blob: fb32d6218e213e3e31c5fe000dcb5ef362b3f176 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/locks.c
3 *
4 * Provide support for fcntl()'s F_GETLK, F_SETLK, and F_SETLKW calls.
5 * Doug Evans (dje@spiff.uucp), August 07, 1992
6 *
7 * Deadlock detection added.
8 * FIXME: one thing isn't handled yet:
9 * - mandatory locks (requires lots of changes elsewhere)
10 * Kelly Carmichael (kelly@[142.24.8.65]), September 17, 1994.
11 *
12 * Miscellaneous edits, and a total rewrite of posix_lock_file() code.
13 * Kai Petzke (wpp@marie.physik.tu-berlin.de), 1994
14 *
15 * Converted file_lock_table to a linked list from an array, which eliminates
16 * the limits on how many active file locks are open.
17 * Chad Page (pageone@netcom.com), November 27, 1994
18 *
19 * Removed dependency on file descriptors. dup()'ed file descriptors now
20 * get the same locks as the original file descriptors, and a close() on
21 * any file descriptor removes ALL the locks on the file for the current
22 * process. Since locks still depend on the process id, locks are inherited
23 * after an exec() but not after a fork(). This agrees with POSIX, and both
24 * BSD and SVR4 practice.
25 * Andy Walker (andy@lysaker.kvaerner.no), February 14, 1995
26 *
27 * Scrapped free list which is redundant now that we allocate locks
28 * dynamically with kmalloc()/kfree().
29 * Andy Walker (andy@lysaker.kvaerner.no), February 21, 1995
30 *
31 * Implemented two lock personalities - FL_FLOCK and FL_POSIX.
32 *
33 * FL_POSIX locks are created with calls to fcntl() and lockf() through the
34 * fcntl() system call. They have the semantics described above.
35 *
36 * FL_FLOCK locks are created with calls to flock(), through the flock()
37 * system call, which is new. Old C libraries implement flock() via fcntl()
38 * and will continue to use the old, broken implementation.
39 *
40 * FL_FLOCK locks follow the 4.4 BSD flock() semantics. They are associated
41 * with a file pointer (filp). As a result they can be shared by a parent
42 * process and its children after a fork(). They are removed when the last
43 * file descriptor referring to the file pointer is closed (unless explicitly
44 * unlocked).
45 *
46 * FL_FLOCK locks never deadlock, an existing lock is always removed before
47 * upgrading from shared to exclusive (or vice versa). When this happens
48 * any processes blocked by the current lock are woken up and allowed to
49 * run before the new lock is applied.
50 * Andy Walker (andy@lysaker.kvaerner.no), June 09, 1995
51 *
52 * Removed some race conditions in flock_lock_file(), marked other possible
53 * races. Just grep for FIXME to see them.
54 * Dmitry Gorodchanin (pgmdsg@ibi.com), February 09, 1996.
55 *
56 * Addressed Dmitry's concerns. Deadlock checking no longer recursive.
57 * Lock allocation changed to GFP_ATOMIC as we can't afford to sleep
58 * once we've checked for blocking and deadlocking.
59 * Andy Walker (andy@lysaker.kvaerner.no), April 03, 1996.
60 *
61 * Initial implementation of mandatory locks. SunOS turned out to be
62 * a rotten model, so I implemented the "obvious" semantics.
63 * See 'Documentation/mandatory.txt' for details.
64 * Andy Walker (andy@lysaker.kvaerner.no), April 06, 1996.
65 *
66 * Don't allow mandatory locks on mmap()'ed files. Added simple functions to
67 * check if a file has mandatory locks, used by mmap(), open() and creat() to
68 * see if system call should be rejected. Ref. HP-UX/SunOS/Solaris Reference
69 * Manual, Section 2.
70 * Andy Walker (andy@lysaker.kvaerner.no), April 09, 1996.
71 *
72 * Tidied up block list handling. Added '/proc/locks' interface.
73 * Andy Walker (andy@lysaker.kvaerner.no), April 24, 1996.
74 *
75 * Fixed deadlock condition for pathological code that mixes calls to
76 * flock() and fcntl().
77 * Andy Walker (andy@lysaker.kvaerner.no), April 29, 1996.
78 *
79 * Allow only one type of locking scheme (FL_POSIX or FL_FLOCK) to be in use
80 * for a given file at a time. Changed the CONFIG_LOCK_MANDATORY scheme to
81 * guarantee sensible behaviour in the case where file system modules might
82 * be compiled with different options than the kernel itself.
83 * Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996.
84 *
85 * Added a couple of missing wake_up() calls. Thanks to Thomas Meckel
86 * (Thomas.Meckel@mni.fh-giessen.de) for spotting this.
87 * Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996.
88 *
89 * Changed FL_POSIX locks to use the block list in the same way as FL_FLOCK
90 * locks. Changed process synchronisation to avoid dereferencing locks that
91 * have already been freed.
92 * Andy Walker (andy@lysaker.kvaerner.no), Sep 21, 1996.
93 *
94 * Made the block list a circular list to minimise searching in the list.
95 * Andy Walker (andy@lysaker.kvaerner.no), Sep 25, 1996.
96 *
97 * Made mandatory locking a mount option. Default is not to allow mandatory
98 * locking.
99 * Andy Walker (andy@lysaker.kvaerner.no), Oct 04, 1996.
100 *
101 * Some adaptations for NFS support.
102 * Olaf Kirch (okir@monad.swb.de), Dec 1996,
103 *
104 * Fixed /proc/locks interface so that we can't overrun the buffer we are handed.
105 * Andy Walker (andy@lysaker.kvaerner.no), May 12, 1997.
106 *
107 * Use slab allocator instead of kmalloc/kfree.
108 * Use generic list implementation from <linux/list.h>.
109 * Sped up posix_locks_deadlock by only considering blocked locks.
110 * Matthew Wilcox <willy@debian.org>, March, 2000.
111 *
112 * Leases and LOCK_MAND
113 * Matthew Wilcox <willy@debian.org>, June, 2000.
114 * Stephen Rothwell <sfr@canb.auug.org.au>, June, 2000.
115 */
116
117#include <linux/capability.h>
118#include <linux/file.h>
119#include <linux/fs.h>
120#include <linux/init.h>
121#include <linux/module.h>
122#include <linux/security.h>
123#include <linux/slab.h>
124#include <linux/smp_lock.h>
125#include <linux/syscalls.h>
126#include <linux/time.h>
Dipankar Sarma4fb3a532005-09-16 19:28:13 -0700127#include <linux/rcupdate.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128
129#include <asm/semaphore.h>
130#include <asm/uaccess.h>
131
132#define IS_POSIX(fl) (fl->fl_flags & FL_POSIX)
133#define IS_FLOCK(fl) (fl->fl_flags & FL_FLOCK)
134#define IS_LEASE(fl) (fl->fl_flags & FL_LEASE)
135
136int leases_enable = 1;
137int lease_break_time = 45;
138
139#define for_each_lock(inode, lockp) \
140 for (lockp = &inode->i_flock; *lockp != NULL; lockp = &(*lockp)->fl_next)
141
142LIST_HEAD(file_lock_list);
143
144EXPORT_SYMBOL(file_lock_list);
145
146static LIST_HEAD(blocked_list);
147
148static kmem_cache_t *filelock_cache;
149
150/* Allocate an empty lock structure. */
151static struct file_lock *locks_alloc_lock(void)
152{
153 return kmem_cache_alloc(filelock_cache, SLAB_KERNEL);
154}
155
156/* Free a lock which is not in use. */
157static inline void locks_free_lock(struct file_lock *fl)
158{
159 if (fl == NULL) {
160 BUG();
161 return;
162 }
163 if (waitqueue_active(&fl->fl_wait))
164 panic("Attempting to free lock with active wait queue");
165
166 if (!list_empty(&fl->fl_block))
167 panic("Attempting to free lock with active block list");
168
169 if (!list_empty(&fl->fl_link))
170 panic("Attempting to free lock on active lock list");
171
172 if (fl->fl_ops) {
173 if (fl->fl_ops->fl_release_private)
174 fl->fl_ops->fl_release_private(fl);
175 fl->fl_ops = NULL;
176 }
177
178 if (fl->fl_lmops) {
179 if (fl->fl_lmops->fl_release_private)
180 fl->fl_lmops->fl_release_private(fl);
181 fl->fl_lmops = NULL;
182 }
183
184 kmem_cache_free(filelock_cache, fl);
185}
186
187void locks_init_lock(struct file_lock *fl)
188{
189 INIT_LIST_HEAD(&fl->fl_link);
190 INIT_LIST_HEAD(&fl->fl_block);
191 init_waitqueue_head(&fl->fl_wait);
192 fl->fl_next = NULL;
193 fl->fl_fasync = NULL;
194 fl->fl_owner = NULL;
195 fl->fl_pid = 0;
196 fl->fl_file = NULL;
197 fl->fl_flags = 0;
198 fl->fl_type = 0;
199 fl->fl_start = fl->fl_end = 0;
200 fl->fl_ops = NULL;
201 fl->fl_lmops = NULL;
202}
203
204EXPORT_SYMBOL(locks_init_lock);
205
206/*
207 * Initialises the fields of the file lock which are invariant for
208 * free file_locks.
209 */
210static void init_once(void *foo, kmem_cache_t *cache, unsigned long flags)
211{
212 struct file_lock *lock = (struct file_lock *) foo;
213
214 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) !=
215 SLAB_CTOR_CONSTRUCTOR)
216 return;
217
218 locks_init_lock(lock);
219}
220
221/*
222 * Initialize a new lock from an existing file_lock structure.
223 */
224void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
225{
226 new->fl_owner = fl->fl_owner;
227 new->fl_pid = fl->fl_pid;
228 new->fl_file = fl->fl_file;
229 new->fl_flags = fl->fl_flags;
230 new->fl_type = fl->fl_type;
231 new->fl_start = fl->fl_start;
232 new->fl_end = fl->fl_end;
233 new->fl_ops = fl->fl_ops;
234 new->fl_lmops = fl->fl_lmops;
235 if (fl->fl_ops && fl->fl_ops->fl_copy_lock)
236 fl->fl_ops->fl_copy_lock(new, fl);
237 if (fl->fl_lmops && fl->fl_lmops->fl_copy_lock)
238 fl->fl_lmops->fl_copy_lock(new, fl);
239}
240
241EXPORT_SYMBOL(locks_copy_lock);
242
243static inline int flock_translate_cmd(int cmd) {
244 if (cmd & LOCK_MAND)
245 return cmd & (LOCK_MAND | LOCK_RW);
246 switch (cmd) {
247 case LOCK_SH:
248 return F_RDLCK;
249 case LOCK_EX:
250 return F_WRLCK;
251 case LOCK_UN:
252 return F_UNLCK;
253 }
254 return -EINVAL;
255}
256
257/* Fill in a file_lock structure with an appropriate FLOCK lock. */
258static int flock_make_lock(struct file *filp, struct file_lock **lock,
259 unsigned int cmd)
260{
261 struct file_lock *fl;
262 int type = flock_translate_cmd(cmd);
263 if (type < 0)
264 return type;
265
266 fl = locks_alloc_lock();
267 if (fl == NULL)
268 return -ENOMEM;
269
270 fl->fl_file = filp;
271 fl->fl_pid = current->tgid;
272 fl->fl_flags = FL_FLOCK;
273 fl->fl_type = type;
274 fl->fl_end = OFFSET_MAX;
275
276 *lock = fl;
277 return 0;
278}
279
280static int assign_type(struct file_lock *fl, int type)
281{
282 switch (type) {
283 case F_RDLCK:
284 case F_WRLCK:
285 case F_UNLCK:
286 fl->fl_type = type;
287 break;
288 default:
289 return -EINVAL;
290 }
291 return 0;
292}
293
294/* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
295 * style lock.
296 */
297static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
298 struct flock *l)
299{
300 off_t start, end;
301
302 switch (l->l_whence) {
303 case 0: /*SEEK_SET*/
304 start = 0;
305 break;
306 case 1: /*SEEK_CUR*/
307 start = filp->f_pos;
308 break;
309 case 2: /*SEEK_END*/
310 start = i_size_read(filp->f_dentry->d_inode);
311 break;
312 default:
313 return -EINVAL;
314 }
315
316 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
317 POSIX-2001 defines it. */
318 start += l->l_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319 if (start < 0)
320 return -EINVAL;
Trond Myklebust4c780a42005-10-18 14:20:21 -0700321 fl->fl_end = OFFSET_MAX;
322 if (l->l_len > 0) {
323 end = start + l->l_len - 1;
324 fl->fl_end = end;
325 } else if (l->l_len < 0) {
326 end = start - 1;
327 fl->fl_end = end;
328 start += l->l_len;
329 if (start < 0)
330 return -EINVAL;
331 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 fl->fl_start = start; /* we record the absolute position */
Trond Myklebust4c780a42005-10-18 14:20:21 -0700333 if (fl->fl_end < fl->fl_start)
334 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335
336 fl->fl_owner = current->files;
337 fl->fl_pid = current->tgid;
338 fl->fl_file = filp;
339 fl->fl_flags = FL_POSIX;
340 fl->fl_ops = NULL;
341 fl->fl_lmops = NULL;
342
343 return assign_type(fl, l->l_type);
344}
345
346#if BITS_PER_LONG == 32
347static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
348 struct flock64 *l)
349{
350 loff_t start;
351
352 switch (l->l_whence) {
353 case 0: /*SEEK_SET*/
354 start = 0;
355 break;
356 case 1: /*SEEK_CUR*/
357 start = filp->f_pos;
358 break;
359 case 2: /*SEEK_END*/
360 start = i_size_read(filp->f_dentry->d_inode);
361 break;
362 default:
363 return -EINVAL;
364 }
365
Trond Myklebust4c780a42005-10-18 14:20:21 -0700366 start += l->l_start;
367 if (start < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 return -EINVAL;
Trond Myklebust4c780a42005-10-18 14:20:21 -0700369 fl->fl_end = OFFSET_MAX;
370 if (l->l_len > 0) {
371 fl->fl_end = start + l->l_len - 1;
372 } else if (l->l_len < 0) {
373 fl->fl_end = start - 1;
374 start += l->l_len;
375 if (start < 0)
376 return -EINVAL;
377 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 fl->fl_start = start; /* we record the absolute position */
Trond Myklebust4c780a42005-10-18 14:20:21 -0700379 if (fl->fl_end < fl->fl_start)
380 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381
382 fl->fl_owner = current->files;
383 fl->fl_pid = current->tgid;
384 fl->fl_file = filp;
385 fl->fl_flags = FL_POSIX;
386 fl->fl_ops = NULL;
387 fl->fl_lmops = NULL;
388
389 switch (l->l_type) {
390 case F_RDLCK:
391 case F_WRLCK:
392 case F_UNLCK:
393 fl->fl_type = l->l_type;
394 break;
395 default:
396 return -EINVAL;
397 }
398
399 return (0);
400}
401#endif
402
403/* default lease lock manager operations */
404static void lease_break_callback(struct file_lock *fl)
405{
406 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
407}
408
409static void lease_release_private_callback(struct file_lock *fl)
410{
411 if (!fl->fl_file)
412 return;
413
414 f_delown(fl->fl_file);
415 fl->fl_file->f_owner.signum = 0;
416}
417
Adrian Bunk75c96f82005-05-05 16:16:09 -0700418static int lease_mylease_callback(struct file_lock *fl, struct file_lock *try)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419{
420 return fl->fl_file == try->fl_file;
421}
422
Adrian Bunk75c96f82005-05-05 16:16:09 -0700423static struct lock_manager_operations lease_manager_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 .fl_break = lease_break_callback,
425 .fl_release_private = lease_release_private_callback,
426 .fl_mylease = lease_mylease_callback,
427 .fl_change = lease_modify,
428};
429
430/*
431 * Initialize a lease, use the default lock manager operations
432 */
433static int lease_init(struct file *filp, int type, struct file_lock *fl)
434 {
435 fl->fl_owner = current->files;
436 fl->fl_pid = current->tgid;
437
438 fl->fl_file = filp;
439 fl->fl_flags = FL_LEASE;
440 if (assign_type(fl, type) != 0) {
441 locks_free_lock(fl);
442 return -EINVAL;
443 }
444 fl->fl_start = 0;
445 fl->fl_end = OFFSET_MAX;
446 fl->fl_ops = NULL;
447 fl->fl_lmops = &lease_manager_ops;
448 return 0;
449}
450
451/* Allocate a file_lock initialised to this type of lease */
452static int lease_alloc(struct file *filp, int type, struct file_lock **flp)
453{
454 struct file_lock *fl = locks_alloc_lock();
455 int error;
456
457 if (fl == NULL)
458 return -ENOMEM;
459
460 error = lease_init(filp, type, fl);
461 if (error)
462 return error;
463 *flp = fl;
464 return 0;
465}
466
467/* Check if two locks overlap each other.
468 */
469static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
470{
471 return ((fl1->fl_end >= fl2->fl_start) &&
472 (fl2->fl_end >= fl1->fl_start));
473}
474
475/*
476 * Check whether two locks have the same owner.
477 */
478static inline int
479posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
480{
481 if (fl1->fl_lmops && fl1->fl_lmops->fl_compare_owner)
482 return fl2->fl_lmops == fl1->fl_lmops &&
483 fl1->fl_lmops->fl_compare_owner(fl1, fl2);
484 return fl1->fl_owner == fl2->fl_owner;
485}
486
487/* Remove waiter from blocker's block list.
488 * When blocker ends up pointing to itself then the list is empty.
489 */
490static inline void __locks_delete_block(struct file_lock *waiter)
491{
492 list_del_init(&waiter->fl_block);
493 list_del_init(&waiter->fl_link);
494 waiter->fl_next = NULL;
495}
496
497/*
498 */
499static void locks_delete_block(struct file_lock *waiter)
500{
501 lock_kernel();
502 __locks_delete_block(waiter);
503 unlock_kernel();
504}
505
506/* Insert waiter into blocker's block list.
507 * We use a circular list so that processes can be easily woken up in
508 * the order they blocked. The documentation doesn't require this but
509 * it seems like the reasonable thing to do.
510 */
511static void locks_insert_block(struct file_lock *blocker,
512 struct file_lock *waiter)
513{
514 if (!list_empty(&waiter->fl_block)) {
515 printk(KERN_ERR "locks_insert_block: removing duplicated lock "
516 "(pid=%d %Ld-%Ld type=%d)\n", waiter->fl_pid,
517 waiter->fl_start, waiter->fl_end, waiter->fl_type);
518 __locks_delete_block(waiter);
519 }
520 list_add_tail(&waiter->fl_block, &blocker->fl_block);
521 waiter->fl_next = blocker;
522 if (IS_POSIX(blocker))
523 list_add(&waiter->fl_link, &blocked_list);
524}
525
526/* Wake up processes blocked waiting for blocker.
527 * If told to wait then schedule the processes until the block list
528 * is empty, otherwise empty the block list ourselves.
529 */
530static void locks_wake_up_blocks(struct file_lock *blocker)
531{
532 while (!list_empty(&blocker->fl_block)) {
533 struct file_lock *waiter = list_entry(blocker->fl_block.next,
534 struct file_lock, fl_block);
535 __locks_delete_block(waiter);
536 if (waiter->fl_lmops && waiter->fl_lmops->fl_notify)
537 waiter->fl_lmops->fl_notify(waiter);
538 else
539 wake_up(&waiter->fl_wait);
540 }
541}
542
543/* Insert file lock fl into an inode's lock list at the position indicated
544 * by pos. At the same time add the lock to the global file lock list.
545 */
546static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
547{
548 list_add(&fl->fl_link, &file_lock_list);
549
550 /* insert into file's list */
551 fl->fl_next = *pos;
552 *pos = fl;
553
554 if (fl->fl_ops && fl->fl_ops->fl_insert)
555 fl->fl_ops->fl_insert(fl);
556}
557
558/*
559 * Delete a lock and then free it.
560 * Wake up processes that are blocked waiting for this lock,
561 * notify the FS that the lock has been cleared and
562 * finally free the lock.
563 */
564static void locks_delete_lock(struct file_lock **thisfl_p)
565{
566 struct file_lock *fl = *thisfl_p;
567
568 *thisfl_p = fl->fl_next;
569 fl->fl_next = NULL;
570 list_del_init(&fl->fl_link);
571
572 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
573 if (fl->fl_fasync != NULL) {
574 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
575 fl->fl_fasync = NULL;
576 }
577
578 if (fl->fl_ops && fl->fl_ops->fl_remove)
579 fl->fl_ops->fl_remove(fl);
580
581 locks_wake_up_blocks(fl);
582 locks_free_lock(fl);
583}
584
585/* Determine if lock sys_fl blocks lock caller_fl. Common functionality
586 * checks for shared/exclusive status of overlapping locks.
587 */
588static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
589{
590 if (sys_fl->fl_type == F_WRLCK)
591 return 1;
592 if (caller_fl->fl_type == F_WRLCK)
593 return 1;
594 return 0;
595}
596
597/* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
598 * checking before calling the locks_conflict().
599 */
600static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
601{
602 /* POSIX locks owned by the same process do not conflict with
603 * each other.
604 */
605 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
606 return (0);
607
608 /* Check whether they overlap */
609 if (!locks_overlap(caller_fl, sys_fl))
610 return 0;
611
612 return (locks_conflict(caller_fl, sys_fl));
613}
614
615/* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
616 * checking before calling the locks_conflict().
617 */
618static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
619{
620 /* FLOCK locks referring to the same filp do not conflict with
621 * each other.
622 */
623 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
624 return (0);
625 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
626 return 0;
627
628 return (locks_conflict(caller_fl, sys_fl));
629}
630
631static int interruptible_sleep_on_locked(wait_queue_head_t *fl_wait, int timeout)
632{
633 int result = 0;
634 DECLARE_WAITQUEUE(wait, current);
635
636 __set_current_state(TASK_INTERRUPTIBLE);
637 add_wait_queue(fl_wait, &wait);
638 if (timeout == 0)
639 schedule();
640 else
641 result = schedule_timeout(timeout);
642 if (signal_pending(current))
643 result = -ERESTARTSYS;
644 remove_wait_queue(fl_wait, &wait);
645 __set_current_state(TASK_RUNNING);
646 return result;
647}
648
649static int locks_block_on_timeout(struct file_lock *blocker, struct file_lock *waiter, int time)
650{
651 int result;
652 locks_insert_block(blocker, waiter);
653 result = interruptible_sleep_on_locked(&waiter->fl_wait, time);
654 __locks_delete_block(waiter);
655 return result;
656}
657
658struct file_lock *
659posix_test_lock(struct file *filp, struct file_lock *fl)
660{
661 struct file_lock *cfl;
662
663 lock_kernel();
664 for (cfl = filp->f_dentry->d_inode->i_flock; cfl; cfl = cfl->fl_next) {
665 if (!IS_POSIX(cfl))
666 continue;
667 if (posix_locks_conflict(cfl, fl))
668 break;
669 }
670 unlock_kernel();
671
672 return (cfl);
673}
674
675EXPORT_SYMBOL(posix_test_lock);
676
677/* This function tests for deadlock condition before putting a process to
678 * sleep. The detection scheme is no longer recursive. Recursive was neat,
679 * but dangerous - we risked stack corruption if the lock data was bad, or
680 * if the recursion was too deep for any other reason.
681 *
682 * We rely on the fact that a task can only be on one lock's wait queue
683 * at a time. When we find blocked_task on a wait queue we can re-search
684 * with blocked_task equal to that queue's owner, until either blocked_task
685 * isn't found, or blocked_task is found on a queue owned by my_task.
686 *
687 * Note: the above assumption may not be true when handling lock requests
688 * from a broken NFS client. But broken NFS clients have a lot more to
689 * worry about than proper deadlock detection anyway... --okir
690 */
691int posix_locks_deadlock(struct file_lock *caller_fl,
692 struct file_lock *block_fl)
693{
694 struct list_head *tmp;
695
696next_task:
697 if (posix_same_owner(caller_fl, block_fl))
698 return 1;
699 list_for_each(tmp, &blocked_list) {
700 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
701 if (posix_same_owner(fl, block_fl)) {
702 fl = fl->fl_next;
703 block_fl = fl;
704 goto next_task;
705 }
706 }
707 return 0;
708}
709
710EXPORT_SYMBOL(posix_locks_deadlock);
711
712/* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
713 * at the head of the list, but that's secret knowledge known only to
714 * flock_lock_file and posix_lock_file.
715 */
716static int flock_lock_file(struct file *filp, struct file_lock *new_fl)
717{
718 struct file_lock **before;
719 struct inode * inode = filp->f_dentry->d_inode;
720 int error = 0;
721 int found = 0;
722
723 lock_kernel();
724 for_each_lock(inode, before) {
725 struct file_lock *fl = *before;
726 if (IS_POSIX(fl))
727 break;
728 if (IS_LEASE(fl))
729 continue;
730 if (filp != fl->fl_file)
731 continue;
732 if (new_fl->fl_type == fl->fl_type)
733 goto out;
734 found = 1;
735 locks_delete_lock(before);
736 break;
737 }
738 unlock_kernel();
739
740 if (new_fl->fl_type == F_UNLCK)
741 return 0;
742
743 /*
744 * If a higher-priority process was blocked on the old file lock,
745 * give it the opportunity to lock the file.
746 */
747 if (found)
748 cond_resched();
749
750 lock_kernel();
751 for_each_lock(inode, before) {
752 struct file_lock *fl = *before;
753 if (IS_POSIX(fl))
754 break;
755 if (IS_LEASE(fl))
756 continue;
757 if (!flock_locks_conflict(new_fl, fl))
758 continue;
759 error = -EAGAIN;
760 if (new_fl->fl_flags & FL_SLEEP) {
761 locks_insert_block(fl, new_fl);
762 }
763 goto out;
764 }
765 locks_insert_lock(&inode->i_flock, new_fl);
766 error = 0;
767
768out:
769 unlock_kernel();
770 return error;
771}
772
773EXPORT_SYMBOL(posix_lock_file);
774
775static int __posix_lock_file(struct inode *inode, struct file_lock *request)
776{
777 struct file_lock *fl;
778 struct file_lock *new_fl, *new_fl2;
779 struct file_lock *left = NULL;
780 struct file_lock *right = NULL;
781 struct file_lock **before;
782 int error, added = 0;
783
784 /*
785 * We may need two file_lock structures for this operation,
786 * so we get them in advance to avoid races.
787 */
788 new_fl = locks_alloc_lock();
789 new_fl2 = locks_alloc_lock();
790
791 lock_kernel();
792 if (request->fl_type != F_UNLCK) {
793 for_each_lock(inode, before) {
794 struct file_lock *fl = *before;
795 if (!IS_POSIX(fl))
796 continue;
797 if (!posix_locks_conflict(request, fl))
798 continue;
799 error = -EAGAIN;
800 if (!(request->fl_flags & FL_SLEEP))
801 goto out;
802 error = -EDEADLK;
803 if (posix_locks_deadlock(request, fl))
804 goto out;
805 error = -EAGAIN;
806 locks_insert_block(fl, request);
807 goto out;
808 }
809 }
810
811 /* If we're just looking for a conflict, we're done. */
812 error = 0;
813 if (request->fl_flags & FL_ACCESS)
814 goto out;
815
816 error = -ENOLCK; /* "no luck" */
817 if (!(new_fl && new_fl2))
818 goto out;
819
820 /*
821 * We've allocated the new locks in advance, so there are no
822 * errors possible (and no blocking operations) from here on.
823 *
824 * Find the first old lock with the same owner as the new lock.
825 */
826
827 before = &inode->i_flock;
828
829 /* First skip locks owned by other processes. */
830 while ((fl = *before) && (!IS_POSIX(fl) ||
831 !posix_same_owner(request, fl))) {
832 before = &fl->fl_next;
833 }
834
835 /* Process locks with this owner. */
836 while ((fl = *before) && posix_same_owner(request, fl)) {
837 /* Detect adjacent or overlapping regions (if same lock type)
838 */
839 if (request->fl_type == fl->fl_type) {
Olaf Kirch449231d2005-08-25 16:25:35 -0700840 /* In all comparisons of start vs end, use
841 * "start - 1" rather than "end + 1". If end
842 * is OFFSET_MAX, end + 1 will become negative.
843 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 if (fl->fl_end < request->fl_start - 1)
845 goto next_lock;
846 /* If the next lock in the list has entirely bigger
847 * addresses than the new one, insert the lock here.
848 */
Olaf Kirch449231d2005-08-25 16:25:35 -0700849 if (fl->fl_start - 1 > request->fl_end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 break;
851
852 /* If we come here, the new and old lock are of the
853 * same type and adjacent or overlapping. Make one
854 * lock yielding from the lower start address of both
855 * locks to the higher end address.
856 */
857 if (fl->fl_start > request->fl_start)
858 fl->fl_start = request->fl_start;
859 else
860 request->fl_start = fl->fl_start;
861 if (fl->fl_end < request->fl_end)
862 fl->fl_end = request->fl_end;
863 else
864 request->fl_end = fl->fl_end;
865 if (added) {
866 locks_delete_lock(before);
867 continue;
868 }
869 request = fl;
870 added = 1;
871 }
872 else {
873 /* Processing for different lock types is a bit
874 * more complex.
875 */
876 if (fl->fl_end < request->fl_start)
877 goto next_lock;
878 if (fl->fl_start > request->fl_end)
879 break;
880 if (request->fl_type == F_UNLCK)
881 added = 1;
882 if (fl->fl_start < request->fl_start)
883 left = fl;
884 /* If the next lock in the list has a higher end
885 * address than the new one, insert the new one here.
886 */
887 if (fl->fl_end > request->fl_end) {
888 right = fl;
889 break;
890 }
891 if (fl->fl_start >= request->fl_start) {
892 /* The new lock completely replaces an old
893 * one (This may happen several times).
894 */
895 if (added) {
896 locks_delete_lock(before);
897 continue;
898 }
899 /* Replace the old lock with the new one.
900 * Wake up anybody waiting for the old one,
901 * as the change in lock type might satisfy
902 * their needs.
903 */
904 locks_wake_up_blocks(fl);
905 fl->fl_start = request->fl_start;
906 fl->fl_end = request->fl_end;
907 fl->fl_type = request->fl_type;
908 fl->fl_u = request->fl_u;
909 request = fl;
910 added = 1;
911 }
912 }
913 /* Go on to next lock.
914 */
915 next_lock:
916 before = &fl->fl_next;
917 }
918
919 error = 0;
920 if (!added) {
921 if (request->fl_type == F_UNLCK)
922 goto out;
923 locks_copy_lock(new_fl, request);
924 locks_insert_lock(before, new_fl);
925 new_fl = NULL;
926 }
927 if (right) {
928 if (left == right) {
929 /* The new lock breaks the old one in two pieces,
930 * so we have to use the second new lock.
931 */
932 left = new_fl2;
933 new_fl2 = NULL;
934 locks_copy_lock(left, right);
935 locks_insert_lock(before, left);
936 }
937 right->fl_start = request->fl_end + 1;
938 locks_wake_up_blocks(right);
939 }
940 if (left) {
941 left->fl_end = request->fl_start - 1;
942 locks_wake_up_blocks(left);
943 }
944 out:
945 unlock_kernel();
946 /*
947 * Free any unused locks.
948 */
949 if (new_fl)
950 locks_free_lock(new_fl);
951 if (new_fl2)
952 locks_free_lock(new_fl2);
953 return error;
954}
955
956/**
957 * posix_lock_file - Apply a POSIX-style lock to a file
958 * @filp: The file to apply the lock to
959 * @fl: The lock to be applied
960 *
961 * Add a POSIX style lock to a file.
962 * We merge adjacent & overlapping locks whenever possible.
963 * POSIX locks are sorted by owner task, then by starting address
964 */
965int posix_lock_file(struct file *filp, struct file_lock *fl)
966{
967 return __posix_lock_file(filp->f_dentry->d_inode, fl);
968}
969
970/**
971 * posix_lock_file_wait - Apply a POSIX-style lock to a file
972 * @filp: The file to apply the lock to
973 * @fl: The lock to be applied
974 *
975 * Add a POSIX style lock to a file.
976 * We merge adjacent & overlapping locks whenever possible.
977 * POSIX locks are sorted by owner task, then by starting address
978 */
979int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
980{
981 int error;
982 might_sleep ();
983 for (;;) {
984 error = __posix_lock_file(filp->f_dentry->d_inode, fl);
985 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
986 break;
987 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
988 if (!error)
989 continue;
990
991 locks_delete_block(fl);
992 break;
993 }
994 return error;
995}
996EXPORT_SYMBOL(posix_lock_file_wait);
997
998/**
999 * locks_mandatory_locked - Check for an active lock
1000 * @inode: the file to check
1001 *
1002 * Searches the inode's list of locks to find any POSIX locks which conflict.
1003 * This function is called from locks_verify_locked() only.
1004 */
1005int locks_mandatory_locked(struct inode *inode)
1006{
1007 fl_owner_t owner = current->files;
1008 struct file_lock *fl;
1009
1010 /*
1011 * Search the lock list for this inode for any POSIX locks.
1012 */
1013 lock_kernel();
1014 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1015 if (!IS_POSIX(fl))
1016 continue;
1017 if (fl->fl_owner != owner)
1018 break;
1019 }
1020 unlock_kernel();
1021 return fl ? -EAGAIN : 0;
1022}
1023
1024/**
1025 * locks_mandatory_area - Check for a conflicting lock
1026 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1027 * for shared
1028 * @inode: the file to check
1029 * @filp: how the file was opened (if it was)
1030 * @offset: start of area to check
1031 * @count: length of area to check
1032 *
1033 * Searches the inode's list of locks to find any POSIX locks which conflict.
1034 * This function is called from rw_verify_area() and
1035 * locks_verify_truncate().
1036 */
1037int locks_mandatory_area(int read_write, struct inode *inode,
1038 struct file *filp, loff_t offset,
1039 size_t count)
1040{
1041 struct file_lock fl;
1042 int error;
1043
1044 locks_init_lock(&fl);
1045 fl.fl_owner = current->files;
1046 fl.fl_pid = current->tgid;
1047 fl.fl_file = filp;
1048 fl.fl_flags = FL_POSIX | FL_ACCESS;
1049 if (filp && !(filp->f_flags & O_NONBLOCK))
1050 fl.fl_flags |= FL_SLEEP;
1051 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1052 fl.fl_start = offset;
1053 fl.fl_end = offset + count - 1;
1054
1055 for (;;) {
1056 error = __posix_lock_file(inode, &fl);
1057 if (error != -EAGAIN)
1058 break;
1059 if (!(fl.fl_flags & FL_SLEEP))
1060 break;
1061 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1062 if (!error) {
1063 /*
1064 * If we've been sleeping someone might have
1065 * changed the permissions behind our back.
1066 */
1067 if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID)
1068 continue;
1069 }
1070
1071 locks_delete_block(&fl);
1072 break;
1073 }
1074
1075 return error;
1076}
1077
1078EXPORT_SYMBOL(locks_mandatory_area);
1079
1080/* We already had a lease on this file; just change its type */
1081int lease_modify(struct file_lock **before, int arg)
1082{
1083 struct file_lock *fl = *before;
1084 int error = assign_type(fl, arg);
1085
1086 if (error)
1087 return error;
1088 locks_wake_up_blocks(fl);
1089 if (arg == F_UNLCK)
1090 locks_delete_lock(before);
1091 return 0;
1092}
1093
1094EXPORT_SYMBOL(lease_modify);
1095
1096static void time_out_leases(struct inode *inode)
1097{
1098 struct file_lock **before;
1099 struct file_lock *fl;
1100
1101 before = &inode->i_flock;
1102 while ((fl = *before) && IS_LEASE(fl) && (fl->fl_type & F_INPROGRESS)) {
1103 if ((fl->fl_break_time == 0)
1104 || time_before(jiffies, fl->fl_break_time)) {
1105 before = &fl->fl_next;
1106 continue;
1107 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108 lease_modify(before, fl->fl_type & ~F_INPROGRESS);
1109 if (fl == *before) /* lease_modify may have freed fl */
1110 before = &fl->fl_next;
1111 }
1112}
1113
1114/**
1115 * __break_lease - revoke all outstanding leases on file
1116 * @inode: the inode of the file to return
1117 * @mode: the open mode (read or write)
1118 *
1119 * break_lease (inlined for speed) has checked there already
1120 * is a lease on this file. Leases are broken on a call to open()
1121 * or truncate(). This function can sleep unless you
1122 * specified %O_NONBLOCK to your open().
1123 */
1124int __break_lease(struct inode *inode, unsigned int mode)
1125{
1126 int error = 0, future;
1127 struct file_lock *new_fl, *flock;
1128 struct file_lock *fl;
1129 int alloc_err;
1130 unsigned long break_time;
1131 int i_have_this_lease = 0;
1132
1133 alloc_err = lease_alloc(NULL, mode & FMODE_WRITE ? F_WRLCK : F_RDLCK,
1134 &new_fl);
1135
1136 lock_kernel();
1137
1138 time_out_leases(inode);
1139
1140 flock = inode->i_flock;
1141 if ((flock == NULL) || !IS_LEASE(flock))
1142 goto out;
1143
1144 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1145 if (fl->fl_owner == current->files)
1146 i_have_this_lease = 1;
1147
1148 if (mode & FMODE_WRITE) {
1149 /* If we want write access, we have to revoke any lease. */
1150 future = F_UNLCK | F_INPROGRESS;
1151 } else if (flock->fl_type & F_INPROGRESS) {
1152 /* If the lease is already being broken, we just leave it */
1153 future = flock->fl_type;
1154 } else if (flock->fl_type & F_WRLCK) {
1155 /* Downgrade the exclusive lease to a read-only lease. */
1156 future = F_RDLCK | F_INPROGRESS;
1157 } else {
1158 /* the existing lease was read-only, so we can read too. */
1159 goto out;
1160 }
1161
1162 if (alloc_err && !i_have_this_lease && ((mode & O_NONBLOCK) == 0)) {
1163 error = alloc_err;
1164 goto out;
1165 }
1166
1167 break_time = 0;
1168 if (lease_break_time > 0) {
1169 break_time = jiffies + lease_break_time * HZ;
1170 if (break_time == 0)
1171 break_time++; /* so that 0 means no break time */
1172 }
1173
1174 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1175 if (fl->fl_type != future) {
1176 fl->fl_type = future;
1177 fl->fl_break_time = break_time;
1178 /* lease must have lmops break callback */
1179 fl->fl_lmops->fl_break(fl);
1180 }
1181 }
1182
1183 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1184 error = -EWOULDBLOCK;
1185 goto out;
1186 }
1187
1188restart:
1189 break_time = flock->fl_break_time;
1190 if (break_time != 0) {
1191 break_time -= jiffies;
1192 if (break_time == 0)
1193 break_time++;
1194 }
1195 error = locks_block_on_timeout(flock, new_fl, break_time);
1196 if (error >= 0) {
1197 if (error == 0)
1198 time_out_leases(inode);
1199 /* Wait for the next lease that has not been broken yet */
1200 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1201 flock = flock->fl_next) {
1202 if (flock->fl_type & F_INPROGRESS)
1203 goto restart;
1204 }
1205 error = 0;
1206 }
1207
1208out:
1209 unlock_kernel();
1210 if (!alloc_err)
1211 locks_free_lock(new_fl);
1212 return error;
1213}
1214
1215EXPORT_SYMBOL(__break_lease);
1216
1217/**
1218 * lease_get_mtime
1219 * @inode: the inode
1220 * @time: pointer to a timespec which will contain the last modified time
1221 *
1222 * This is to force NFS clients to flush their caches for files with
1223 * exclusive leases. The justification is that if someone has an
1224 * exclusive lease, then they could be modifiying it.
1225 */
1226void lease_get_mtime(struct inode *inode, struct timespec *time)
1227{
1228 struct file_lock *flock = inode->i_flock;
1229 if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK))
1230 *time = current_fs_time(inode->i_sb);
1231 else
1232 *time = inode->i_mtime;
1233}
1234
1235EXPORT_SYMBOL(lease_get_mtime);
1236
1237/**
1238 * fcntl_getlease - Enquire what lease is currently active
1239 * @filp: the file
1240 *
1241 * The value returned by this function will be one of
1242 * (if no lease break is pending):
1243 *
1244 * %F_RDLCK to indicate a shared lease is held.
1245 *
1246 * %F_WRLCK to indicate an exclusive lease is held.
1247 *
1248 * %F_UNLCK to indicate no lease is held.
1249 *
1250 * (if a lease break is pending):
1251 *
1252 * %F_RDLCK to indicate an exclusive lease needs to be
1253 * changed to a shared lease (or removed).
1254 *
1255 * %F_UNLCK to indicate the lease needs to be removed.
1256 *
1257 * XXX: sfr & willy disagree over whether F_INPROGRESS
1258 * should be returned to userspace.
1259 */
1260int fcntl_getlease(struct file *filp)
1261{
1262 struct file_lock *fl;
1263 int type = F_UNLCK;
1264
1265 lock_kernel();
1266 time_out_leases(filp->f_dentry->d_inode);
1267 for (fl = filp->f_dentry->d_inode->i_flock; fl && IS_LEASE(fl);
1268 fl = fl->fl_next) {
1269 if (fl->fl_file == filp) {
1270 type = fl->fl_type & ~F_INPROGRESS;
1271 break;
1272 }
1273 }
1274 unlock_kernel();
1275 return type;
1276}
1277
1278/**
1279 * __setlease - sets a lease on an open file
1280 * @filp: file pointer
1281 * @arg: type of lease to obtain
1282 * @flp: input - file_lock to use, output - file_lock inserted
1283 *
1284 * The (input) flp->fl_lmops->fl_break function is required
1285 * by break_lease().
1286 *
1287 * Called with kernel lock held.
1288 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07001289static int __setlease(struct file *filp, long arg, struct file_lock **flp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290{
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001291 struct file_lock *fl, **before, **my_before = NULL, *lease;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292 struct dentry *dentry = filp->f_dentry;
1293 struct inode *inode = dentry->d_inode;
1294 int error, rdlease_count = 0, wrlease_count = 0;
1295
1296 time_out_leases(inode);
1297
1298 error = -EINVAL;
1299 if (!flp || !(*flp) || !(*flp)->fl_lmops || !(*flp)->fl_lmops->fl_break)
1300 goto out;
1301
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001302 lease = *flp;
1303
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 error = -EAGAIN;
1305 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1306 goto out;
1307 if ((arg == F_WRLCK)
1308 && ((atomic_read(&dentry->d_count) > 1)
1309 || (atomic_read(&inode->i_count) > 1)))
1310 goto out;
1311
1312 /*
1313 * At this point, we know that if there is an exclusive
1314 * lease on this file, then we hold it on this filp
1315 * (otherwise our open of this file would have blocked).
1316 * And if we are trying to acquire an exclusive lease,
1317 * then the file is not open by anyone (including us)
1318 * except for this filp.
1319 */
1320 for (before = &inode->i_flock;
1321 ((fl = *before) != NULL) && IS_LEASE(fl);
1322 before = &fl->fl_next) {
1323 if (lease->fl_lmops->fl_mylease(fl, lease))
1324 my_before = before;
1325 else if (fl->fl_type == (F_INPROGRESS | F_UNLCK))
1326 /*
1327 * Someone is in the process of opening this
1328 * file for writing so we may not take an
1329 * exclusive lease on it.
1330 */
1331 wrlease_count++;
1332 else
1333 rdlease_count++;
1334 }
1335
1336 if ((arg == F_RDLCK && (wrlease_count > 0)) ||
1337 (arg == F_WRLCK && ((rdlease_count + wrlease_count) > 0)))
1338 goto out;
1339
1340 if (my_before != NULL) {
1341 error = lease->fl_lmops->fl_change(my_before, arg);
1342 goto out;
1343 }
1344
1345 error = 0;
1346 if (arg == F_UNLCK)
1347 goto out;
1348
1349 error = -EINVAL;
1350 if (!leases_enable)
1351 goto out;
1352
1353 error = lease_alloc(filp, arg, &fl);
1354 if (error)
1355 goto out;
1356
1357 locks_copy_lock(fl, lease);
1358
1359 locks_insert_lock(before, fl);
1360
1361 *flp = fl;
1362out:
1363 return error;
1364}
1365
1366 /**
1367 * setlease - sets a lease on an open file
1368 * @filp: file pointer
1369 * @arg: type of lease to obtain
1370 * @lease: file_lock to use
1371 *
1372 * Call this to establish a lease on the file.
1373 * The fl_lmops fl_break function is required by break_lease
1374 */
1375
1376int setlease(struct file *filp, long arg, struct file_lock **lease)
1377{
1378 struct dentry *dentry = filp->f_dentry;
1379 struct inode *inode = dentry->d_inode;
1380 int error;
1381
1382 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1383 return -EACCES;
1384 if (!S_ISREG(inode->i_mode))
1385 return -EINVAL;
1386 error = security_file_lock(filp, arg);
1387 if (error)
1388 return error;
1389
1390 lock_kernel();
1391 error = __setlease(filp, arg, lease);
1392 unlock_kernel();
1393
1394 return error;
1395}
1396
1397EXPORT_SYMBOL(setlease);
1398
1399/**
1400 * fcntl_setlease - sets a lease on an open file
1401 * @fd: open file descriptor
1402 * @filp: file pointer
1403 * @arg: type of lease to obtain
1404 *
1405 * Call this fcntl to establish a lease on the file.
1406 * Note that you also need to call %F_SETSIG to
1407 * receive a signal when the lease is broken.
1408 */
1409int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1410{
1411 struct file_lock fl, *flp = &fl;
1412 struct dentry *dentry = filp->f_dentry;
1413 struct inode *inode = dentry->d_inode;
1414 int error;
1415
1416 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1417 return -EACCES;
1418 if (!S_ISREG(inode->i_mode))
1419 return -EINVAL;
1420 error = security_file_lock(filp, arg);
1421 if (error)
1422 return error;
1423
1424 locks_init_lock(&fl);
1425 error = lease_init(filp, arg, &fl);
1426 if (error)
1427 return error;
1428
1429 lock_kernel();
1430
1431 error = __setlease(filp, arg, &flp);
J. Bruce Fieldsdc15ae12005-11-10 19:08:00 -05001432 if (error || arg == F_UNLCK)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433 goto out_unlock;
1434
1435 error = fasync_helper(fd, filp, 1, &flp->fl_fasync);
1436 if (error < 0) {
1437 /* remove lease just inserted by __setlease */
1438 flp->fl_type = F_UNLCK | F_INPROGRESS;
1439 flp->fl_break_time = jiffies- 10;
1440 time_out_leases(inode);
1441 goto out_unlock;
1442 }
1443
1444 error = f_setown(filp, current->pid, 0);
1445out_unlock:
1446 unlock_kernel();
1447 return error;
1448}
1449
1450/**
1451 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1452 * @filp: The file to apply the lock to
1453 * @fl: The lock to be applied
1454 *
1455 * Add a FLOCK style lock to a file.
1456 */
1457int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1458{
1459 int error;
1460 might_sleep();
1461 for (;;) {
1462 error = flock_lock_file(filp, fl);
1463 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1464 break;
1465 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1466 if (!error)
1467 continue;
1468
1469 locks_delete_block(fl);
1470 break;
1471 }
1472 return error;
1473}
1474
1475EXPORT_SYMBOL(flock_lock_file_wait);
1476
1477/**
1478 * sys_flock: - flock() system call.
1479 * @fd: the file descriptor to lock.
1480 * @cmd: the type of lock to apply.
1481 *
1482 * Apply a %FL_FLOCK style lock to an open file descriptor.
1483 * The @cmd can be one of
1484 *
1485 * %LOCK_SH -- a shared lock.
1486 *
1487 * %LOCK_EX -- an exclusive lock.
1488 *
1489 * %LOCK_UN -- remove an existing lock.
1490 *
1491 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1492 *
1493 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1494 * processes read and write access respectively.
1495 */
1496asmlinkage long sys_flock(unsigned int fd, unsigned int cmd)
1497{
1498 struct file *filp;
1499 struct file_lock *lock;
1500 int can_sleep, unlock;
1501 int error;
1502
1503 error = -EBADF;
1504 filp = fget(fd);
1505 if (!filp)
1506 goto out;
1507
1508 can_sleep = !(cmd & LOCK_NB);
1509 cmd &= ~LOCK_NB;
1510 unlock = (cmd == LOCK_UN);
1511
1512 if (!unlock && !(cmd & LOCK_MAND) && !(filp->f_mode & 3))
1513 goto out_putf;
1514
1515 error = flock_make_lock(filp, &lock, cmd);
1516 if (error)
1517 goto out_putf;
1518 if (can_sleep)
1519 lock->fl_flags |= FL_SLEEP;
1520
1521 error = security_file_lock(filp, cmd);
1522 if (error)
1523 goto out_free;
1524
1525 if (filp->f_op && filp->f_op->flock)
1526 error = filp->f_op->flock(filp,
1527 (can_sleep) ? F_SETLKW : F_SETLK,
1528 lock);
1529 else
1530 error = flock_lock_file_wait(filp, lock);
1531
1532 out_free:
1533 if (list_empty(&lock->fl_link)) {
1534 locks_free_lock(lock);
1535 }
1536
1537 out_putf:
1538 fput(filp);
1539 out:
1540 return error;
1541}
1542
1543/* Report the first existing lock that would conflict with l.
1544 * This implements the F_GETLK command of fcntl().
1545 */
1546int fcntl_getlk(struct file *filp, struct flock __user *l)
1547{
1548 struct file_lock *fl, file_lock;
1549 struct flock flock;
1550 int error;
1551
1552 error = -EFAULT;
1553 if (copy_from_user(&flock, l, sizeof(flock)))
1554 goto out;
1555 error = -EINVAL;
1556 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1557 goto out;
1558
1559 error = flock_to_posix_lock(filp, &file_lock, &flock);
1560 if (error)
1561 goto out;
1562
1563 if (filp->f_op && filp->f_op->lock) {
1564 error = filp->f_op->lock(filp, F_GETLK, &file_lock);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00001565 if (file_lock.fl_ops && file_lock.fl_ops->fl_release_private)
1566 file_lock.fl_ops->fl_release_private(&file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567 if (error < 0)
1568 goto out;
1569 else
1570 fl = (file_lock.fl_type == F_UNLCK ? NULL : &file_lock);
1571 } else {
1572 fl = posix_test_lock(filp, &file_lock);
1573 }
1574
1575 flock.l_type = F_UNLCK;
1576 if (fl != NULL) {
1577 flock.l_pid = fl->fl_pid;
1578#if BITS_PER_LONG == 32
1579 /*
1580 * Make sure we can represent the posix lock via
1581 * legacy 32bit flock.
1582 */
1583 error = -EOVERFLOW;
1584 if (fl->fl_start > OFFT_OFFSET_MAX)
1585 goto out;
1586 if ((fl->fl_end != OFFSET_MAX)
1587 && (fl->fl_end > OFFT_OFFSET_MAX))
1588 goto out;
1589#endif
1590 flock.l_start = fl->fl_start;
1591 flock.l_len = fl->fl_end == OFFSET_MAX ? 0 :
1592 fl->fl_end - fl->fl_start + 1;
1593 flock.l_whence = 0;
1594 flock.l_type = fl->fl_type;
1595 }
1596 error = -EFAULT;
1597 if (!copy_to_user(l, &flock, sizeof(flock)))
1598 error = 0;
1599out:
1600 return error;
1601}
1602
1603/* Apply the lock described by l to an open file descriptor.
1604 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1605 */
Peter Staubachc2936212005-07-27 11:45:09 -07001606int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1607 struct flock __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608{
1609 struct file_lock *file_lock = locks_alloc_lock();
1610 struct flock flock;
1611 struct inode *inode;
1612 int error;
1613
1614 if (file_lock == NULL)
1615 return -ENOLCK;
1616
1617 /*
1618 * This might block, so we do it before checking the inode.
1619 */
1620 error = -EFAULT;
1621 if (copy_from_user(&flock, l, sizeof(flock)))
1622 goto out;
1623
1624 inode = filp->f_dentry->d_inode;
1625
1626 /* Don't allow mandatory locks on files that may be memory mapped
1627 * and shared.
1628 */
1629 if (IS_MANDLOCK(inode) &&
1630 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1631 mapping_writably_mapped(filp->f_mapping)) {
1632 error = -EAGAIN;
1633 goto out;
1634 }
1635
Peter Staubachc2936212005-07-27 11:45:09 -07001636again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 error = flock_to_posix_lock(filp, file_lock, &flock);
1638 if (error)
1639 goto out;
1640 if (cmd == F_SETLKW) {
1641 file_lock->fl_flags |= FL_SLEEP;
1642 }
1643
1644 error = -EBADF;
1645 switch (flock.l_type) {
1646 case F_RDLCK:
1647 if (!(filp->f_mode & FMODE_READ))
1648 goto out;
1649 break;
1650 case F_WRLCK:
1651 if (!(filp->f_mode & FMODE_WRITE))
1652 goto out;
1653 break;
1654 case F_UNLCK:
1655 break;
1656 default:
1657 error = -EINVAL;
1658 goto out;
1659 }
1660
1661 error = security_file_lock(filp, file_lock->fl_type);
1662 if (error)
1663 goto out;
1664
Peter Staubachc2936212005-07-27 11:45:09 -07001665 if (filp->f_op && filp->f_op->lock != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666 error = filp->f_op->lock(filp, cmd, file_lock);
Peter Staubachc2936212005-07-27 11:45:09 -07001667 else {
1668 for (;;) {
1669 error = __posix_lock_file(inode, file_lock);
1670 if ((error != -EAGAIN) || (cmd == F_SETLK))
1671 break;
1672 error = wait_event_interruptible(file_lock->fl_wait,
1673 !file_lock->fl_next);
1674 if (!error)
1675 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676
Peter Staubachc2936212005-07-27 11:45:09 -07001677 locks_delete_block(file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678 break;
Peter Staubachc2936212005-07-27 11:45:09 -07001679 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 }
1681
Peter Staubachc2936212005-07-27 11:45:09 -07001682 /*
1683 * Attempt to detect a close/fcntl race and recover by
1684 * releasing the lock that was just acquired.
1685 */
1686 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1687 flock.l_type = F_UNLCK;
1688 goto again;
1689 }
1690
1691out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692 locks_free_lock(file_lock);
1693 return error;
1694}
1695
1696#if BITS_PER_LONG == 32
1697/* Report the first existing lock that would conflict with l.
1698 * This implements the F_GETLK command of fcntl().
1699 */
1700int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1701{
1702 struct file_lock *fl, file_lock;
1703 struct flock64 flock;
1704 int error;
1705
1706 error = -EFAULT;
1707 if (copy_from_user(&flock, l, sizeof(flock)))
1708 goto out;
1709 error = -EINVAL;
1710 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1711 goto out;
1712
1713 error = flock64_to_posix_lock(filp, &file_lock, &flock);
1714 if (error)
1715 goto out;
1716
1717 if (filp->f_op && filp->f_op->lock) {
1718 error = filp->f_op->lock(filp, F_GETLK, &file_lock);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00001719 if (file_lock.fl_ops && file_lock.fl_ops->fl_release_private)
1720 file_lock.fl_ops->fl_release_private(&file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721 if (error < 0)
1722 goto out;
1723 else
1724 fl = (file_lock.fl_type == F_UNLCK ? NULL : &file_lock);
1725 } else {
1726 fl = posix_test_lock(filp, &file_lock);
1727 }
1728
1729 flock.l_type = F_UNLCK;
1730 if (fl != NULL) {
1731 flock.l_pid = fl->fl_pid;
1732 flock.l_start = fl->fl_start;
1733 flock.l_len = fl->fl_end == OFFSET_MAX ? 0 :
1734 fl->fl_end - fl->fl_start + 1;
1735 flock.l_whence = 0;
1736 flock.l_type = fl->fl_type;
1737 }
1738 error = -EFAULT;
1739 if (!copy_to_user(l, &flock, sizeof(flock)))
1740 error = 0;
1741
1742out:
1743 return error;
1744}
1745
1746/* Apply the lock described by l to an open file descriptor.
1747 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1748 */
Peter Staubachc2936212005-07-27 11:45:09 -07001749int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
1750 struct flock64 __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751{
1752 struct file_lock *file_lock = locks_alloc_lock();
1753 struct flock64 flock;
1754 struct inode *inode;
1755 int error;
1756
1757 if (file_lock == NULL)
1758 return -ENOLCK;
1759
1760 /*
1761 * This might block, so we do it before checking the inode.
1762 */
1763 error = -EFAULT;
1764 if (copy_from_user(&flock, l, sizeof(flock)))
1765 goto out;
1766
1767 inode = filp->f_dentry->d_inode;
1768
1769 /* Don't allow mandatory locks on files that may be memory mapped
1770 * and shared.
1771 */
1772 if (IS_MANDLOCK(inode) &&
1773 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1774 mapping_writably_mapped(filp->f_mapping)) {
1775 error = -EAGAIN;
1776 goto out;
1777 }
1778
Peter Staubachc2936212005-07-27 11:45:09 -07001779again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780 error = flock64_to_posix_lock(filp, file_lock, &flock);
1781 if (error)
1782 goto out;
1783 if (cmd == F_SETLKW64) {
1784 file_lock->fl_flags |= FL_SLEEP;
1785 }
1786
1787 error = -EBADF;
1788 switch (flock.l_type) {
1789 case F_RDLCK:
1790 if (!(filp->f_mode & FMODE_READ))
1791 goto out;
1792 break;
1793 case F_WRLCK:
1794 if (!(filp->f_mode & FMODE_WRITE))
1795 goto out;
1796 break;
1797 case F_UNLCK:
1798 break;
1799 default:
1800 error = -EINVAL;
1801 goto out;
1802 }
1803
1804 error = security_file_lock(filp, file_lock->fl_type);
1805 if (error)
1806 goto out;
1807
Peter Staubachc2936212005-07-27 11:45:09 -07001808 if (filp->f_op && filp->f_op->lock != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809 error = filp->f_op->lock(filp, cmd, file_lock);
Peter Staubachc2936212005-07-27 11:45:09 -07001810 else {
1811 for (;;) {
1812 error = __posix_lock_file(inode, file_lock);
1813 if ((error != -EAGAIN) || (cmd == F_SETLK64))
1814 break;
1815 error = wait_event_interruptible(file_lock->fl_wait,
1816 !file_lock->fl_next);
1817 if (!error)
1818 continue;
1819
1820 locks_delete_block(file_lock);
1821 break;
1822 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823 }
1824
Peter Staubachc2936212005-07-27 11:45:09 -07001825 /*
1826 * Attempt to detect a close/fcntl race and recover by
1827 * releasing the lock that was just acquired.
1828 */
1829 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1830 flock.l_type = F_UNLCK;
1831 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832 }
1833
1834out:
1835 locks_free_lock(file_lock);
1836 return error;
1837}
1838#endif /* BITS_PER_LONG == 32 */
1839
1840/*
1841 * This function is called when the file is being removed
1842 * from the task's fd array. POSIX locks belonging to this task
1843 * are deleted at this time.
1844 */
1845void locks_remove_posix(struct file *filp, fl_owner_t owner)
1846{
1847 struct file_lock lock, **before;
1848
1849 /*
1850 * If there are no locks held on this file, we don't need to call
1851 * posix_lock_file(). Another process could be setting a lock on this
1852 * file at the same time, but we wouldn't remove that lock anyway.
1853 */
1854 before = &filp->f_dentry->d_inode->i_flock;
1855 if (*before == NULL)
1856 return;
1857
1858 lock.fl_type = F_UNLCK;
1859 lock.fl_flags = FL_POSIX;
1860 lock.fl_start = 0;
1861 lock.fl_end = OFFSET_MAX;
1862 lock.fl_owner = owner;
1863 lock.fl_pid = current->tgid;
1864 lock.fl_file = filp;
1865 lock.fl_ops = NULL;
1866 lock.fl_lmops = NULL;
1867
1868 if (filp->f_op && filp->f_op->lock != NULL) {
1869 filp->f_op->lock(filp, F_SETLK, &lock);
1870 goto out;
1871 }
1872
1873 /* Can't use posix_lock_file here; we need to remove it no matter
1874 * which pid we have.
1875 */
1876 lock_kernel();
1877 while (*before != NULL) {
1878 struct file_lock *fl = *before;
1879 if (IS_POSIX(fl) && posix_same_owner(fl, &lock)) {
1880 locks_delete_lock(before);
1881 continue;
1882 }
1883 before = &fl->fl_next;
1884 }
1885 unlock_kernel();
1886out:
1887 if (lock.fl_ops && lock.fl_ops->fl_release_private)
1888 lock.fl_ops->fl_release_private(&lock);
1889}
1890
1891EXPORT_SYMBOL(locks_remove_posix);
1892
1893/*
1894 * This function is called on the last close of an open file.
1895 */
1896void locks_remove_flock(struct file *filp)
1897{
1898 struct inode * inode = filp->f_dentry->d_inode;
1899 struct file_lock *fl;
1900 struct file_lock **before;
1901
1902 if (!inode->i_flock)
1903 return;
1904
1905 if (filp->f_op && filp->f_op->flock) {
1906 struct file_lock fl = {
1907 .fl_pid = current->tgid,
1908 .fl_file = filp,
1909 .fl_flags = FL_FLOCK,
1910 .fl_type = F_UNLCK,
1911 .fl_end = OFFSET_MAX,
1912 };
1913 filp->f_op->flock(filp, F_SETLKW, &fl);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00001914 if (fl.fl_ops && fl.fl_ops->fl_release_private)
1915 fl.fl_ops->fl_release_private(&fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916 }
1917
1918 lock_kernel();
1919 before = &inode->i_flock;
1920
1921 while ((fl = *before) != NULL) {
1922 if (fl->fl_file == filp) {
Peter Staubachc2936212005-07-27 11:45:09 -07001923 if (IS_FLOCK(fl)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924 locks_delete_lock(before);
1925 continue;
1926 }
1927 if (IS_LEASE(fl)) {
1928 lease_modify(before, F_UNLCK);
1929 continue;
1930 }
1931 /* What? */
1932 BUG();
1933 }
1934 before = &fl->fl_next;
1935 }
1936 unlock_kernel();
1937}
1938
1939/**
1940 * posix_block_lock - blocks waiting for a file lock
1941 * @blocker: the lock which is blocking
1942 * @waiter: the lock which conflicts and has to wait
1943 *
1944 * lockd needs to block waiting for locks.
1945 */
1946void
1947posix_block_lock(struct file_lock *blocker, struct file_lock *waiter)
1948{
1949 locks_insert_block(blocker, waiter);
1950}
1951
1952EXPORT_SYMBOL(posix_block_lock);
1953
1954/**
1955 * posix_unblock_lock - stop waiting for a file lock
1956 * @filp: how the file was opened
1957 * @waiter: the lock which was waiting
1958 *
1959 * lockd needs to block waiting for locks.
1960 */
J. Bruce Fields64a318e2006-01-03 09:55:46 +01001961int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962posix_unblock_lock(struct file *filp, struct file_lock *waiter)
1963{
J. Bruce Fields64a318e2006-01-03 09:55:46 +01001964 int status = 0;
1965
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966 lock_kernel();
J. Bruce Fields5996a292006-01-03 09:55:44 +01001967 if (waiter->fl_next)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968 __locks_delete_block(waiter);
J. Bruce Fields64a318e2006-01-03 09:55:46 +01001969 else
1970 status = -ENOENT;
J. Bruce Fields5996a292006-01-03 09:55:44 +01001971 unlock_kernel();
J. Bruce Fields64a318e2006-01-03 09:55:46 +01001972 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973}
1974
1975EXPORT_SYMBOL(posix_unblock_lock);
1976
1977static void lock_get_status(char* out, struct file_lock *fl, int id, char *pfx)
1978{
1979 struct inode *inode = NULL;
1980
1981 if (fl->fl_file != NULL)
1982 inode = fl->fl_file->f_dentry->d_inode;
1983
1984 out += sprintf(out, "%d:%s ", id, pfx);
1985 if (IS_POSIX(fl)) {
1986 out += sprintf(out, "%6s %s ",
1987 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
1988 (inode == NULL) ? "*NOINODE*" :
1989 (IS_MANDLOCK(inode) &&
1990 (inode->i_mode & (S_IXGRP | S_ISGID)) == S_ISGID) ?
1991 "MANDATORY" : "ADVISORY ");
1992 } else if (IS_FLOCK(fl)) {
1993 if (fl->fl_type & LOCK_MAND) {
1994 out += sprintf(out, "FLOCK MSNFS ");
1995 } else {
1996 out += sprintf(out, "FLOCK ADVISORY ");
1997 }
1998 } else if (IS_LEASE(fl)) {
1999 out += sprintf(out, "LEASE ");
2000 if (fl->fl_type & F_INPROGRESS)
2001 out += sprintf(out, "BREAKING ");
2002 else if (fl->fl_file)
2003 out += sprintf(out, "ACTIVE ");
2004 else
2005 out += sprintf(out, "BREAKER ");
2006 } else {
2007 out += sprintf(out, "UNKNOWN UNKNOWN ");
2008 }
2009 if (fl->fl_type & LOCK_MAND) {
2010 out += sprintf(out, "%s ",
2011 (fl->fl_type & LOCK_READ)
2012 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2013 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2014 } else {
2015 out += sprintf(out, "%s ",
2016 (fl->fl_type & F_INPROGRESS)
2017 ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ "
2018 : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ ");
2019 }
2020 if (inode) {
2021#ifdef WE_CAN_BREAK_LSLK_NOW
2022 out += sprintf(out, "%d %s:%ld ", fl->fl_pid,
2023 inode->i_sb->s_id, inode->i_ino);
2024#else
2025 /* userspace relies on this representation of dev_t ;-( */
2026 out += sprintf(out, "%d %02x:%02x:%ld ", fl->fl_pid,
2027 MAJOR(inode->i_sb->s_dev),
2028 MINOR(inode->i_sb->s_dev), inode->i_ino);
2029#endif
2030 } else {
2031 out += sprintf(out, "%d <none>:0 ", fl->fl_pid);
2032 }
2033 if (IS_POSIX(fl)) {
2034 if (fl->fl_end == OFFSET_MAX)
2035 out += sprintf(out, "%Ld EOF\n", fl->fl_start);
2036 else
2037 out += sprintf(out, "%Ld %Ld\n", fl->fl_start,
2038 fl->fl_end);
2039 } else {
2040 out += sprintf(out, "0 EOF\n");
2041 }
2042}
2043
2044static void move_lock_status(char **p, off_t* pos, off_t offset)
2045{
2046 int len;
2047 len = strlen(*p);
2048 if(*pos >= offset) {
2049 /* the complete line is valid */
2050 *p += len;
2051 *pos += len;
2052 return;
2053 }
2054 if(*pos+len > offset) {
2055 /* use the second part of the line */
2056 int i = offset-*pos;
2057 memmove(*p,*p+i,len-i);
2058 *p += len-i;
2059 *pos += len;
2060 return;
2061 }
2062 /* discard the complete line */
2063 *pos += len;
2064}
2065
2066/**
2067 * get_locks_status - reports lock usage in /proc/locks
2068 * @buffer: address in userspace to write into
2069 * @start: ?
2070 * @offset: how far we are through the buffer
2071 * @length: how much to read
2072 */
2073
2074int get_locks_status(char *buffer, char **start, off_t offset, int length)
2075{
2076 struct list_head *tmp;
2077 char *q = buffer;
2078 off_t pos = 0;
2079 int i = 0;
2080
2081 lock_kernel();
2082 list_for_each(tmp, &file_lock_list) {
2083 struct list_head *btmp;
2084 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
2085 lock_get_status(q, fl, ++i, "");
2086 move_lock_status(&q, &pos, offset);
2087
2088 if(pos >= offset+length)
2089 goto done;
2090
2091 list_for_each(btmp, &fl->fl_block) {
2092 struct file_lock *bfl = list_entry(btmp,
2093 struct file_lock, fl_block);
2094 lock_get_status(q, bfl, i, " ->");
2095 move_lock_status(&q, &pos, offset);
2096
2097 if(pos >= offset+length)
2098 goto done;
2099 }
2100 }
2101done:
2102 unlock_kernel();
2103 *start = buffer;
2104 if(q-buffer < length)
2105 return (q-buffer);
2106 return length;
2107}
2108
2109/**
2110 * lock_may_read - checks that the region is free of locks
2111 * @inode: the inode that is being read
2112 * @start: the first byte to read
2113 * @len: the number of bytes to read
2114 *
2115 * Emulates Windows locking requirements. Whole-file
2116 * mandatory locks (share modes) can prohibit a read and
2117 * byte-range POSIX locks can prohibit a read if they overlap.
2118 *
2119 * N.B. this function is only ever called
2120 * from knfsd and ownership of locks is never checked.
2121 */
2122int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2123{
2124 struct file_lock *fl;
2125 int result = 1;
2126 lock_kernel();
2127 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2128 if (IS_POSIX(fl)) {
2129 if (fl->fl_type == F_RDLCK)
2130 continue;
2131 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2132 continue;
2133 } else if (IS_FLOCK(fl)) {
2134 if (!(fl->fl_type & LOCK_MAND))
2135 continue;
2136 if (fl->fl_type & LOCK_READ)
2137 continue;
2138 } else
2139 continue;
2140 result = 0;
2141 break;
2142 }
2143 unlock_kernel();
2144 return result;
2145}
2146
2147EXPORT_SYMBOL(lock_may_read);
2148
2149/**
2150 * lock_may_write - checks that the region is free of locks
2151 * @inode: the inode that is being written
2152 * @start: the first byte to write
2153 * @len: the number of bytes to write
2154 *
2155 * Emulates Windows locking requirements. Whole-file
2156 * mandatory locks (share modes) can prohibit a write and
2157 * byte-range POSIX locks can prohibit a write if they overlap.
2158 *
2159 * N.B. this function is only ever called
2160 * from knfsd and ownership of locks is never checked.
2161 */
2162int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2163{
2164 struct file_lock *fl;
2165 int result = 1;
2166 lock_kernel();
2167 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2168 if (IS_POSIX(fl)) {
2169 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2170 continue;
2171 } else if (IS_FLOCK(fl)) {
2172 if (!(fl->fl_type & LOCK_MAND))
2173 continue;
2174 if (fl->fl_type & LOCK_WRITE)
2175 continue;
2176 } else
2177 continue;
2178 result = 0;
2179 break;
2180 }
2181 unlock_kernel();
2182 return result;
2183}
2184
2185EXPORT_SYMBOL(lock_may_write);
2186
2187static inline void __steal_locks(struct file *file, fl_owner_t from)
2188{
2189 struct inode *inode = file->f_dentry->d_inode;
2190 struct file_lock *fl = inode->i_flock;
2191
2192 while (fl) {
2193 if (fl->fl_file == file && fl->fl_owner == from)
2194 fl->fl_owner = current->files;
2195 fl = fl->fl_next;
2196 }
2197}
2198
2199/* When getting ready for executing a binary, we make sure that current
2200 * has a files_struct on its own. Before dropping the old files_struct,
2201 * we take over ownership of all locks for all file descriptors we own.
2202 * Note that we may accidentally steal a lock for a file that a sibling
2203 * has created since the unshare_files() call.
2204 */
2205void steal_locks(fl_owner_t from)
2206{
2207 struct files_struct *files = current->files;
2208 int i, j;
Dipankar Sarmabadf1662005-09-09 13:04:10 -07002209 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002210
2211 if (from == files)
2212 return;
2213
2214 lock_kernel();
2215 j = 0;
Dipankar Sarma4fb3a532005-09-16 19:28:13 -07002216 rcu_read_lock();
Dipankar Sarmabadf1662005-09-09 13:04:10 -07002217 fdt = files_fdtable(files);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218 for (;;) {
2219 unsigned long set;
2220 i = j * __NFDBITS;
Dipankar Sarmabadf1662005-09-09 13:04:10 -07002221 if (i >= fdt->max_fdset || i >= fdt->max_fds)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222 break;
Dipankar Sarmabadf1662005-09-09 13:04:10 -07002223 set = fdt->open_fds->fds_bits[j++];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224 while (set) {
2225 if (set & 1) {
Dipankar Sarmabadf1662005-09-09 13:04:10 -07002226 struct file *file = fdt->fd[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227 if (file)
2228 __steal_locks(file, from);
2229 }
2230 i++;
2231 set >>= 1;
2232 }
2233 }
Dipankar Sarma4fb3a532005-09-16 19:28:13 -07002234 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002235 unlock_kernel();
2236}
2237EXPORT_SYMBOL(steal_locks);
2238
2239static int __init filelock_init(void)
2240{
2241 filelock_cache = kmem_cache_create("file_lock_cache",
2242 sizeof(struct file_lock), 0, SLAB_PANIC,
2243 init_once, NULL);
2244 return 0;
2245}
2246
2247core_initcall(filelock_init);