */
733static int flock_lock_file(struct file *filp, struct file_lock *new_fl)
734{
735 struct file_lock **before;
736 struct inode * inode = filp->f_dentry->d_inode;
737 int error = 0;
738 int found = 0;
739
740 lock_kernel();
741 for_each_lock(inode, before) {
742 struct file_lock *fl = *before;
743 if (IS_POSIX(fl))
744 break;
745 if (IS_LEASE(fl))
746 continue;
747 if (filp != fl->fl_file)
748 continue;
749 if (new_fl->fl_type == fl->fl_type)
750 goto out;
751 found = 1;
752 locks_delete_lock(before);
753 break;
754 }
755 unlock_kernel();
756
757 if (new_fl->fl_type == F_UNLCK)
758 return 0;
759
760 /*
761 * If a higher-priority process was blocked on the old file lock,
762 * give it the opportunity to lock the file.
763 */
764 if (found)
765 cond_resched();
766
767 lock_kernel();
768 for_each_lock(inode, before) {
769 struct file_lock *fl = *before;
770 if (IS_POSIX(fl))
771 break;
772 if (IS_LEASE(fl))
773 continue;
774 if (!flock_locks_conflict(new_fl, fl))
775 continue;
776 error = -EAGAIN;
777 if (new_fl->fl_flags & FL_SLEEP) {
778 locks_insert_block(fl, new_fl);
779 }
780 goto out;
781 }
782 locks_insert_lock(&inode->i_flock, new_fl);
783 error = 0;
784
785out:
786 unlock_kernel();
787 return error;
788}
789
790EXPORT_SYMBOL(posix_lock_file);
791
792static int __posix_lock_file(struct inode *inode, struct file_lock *request)
793{
794 struct file_lock *fl;
795 struct file_lock *new_fl, *new_fl2;
796 struct file_lock *left = NULL;
797 struct file_lock *right = NULL;
798 struct file_lock **before;
799 int error, added = 0;
800
801 /*
802 * We may need two file_lock structures for this operation,
803 * so we get them in advance to avoid races.
804 */
805 new_fl = locks_alloc_lock();
806 new_fl2 = locks_alloc_lock();
807
808 lock_kernel();
809 if (request->fl_type != F_UNLCK) {
810 for_each_lock(inode, before) {
811 struct file_lock *fl = *before;
812 if (!IS_POSIX(fl))
813 continue;
814 if (!posix_locks_conflict(request, fl))
815 continue;
816 error = -EAGAIN;
817 if (!(request->fl_flags & FL_SLEEP))
818 goto out;
819 error = -EDEADLK;
820 if (posix_locks_deadlock(request, fl))
821 goto out;
822 error = -EAGAIN;
823 locks_insert_block(fl, request);
824 goto out;
825 }
826 }
827
828 /* If we're just looking for a conflict, we're done. */
829 error = 0;
830 if (request->fl_flags & FL_ACCESS)
831 goto out;
832
833 error = -ENOLCK; /* "no luck" */
834 if (!(new_fl && new_fl2))
835 goto out;
836
837 /*
838 * We've allocated the new locks in advance, so there are no
839 * errors possible (and no blocking operations) from here on.
840 *
841 * Find the first old lock with the same owner as the new lock.
842 */
843
844 before = &inode->i_flock;
845
846 /* First skip locks owned by other processes. */
847 while ((fl = *before) && (!IS_POSIX(fl) ||
848 !posix_same_owner(request, fl))) {
849 before = &fl->fl_next;
850 }
851
852 /* Process locks with this owner. */
853 while ((fl = *before) && posix_same_owner(request, fl)) {
854 /* Detect adjacent or overlapping regions (if same lock type)
855 */
856 if (request->fl_type == fl->fl_type) {
Olaf Kirch449231d2005-08-25 16:25:35 -0700857 /* In all comparisons of start vs end, use
858 * "start - 1" rather than "end + 1". If end
859 * is OFFSET_MAX, end + 1 will become negative.
860 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 if (fl->fl_end < request->fl_start - 1)
862 goto next_lock;
863 /* If the next lock in the list has entirely bigger
864 * addresses than the new one, insert the lock here.
865 */
Olaf Kirch449231d2005-08-25 16:25:35 -0700866 if (fl->fl_start - 1 > request->fl_end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 break;
868
869 /* If we come here, the new and old lock are of the
870 * same type and adjacent or overlapping. Make one
871 * lock yielding from the lower start address of both
872 * locks to the higher end address.
873 */
874 if (fl->fl_start > request->fl_start)
875 fl->fl_start = request->fl_start;
876 else
877 request->fl_start = fl->fl_start;
878 if (fl->fl_end < request->fl_end)
879 fl->fl_end = request->fl_end;
880 else
881 request->fl_end = fl->fl_end;
882 if (added) {
883 locks_delete_lock(before);
884 continue;
885 }
886 request = fl;
887 added = 1;
888 }
889 else {
890 /* Processing for different lock types is a bit
891 * more complex.
892 */
893 if (fl->fl_end < request->fl_start)
894 goto next_lock;
895 if (fl->fl_start > request->fl_end)
896 break;
897 if (request->fl_type == F_UNLCK)
898 added = 1;
899 if (fl->fl_start < request->fl_start)
900 left = fl;
901 /* If the next lock in the list has a higher end
902 * address than the new one, insert the new one here.
903 */
904 if (fl->fl_end > request->fl_end) {
905 right = fl;
906 break;
907 }
908 if (fl->fl_start >= request->fl_start) {
909 /* The new lock completely replaces an old
910 * one (This may happen several times).
911 */
912 if (added) {
913 locks_delete_lock(before);
914 continue;
915 }
916 /* Replace the old lock with the new one.
917 * Wake up anybody waiting for the old one,
918 * as the change in lock type might satisfy
919 * their needs.
920 */
921 locks_wake_up_blocks(fl);
922 fl->fl_start = request->fl_start;
923 fl->fl_end = request->fl_end;
924 fl->fl_type = request->fl_type;
Trond Myklebust47831f32006-03-20 13:44:05 -0500925 locks_release_private(fl);
926 locks_copy_private(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927 request = fl;
928 added = 1;
929 }
930 }
931 /* Go on to next lock.
932 */
933 next_lock:
934 before = &fl->fl_next;
935 }
936
937 error = 0;
938 if (!added) {
939 if (request->fl_type == F_UNLCK)
940 goto out;
941 locks_copy_lock(new_fl, request);
942 locks_insert_lock(before, new_fl);
943 new_fl = NULL;
944 }
945 if (right) {
946 if (left == right) {
947 /* The new lock breaks the old one in two pieces,
948 * so we have to use the second new lock.
949 */
950 left = new_fl2;
951 new_fl2 = NULL;
952 locks_copy_lock(left, right);
953 locks_insert_lock(before, left);
954 }
955 right->fl_start = request->fl_end + 1;
956 locks_wake_up_blocks(right);
957 }
958 if (left) {
959 left->fl_end = request->fl_start - 1;
960 locks_wake_up_blocks(left);
961 }
962 out:
963 unlock_kernel();
964 /*
965 * Free any unused locks.
966 */
967 if (new_fl)
968 locks_free_lock(new_fl);
969 if (new_fl2)
970 locks_free_lock(new_fl2);
971 return error;
972}
973
974/**
975 * posix_lock_file - Apply a POSIX-style lock to a file
976 * @filp: The file to apply the lock to
977 * @fl: The lock to be applied
978 *
979 * Add a POSIX style lock to a file.
980 * We merge adjacent & overlapping locks whenever possible.
981 * POSIX locks are sorted by owner task, then by starting address
982 */
983int posix_lock_file(struct file *filp, struct file_lock *fl)
984{
985 return __posix_lock_file(filp->f_dentry->d_inode, fl);
986}
987
988/**
989 * posix_lock_file_wait - Apply a POSIX-style lock to a file
990 * @filp: The file to apply the lock to
991 * @fl: The lock to be applied
992 *
993 * Add a POSIX style lock to a file.
994 * We merge adjacent & overlapping locks whenever possible.
995 * POSIX locks are sorted by owner task, then by starting address
996 */
997int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
998{
999 int error;
1000 might_sleep ();
1001 for (;;) {
1002 error = __posix_lock_file(filp->f_dentry->d_inode, fl);
1003 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1004 break;
1005 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1006 if (!error)
1007 continue;
1008
1009 locks_delete_block(fl);
1010 break;
1011 }
1012 return error;
1013}
1014EXPORT_SYMBOL(posix_lock_file_wait);
1015
1016/**
1017 * locks_mandatory_locked - Check for an active lock
1018 * @inode: the file to check
1019 *
1020 * Searches the inode's list of locks to find any POSIX locks which conflict.
1021 * This function is called from locks_verify_locked() only.
1022 */
1023int locks_mandatory_locked(struct inode *inode)
1024{
1025 fl_owner_t owner = current->files;
1026 struct file_lock *fl;
1027
1028 /*
1029 * Search the lock list for this inode for any POSIX locks.
1030 */
1031 lock_kernel();
1032 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1033 if (!IS_POSIX(fl))
1034 continue;
1035 if (fl->fl_owner != owner)
1036 break;
1037 }
1038 unlock_kernel();
1039 return fl ? -EAGAIN : 0;
1040}
1041
1042/**
1043 * locks_mandatory_area - Check for a conflicting lock
1044 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1045 * for shared
1046 * @inode: the file to check
1047 * @filp: how the file was opened (if it was)
1048 * @offset: start of area to check
1049 * @count: length of area to check
1050 *
1051 * Searches the inode's list of locks to find any POSIX locks which conflict.
1052 * This function is called from rw_verify_area() and
1053 * locks_verify_truncate().
1054 */
1055int locks_mandatory_area(int read_write, struct inode *inode,
1056 struct file *filp, loff_t offset,
1057 size_t count)
1058{
1059 struct file_lock fl;
1060 int error;
1061
1062 locks_init_lock(&fl);
1063 fl.fl_owner = current->files;
1064 fl.fl_pid = current->tgid;
1065 fl.fl_file = filp;
1066 fl.fl_flags = FL_POSIX | FL_ACCESS;
1067 if (filp && !(filp->f_flags & O_NONBLOCK))
1068 fl.fl_flags |= FL_SLEEP;
1069 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1070 fl.fl_start = offset;
1071 fl.fl_end = offset + count - 1;
1072
1073 for (;;) {
1074 error = __posix_lock_file(inode, &fl);
1075 if (error != -EAGAIN)
1076 break;
1077 if (!(fl.fl_flags & FL_SLEEP))
1078 break;
1079 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1080 if (!error) {
1081 /*
1082 * If we've been sleeping someone might have
1083 * changed the permissions behind our back.
1084 */
1085 if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID)
1086 continue;
1087 }
1088
1089 locks_delete_block(&fl);
1090 break;
1091 }
1092
1093 return error;
1094}
1095
1096EXPORT_SYMBOL(locks_mandatory_area);
1097
1098/* We already had a lease on this file; just change its type */
1099int lease_modify(struct file_lock **before, int arg)
1100{
1101 struct file_lock *fl = *before;
1102 int error = assign_type(fl, arg);
1103
1104 if (error)
1105 return error;
1106 locks_wake_up_blocks(fl);
1107 if (arg == F_UNLCK)
1108 locks_delete_lock(before);
1109 return 0;
1110}
1111
1112EXPORT_SYMBOL(lease_modify);
1113
1114static void time_out_leases(struct inode *inode)
1115{
1116 struct file_lock **before;
1117 struct file_lock *fl;
1118
1119 before = &inode->i_flock;
1120 while ((fl = *before) && IS_LEASE(fl) && (fl->fl_type & F_INPROGRESS)) {
1121 if ((fl->fl_break_time == 0)
1122 || time_before(jiffies, fl->fl_break_time)) {
1123 before = &fl->fl_next;
1124 continue;
1125 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 lease_modify(before, fl->fl_type & ~F_INPROGRESS);
1127 if (fl == *before) /* lease_modify may have freed fl */
1128 before = &fl->fl_next;
1129 }
1130}
1131
1132/**
1133 * __break_lease - revoke all outstanding leases on file
1134 * @inode: the inode of the file to return
1135 * @mode: the open mode (read or write)
1136 *
1137 * break_lease (inlined for speed) has checked there already
1138 * is a lease on this file. Leases are broken on a call to open()
1139 * or truncate(). This function can sleep unless you
1140 * specified %O_NONBLOCK to your open().
1141 */
1142int __break_lease(struct inode *inode, unsigned int mode)
1143{
1144 int error = 0, future;
1145 struct file_lock *new_fl, *flock;
1146 struct file_lock *fl;
1147 int alloc_err;
1148 unsigned long break_time;
1149 int i_have_this_lease = 0;
1150
1151 alloc_err = lease_alloc(NULL, mode & FMODE_WRITE ? F_WRLCK : F_RDLCK,
1152 &new_fl);
1153
1154 lock_kernel();
1155
1156 time_out_leases(inode);
1157
1158 flock = inode->i_flock;
1159 if ((flock == NULL) || !IS_LEASE(flock))
1160 goto out;
1161
1162 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1163 if (fl->fl_owner == current->files)
1164 i_have_this_lease = 1;
1165
1166 if (mode & FMODE_WRITE) {
1167 /* If we want write access, we have to revoke any lease. */
1168 future = F_UNLCK | F_INPROGRESS;
1169 } else if (flock->fl_type & F_INPROGRESS) {
1170 /* If the lease is already being broken, we just leave it */
1171 future = flock->fl_type;
1172 } else if (flock->fl_type & F_WRLCK) {
1173 /* Downgrade the exclusive lease to a read-only lease. */
1174 future = F_RDLCK | F_INPROGRESS;
1175 } else {
1176 /* the existing lease was read-only, so we can read too. */
1177 goto out;
1178 }
1179
1180 if (alloc_err && !i_have_this_lease && ((mode & O_NONBLOCK) == 0)) {
1181 error = alloc_err;
1182 goto out;
1183 }
1184
1185 break_time = 0;
1186 if (lease_break_time > 0) {
1187 break_time = jiffies + lease_break_time * HZ;
1188 if (break_time == 0)
1189 break_time++; /* so that 0 means no break time */
1190 }
1191
1192 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1193 if (fl->fl_type != future) {
1194 fl->fl_type = future;
1195 fl->fl_break_time = break_time;
1196 /* lease must have lmops break callback */
1197 fl->fl_lmops->fl_break(fl);
1198 }
1199 }
1200
1201 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1202 error = -EWOULDBLOCK;
1203 goto out;
1204 }
1205
1206restart:
1207 break_time = flock->fl_break_time;
1208 if (break_time != 0) {
1209 break_time -= jiffies;
1210 if (break_time == 0)
1211 break_time++;
1212 }
1213 error = locks_block_on_timeout(flock, new_fl, break_time);
1214 if (error >= 0) {
1215 if (error == 0)
1216 time_out_leases(inode);
1217 /* Wait for the next lease that has not been broken yet */
1218 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1219 flock = flock->fl_next) {
1220 if (flock->fl_type & F_INPROGRESS)
1221 goto restart;
1222 }
1223 error = 0;
1224 }
1225
1226out:
1227 unlock_kernel();
1228 if (!alloc_err)
1229 locks_free_lock(new_fl);
1230 return error;
1231}
1232
1233EXPORT_SYMBOL(__break_lease);
1234
1235/**
1236 * lease_get_mtime
1237 * @inode: the inode
1238 * @time: pointer to a timespec which will contain the last modified time
1239 *
1240 * This is to force NFS clients to flush their caches for files with
1241 * exclusive leases. The justification is that if someone has an
1242 * exclusive lease, then they could be modifiying it.
1243 */
1244void lease_get_mtime(struct inode *inode, struct timespec *time)
1245{
1246 struct file_lock *flock = inode->i_flock;
1247 if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK))
1248 *time = current_fs_time(inode->i_sb);
1249 else
1250 *time = inode->i_mtime;
1251}
1252
1253EXPORT_SYMBOL(lease_get_mtime);
1254
1255/**
1256 * fcntl_getlease - Enquire what lease is currently active
1257 * @filp: the file
1258 *
1259 * The value returned by this function will be one of
1260 * (if no lease break is pending):
1261 *
1262 * %F_RDLCK to indicate a shared lease is held.
1263 *
1264 * %F_WRLCK to indicate an exclusive lease is held.
1265 *
1266 * %F_UNLCK to indicate no lease is held.
1267 *
1268 * (if a lease break is pending):
1269 *
1270 * %F_RDLCK to indicate an exclusive lease needs to be
1271 * changed to a shared lease (or removed).
1272 *
1273 * %F_UNLCK to indicate the lease needs to be removed.
1274 *
1275 * XXX: sfr & willy disagree over whether F_INPROGRESS
1276 * should be returned to userspace.
1277 */
1278int fcntl_getlease(struct file *filp)
1279{
1280 struct file_lock *fl;
1281 int type = F_UNLCK;
1282
1283 lock_kernel();
1284 time_out_leases(filp->f_dentry->d_inode);
1285 for (fl = filp->f_dentry->d_inode->i_flock; fl && IS_LEASE(fl);
1286 fl = fl->fl_next) {
1287 if (fl->fl_file == filp) {
1288 type = fl->fl_type & ~F_INPROGRESS;
1289 break;
1290 }
1291 }
1292 unlock_kernel();
1293 return type;
1294}
1295
1296/**
1297 * __setlease - sets a lease on an open file
1298 * @filp: file pointer
1299 * @arg: type of lease to obtain
1300 * @flp: input - file_lock to use, output - file_lock inserted
1301 *
1302 * The (input) flp->fl_lmops->fl_break function is required
1303 * by break_lease().
1304 *
1305 * Called with kernel lock held.
1306 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07001307static int __setlease(struct file *filp, long arg, struct file_lock **flp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308{
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001309 struct file_lock *fl, **before, **my_before = NULL, *lease;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310 struct dentry *dentry = filp->f_dentry;
1311 struct inode *inode = dentry->d_inode;
1312 int error, rdlease_count = 0, wrlease_count = 0;
1313
1314 time_out_leases(inode);
1315
1316 error = -EINVAL;
1317 if (!flp || !(*flp) || !(*flp)->fl_lmops || !(*flp)->fl_lmops->fl_break)
1318 goto out;
1319
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001320 lease = *flp;
1321
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322 error = -EAGAIN;
1323 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1324 goto out;
1325 if ((arg == F_WRLCK)
1326 && ((atomic_read(&dentry->d_count) > 1)
1327 || (atomic_read(&inode->i_count) > 1)))
1328 goto out;
1329
1330 /*
1331 * At this point, we know that if there is an exclusive
1332 * lease on this file, then we hold it on this filp
1333 * (otherwise our open of this file would have blocked).
1334 * And if we are trying to acquire an exclusive lease,
1335 * then the file is not open by anyone (including us)
1336 * except for this filp.
1337 */
1338 for (before = &inode->i_flock;
1339 ((fl = *before) != NULL) && IS_LEASE(fl);
1340 before = &fl->fl_next) {
1341 if (lease->fl_lmops->fl_mylease(fl, lease))
1342 my_before = before;
1343 else if (fl->fl_type == (F_INPROGRESS | F_UNLCK))
1344 /*
1345 * Someone is in the process of opening this
1346 * file for writing so we may not take an
1347 * exclusive lease on it.
1348 */
1349 wrlease_count++;
1350 else
1351 rdlease_count++;
1352 }
1353
1354 if ((arg == F_RDLCK && (wrlease_count > 0)) ||
1355 (arg == F_WRLCK && ((rdlease_count + wrlease_count) > 0)))
1356 goto out;
1357
1358 if (my_before != NULL) {
1359 error = lease->fl_lmops->fl_change(my_before, arg);
1360 goto out;
1361 }
1362
1363 error = 0;
1364 if (arg == F_UNLCK)
1365 goto out;
1366
1367 error = -EINVAL;
1368 if (!leases_enable)
1369 goto out;
1370
1371 error = lease_alloc(filp, arg, &fl);
1372 if (error)
1373 goto out;
1374
1375 locks_copy_lock(fl, lease);
1376
1377 locks_insert_lock(before, fl);
1378
1379 *flp = fl;
1380out:
1381 return error;
1382}
1383
1384 /**
1385 * setlease - sets a lease on an open file
1386 * @filp: file pointer
1387 * @arg: type of lease to obtain
1388 * @lease: file_lock to use
1389 *
1390 * Call this to establish a lease on the file.
1391 * The fl_lmops fl_break function is required by break_lease
1392 */
1393
1394int setlease(struct file *filp, long arg, struct file_lock **lease)
1395{
1396 struct dentry *dentry = filp->f_dentry;
1397 struct inode *inode = dentry->d_inode;
1398 int error;
1399
1400 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1401 return -EACCES;
1402 if (!S_ISREG(inode->i_mode))
1403 return -EINVAL;
1404 error = security_file_lock(filp, arg);
1405 if (error)
1406 return error;
1407
1408 lock_kernel();
1409 error = __setlease(filp, arg, lease);
1410 unlock_kernel();
1411
1412 return error;
1413}
1414
1415EXPORT_SYMBOL(setlease);
1416
1417/**
1418 * fcntl_setlease - sets a lease on an open file
1419 * @fd: open file descriptor
1420 * @filp: file pointer
1421 * @arg: type of lease to obtain
1422 *
1423 * Call this fcntl to establish a lease on the file.
1424 * Note that you also need to call %F_SETSIG to
1425 * receive a signal when the lease is broken.
1426 */
1427int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1428{
1429 struct file_lock fl, *flp = &fl;
1430 struct dentry *dentry = filp->f_dentry;
1431 struct inode *inode = dentry->d_inode;
1432 int error;
1433
1434 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1435 return -EACCES;
1436 if (!S_ISREG(inode->i_mode))
1437 return -EINVAL;
1438 error = security_file_lock(filp, arg);
1439 if (error)
1440 return error;
1441
1442 locks_init_lock(&fl);
1443 error = lease_init(filp, arg, &fl);
1444 if (error)
1445 return error;
1446
1447 lock_kernel();
1448
1449 error = __setlease(filp, arg, &flp);
J. Bruce Fieldsdc15ae12005-11-10 19:08:00 -05001450 if (error || arg == F_UNLCK)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451 goto out_unlock;
1452
1453 error = fasync_helper(fd, filp, 1, &flp->fl_fasync);
1454 if (error < 0) {
1455 /* remove lease just inserted by __setlease */
1456 flp->fl_type = F_UNLCK | F_INPROGRESS;
1457 flp->fl_break_time = jiffies- 10;
1458 time_out_leases(inode);
1459 goto out_unlock;
1460 }
1461
1462 error = f_setown(filp, current->pid, 0);
1463out_unlock:
1464 unlock_kernel();
1465 return error;
1466}
1467
1468/**
1469 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1470 * @filp: The file to apply the lock to
1471 * @fl: The lock to be applied
1472 *
1473 * Add a FLOCK style lock to a file.
1474 */
1475int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1476{
1477 int error;
1478 might_sleep();
1479 for (;;) {
1480 error = flock_lock_file(filp, fl);
1481 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1482 break;
1483 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1484 if (!error)
1485 continue;
1486
1487 locks_delete_block(fl);
1488 break;
1489 }
1490 return error;
1491}
1492
1493EXPORT_SYMBOL(flock_lock_file_wait);
1494
1495/**
1496 * sys_flock: - flock() system call.
1497 * @fd: the file descriptor to lock.
1498 * @cmd: the type of lock to apply.
1499 *
1500 * Apply a %FL_FLOCK style lock to an open file descriptor.
1501 * The @cmd can be one of
1502 *
1503 * %LOCK_SH -- a shared lock.
1504 *
1505 * %LOCK_EX -- an exclusive lock.
1506 *
1507 * %LOCK_UN -- remove an existing lock.
1508 *
1509 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1510 *
1511 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1512 * processes read and write access respectively.
1513 */
1514asmlinkage long sys_flock(unsigned int fd, unsigned int cmd)
1515{
1516 struct file *filp;
1517 struct file_lock *lock;
1518 int can_sleep, unlock;
1519 int error;
1520
1521 error = -EBADF;
1522 filp = fget(fd);
1523 if (!filp)
1524 goto out;
1525
1526 can_sleep = !(cmd & LOCK_NB);
1527 cmd &= ~LOCK_NB;
1528 unlock = (cmd == LOCK_UN);
1529
1530 if (!unlock && !(cmd & LOCK_MAND) && !(filp->f_mode & 3))
1531 goto out_putf;
1532
1533 error = flock_make_lock(filp, &lock, cmd);
1534 if (error)
1535 goto out_putf;
1536 if (can_sleep)
1537 lock->fl_flags |= FL_SLEEP;
1538
1539 error = security_file_lock(filp, cmd);
1540 if (error)
1541 goto out_free;
1542
1543 if (filp->f_op && filp->f_op->flock)
1544 error = filp->f_op->flock(filp,
1545 (can_sleep) ? F_SETLKW : F_SETLK,
1546 lock);
1547 else
1548 error = flock_lock_file_wait(filp, lock);
1549
1550 out_free:
1551 if (list_empty(&lock->fl_link)) {
1552 locks_free_lock(lock);
1553 }
1554
1555 out_putf:
1556 fput(filp);
1557 out:
1558 return error;
1559}
1560
1561/* Report the first existing lock that would conflict with l.
1562 * This implements the F_GETLK command of fcntl().
1563 */
1564int fcntl_getlk(struct file *filp, struct flock __user *l)
1565{
1566 struct file_lock *fl, file_lock;
1567 struct flock flock;
1568 int error;
1569
1570 error = -EFAULT;
1571 if (copy_from_user(&flock, l, sizeof(flock)))
1572 goto out;
1573 error = -EINVAL;
1574 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1575 goto out;
1576
1577 error = flock_to_posix_lock(filp, &file_lock, &flock);
1578 if (error)
1579 goto out;
1580
1581 if (filp->f_op && filp->f_op->lock) {
1582 error = filp->f_op->lock(filp, F_GETLK, &file_lock);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00001583 if (file_lock.fl_ops && file_lock.fl_ops->fl_release_private)
1584 file_lock.fl_ops->fl_release_private(&file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585 if (error < 0)
1586 goto out;
1587 else
1588 fl = (file_lock.fl_type == F_UNLCK ? NULL : &file_lock);
1589 } else {
1590 fl = posix_test_lock(filp, &file_lock);
1591 }
1592
1593 flock.l_type = F_UNLCK;
1594 if (fl != NULL) {
1595 flock.l_pid = fl->fl_pid;
1596#if BITS_PER_LONG == 32
1597 /*
1598 * Make sure we can represent the posix lock via
1599 * legacy 32bit flock.
1600 */
1601 error = -EOVERFLOW;
1602 if (fl->fl_start > OFFT_OFFSET_MAX)
1603 goto out;
1604 if ((fl->fl_end != OFFSET_MAX)
1605 && (fl->fl_end > OFFT_OFFSET_MAX))
1606 goto out;
1607#endif
1608 flock.l_start = fl->fl_start;
1609 flock.l_len = fl->fl_end == OFFSET_MAX ? 0 :
1610 fl->fl_end - fl->fl_start + 1;
1611 flock.l_whence = 0;
1612 flock.l_type = fl->fl_type;
1613 }
1614 error = -EFAULT;
1615 if (!copy_to_user(l, &flock, sizeof(flock)))
1616 error = 0;
1617out:
1618 return error;
1619}
1620
1621/* Apply the lock described by l to an open file descriptor.
1622 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1623 */
Peter Staubachc2936212005-07-27 11:45:09 -07001624int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1625 struct flock __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626{
1627 struct file_lock *file_lock = locks_alloc_lock();
1628 struct flock flock;
1629 struct inode *inode;
1630 int error;
1631
1632 if (file_lock == NULL)
1633 return -ENOLCK;
1634
1635 /*
1636 * This might block, so we do it before checking the inode.
1637 */
1638 error = -EFAULT;
1639 if (copy_from_user(&flock, l, sizeof(flock)))
1640 goto out;
1641
1642 inode = filp->f_dentry->d_inode;
1643
1644 /* Don't allow mandatory locks on files that may be memory mapped
1645 * and shared.
1646 */
1647 if (IS_MANDLOCK(inode) &&
1648 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1649 mapping_writably_mapped(filp->f_mapping)) {
1650 error = -EAGAIN;
1651 goto out;
1652 }
1653
Peter Staubachc2936212005-07-27 11:45:09 -07001654again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655 error = flock_to_posix_lock(filp, file_lock, &flock);
1656 if (error)
1657 goto out;
1658 if (cmd == F_SETLKW) {
1659 file_lock->fl_flags |= FL_SLEEP;
1660 }
1661
1662 error = -EBADF;
1663 switch (flock.l_type) {
1664 case F_RDLCK:
1665 if (!(filp->f_mode & FMODE_READ))
1666 goto out;
1667 break;
1668 case F_WRLCK:
1669 if (!(filp->f_mode & FMODE_WRITE))
1670 goto out;
1671 break;
1672 case F_UNLCK:
1673 break;
1674 default:
1675 error = -EINVAL;
1676 goto out;
1677 }
1678
1679 error = security_file_lock(filp, file_lock->fl_type);
1680 if (error)
1681 goto out;
1682
Peter Staubachc2936212005-07-27 11:45:09 -07001683 if (filp->f_op && filp->f_op->lock != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684 error = filp->f_op->lock(filp, cmd, file_lock);
Peter Staubachc2936212005-07-27 11:45:09 -07001685 else {
1686 for (;;) {
1687 error = __posix_lock_file(inode, file_lock);
1688 if ((error != -EAGAIN) || (cmd == F_SETLK))
1689 break;
1690 error = wait_event_interruptible(file_lock->fl_wait,
1691 !file_lock->fl_next);
1692 if (!error)
1693 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694
Peter Staubachc2936212005-07-27 11:45:09 -07001695 locks_delete_block(file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696 break;
Peter Staubachc2936212005-07-27 11:45:09 -07001697 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698 }
1699
Peter Staubachc2936212005-07-27 11:45:09 -07001700 /*
1701 * Attempt to detect a close/fcntl race and recover by
1702 * releasing the lock that was just acquired.
1703 */
1704 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1705 flock.l_type = F_UNLCK;
1706 goto again;
1707 }
1708
1709out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 locks_free_lock(file_lock);
1711 return error;
1712}
1713
1714#if BITS_PER_LONG == 32
1715/* Report the first existing lock that would conflict with l.
1716 * This implements the F_GETLK command of fcntl().
1717 */
1718int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1719{
1720 struct file_lock *fl, file_lock;
1721 struct flock64 flock;
1722 int error;
1723
1724 error = -EFAULT;
1725 if (copy_from_user(&flock, l, sizeof(flock)))
1726 goto out;
1727 error = -EINVAL;
1728 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1729 goto out;
1730
1731 error = flock64_to_posix_lock(filp, &file_lock, &flock);
1732 if (error)
1733 goto out;
1734
1735 if (filp->f_op && filp->f_op->lock) {
1736 error = filp->f_op->lock(filp, F_GETLK, &file_lock);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00001737 if (file_lock.fl_ops && file_lock.fl_ops->fl_release_private)
1738 file_lock.fl_ops->fl_release_private(&file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739 if (error < 0)
1740 goto out;
1741 else
1742 fl = (file_lock.fl_type == F_UNLCK ? NULL : &file_lock);
1743 } else {
1744 fl = posix_test_lock(filp, &file_lock);
1745 }
1746
1747 flock.l_type = F_UNLCK;
1748 if (fl != NULL) {
1749 flock.l_pid = fl->fl_pid;
1750 flock.l_start = fl->fl_start;
1751 flock.l_len = fl->fl_end == OFFSET_MAX ? 0 :
1752 fl->fl_end - fl->fl_start + 1;
1753 flock.l_whence = 0;
1754 flock.l_type = fl->fl_type;
1755 }
1756 error = -EFAULT;
1757 if (!copy_to_user(l, &flock, sizeof(flock)))
1758 error = 0;
1759
1760out:
1761 return error;
1762}
1763
1764/* Apply the lock described by l to an open file descriptor.
1765 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1766 */
Peter Staubachc2936212005-07-27 11:45:09 -07001767int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
1768 struct flock64 __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769{
1770 struct file_lock *file_lock = locks_alloc_lock();
1771 struct flock64 flock;
1772 struct inode *inode;
1773 int error;
1774
1775 if (file_lock == NULL)
1776 return -ENOLCK;
1777
1778 /*
1779 * This might block, so we do it before checking the inode.
1780 */
1781 error = -EFAULT;
1782 if (copy_from_user(&flock, l, sizeof(flock)))
1783 goto out;
1784
1785 inode = filp->f_dentry->d_inode;
1786
1787 /* Don't allow mandatory locks on files that may be memory mapped
1788 * and shared.
1789 */
1790 if (IS_MANDLOCK(inode) &&
1791 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1792 mapping_writably_mapped(filp->f_mapping)) {
1793 error = -EAGAIN;
1794 goto out;
1795 }
1796
Peter Staubachc2936212005-07-27 11:45:09 -07001797again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798 error = flock64_to_posix_lock(filp, file_lock, &flock);
1799 if (error)
1800 goto out;
1801 if (cmd == F_SETLKW64) {
1802 file_lock->fl_flags |= FL_SLEEP;
1803 }
1804
1805 error = -EBADF;
1806 switch (flock.l_type) {
1807 case F_RDLCK:
1808 if (!(filp->f_mode & FMODE_READ))
1809 goto out;
1810 break;
1811 case F_WRLCK:
1812 if (!(filp->f_mode & FMODE_WRITE))
1813 goto out;
1814 break;
1815 case F_UNLCK:
1816 break;
1817 default:
1818 error = -EINVAL;
1819 goto out;
1820 }
1821
1822 error = security_file_lock(filp, file_lock->fl_type);
1823 if (error)
1824 goto out;
1825
Peter Staubachc2936212005-07-27 11:45:09 -07001826 if (filp->f_op && filp->f_op->lock != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827 error = filp->f_op->lock(filp, cmd, file_lock);
Peter Staubachc2936212005-07-27 11:45:09 -07001828 else {
1829 for (;;) {
1830 error = __posix_lock_file(inode, file_lock);
1831 if ((error != -EAGAIN) || (cmd == F_SETLK64))
1832 break;
1833 error = wait_event_interruptible(file_lock->fl_wait,
1834 !file_lock->fl_next);
1835 if (!error)
1836 continue;
1837
1838 locks_delete_block(file_lock);
1839 break;
1840 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841 }
1842
Peter Staubachc2936212005-07-27 11:45:09 -07001843 /*
1844 * Attempt to detect a close/fcntl race and recover by
1845 * releasing the lock that was just acquired.
1846 */
1847 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1848 flock.l_type = F_UNLCK;
1849 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850 }
1851
1852out:
1853 locks_free_lock(file_lock);
1854 return error;
1855}
1856#endif /* BITS_PER_LONG == 32 */
1857
1858/*
1859 * This function is called when the file is being removed
1860 * from the task's fd array. POSIX locks belonging to this task
1861 * are deleted at this time.
1862 */
1863void locks_remove_posix(struct file *filp, fl_owner_t owner)
1864{
1865 struct file_lock lock, **before;
1866
1867 /*
1868 * If there are no locks held on this file, we don't need to call
1869 * posix_lock_file(). Another process could be setting a lock on this
1870 * file at the same time, but we wouldn't remove that lock anyway.
1871 */
1872 before = &filp->f_dentry->d_inode->i_flock;
1873 if (*before == NULL)
1874 return;
1875
1876 lock.fl_type = F_UNLCK;
1877 lock.fl_flags = FL_POSIX;
1878 lock.fl_start = 0;
1879 lock.fl_end = OFFSET_MAX;
1880 lock.fl_owner = owner;
1881 lock.fl_pid = current->tgid;
1882 lock.fl_file = filp;
1883 lock.fl_ops = NULL;
1884 lock.fl_lmops = NULL;
1885
1886 if (filp->f_op && filp->f_op->lock != NULL) {
1887 filp->f_op->lock(filp, F_SETLK, &lock);
1888 goto out;
1889 }
1890
1891 /* Can't use posix_lock_file here; we need to remove it no matter
1892 * which pid we have.
1893 */
1894 lock_kernel();
1895 while (*before != NULL) {
1896 struct file_lock *fl = *before;
1897 if (IS_POSIX(fl) && posix_same_owner(fl, &lock)) {
1898 locks_delete_lock(before);
1899 continue;
1900 }
1901 before = &fl->fl_next;
1902 }
1903 unlock_kernel();
1904out:
1905 if (lock.fl_ops && lock.fl_ops->fl_release_private)
1906 lock.fl_ops->fl_release_private(&lock);
1907}
1908
1909EXPORT_SYMBOL(locks_remove_posix);
1910
1911/*
1912 * This function is called on the last close of an open file.
1913 */
1914void locks_remove_flock(struct file *filp)
1915{
1916 struct inode * inode = filp->f_dentry->d_inode;
1917 struct file_lock *fl;
1918 struct file_lock **before;
1919
1920 if (!inode->i_flock)
1921 return;
1922
1923 if (filp->f_op && filp->f_op->flock) {
1924 struct file_lock fl = {
1925 .fl_pid = current->tgid,
1926 .fl_file = filp,
1927 .fl_flags = FL_FLOCK,
1928 .fl_type = F_UNLCK,
1929 .fl_end = OFFSET_MAX,
1930 };
1931 filp->f_op->flock(filp, F_SETLKW, &fl);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00001932 if (fl.fl_ops && fl.fl_ops->fl_release_private)
1933 fl.fl_ops->fl_release_private(&fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934 }
1935
1936 lock_kernel();
1937 before = &inode->i_flock;
1938
1939 while ((fl = *before) != NULL) {
1940 if (fl->fl_file == filp) {
Peter Staubachc2936212005-07-27 11:45:09 -07001941 if (IS_FLOCK(fl)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942 locks_delete_lock(before);
1943 continue;
1944 }
1945 if (IS_LEASE(fl)) {
1946 lease_modify(before, F_UNLCK);
1947 continue;
1948 }
1949 /* What? */
1950 BUG();
1951 }
1952 before = &fl->fl_next;
1953 }
1954 unlock_kernel();
1955}
1956
1957/**
1958 * posix_block_lock - blocks waiting for a file lock
1959 * @blocker: the lock which is blocking
1960 * @waiter: the lock which conflicts and has to wait
1961 *
1962 * lockd needs to block waiting for locks.
1963 */
1964void
1965posix_block_lock(struct file_lock *blocker, struct file_lock *waiter)
1966{
1967 locks_insert_block(blocker, waiter);
1968}
1969
1970EXPORT_SYMBOL(posix_block_lock);
1971
1972/**
1973 * posix_unblock_lock - stop waiting for a file lock
1974 * @filp: how the file was opened
1975 * @waiter: the lock which was waiting
1976 *
1977 * lockd needs to block waiting for locks.
1978 */
J. Bruce Fields64a318e2006-01-03 09:55:46 +01001979int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980posix_unblock_lock(struct file *filp, struct file_lock *waiter)
1981{
J. Bruce Fields64a318e2006-01-03 09:55:46 +01001982 int status = 0;
1983
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984 lock_kernel();
J. Bruce Fields5996a292006-01-03 09:55:44 +01001985 if (waiter->fl_next)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986 __locks_delete_block(waiter);
J. Bruce Fields64a318e2006-01-03 09:55:46 +01001987 else
1988 status = -ENOENT;
J. Bruce Fields5996a292006-01-03 09:55:44 +01001989 unlock_kernel();
J. Bruce Fields64a318e2006-01-03 09:55:46 +01001990 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991}
1992
1993EXPORT_SYMBOL(posix_unblock_lock);
1994
1995static void lock_get_status(char* out, struct file_lock *fl, int id, char *pfx)
1996{
1997 struct inode *inode = NULL;
1998
1999 if (fl->fl_file != NULL)
2000 inode = fl->fl_file->f_dentry->d_inode;
2001
2002 out += sprintf(out, "%d:%s ", id, pfx);
2003 if (IS_POSIX(fl)) {
2004 out += sprintf(out, "%6s %s ",
2005 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2006 (inode == NULL) ? "*NOINODE*" :
2007 (IS_MANDLOCK(inode) &&
2008 (inode->i_mode & (S_IXGRP | S_ISGID)) == S_ISGID) ?
2009 "MANDATORY" : "ADVISORY ");
2010 } else if (IS_FLOCK(fl)) {
2011 if (fl->fl_type & LOCK_MAND) {
2012 out += sprintf(out, "FLOCK MSNFS ");
2013 } else {
2014 out += sprintf(out, "FLOCK ADVISORY ");
2015 }
2016 } else if (IS_LEASE(fl)) {
2017 out += sprintf(out, "LEASE ");
2018 if (fl->fl_type & F_INPROGRESS)
2019 out += sprintf(out, "BREAKING ");
2020 else if (fl->fl_file)
2021 out += sprintf(out, "ACTIVE ");
2022 else
2023 out += sprintf(out, "BREAKER ");
2024 } else {
2025 out += sprintf(out, "UNKNOWN UNKNOWN ");
2026 }
2027 if (fl->fl_type & LOCK_MAND) {
2028 out += sprintf(out, "%s ",
2029 (fl->fl_type & LOCK_READ)
2030 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2031 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2032 } else {
2033 out += sprintf(out, "%s ",
2034 (fl->fl_type & F_INPROGRESS)
2035 ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ "
2036 : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ ");
2037 }
2038 if (inode) {
2039#ifdef WE_CAN_BREAK_LSLK_NOW
2040 out += sprintf(out, "%d %s:%ld ", fl->fl_pid,
2041 inode->i_sb->s_id, inode->i_ino);
2042#else
2043 /* userspace relies on this representation of dev_t ;-( */
2044 out += sprintf(out, "%d %02x:%02x:%ld ", fl->fl_pid,
2045 MAJOR(inode->i_sb->s_dev),
2046 MINOR(inode->i_sb->s_dev), inode->i_ino);
2047#endif
2048 } else {
2049 out += sprintf(out, "%d <none>:0 ", fl->fl_pid);
2050 }
2051 if (IS_POSIX(fl)) {
2052 if (fl->fl_end == OFFSET_MAX)
2053 out += sprintf(out, "%Ld EOF\n", fl->fl_start);
2054 else
2055 out += sprintf(out, "%Ld %Ld\n", fl->fl_start,
2056 fl->fl_end);
2057 } else {
2058 out += sprintf(out, "0 EOF\n");
2059 }
2060}
2061
2062static void move_lock_status(char **p, off_t* pos, off_t offset)
2063{
2064 int len;
2065 len = strlen(*p);
2066 if(*pos >= offset) {
2067 /* the complete line is valid */
2068 *p += len;
2069 *pos += len;
2070 return;
2071 }
2072 if(*pos+len > offset) {
2073 /* use the second part of the line */
2074 int i = offset-*pos;
2075 memmove(*p,*p+i,len-i);
2076 *p += len-i;
2077 *pos += len;
2078 return;
2079 }
2080 /* discard the complete line */
2081 *pos += len;
2082}
2083
2084/**
2085 * get_locks_status - reports lock usage in /proc/locks
2086 * @buffer: address in userspace to write into
2087 * @start: ?
2088 * @offset: how far we are through the buffer
2089 * @length: how much to read
2090 */
2091
2092int get_locks_status(char *buffer, char **start, off_t offset, int length)
2093{
2094 struct list_head *tmp;
2095 char *q = buffer;
2096 off_t pos = 0;
2097 int i = 0;
2098
2099 lock_kernel();
2100 list_for_each(tmp, &file_lock_list) {
2101 struct list_head *btmp;
2102 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
2103 lock_get_status(q, fl, ++i, "");
2104 move_lock_status(&q, &pos, offset);
2105
2106 if(pos >= offset+length)
2107 goto done;
2108
2109 list_for_each(btmp, &fl->fl_block) {
2110 struct file_lock *bfl = list_entry(btmp,
2111 struct file_lock, fl_block);
2112 lock_get_status(q, bfl, i, " ->");
2113 move_lock_status(&q, &pos, offset);
2114
2115 if(pos >= offset+length)
2116 goto done;
2117 }
2118 }
2119done:
2120 unlock_kernel();
2121 *start = buffer;
2122 if(q-buffer < length)
2123 return (q-buffer);
2124 return length;
2125}
2126
2127/**
2128 * lock_may_read - checks that the region is free of locks
2129 * @inode: the inode that is being read
2130 * @start: the first byte to read
2131 * @len: the number of bytes to read
2132 *
2133 * Emulates Windows locking requirements. Whole-file
2134 * mandatory locks (share modes) can prohibit a read and
2135 * byte-range POSIX locks can prohibit a read if they overlap.
2136 *
2137 * N.B. this function is only ever called
2138 * from knfsd and ownership of locks is never checked.
2139 */
2140int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2141{
2142 struct file_lock *fl;
2143 int result = 1;
2144 lock_kernel();
2145 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2146 if (IS_POSIX(fl)) {
2147 if (fl->fl_type == F_RDLCK)
2148 continue;
2149 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2150 continue;
2151 } else if (IS_FLOCK(fl)) {
2152 if (!(fl->fl_type & LOCK_MAND))
2153 continue;
2154 if (fl->fl_type & LOCK_READ)
2155 continue;
2156 } else
2157 continue;
2158 result = 0;
2159 break;
2160 }
2161 unlock_kernel();
2162 return result;
2163}
2164
2165EXPORT_SYMBOL(lock_may_read);
2166
2167/**
2168 * lock_may_write - checks that the region is free of locks
2169 * @inode: the inode that is being written
2170 * @start: the first byte to write
2171 * @len: the number of bytes to write
2172 *
2173 * Emulates Windows locking requirements. Whole-file
2174 * mandatory locks (share modes) can prohibit a write and
2175 * byte-range POSIX locks can prohibit a write if they overlap.
2176 *
2177 * N.B. this function is only ever called
2178 * from knfsd and ownership of locks is never checked.
2179 */
2180int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2181{
2182 struct file_lock *fl;
2183 int result = 1;
2184 lock_kernel();
2185 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2186 if (IS_POSIX(fl)) {
2187 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2188 continue;
2189 } else if (IS_FLOCK(fl)) {
2190 if (!(fl->fl_type & LOCK_MAND))
2191 continue;
2192 if (fl->fl_type & LOCK_WRITE)
2193 continue;
2194 } else
2195 continue;
2196 result = 0;
2197 break;
2198 }
2199 unlock_kernel();
2200 return result;
2201}
2202
2203EXPORT_SYMBOL(lock_may_write);
2204
2205static inline void __steal_locks(struct file *file, fl_owner_t from)
2206{
2207 struct inode *inode = file->f_dentry->d_inode;
2208 struct file_lock *fl = inode->i_flock;
2209
2210 while (fl) {
2211 if (fl->fl_file == file && fl->fl_owner == from)
2212 fl->fl_owner = current->files;
2213 fl = fl->fl_next;
2214 }
2215}
2216
2217/* When getting ready for executing a binary, we make sure that current
2218 * has a files_struct on its own. Before dropping the old files_struct,
2219 * we take over ownership of all locks for all file descriptors we own.
2220 * Note that we may accidentally steal a lock for a file that a sibling
2221 * has created since the unshare_files() call.
2222 */
2223void steal_locks(fl_owner_t from)
2224{
2225 struct files_struct *files = current->files;
2226 int i, j;
Dipankar Sarmabadf1662005-09-09 13:04:10 -07002227 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228
2229 if (from == files)
2230 return;
2231
2232 lock_kernel();
2233 j = 0;
Dipankar Sarma4fb3a532005-09-16 19:28:13 -07002234 rcu_read_lock();
Dipankar Sarmabadf1662005-09-09 13:04:10 -07002235 fdt = files_fdtable(files);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002236 for (;;) {
2237 unsigned long set;
2238 i = j * __NFDBITS;
Dipankar Sarmabadf1662005-09-09 13:04:10 -07002239 if (i >= fdt->max_fdset || i >= fdt->max_fds)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240 break;
Dipankar Sarmabadf1662005-09-09 13:04:10 -07002241 set = fdt->open_fds->fds_bits[j++];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242 while (set) {
2243 if (set & 1) {
Dipankar Sarmabadf1662005-09-09 13:04:10 -07002244 struct file *file = fdt->fd[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245 if (file)
2246 __steal_locks(file, from);
2247 }
2248 i++;
2249 set >>= 1;
2250 }
2251 }
Dipankar Sarma4fb3a532005-09-16 19:28:13 -07002252 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253 unlock_kernel();
2254}
2255EXPORT_SYMBOL(steal_locks);
2256
2257static int __init filelock_init(void)
2258{
2259 filelock_cache = kmem_cache_create("file_lock_cache",
2260 sizeof(struct file_lock), 0, SLAB_PANIC,
2261 init_once, NULL);
2262 return 0;
2263}
2264
2265core_initcall(filelock_init);