blob: 671a034dc999fffdf88856401694fda0cef87147 [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
Christoph Hellwig26bcbf92006-03-20 13:44:40 -0500142static LIST_HEAD(file_lock_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143static LIST_HEAD(blocked_list);
144
Christoph Lametere18b8902006-12-06 20:33:20 -0800145static struct kmem_cache *filelock_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146
147/* Allocate an empty lock structure. */
148static struct file_lock *locks_alloc_lock(void)
149{
Christoph Lametere94b1762006-12-06 20:33:17 -0800150 return kmem_cache_alloc(filelock_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151}
152
Trond Myklebust47831f32006-03-20 13:44:05 -0500153static void locks_release_private(struct file_lock *fl)
154{
155 if (fl->fl_ops) {
156 if (fl->fl_ops->fl_release_private)
157 fl->fl_ops->fl_release_private(fl);
158 fl->fl_ops = NULL;
159 }
160 if (fl->fl_lmops) {
161 if (fl->fl_lmops->fl_release_private)
162 fl->fl_lmops->fl_release_private(fl);
163 fl->fl_lmops = NULL;
164 }
165
166}
167
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168/* Free a lock which is not in use. */
Matt Mackall33443c42006-01-08 01:05:22 -0800169static void locks_free_lock(struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170{
Miklos Szeredi5ce29642006-03-31 02:30:29 -0800171 BUG_ON(waitqueue_active(&fl->fl_wait));
172 BUG_ON(!list_empty(&fl->fl_block));
173 BUG_ON(!list_empty(&fl->fl_link));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174
Trond Myklebust47831f32006-03-20 13:44:05 -0500175 locks_release_private(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 kmem_cache_free(filelock_cache, fl);
177}
178
179void locks_init_lock(struct file_lock *fl)
180{
181 INIT_LIST_HEAD(&fl->fl_link);
182 INIT_LIST_HEAD(&fl->fl_block);
183 init_waitqueue_head(&fl->fl_wait);
184 fl->fl_next = NULL;
185 fl->fl_fasync = NULL;
186 fl->fl_owner = NULL;
187 fl->fl_pid = 0;
188 fl->fl_file = NULL;
189 fl->fl_flags = 0;
190 fl->fl_type = 0;
191 fl->fl_start = fl->fl_end = 0;
192 fl->fl_ops = NULL;
193 fl->fl_lmops = NULL;
194}
195
196EXPORT_SYMBOL(locks_init_lock);
197
198/*
199 * Initialises the fields of the file lock which are invariant for
200 * free file_locks.
201 */
Christoph Lametere18b8902006-12-06 20:33:20 -0800202static void init_once(void *foo, struct kmem_cache *cache, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203{
204 struct file_lock *lock = (struct file_lock *) foo;
205
Christoph Lameter50953fe2007-05-06 14:50:16 -0700206 if (!(flags & SLAB_CTOR_CONSTRUCTOR))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 return;
208
209 locks_init_lock(lock);
210}
211
Trond Myklebust47831f32006-03-20 13:44:05 -0500212static void locks_copy_private(struct file_lock *new, struct file_lock *fl)
213{
214 if (fl->fl_ops) {
215 if (fl->fl_ops->fl_copy_lock)
216 fl->fl_ops->fl_copy_lock(new, fl);
217 new->fl_ops = fl->fl_ops;
218 }
219 if (fl->fl_lmops) {
220 if (fl->fl_lmops->fl_copy_lock)
221 fl->fl_lmops->fl_copy_lock(new, fl);
222 new->fl_lmops = fl->fl_lmops;
223 }
224}
225
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226/*
227 * Initialize a new lock from an existing file_lock structure.
228 */
Trond Myklebust09969052006-03-20 13:44:38 -0500229static void __locks_copy_lock(struct file_lock *new, const struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230{
231 new->fl_owner = fl->fl_owner;
232 new->fl_pid = fl->fl_pid;
Trond Myklebust09969052006-03-20 13:44:38 -0500233 new->fl_file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 new->fl_flags = fl->fl_flags;
235 new->fl_type = fl->fl_type;
236 new->fl_start = fl->fl_start;
237 new->fl_end = fl->fl_end;
Trond Myklebust09969052006-03-20 13:44:38 -0500238 new->fl_ops = NULL;
239 new->fl_lmops = NULL;
240}
241
242void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
243{
244 locks_release_private(new);
245
246 __locks_copy_lock(new, fl);
247 new->fl_file = fl->fl_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 new->fl_ops = fl->fl_ops;
249 new->fl_lmops = fl->fl_lmops;
Trond Myklebust47831f32006-03-20 13:44:05 -0500250
251 locks_copy_private(new, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252}
253
254EXPORT_SYMBOL(locks_copy_lock);
255
256static inline int flock_translate_cmd(int cmd) {
257 if (cmd & LOCK_MAND)
258 return cmd & (LOCK_MAND | LOCK_RW);
259 switch (cmd) {
260 case LOCK_SH:
261 return F_RDLCK;
262 case LOCK_EX:
263 return F_WRLCK;
264 case LOCK_UN:
265 return F_UNLCK;
266 }
267 return -EINVAL;
268}
269
270/* Fill in a file_lock structure with an appropriate FLOCK lock. */
271static int flock_make_lock(struct file *filp, struct file_lock **lock,
272 unsigned int cmd)
273{
274 struct file_lock *fl;
275 int type = flock_translate_cmd(cmd);
276 if (type < 0)
277 return type;
278
279 fl = locks_alloc_lock();
280 if (fl == NULL)
281 return -ENOMEM;
282
283 fl->fl_file = filp;
284 fl->fl_pid = current->tgid;
285 fl->fl_flags = FL_FLOCK;
286 fl->fl_type = type;
287 fl->fl_end = OFFSET_MAX;
288
289 *lock = fl;
290 return 0;
291}
292
293static int assign_type(struct file_lock *fl, int type)
294{
295 switch (type) {
296 case F_RDLCK:
297 case F_WRLCK:
298 case F_UNLCK:
299 fl->fl_type = type;
300 break;
301 default:
302 return -EINVAL;
303 }
304 return 0;
305}
306
307/* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
308 * style lock.
309 */
310static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
311 struct flock *l)
312{
313 off_t start, end;
314
315 switch (l->l_whence) {
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700316 case SEEK_SET:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 start = 0;
318 break;
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700319 case SEEK_CUR:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 start = filp->f_pos;
321 break;
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700322 case SEEK_END:
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800323 start = i_size_read(filp->f_path.dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 break;
325 default:
326 return -EINVAL;
327 }
328
329 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
330 POSIX-2001 defines it. */
331 start += l->l_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 if (start < 0)
333 return -EINVAL;
Trond Myklebust4c780a42005-10-18 14:20:21 -0700334 fl->fl_end = OFFSET_MAX;
335 if (l->l_len > 0) {
336 end = start + l->l_len - 1;
337 fl->fl_end = end;
338 } else if (l->l_len < 0) {
339 end = start - 1;
340 fl->fl_end = end;
341 start += l->l_len;
342 if (start < 0)
343 return -EINVAL;
344 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 fl->fl_start = start; /* we record the absolute position */
Trond Myklebust4c780a42005-10-18 14:20:21 -0700346 if (fl->fl_end < fl->fl_start)
347 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348
349 fl->fl_owner = current->files;
350 fl->fl_pid = current->tgid;
351 fl->fl_file = filp;
352 fl->fl_flags = FL_POSIX;
353 fl->fl_ops = NULL;
354 fl->fl_lmops = NULL;
355
356 return assign_type(fl, l->l_type);
357}
358
359#if BITS_PER_LONG == 32
360static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
361 struct flock64 *l)
362{
363 loff_t start;
364
365 switch (l->l_whence) {
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700366 case SEEK_SET:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 start = 0;
368 break;
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700369 case SEEK_CUR:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 start = filp->f_pos;
371 break;
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700372 case SEEK_END:
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800373 start = i_size_read(filp->f_path.dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 break;
375 default:
376 return -EINVAL;
377 }
378
Trond Myklebust4c780a42005-10-18 14:20:21 -0700379 start += l->l_start;
380 if (start < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 return -EINVAL;
Trond Myklebust4c780a42005-10-18 14:20:21 -0700382 fl->fl_end = OFFSET_MAX;
383 if (l->l_len > 0) {
384 fl->fl_end = start + l->l_len - 1;
385 } else if (l->l_len < 0) {
386 fl->fl_end = start - 1;
387 start += l->l_len;
388 if (start < 0)
389 return -EINVAL;
390 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 fl->fl_start = start; /* we record the absolute position */
Trond Myklebust4c780a42005-10-18 14:20:21 -0700392 if (fl->fl_end < fl->fl_start)
393 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394
395 fl->fl_owner = current->files;
396 fl->fl_pid = current->tgid;
397 fl->fl_file = filp;
398 fl->fl_flags = FL_POSIX;
399 fl->fl_ops = NULL;
400 fl->fl_lmops = NULL;
401
402 switch (l->l_type) {
403 case F_RDLCK:
404 case F_WRLCK:
405 case F_UNLCK:
406 fl->fl_type = l->l_type;
407 break;
408 default:
409 return -EINVAL;
410 }
411
412 return (0);
413}
414#endif
415
416/* default lease lock manager operations */
417static void lease_break_callback(struct file_lock *fl)
418{
419 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
420}
421
422static void lease_release_private_callback(struct file_lock *fl)
423{
424 if (!fl->fl_file)
425 return;
426
427 f_delown(fl->fl_file);
428 fl->fl_file->f_owner.signum = 0;
429}
430
Adrian Bunk75c96f82005-05-05 16:16:09 -0700431static int lease_mylease_callback(struct file_lock *fl, struct file_lock *try)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432{
433 return fl->fl_file == try->fl_file;
434}
435
Adrian Bunk75c96f82005-05-05 16:16:09 -0700436static struct lock_manager_operations lease_manager_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 .fl_break = lease_break_callback,
438 .fl_release_private = lease_release_private_callback,
439 .fl_mylease = lease_mylease_callback,
440 .fl_change = lease_modify,
441};
442
443/*
444 * Initialize a lease, use the default lock manager operations
445 */
446static int lease_init(struct file *filp, int type, struct file_lock *fl)
447 {
Trond Myklebust75dff552006-05-07 23:02:42 -0400448 if (assign_type(fl, type) != 0)
449 return -EINVAL;
450
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 fl->fl_owner = current->files;
452 fl->fl_pid = current->tgid;
453
454 fl->fl_file = filp;
455 fl->fl_flags = FL_LEASE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 fl->fl_start = 0;
457 fl->fl_end = OFFSET_MAX;
458 fl->fl_ops = NULL;
459 fl->fl_lmops = &lease_manager_ops;
460 return 0;
461}
462
463/* Allocate a file_lock initialised to this type of lease */
464static int lease_alloc(struct file *filp, int type, struct file_lock **flp)
465{
466 struct file_lock *fl = locks_alloc_lock();
Trond Myklebust75dff552006-05-07 23:02:42 -0400467 int error = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468
469 if (fl == NULL)
Trond Myklebust75dff552006-05-07 23:02:42 -0400470 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471
472 error = lease_init(filp, type, fl);
Trond Myklebust75dff552006-05-07 23:02:42 -0400473 if (error) {
474 locks_free_lock(fl);
475 fl = NULL;
476 }
477out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 *flp = fl;
Trond Myklebust75dff552006-05-07 23:02:42 -0400479 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480}
481
482/* Check if two locks overlap each other.
483 */
484static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
485{
486 return ((fl1->fl_end >= fl2->fl_start) &&
487 (fl2->fl_end >= fl1->fl_start));
488}
489
490/*
491 * Check whether two locks have the same owner.
492 */
Matt Mackall33443c42006-01-08 01:05:22 -0800493static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494{
495 if (fl1->fl_lmops && fl1->fl_lmops->fl_compare_owner)
496 return fl2->fl_lmops == fl1->fl_lmops &&
497 fl1->fl_lmops->fl_compare_owner(fl1, fl2);
498 return fl1->fl_owner == fl2->fl_owner;
499}
500
501/* Remove waiter from blocker's block list.
502 * When blocker ends up pointing to itself then the list is empty.
503 */
Matt Mackall33443c42006-01-08 01:05:22 -0800504static void __locks_delete_block(struct file_lock *waiter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505{
506 list_del_init(&waiter->fl_block);
507 list_del_init(&waiter->fl_link);
508 waiter->fl_next = NULL;
509}
510
511/*
512 */
513static void locks_delete_block(struct file_lock *waiter)
514{
515 lock_kernel();
516 __locks_delete_block(waiter);
517 unlock_kernel();
518}
519
520/* Insert waiter into blocker's block list.
521 * We use a circular list so that processes can be easily woken up in
522 * the order they blocked. The documentation doesn't require this but
523 * it seems like the reasonable thing to do.
524 */
525static void locks_insert_block(struct file_lock *blocker,
526 struct file_lock *waiter)
527{
J. Bruce Fields6dc0fe82006-03-26 01:37:24 -0800528 BUG_ON(!list_empty(&waiter->fl_block));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 list_add_tail(&waiter->fl_block, &blocker->fl_block);
530 waiter->fl_next = blocker;
531 if (IS_POSIX(blocker))
532 list_add(&waiter->fl_link, &blocked_list);
533}
534
535/* Wake up processes blocked waiting for blocker.
536 * If told to wait then schedule the processes until the block list
537 * is empty, otherwise empty the block list ourselves.
538 */
539static void locks_wake_up_blocks(struct file_lock *blocker)
540{
541 while (!list_empty(&blocker->fl_block)) {
542 struct file_lock *waiter = list_entry(blocker->fl_block.next,
543 struct file_lock, fl_block);
544 __locks_delete_block(waiter);
545 if (waiter->fl_lmops && waiter->fl_lmops->fl_notify)
546 waiter->fl_lmops->fl_notify(waiter);
547 else
548 wake_up(&waiter->fl_wait);
549 }
550}
551
552/* Insert file lock fl into an inode's lock list at the position indicated
553 * by pos. At the same time add the lock to the global file lock list.
554 */
555static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
556{
557 list_add(&fl->fl_link, &file_lock_list);
558
559 /* insert into file's list */
560 fl->fl_next = *pos;
561 *pos = fl;
562
563 if (fl->fl_ops && fl->fl_ops->fl_insert)
564 fl->fl_ops->fl_insert(fl);
565}
566
567/*
568 * Delete a lock and then free it.
569 * Wake up processes that are blocked waiting for this lock,
570 * notify the FS that the lock has been cleared and
571 * finally free the lock.
572 */
573static void locks_delete_lock(struct file_lock **thisfl_p)
574{
575 struct file_lock *fl = *thisfl_p;
576
577 *thisfl_p = fl->fl_next;
578 fl->fl_next = NULL;
579 list_del_init(&fl->fl_link);
580
581 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
582 if (fl->fl_fasync != NULL) {
583 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
584 fl->fl_fasync = NULL;
585 }
586
587 if (fl->fl_ops && fl->fl_ops->fl_remove)
588 fl->fl_ops->fl_remove(fl);
589
590 locks_wake_up_blocks(fl);
591 locks_free_lock(fl);
592}
593
594/* Determine if lock sys_fl blocks lock caller_fl. Common functionality
595 * checks for shared/exclusive status of overlapping locks.
596 */
597static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
598{
599 if (sys_fl->fl_type == F_WRLCK)
600 return 1;
601 if (caller_fl->fl_type == F_WRLCK)
602 return 1;
603 return 0;
604}
605
606/* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
607 * checking before calling the locks_conflict().
608 */
609static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
610{
611 /* POSIX locks owned by the same process do not conflict with
612 * each other.
613 */
614 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
615 return (0);
616
617 /* Check whether they overlap */
618 if (!locks_overlap(caller_fl, sys_fl))
619 return 0;
620
621 return (locks_conflict(caller_fl, sys_fl));
622}
623
624/* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
625 * checking before calling the locks_conflict().
626 */
627static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
628{
629 /* FLOCK locks referring to the same filp do not conflict with
630 * each other.
631 */
632 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
633 return (0);
634 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
635 return 0;
636
637 return (locks_conflict(caller_fl, sys_fl));
638}
639
640static int interruptible_sleep_on_locked(wait_queue_head_t *fl_wait, int timeout)
641{
642 int result = 0;
643 DECLARE_WAITQUEUE(wait, current);
644
645 __set_current_state(TASK_INTERRUPTIBLE);
646 add_wait_queue(fl_wait, &wait);
647 if (timeout == 0)
648 schedule();
649 else
650 result = schedule_timeout(timeout);
651 if (signal_pending(current))
652 result = -ERESTARTSYS;
653 remove_wait_queue(fl_wait, &wait);
654 __set_current_state(TASK_RUNNING);
655 return result;
656}
657
658static int locks_block_on_timeout(struct file_lock *blocker, struct file_lock *waiter, int time)
659{
660 int result;
661 locks_insert_block(blocker, waiter);
662 result = interruptible_sleep_on_locked(&waiter->fl_wait, time);
663 __locks_delete_block(waiter);
664 return result;
665}
666
Andy Adamson8dc7c312006-03-20 13:44:26 -0500667int
Marc Eshel9d6a8c52007-02-21 00:55:18 -0500668posix_test_lock(struct file *filp, struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669{
670 struct file_lock *cfl;
671
Marc Eshel9d6a8c52007-02-21 00:55:18 -0500672 fl->fl_type = F_UNLCK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 lock_kernel();
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800674 for (cfl = filp->f_path.dentry->d_inode->i_flock; cfl; cfl = cfl->fl_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 if (!IS_POSIX(cfl))
676 continue;
677 if (posix_locks_conflict(cfl, fl))
678 break;
679 }
Andy Adamson8dc7c312006-03-20 13:44:26 -0500680 if (cfl) {
Marc Eshel9d6a8c52007-02-21 00:55:18 -0500681 __locks_copy_lock(fl, cfl);
Andy Adamson8dc7c312006-03-20 13:44:26 -0500682 unlock_kernel();
683 return 1;
684 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 unlock_kernel();
Andy Adamson8dc7c312006-03-20 13:44:26 -0500686 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687}
688
689EXPORT_SYMBOL(posix_test_lock);
690
691/* This function tests for deadlock condition before putting a process to
692 * sleep. The detection scheme is no longer recursive. Recursive was neat,
693 * but dangerous - we risked stack corruption if the lock data was bad, or
694 * if the recursion was too deep for any other reason.
695 *
696 * We rely on the fact that a task can only be on one lock's wait queue
697 * at a time. When we find blocked_task on a wait queue we can re-search
698 * with blocked_task equal to that queue's owner, until either blocked_task
699 * isn't found, or blocked_task is found on a queue owned by my_task.
700 *
701 * Note: the above assumption may not be true when handling lock requests
702 * from a broken NFS client. But broken NFS clients have a lot more to
703 * worry about than proper deadlock detection anyway... --okir
704 */
Adrian Bunkb0904e12006-06-23 02:05:13 -0700705static int posix_locks_deadlock(struct file_lock *caller_fl,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 struct file_lock *block_fl)
707{
708 struct list_head *tmp;
709
710next_task:
711 if (posix_same_owner(caller_fl, block_fl))
712 return 1;
713 list_for_each(tmp, &blocked_list) {
714 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
715 if (posix_same_owner(fl, block_fl)) {
716 fl = fl->fl_next;
717 block_fl = fl;
718 goto next_task;
719 }
720 }
721 return 0;
722}
723
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724/* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
725 * at the head of the list, but that's secret knowledge known only to
726 * flock_lock_file and posix_lock_file.
Trond Myklebustf475ae92006-06-29 16:38:32 -0400727 *
728 * Note that if called with an FL_EXISTS argument, the caller may determine
729 * whether or not a lock was successfully freed by testing the return
730 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 */
Trond Myklebust993dfa82006-03-31 02:30:55 -0800732static int flock_lock_file(struct file *filp, struct file_lock *request)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733{
Trond Myklebust993dfa82006-03-31 02:30:55 -0800734 struct file_lock *new_fl = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 struct file_lock **before;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800736 struct inode * inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 int error = 0;
738 int found = 0;
739
740 lock_kernel();
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400741 if (request->fl_flags & FL_ACCESS)
742 goto find_conflict;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 for_each_lock(inode, before) {
744 struct file_lock *fl = *before;
745 if (IS_POSIX(fl))
746 break;
747 if (IS_LEASE(fl))
748 continue;
749 if (filp != fl->fl_file)
750 continue;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800751 if (request->fl_type == fl->fl_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 goto out;
753 found = 1;
754 locks_delete_lock(before);
755 break;
756 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757
Trond Myklebustf475ae92006-06-29 16:38:32 -0400758 if (request->fl_type == F_UNLCK) {
759 if ((request->fl_flags & FL_EXISTS) && !found)
760 error = -ENOENT;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800761 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -0400762 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763
Kirill Korotaev9cedc192006-06-14 17:59:35 +0400764 error = -ENOMEM;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800765 new_fl = locks_alloc_lock();
766 if (new_fl == NULL)
767 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 /*
769 * If a higher-priority process was blocked on the old file lock,
770 * give it the opportunity to lock the file.
771 */
772 if (found)
773 cond_resched();
774
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400775find_conflict:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 for_each_lock(inode, before) {
777 struct file_lock *fl = *before;
778 if (IS_POSIX(fl))
779 break;
780 if (IS_LEASE(fl))
781 continue;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800782 if (!flock_locks_conflict(request, fl))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 continue;
784 error = -EAGAIN;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800785 if (request->fl_flags & FL_SLEEP)
786 locks_insert_block(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 goto out;
788 }
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400789 if (request->fl_flags & FL_ACCESS)
790 goto out;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800791 locks_copy_lock(new_fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 locks_insert_lock(&inode->i_flock, new_fl);
Trond Myklebust993dfa82006-03-31 02:30:55 -0800793 new_fl = NULL;
Kirill Korotaev9cedc192006-06-14 17:59:35 +0400794 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795
796out:
797 unlock_kernel();
Trond Myklebust993dfa82006-03-31 02:30:55 -0800798 if (new_fl)
799 locks_free_lock(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 return error;
801}
802
Marc Eshel150b3932007-01-18 16:15:35 -0500803static int __posix_lock_file(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804{
805 struct file_lock *fl;
Miklos Szeredi39005d02006-06-23 02:05:10 -0700806 struct file_lock *new_fl = NULL;
807 struct file_lock *new_fl2 = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 struct file_lock *left = NULL;
809 struct file_lock *right = NULL;
810 struct file_lock **before;
811 int error, added = 0;
812
813 /*
814 * We may need two file_lock structures for this operation,
815 * so we get them in advance to avoid races.
Miklos Szeredi39005d02006-06-23 02:05:10 -0700816 *
817 * In some cases we can be sure, that no new locks will be needed
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 */
Miklos Szeredi39005d02006-06-23 02:05:10 -0700819 if (!(request->fl_flags & FL_ACCESS) &&
820 (request->fl_type != F_UNLCK ||
821 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
822 new_fl = locks_alloc_lock();
823 new_fl2 = locks_alloc_lock();
824 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825
826 lock_kernel();
827 if (request->fl_type != F_UNLCK) {
828 for_each_lock(inode, before) {
829 struct file_lock *fl = *before;
830 if (!IS_POSIX(fl))
831 continue;
832 if (!posix_locks_conflict(request, fl))
833 continue;
Andy Adamson5842add2006-03-26 01:37:26 -0800834 if (conflock)
835 locks_copy_lock(conflock, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 error = -EAGAIN;
837 if (!(request->fl_flags & FL_SLEEP))
838 goto out;
839 error = -EDEADLK;
840 if (posix_locks_deadlock(request, fl))
841 goto out;
842 error = -EAGAIN;
843 locks_insert_block(fl, request);
844 goto out;
845 }
846 }
847
848 /* If we're just looking for a conflict, we're done. */
849 error = 0;
850 if (request->fl_flags & FL_ACCESS)
851 goto out;
852
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 * Find the first old lock with the same owner as the new lock.
855 */
856
857 before = &inode->i_flock;
858
859 /* First skip locks owned by other processes. */
860 while ((fl = *before) && (!IS_POSIX(fl) ||
861 !posix_same_owner(request, fl))) {
862 before = &fl->fl_next;
863 }
864
865 /* Process locks with this owner. */
866 while ((fl = *before) && posix_same_owner(request, fl)) {
867 /* Detect adjacent or overlapping regions (if same lock type)
868 */
869 if (request->fl_type == fl->fl_type) {
Olaf Kirch449231d2005-08-25 16:25:35 -0700870 /* In all comparisons of start vs end, use
871 * "start - 1" rather than "end + 1". If end
872 * is OFFSET_MAX, end + 1 will become negative.
873 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 if (fl->fl_end < request->fl_start - 1)
875 goto next_lock;
876 /* If the next lock in the list has entirely bigger
877 * addresses than the new one, insert the lock here.
878 */
Olaf Kirch449231d2005-08-25 16:25:35 -0700879 if (fl->fl_start - 1 > request->fl_end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880 break;
881
882 /* If we come here, the new and old lock are of the
883 * same type and adjacent or overlapping. Make one
884 * lock yielding from the lower start address of both
885 * locks to the higher end address.
886 */
887 if (fl->fl_start > request->fl_start)
888 fl->fl_start = request->fl_start;
889 else
890 request->fl_start = fl->fl_start;
891 if (fl->fl_end < request->fl_end)
892 fl->fl_end = request->fl_end;
893 else
894 request->fl_end = fl->fl_end;
895 if (added) {
896 locks_delete_lock(before);
897 continue;
898 }
899 request = fl;
900 added = 1;
901 }
902 else {
903 /* Processing for different lock types is a bit
904 * more complex.
905 */
906 if (fl->fl_end < request->fl_start)
907 goto next_lock;
908 if (fl->fl_start > request->fl_end)
909 break;
910 if (request->fl_type == F_UNLCK)
911 added = 1;
912 if (fl->fl_start < request->fl_start)
913 left = fl;
914 /* If the next lock in the list has a higher end
915 * address than the new one, insert the new one here.
916 */
917 if (fl->fl_end > request->fl_end) {
918 right = fl;
919 break;
920 }
921 if (fl->fl_start >= request->fl_start) {
922 /* The new lock completely replaces an old
923 * one (This may happen several times).
924 */
925 if (added) {
926 locks_delete_lock(before);
927 continue;
928 }
929 /* Replace the old lock with the new one.
930 * Wake up anybody waiting for the old one,
931 * as the change in lock type might satisfy
932 * their needs.
933 */
934 locks_wake_up_blocks(fl);
935 fl->fl_start = request->fl_start;
936 fl->fl_end = request->fl_end;
937 fl->fl_type = request->fl_type;
Trond Myklebust47831f32006-03-20 13:44:05 -0500938 locks_release_private(fl);
939 locks_copy_private(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 request = fl;
941 added = 1;
942 }
943 }
944 /* Go on to next lock.
945 */
946 next_lock:
947 before = &fl->fl_next;
948 }
949
Miklos Szeredi0d9a4902006-06-23 02:05:09 -0700950 /*
951 * The above code only modifies existing locks in case of
952 * merging or replacing. If new lock(s) need to be inserted
953 * all modifications are done bellow this, so it's safe yet to
954 * bail out.
955 */
956 error = -ENOLCK; /* "no luck" */
957 if (right && left == right && !new_fl2)
958 goto out;
959
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 error = 0;
961 if (!added) {
Trond Myklebustf475ae92006-06-29 16:38:32 -0400962 if (request->fl_type == F_UNLCK) {
963 if (request->fl_flags & FL_EXISTS)
964 error = -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -0400966 }
Miklos Szeredi0d9a4902006-06-23 02:05:09 -0700967
968 if (!new_fl) {
969 error = -ENOLCK;
970 goto out;
971 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 locks_copy_lock(new_fl, request);
973 locks_insert_lock(before, new_fl);
974 new_fl = NULL;
975 }
976 if (right) {
977 if (left == right) {
978 /* The new lock breaks the old one in two pieces,
979 * so we have to use the second new lock.
980 */
981 left = new_fl2;
982 new_fl2 = NULL;
983 locks_copy_lock(left, right);
984 locks_insert_lock(before, left);
985 }
986 right->fl_start = request->fl_end + 1;
987 locks_wake_up_blocks(right);
988 }
989 if (left) {
990 left->fl_end = request->fl_start - 1;
991 locks_wake_up_blocks(left);
992 }
993 out:
994 unlock_kernel();
995 /*
996 * Free any unused locks.
997 */
998 if (new_fl)
999 locks_free_lock(new_fl);
1000 if (new_fl2)
1001 locks_free_lock(new_fl2);
1002 return error;
1003}
1004
1005/**
1006 * posix_lock_file - Apply a POSIX-style lock to a file
1007 * @filp: The file to apply the lock to
1008 * @fl: The lock to be applied
Marc Eshel150b3932007-01-18 16:15:35 -05001009 * @conflock: Place to return a copy of the conflicting lock, if found.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 *
1011 * Add a POSIX style lock to a file.
1012 * We merge adjacent & overlapping locks whenever possible.
1013 * POSIX locks are sorted by owner task, then by starting address
Trond Myklebustf475ae92006-06-29 16:38:32 -04001014 *
1015 * Note that if called with an FL_EXISTS argument, the caller may determine
1016 * whether or not a lock was successfully freed by testing the return
1017 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 */
Marc Eshel150b3932007-01-18 16:15:35 -05001019int posix_lock_file(struct file *filp, struct file_lock *fl,
Andy Adamson5842add2006-03-26 01:37:26 -08001020 struct file_lock *conflock)
1021{
Marc Eshel150b3932007-01-18 16:15:35 -05001022 return __posix_lock_file(filp->f_path.dentry->d_inode, fl, conflock);
Andy Adamson5842add2006-03-26 01:37:26 -08001023}
Marc Eshel150b3932007-01-18 16:15:35 -05001024EXPORT_SYMBOL(posix_lock_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025
1026/**
1027 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1028 * @filp: The file to apply the lock to
1029 * @fl: The lock to be applied
1030 *
1031 * Add a POSIX style lock to a file.
1032 * We merge adjacent & overlapping locks whenever possible.
1033 * POSIX locks are sorted by owner task, then by starting address
1034 */
1035int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1036{
1037 int error;
1038 might_sleep ();
1039 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001040 error = posix_lock_file(filp, fl, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1042 break;
1043 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1044 if (!error)
1045 continue;
1046
1047 locks_delete_block(fl);
1048 break;
1049 }
1050 return error;
1051}
1052EXPORT_SYMBOL(posix_lock_file_wait);
1053
1054/**
1055 * locks_mandatory_locked - Check for an active lock
1056 * @inode: the file to check
1057 *
1058 * Searches the inode's list of locks to find any POSIX locks which conflict.
1059 * This function is called from locks_verify_locked() only.
1060 */
1061int locks_mandatory_locked(struct inode *inode)
1062{
1063 fl_owner_t owner = current->files;
1064 struct file_lock *fl;
1065
1066 /*
1067 * Search the lock list for this inode for any POSIX locks.
1068 */
1069 lock_kernel();
1070 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1071 if (!IS_POSIX(fl))
1072 continue;
1073 if (fl->fl_owner != owner)
1074 break;
1075 }
1076 unlock_kernel();
1077 return fl ? -EAGAIN : 0;
1078}
1079
1080/**
1081 * locks_mandatory_area - Check for a conflicting lock
1082 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1083 * for shared
1084 * @inode: the file to check
1085 * @filp: how the file was opened (if it was)
1086 * @offset: start of area to check
1087 * @count: length of area to check
1088 *
1089 * Searches the inode's list of locks to find any POSIX locks which conflict.
1090 * This function is called from rw_verify_area() and
1091 * locks_verify_truncate().
1092 */
1093int locks_mandatory_area(int read_write, struct inode *inode,
1094 struct file *filp, loff_t offset,
1095 size_t count)
1096{
1097 struct file_lock fl;
1098 int error;
1099
1100 locks_init_lock(&fl);
1101 fl.fl_owner = current->files;
1102 fl.fl_pid = current->tgid;
1103 fl.fl_file = filp;
1104 fl.fl_flags = FL_POSIX | FL_ACCESS;
1105 if (filp && !(filp->f_flags & O_NONBLOCK))
1106 fl.fl_flags |= FL_SLEEP;
1107 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1108 fl.fl_start = offset;
1109 fl.fl_end = offset + count - 1;
1110
1111 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001112 error = __posix_lock_file(inode, &fl, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113 if (error != -EAGAIN)
1114 break;
1115 if (!(fl.fl_flags & FL_SLEEP))
1116 break;
1117 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1118 if (!error) {
1119 /*
1120 * If we've been sleeping someone might have
1121 * changed the permissions behind our back.
1122 */
1123 if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID)
1124 continue;
1125 }
1126
1127 locks_delete_block(&fl);
1128 break;
1129 }
1130
1131 return error;
1132}
1133
1134EXPORT_SYMBOL(locks_mandatory_area);
1135
1136/* We already had a lease on this file; just change its type */
1137int lease_modify(struct file_lock **before, int arg)
1138{
1139 struct file_lock *fl = *before;
1140 int error = assign_type(fl, arg);
1141
1142 if (error)
1143 return error;
1144 locks_wake_up_blocks(fl);
1145 if (arg == F_UNLCK)
1146 locks_delete_lock(before);
1147 return 0;
1148}
1149
1150EXPORT_SYMBOL(lease_modify);
1151
1152static void time_out_leases(struct inode *inode)
1153{
1154 struct file_lock **before;
1155 struct file_lock *fl;
1156
1157 before = &inode->i_flock;
1158 while ((fl = *before) && IS_LEASE(fl) && (fl->fl_type & F_INPROGRESS)) {
1159 if ((fl->fl_break_time == 0)
1160 || time_before(jiffies, fl->fl_break_time)) {
1161 before = &fl->fl_next;
1162 continue;
1163 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 lease_modify(before, fl->fl_type & ~F_INPROGRESS);
1165 if (fl == *before) /* lease_modify may have freed fl */
1166 before = &fl->fl_next;
1167 }
1168}
1169
1170/**
1171 * __break_lease - revoke all outstanding leases on file
1172 * @inode: the inode of the file to return
1173 * @mode: the open mode (read or write)
1174 *
1175 * break_lease (inlined for speed) has checked there already
1176 * is a lease on this file. Leases are broken on a call to open()
1177 * or truncate(). This function can sleep unless you
1178 * specified %O_NONBLOCK to your open().
1179 */
1180int __break_lease(struct inode *inode, unsigned int mode)
1181{
1182 int error = 0, future;
1183 struct file_lock *new_fl, *flock;
1184 struct file_lock *fl;
1185 int alloc_err;
1186 unsigned long break_time;
1187 int i_have_this_lease = 0;
1188
1189 alloc_err = lease_alloc(NULL, mode & FMODE_WRITE ? F_WRLCK : F_RDLCK,
1190 &new_fl);
1191
1192 lock_kernel();
1193
1194 time_out_leases(inode);
1195
1196 flock = inode->i_flock;
1197 if ((flock == NULL) || !IS_LEASE(flock))
1198 goto out;
1199
1200 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1201 if (fl->fl_owner == current->files)
1202 i_have_this_lease = 1;
1203
1204 if (mode & FMODE_WRITE) {
1205 /* If we want write access, we have to revoke any lease. */
1206 future = F_UNLCK | F_INPROGRESS;
1207 } else if (flock->fl_type & F_INPROGRESS) {
1208 /* If the lease is already being broken, we just leave it */
1209 future = flock->fl_type;
1210 } else if (flock->fl_type & F_WRLCK) {
1211 /* Downgrade the exclusive lease to a read-only lease. */
1212 future = F_RDLCK | F_INPROGRESS;
1213 } else {
1214 /* the existing lease was read-only, so we can read too. */
1215 goto out;
1216 }
1217
1218 if (alloc_err && !i_have_this_lease && ((mode & O_NONBLOCK) == 0)) {
1219 error = alloc_err;
1220 goto out;
1221 }
1222
1223 break_time = 0;
1224 if (lease_break_time > 0) {
1225 break_time = jiffies + lease_break_time * HZ;
1226 if (break_time == 0)
1227 break_time++; /* so that 0 means no break time */
1228 }
1229
1230 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1231 if (fl->fl_type != future) {
1232 fl->fl_type = future;
1233 fl->fl_break_time = break_time;
1234 /* lease must have lmops break callback */
1235 fl->fl_lmops->fl_break(fl);
1236 }
1237 }
1238
1239 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1240 error = -EWOULDBLOCK;
1241 goto out;
1242 }
1243
1244restart:
1245 break_time = flock->fl_break_time;
1246 if (break_time != 0) {
1247 break_time -= jiffies;
1248 if (break_time == 0)
1249 break_time++;
1250 }
1251 error = locks_block_on_timeout(flock, new_fl, break_time);
1252 if (error >= 0) {
1253 if (error == 0)
1254 time_out_leases(inode);
1255 /* Wait for the next lease that has not been broken yet */
1256 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1257 flock = flock->fl_next) {
1258 if (flock->fl_type & F_INPROGRESS)
1259 goto restart;
1260 }
1261 error = 0;
1262 }
1263
1264out:
1265 unlock_kernel();
1266 if (!alloc_err)
1267 locks_free_lock(new_fl);
1268 return error;
1269}
1270
1271EXPORT_SYMBOL(__break_lease);
1272
1273/**
1274 * lease_get_mtime
1275 * @inode: the inode
1276 * @time: pointer to a timespec which will contain the last modified time
1277 *
1278 * This is to force NFS clients to flush their caches for files with
1279 * exclusive leases. The justification is that if someone has an
1280 * exclusive lease, then they could be modifiying it.
1281 */
1282void lease_get_mtime(struct inode *inode, struct timespec *time)
1283{
1284 struct file_lock *flock = inode->i_flock;
1285 if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK))
1286 *time = current_fs_time(inode->i_sb);
1287 else
1288 *time = inode->i_mtime;
1289}
1290
1291EXPORT_SYMBOL(lease_get_mtime);
1292
1293/**
1294 * fcntl_getlease - Enquire what lease is currently active
1295 * @filp: the file
1296 *
1297 * The value returned by this function will be one of
1298 * (if no lease break is pending):
1299 *
1300 * %F_RDLCK to indicate a shared lease is held.
1301 *
1302 * %F_WRLCK to indicate an exclusive lease is held.
1303 *
1304 * %F_UNLCK to indicate no lease is held.
1305 *
1306 * (if a lease break is pending):
1307 *
1308 * %F_RDLCK to indicate an exclusive lease needs to be
1309 * changed to a shared lease (or removed).
1310 *
1311 * %F_UNLCK to indicate the lease needs to be removed.
1312 *
1313 * XXX: sfr & willy disagree over whether F_INPROGRESS
1314 * should be returned to userspace.
1315 */
1316int fcntl_getlease(struct file *filp)
1317{
1318 struct file_lock *fl;
1319 int type = F_UNLCK;
1320
1321 lock_kernel();
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001322 time_out_leases(filp->f_path.dentry->d_inode);
1323 for (fl = filp->f_path.dentry->d_inode->i_flock; fl && IS_LEASE(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324 fl = fl->fl_next) {
1325 if (fl->fl_file == filp) {
1326 type = fl->fl_type & ~F_INPROGRESS;
1327 break;
1328 }
1329 }
1330 unlock_kernel();
1331 return type;
1332}
1333
1334/**
1335 * __setlease - sets a lease on an open file
1336 * @filp: file pointer
1337 * @arg: type of lease to obtain
1338 * @flp: input - file_lock to use, output - file_lock inserted
1339 *
1340 * The (input) flp->fl_lmops->fl_break function is required
1341 * by break_lease().
1342 *
1343 * Called with kernel lock held.
1344 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07001345static int __setlease(struct file *filp, long arg, struct file_lock **flp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346{
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001347 struct file_lock *fl, **before, **my_before = NULL, *lease;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001348 struct dentry *dentry = filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 struct inode *inode = dentry->d_inode;
1350 int error, rdlease_count = 0, wrlease_count = 0;
1351
1352 time_out_leases(inode);
1353
1354 error = -EINVAL;
1355 if (!flp || !(*flp) || !(*flp)->fl_lmops || !(*flp)->fl_lmops->fl_break)
1356 goto out;
1357
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001358 lease = *flp;
1359
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360 error = -EAGAIN;
1361 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1362 goto out;
1363 if ((arg == F_WRLCK)
1364 && ((atomic_read(&dentry->d_count) > 1)
1365 || (atomic_read(&inode->i_count) > 1)))
1366 goto out;
1367
1368 /*
1369 * At this point, we know that if there is an exclusive
1370 * lease on this file, then we hold it on this filp
1371 * (otherwise our open of this file would have blocked).
1372 * And if we are trying to acquire an exclusive lease,
1373 * then the file is not open by anyone (including us)
1374 * except for this filp.
1375 */
1376 for (before = &inode->i_flock;
1377 ((fl = *before) != NULL) && IS_LEASE(fl);
1378 before = &fl->fl_next) {
1379 if (lease->fl_lmops->fl_mylease(fl, lease))
1380 my_before = before;
1381 else if (fl->fl_type == (F_INPROGRESS | F_UNLCK))
1382 /*
1383 * Someone is in the process of opening this
1384 * file for writing so we may not take an
1385 * exclusive lease on it.
1386 */
1387 wrlease_count++;
1388 else
1389 rdlease_count++;
1390 }
1391
1392 if ((arg == F_RDLCK && (wrlease_count > 0)) ||
1393 (arg == F_WRLCK && ((rdlease_count + wrlease_count) > 0)))
1394 goto out;
1395
1396 if (my_before != NULL) {
Trond Myklebust75dff552006-05-07 23:02:42 -04001397 *flp = *my_before;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398 error = lease->fl_lmops->fl_change(my_before, arg);
1399 goto out;
1400 }
1401
1402 error = 0;
1403 if (arg == F_UNLCK)
1404 goto out;
1405
1406 error = -EINVAL;
1407 if (!leases_enable)
1408 goto out;
1409
Trond Myklebust74361cb2006-08-14 08:54:48 -07001410 error = -ENOMEM;
1411 fl = locks_alloc_lock();
1412 if (fl == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 goto out;
1414
1415 locks_copy_lock(fl, lease);
1416
1417 locks_insert_lock(before, fl);
1418
1419 *flp = fl;
Trond Myklebust74361cb2006-08-14 08:54:48 -07001420 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421out:
1422 return error;
1423}
1424
1425 /**
1426 * setlease - sets a lease on an open file
1427 * @filp: file pointer
1428 * @arg: type of lease to obtain
1429 * @lease: file_lock to use
1430 *
1431 * Call this to establish a lease on the file.
1432 * The fl_lmops fl_break function is required by break_lease
1433 */
1434
1435int setlease(struct file *filp, long arg, struct file_lock **lease)
1436{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001437 struct dentry *dentry = filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 struct inode *inode = dentry->d_inode;
1439 int error;
1440
1441 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1442 return -EACCES;
1443 if (!S_ISREG(inode->i_mode))
1444 return -EINVAL;
1445 error = security_file_lock(filp, arg);
1446 if (error)
1447 return error;
1448
1449 lock_kernel();
1450 error = __setlease(filp, arg, lease);
1451 unlock_kernel();
1452
1453 return error;
1454}
1455
1456EXPORT_SYMBOL(setlease);
1457
1458/**
1459 * fcntl_setlease - sets a lease on an open file
1460 * @fd: open file descriptor
1461 * @filp: file pointer
1462 * @arg: type of lease to obtain
1463 *
1464 * Call this fcntl to establish a lease on the file.
1465 * Note that you also need to call %F_SETSIG to
1466 * receive a signal when the lease is broken.
1467 */
1468int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1469{
1470 struct file_lock fl, *flp = &fl;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001471 struct dentry *dentry = filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472 struct inode *inode = dentry->d_inode;
1473 int error;
1474
1475 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1476 return -EACCES;
1477 if (!S_ISREG(inode->i_mode))
1478 return -EINVAL;
1479 error = security_file_lock(filp, arg);
1480 if (error)
1481 return error;
1482
1483 locks_init_lock(&fl);
1484 error = lease_init(filp, arg, &fl);
1485 if (error)
1486 return error;
1487
1488 lock_kernel();
1489
1490 error = __setlease(filp, arg, &flp);
J. Bruce Fieldsdc15ae12005-11-10 19:08:00 -05001491 if (error || arg == F_UNLCK)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492 goto out_unlock;
1493
1494 error = fasync_helper(fd, filp, 1, &flp->fl_fasync);
1495 if (error < 0) {
1496 /* remove lease just inserted by __setlease */
1497 flp->fl_type = F_UNLCK | F_INPROGRESS;
1498 flp->fl_break_time = jiffies- 10;
1499 time_out_leases(inode);
1500 goto out_unlock;
1501 }
1502
Eric W. Biederman609d7fa2006-10-02 02:17:15 -07001503 error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504out_unlock:
1505 unlock_kernel();
1506 return error;
1507}
1508
1509/**
1510 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1511 * @filp: The file to apply the lock to
1512 * @fl: The lock to be applied
1513 *
1514 * Add a FLOCK style lock to a file.
1515 */
1516int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1517{
1518 int error;
1519 might_sleep();
1520 for (;;) {
1521 error = flock_lock_file(filp, fl);
1522 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1523 break;
1524 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1525 if (!error)
1526 continue;
1527
1528 locks_delete_block(fl);
1529 break;
1530 }
1531 return error;
1532}
1533
1534EXPORT_SYMBOL(flock_lock_file_wait);
1535
1536/**
1537 * sys_flock: - flock() system call.
1538 * @fd: the file descriptor to lock.
1539 * @cmd: the type of lock to apply.
1540 *
1541 * Apply a %FL_FLOCK style lock to an open file descriptor.
1542 * The @cmd can be one of
1543 *
1544 * %LOCK_SH -- a shared lock.
1545 *
1546 * %LOCK_EX -- an exclusive lock.
1547 *
1548 * %LOCK_UN -- remove an existing lock.
1549 *
1550 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1551 *
1552 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1553 * processes read and write access respectively.
1554 */
1555asmlinkage long sys_flock(unsigned int fd, unsigned int cmd)
1556{
1557 struct file *filp;
1558 struct file_lock *lock;
1559 int can_sleep, unlock;
1560 int error;
1561
1562 error = -EBADF;
1563 filp = fget(fd);
1564 if (!filp)
1565 goto out;
1566
1567 can_sleep = !(cmd & LOCK_NB);
1568 cmd &= ~LOCK_NB;
1569 unlock = (cmd == LOCK_UN);
1570
1571 if (!unlock && !(cmd & LOCK_MAND) && !(filp->f_mode & 3))
1572 goto out_putf;
1573
1574 error = flock_make_lock(filp, &lock, cmd);
1575 if (error)
1576 goto out_putf;
1577 if (can_sleep)
1578 lock->fl_flags |= FL_SLEEP;
1579
1580 error = security_file_lock(filp, cmd);
1581 if (error)
1582 goto out_free;
1583
1584 if (filp->f_op && filp->f_op->flock)
1585 error = filp->f_op->flock(filp,
1586 (can_sleep) ? F_SETLKW : F_SETLK,
1587 lock);
1588 else
1589 error = flock_lock_file_wait(filp, lock);
1590
1591 out_free:
Trond Myklebust993dfa82006-03-31 02:30:55 -08001592 locks_free_lock(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593
1594 out_putf:
1595 fput(filp);
1596 out:
1597 return error;
1598}
1599
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001600/**
1601 * vfs_test_lock - test file byte range lock
1602 * @filp: The file to test lock for
1603 * @fl: The lock to test
1604 * @conf: Place to return a copy of the conflicting lock, if found
1605 *
1606 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
1607 * setting conf->fl_type to something other than F_UNLCK.
1608 */
1609int vfs_test_lock(struct file *filp, struct file_lock *fl)
1610{
1611 if (filp->f_op && filp->f_op->lock)
1612 return filp->f_op->lock(filp, F_GETLK, fl);
1613 posix_test_lock(filp, fl);
1614 return 0;
1615}
1616EXPORT_SYMBOL_GPL(vfs_test_lock);
1617
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001618static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
1619{
1620 flock->l_pid = fl->fl_pid;
1621#if BITS_PER_LONG == 32
1622 /*
1623 * Make sure we can represent the posix lock via
1624 * legacy 32bit flock.
1625 */
1626 if (fl->fl_start > OFFT_OFFSET_MAX)
1627 return -EOVERFLOW;
1628 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
1629 return -EOVERFLOW;
1630#endif
1631 flock->l_start = fl->fl_start;
1632 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1633 fl->fl_end - fl->fl_start + 1;
1634 flock->l_whence = 0;
1635 return 0;
1636}
1637
1638#if BITS_PER_LONG == 32
1639static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
1640{
1641 flock->l_pid = fl->fl_pid;
1642 flock->l_start = fl->fl_start;
1643 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1644 fl->fl_end - fl->fl_start + 1;
1645 flock->l_whence = 0;
1646 flock->l_type = fl->fl_type;
1647}
1648#endif
1649
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650/* Report the first existing lock that would conflict with l.
1651 * This implements the F_GETLK command of fcntl().
1652 */
1653int fcntl_getlk(struct file *filp, struct flock __user *l)
1654{
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001655 struct file_lock file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656 struct flock flock;
1657 int error;
1658
1659 error = -EFAULT;
1660 if (copy_from_user(&flock, l, sizeof(flock)))
1661 goto out;
1662 error = -EINVAL;
1663 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1664 goto out;
1665
1666 error = flock_to_posix_lock(filp, &file_lock, &flock);
1667 if (error)
1668 goto out;
1669
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001670 error = vfs_test_lock(filp, &file_lock);
1671 if (error)
1672 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001674 flock.l_type = file_lock.fl_type;
1675 if (file_lock.fl_type != F_UNLCK) {
1676 error = posix_lock_to_flock(&flock, &file_lock);
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001677 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679 }
1680 error = -EFAULT;
1681 if (!copy_to_user(l, &flock, sizeof(flock)))
1682 error = 0;
1683out:
1684 return error;
1685}
1686
Marc Eshel7723ec92007-01-18 15:08:55 -05001687/**
1688 * vfs_lock_file - file byte range lock
1689 * @filp: The file to apply the lock to
1690 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
1691 * @fl: The lock to be applied
Marc Eshel150b3932007-01-18 16:15:35 -05001692 * @conf: Place to return a copy of the conflicting lock, if found.
1693 *
1694 * A caller that doesn't care about the conflicting lock may pass NULL
1695 * as the final argument.
1696 *
1697 * If the filesystem defines a private ->lock() method, then @conf will
1698 * be left unchanged; so a caller that cares should initialize it to
1699 * some acceptable default.
Marc Eshel2beb6612006-12-05 23:31:28 -05001700 *
1701 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
1702 * locks, the ->lock() interface may return asynchronously, before the lock has
1703 * been granted or denied by the underlying filesystem, if (and only if)
1704 * fl_grant is set. Callers expecting ->lock() to return asynchronously
1705 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
1706 * the request is for a blocking lock. When ->lock() does return asynchronously,
1707 * it must return -EINPROGRESS, and call ->fl_grant() when the lock
1708 * request completes.
1709 * If the request is for non-blocking lock the file system should return
1710 * -EINPROGRESS then try to get the lock and call the callback routine with
1711 * the result. If the request timed out the callback routine will return a
1712 * nonzero return code and the file system should release the lock. The file
1713 * system is also responsible to keep a corresponding posix lock when it
1714 * grants a lock so the VFS can find out which locks are locally held and do
1715 * the correct lock cleanup when required.
1716 * The underlying filesystem must not drop the kernel lock or call
1717 * ->fl_grant() before returning to the caller with a -EINPROGRESS
1718 * return code.
Marc Eshel7723ec92007-01-18 15:08:55 -05001719 */
Marc Eshel150b3932007-01-18 16:15:35 -05001720int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
Marc Eshel7723ec92007-01-18 15:08:55 -05001721{
1722 if (filp->f_op && filp->f_op->lock)
1723 return filp->f_op->lock(filp, cmd, fl);
1724 else
Marc Eshel150b3932007-01-18 16:15:35 -05001725 return posix_lock_file(filp, fl, conf);
Marc Eshel7723ec92007-01-18 15:08:55 -05001726}
1727EXPORT_SYMBOL_GPL(vfs_lock_file);
1728
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729/* Apply the lock described by l to an open file descriptor.
1730 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1731 */
Peter Staubachc2936212005-07-27 11:45:09 -07001732int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1733 struct flock __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734{
1735 struct file_lock *file_lock = locks_alloc_lock();
1736 struct flock flock;
1737 struct inode *inode;
1738 int error;
1739
1740 if (file_lock == NULL)
1741 return -ENOLCK;
1742
1743 /*
1744 * This might block, so we do it before checking the inode.
1745 */
1746 error = -EFAULT;
1747 if (copy_from_user(&flock, l, sizeof(flock)))
1748 goto out;
1749
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001750 inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751
1752 /* Don't allow mandatory locks on files that may be memory mapped
1753 * and shared.
1754 */
1755 if (IS_MANDLOCK(inode) &&
1756 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1757 mapping_writably_mapped(filp->f_mapping)) {
1758 error = -EAGAIN;
1759 goto out;
1760 }
1761
Peter Staubachc2936212005-07-27 11:45:09 -07001762again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763 error = flock_to_posix_lock(filp, file_lock, &flock);
1764 if (error)
1765 goto out;
1766 if (cmd == F_SETLKW) {
1767 file_lock->fl_flags |= FL_SLEEP;
1768 }
1769
1770 error = -EBADF;
1771 switch (flock.l_type) {
1772 case F_RDLCK:
1773 if (!(filp->f_mode & FMODE_READ))
1774 goto out;
1775 break;
1776 case F_WRLCK:
1777 if (!(filp->f_mode & FMODE_WRITE))
1778 goto out;
1779 break;
1780 case F_UNLCK:
1781 break;
1782 default:
1783 error = -EINVAL;
1784 goto out;
1785 }
1786
1787 error = security_file_lock(filp, file_lock->fl_type);
1788 if (error)
1789 goto out;
1790
Marc Eshel7723ec92007-01-18 15:08:55 -05001791 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001792 error = vfs_lock_file(filp, cmd, file_lock, NULL);
Marc Eshel7723ec92007-01-18 15:08:55 -05001793 if (error != -EAGAIN || cmd == F_SETLK)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794 break;
Marc Eshel7723ec92007-01-18 15:08:55 -05001795 error = wait_event_interruptible(file_lock->fl_wait,
1796 !file_lock->fl_next);
1797 if (!error)
1798 continue;
1799
1800 locks_delete_block(file_lock);
1801 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802 }
1803
Peter Staubachc2936212005-07-27 11:45:09 -07001804 /*
1805 * Attempt to detect a close/fcntl race and recover by
1806 * releasing the lock that was just acquired.
1807 */
1808 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1809 flock.l_type = F_UNLCK;
1810 goto again;
1811 }
1812
1813out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814 locks_free_lock(file_lock);
1815 return error;
1816}
1817
1818#if BITS_PER_LONG == 32
1819/* Report the first existing lock that would conflict with l.
1820 * This implements the F_GETLK command of fcntl().
1821 */
1822int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1823{
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001824 struct file_lock file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825 struct flock64 flock;
1826 int error;
1827
1828 error = -EFAULT;
1829 if (copy_from_user(&flock, l, sizeof(flock)))
1830 goto out;
1831 error = -EINVAL;
1832 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1833 goto out;
1834
1835 error = flock64_to_posix_lock(filp, &file_lock, &flock);
1836 if (error)
1837 goto out;
1838
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001839 error = vfs_test_lock(filp, &file_lock);
1840 if (error)
1841 goto out;
1842
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001843 flock.l_type = file_lock.fl_type;
1844 if (file_lock.fl_type != F_UNLCK)
1845 posix_lock_to_flock64(&flock, &file_lock);
1846
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847 error = -EFAULT;
1848 if (!copy_to_user(l, &flock, sizeof(flock)))
1849 error = 0;
1850
1851out:
1852 return error;
1853}
1854
1855/* Apply the lock described by l to an open file descriptor.
1856 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1857 */
Peter Staubachc2936212005-07-27 11:45:09 -07001858int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
1859 struct flock64 __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860{
1861 struct file_lock *file_lock = locks_alloc_lock();
1862 struct flock64 flock;
1863 struct inode *inode;
1864 int error;
1865
1866 if (file_lock == NULL)
1867 return -ENOLCK;
1868
1869 /*
1870 * This might block, so we do it before checking the inode.
1871 */
1872 error = -EFAULT;
1873 if (copy_from_user(&flock, l, sizeof(flock)))
1874 goto out;
1875
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001876 inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877
1878 /* Don't allow mandatory locks on files that may be memory mapped
1879 * and shared.
1880 */
1881 if (IS_MANDLOCK(inode) &&
1882 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1883 mapping_writably_mapped(filp->f_mapping)) {
1884 error = -EAGAIN;
1885 goto out;
1886 }
1887
Peter Staubachc2936212005-07-27 11:45:09 -07001888again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889 error = flock64_to_posix_lock(filp, file_lock, &flock);
1890 if (error)
1891 goto out;
1892 if (cmd == F_SETLKW64) {
1893 file_lock->fl_flags |= FL_SLEEP;
1894 }
1895
1896 error = -EBADF;
1897 switch (flock.l_type) {
1898 case F_RDLCK:
1899 if (!(filp->f_mode & FMODE_READ))
1900 goto out;
1901 break;
1902 case F_WRLCK:
1903 if (!(filp->f_mode & FMODE_WRITE))
1904 goto out;
1905 break;
1906 case F_UNLCK:
1907 break;
1908 default:
1909 error = -EINVAL;
1910 goto out;
1911 }
1912
1913 error = security_file_lock(filp, file_lock->fl_type);
1914 if (error)
1915 goto out;
1916
Marc Eshel7723ec92007-01-18 15:08:55 -05001917 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001918 error = vfs_lock_file(filp, cmd, file_lock, NULL);
Marc Eshel7723ec92007-01-18 15:08:55 -05001919 if (error != -EAGAIN || cmd == F_SETLK64)
Peter Staubachc2936212005-07-27 11:45:09 -07001920 break;
Marc Eshel7723ec92007-01-18 15:08:55 -05001921 error = wait_event_interruptible(file_lock->fl_wait,
1922 !file_lock->fl_next);
1923 if (!error)
1924 continue;
1925
1926 locks_delete_block(file_lock);
1927 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001928 }
1929
Peter Staubachc2936212005-07-27 11:45:09 -07001930 /*
1931 * Attempt to detect a close/fcntl race and recover by
1932 * releasing the lock that was just acquired.
1933 */
1934 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1935 flock.l_type = F_UNLCK;
1936 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937 }
1938
1939out:
1940 locks_free_lock(file_lock);
1941 return error;
1942}
1943#endif /* BITS_PER_LONG == 32 */
1944
1945/*
1946 * This function is called when the file is being removed
1947 * from the task's fd array. POSIX locks belonging to this task
1948 * are deleted at this time.
1949 */
1950void locks_remove_posix(struct file *filp, fl_owner_t owner)
1951{
Miklos Szerediff7b86b2006-06-23 02:05:11 -07001952 struct file_lock lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953
1954 /*
1955 * If there are no locks held on this file, we don't need to call
1956 * posix_lock_file(). Another process could be setting a lock on this
1957 * file at the same time, but we wouldn't remove that lock anyway.
1958 */
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001959 if (!filp->f_path.dentry->d_inode->i_flock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960 return;
1961
1962 lock.fl_type = F_UNLCK;
Miklos Szeredi75e1fcc2006-06-23 02:05:12 -07001963 lock.fl_flags = FL_POSIX | FL_CLOSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964 lock.fl_start = 0;
1965 lock.fl_end = OFFSET_MAX;
1966 lock.fl_owner = owner;
1967 lock.fl_pid = current->tgid;
1968 lock.fl_file = filp;
1969 lock.fl_ops = NULL;
1970 lock.fl_lmops = NULL;
1971
Marc Eshel150b3932007-01-18 16:15:35 -05001972 vfs_lock_file(filp, F_SETLK, &lock, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974 if (lock.fl_ops && lock.fl_ops->fl_release_private)
1975 lock.fl_ops->fl_release_private(&lock);
1976}
1977
1978EXPORT_SYMBOL(locks_remove_posix);
1979
1980/*
1981 * This function is called on the last close of an open file.
1982 */
1983void locks_remove_flock(struct file *filp)
1984{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001985 struct inode * inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986 struct file_lock *fl;
1987 struct file_lock **before;
1988
1989 if (!inode->i_flock)
1990 return;
1991
1992 if (filp->f_op && filp->f_op->flock) {
1993 struct file_lock fl = {
1994 .fl_pid = current->tgid,
1995 .fl_file = filp,
1996 .fl_flags = FL_FLOCK,
1997 .fl_type = F_UNLCK,
1998 .fl_end = OFFSET_MAX,
1999 };
2000 filp->f_op->flock(filp, F_SETLKW, &fl);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00002001 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2002 fl.fl_ops->fl_release_private(&fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002003 }
2004
2005 lock_kernel();
2006 before = &inode->i_flock;
2007
2008 while ((fl = *before) != NULL) {
2009 if (fl->fl_file == filp) {
Peter Staubachc2936212005-07-27 11:45:09 -07002010 if (IS_FLOCK(fl)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011 locks_delete_lock(before);
2012 continue;
2013 }
2014 if (IS_LEASE(fl)) {
2015 lease_modify(before, F_UNLCK);
2016 continue;
2017 }
2018 /* What? */
2019 BUG();
2020 }
2021 before = &fl->fl_next;
2022 }
2023 unlock_kernel();
2024}
2025
2026/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027 * posix_unblock_lock - stop waiting for a file lock
2028 * @filp: how the file was opened
2029 * @waiter: the lock which was waiting
2030 *
2031 * lockd needs to block waiting for locks.
2032 */
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002033int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034posix_unblock_lock(struct file *filp, struct file_lock *waiter)
2035{
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002036 int status = 0;
2037
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038 lock_kernel();
J. Bruce Fields5996a292006-01-03 09:55:44 +01002039 if (waiter->fl_next)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040 __locks_delete_block(waiter);
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002041 else
2042 status = -ENOENT;
J. Bruce Fields5996a292006-01-03 09:55:44 +01002043 unlock_kernel();
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002044 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045}
2046
2047EXPORT_SYMBOL(posix_unblock_lock);
2048
Marc Eshel9b9d2ab2007-01-18 17:52:58 -05002049/**
2050 * vfs_cancel_lock - file byte range unblock lock
2051 * @filp: The file to apply the unblock to
2052 * @fl: The lock to be unblocked
2053 *
2054 * Used by lock managers to cancel blocked requests
2055 */
2056int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2057{
2058 if (filp->f_op && filp->f_op->lock)
2059 return filp->f_op->lock(filp, F_CANCELLK, fl);
2060 return 0;
2061}
2062
2063EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2064
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065static void lock_get_status(char* out, struct file_lock *fl, int id, char *pfx)
2066{
2067 struct inode *inode = NULL;
2068
2069 if (fl->fl_file != NULL)
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08002070 inode = fl->fl_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071
2072 out += sprintf(out, "%d:%s ", id, pfx);
2073 if (IS_POSIX(fl)) {
2074 out += sprintf(out, "%6s %s ",
2075 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2076 (inode == NULL) ? "*NOINODE*" :
2077 (IS_MANDLOCK(inode) &&
2078 (inode->i_mode & (S_IXGRP | S_ISGID)) == S_ISGID) ?
2079 "MANDATORY" : "ADVISORY ");
2080 } else if (IS_FLOCK(fl)) {
2081 if (fl->fl_type & LOCK_MAND) {
2082 out += sprintf(out, "FLOCK MSNFS ");
2083 } else {
2084 out += sprintf(out, "FLOCK ADVISORY ");
2085 }
2086 } else if (IS_LEASE(fl)) {
2087 out += sprintf(out, "LEASE ");
2088 if (fl->fl_type & F_INPROGRESS)
2089 out += sprintf(out, "BREAKING ");
2090 else if (fl->fl_file)
2091 out += sprintf(out, "ACTIVE ");
2092 else
2093 out += sprintf(out, "BREAKER ");
2094 } else {
2095 out += sprintf(out, "UNKNOWN UNKNOWN ");
2096 }
2097 if (fl->fl_type & LOCK_MAND) {
2098 out += sprintf(out, "%s ",
2099 (fl->fl_type & LOCK_READ)
2100 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2101 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2102 } else {
2103 out += sprintf(out, "%s ",
2104 (fl->fl_type & F_INPROGRESS)
2105 ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ "
2106 : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ ");
2107 }
2108 if (inode) {
2109#ifdef WE_CAN_BREAK_LSLK_NOW
2110 out += sprintf(out, "%d %s:%ld ", fl->fl_pid,
2111 inode->i_sb->s_id, inode->i_ino);
2112#else
2113 /* userspace relies on this representation of dev_t ;-( */
2114 out += sprintf(out, "%d %02x:%02x:%ld ", fl->fl_pid,
2115 MAJOR(inode->i_sb->s_dev),
2116 MINOR(inode->i_sb->s_dev), inode->i_ino);
2117#endif
2118 } else {
2119 out += sprintf(out, "%d <none>:0 ", fl->fl_pid);
2120 }
2121 if (IS_POSIX(fl)) {
2122 if (fl->fl_end == OFFSET_MAX)
2123 out += sprintf(out, "%Ld EOF\n", fl->fl_start);
2124 else
2125 out += sprintf(out, "%Ld %Ld\n", fl->fl_start,
2126 fl->fl_end);
2127 } else {
2128 out += sprintf(out, "0 EOF\n");
2129 }
2130}
2131
2132static void move_lock_status(char **p, off_t* pos, off_t offset)
2133{
2134 int len;
2135 len = strlen(*p);
2136 if(*pos >= offset) {
2137 /* the complete line is valid */
2138 *p += len;
2139 *pos += len;
2140 return;
2141 }
2142 if(*pos+len > offset) {
2143 /* use the second part of the line */
2144 int i = offset-*pos;
2145 memmove(*p,*p+i,len-i);
2146 *p += len-i;
2147 *pos += len;
2148 return;
2149 }
2150 /* discard the complete line */
2151 *pos += len;
2152}
2153
2154/**
2155 * get_locks_status - reports lock usage in /proc/locks
2156 * @buffer: address in userspace to write into
2157 * @start: ?
2158 * @offset: how far we are through the buffer
2159 * @length: how much to read
2160 */
2161
2162int get_locks_status(char *buffer, char **start, off_t offset, int length)
2163{
2164 struct list_head *tmp;
2165 char *q = buffer;
2166 off_t pos = 0;
2167 int i = 0;
2168
2169 lock_kernel();
2170 list_for_each(tmp, &file_lock_list) {
2171 struct list_head *btmp;
2172 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
2173 lock_get_status(q, fl, ++i, "");
2174 move_lock_status(&q, &pos, offset);
2175
2176 if(pos >= offset+length)
2177 goto done;
2178
2179 list_for_each(btmp, &fl->fl_block) {
2180 struct file_lock *bfl = list_entry(btmp,
2181 struct file_lock, fl_block);
2182 lock_get_status(q, bfl, i, " ->");
2183 move_lock_status(&q, &pos, offset);
2184
2185 if(pos >= offset+length)
2186 goto done;
2187 }
2188 }
2189done:
2190 unlock_kernel();
2191 *start = buffer;
2192 if(q-buffer < length)
2193 return (q-buffer);
2194 return length;
2195}
2196
2197/**
2198 * lock_may_read - checks that the region is free of locks
2199 * @inode: the inode that is being read
2200 * @start: the first byte to read
2201 * @len: the number of bytes to read
2202 *
2203 * Emulates Windows locking requirements. Whole-file
2204 * mandatory locks (share modes) can prohibit a read and
2205 * byte-range POSIX locks can prohibit a read if they overlap.
2206 *
2207 * N.B. this function is only ever called
2208 * from knfsd and ownership of locks is never checked.
2209 */
2210int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2211{
2212 struct file_lock *fl;
2213 int result = 1;
2214 lock_kernel();
2215 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2216 if (IS_POSIX(fl)) {
2217 if (fl->fl_type == F_RDLCK)
2218 continue;
2219 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2220 continue;
2221 } else if (IS_FLOCK(fl)) {
2222 if (!(fl->fl_type & LOCK_MAND))
2223 continue;
2224 if (fl->fl_type & LOCK_READ)
2225 continue;
2226 } else
2227 continue;
2228 result = 0;
2229 break;
2230 }
2231 unlock_kernel();
2232 return result;
2233}
2234
2235EXPORT_SYMBOL(lock_may_read);
2236
2237/**
2238 * lock_may_write - checks that the region is free of locks
2239 * @inode: the inode that is being written
2240 * @start: the first byte to write
2241 * @len: the number of bytes to write
2242 *
2243 * Emulates Windows locking requirements. Whole-file
2244 * mandatory locks (share modes) can prohibit a write and
2245 * byte-range POSIX locks can prohibit a write if they overlap.
2246 *
2247 * N.B. this function is only ever called
2248 * from knfsd and ownership of locks is never checked.
2249 */
2250int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2251{
2252 struct file_lock *fl;
2253 int result = 1;
2254 lock_kernel();
2255 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2256 if (IS_POSIX(fl)) {
2257 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2258 continue;
2259 } else if (IS_FLOCK(fl)) {
2260 if (!(fl->fl_type & LOCK_MAND))
2261 continue;
2262 if (fl->fl_type & LOCK_WRITE)
2263 continue;
2264 } else
2265 continue;
2266 result = 0;
2267 break;
2268 }
2269 unlock_kernel();
2270 return result;
2271}
2272
2273EXPORT_SYMBOL(lock_may_write);
2274
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275static int __init filelock_init(void)
2276{
2277 filelock_cache = kmem_cache_create("file_lock_cache",
2278 sizeof(struct file_lock), 0, SLAB_PANIC,
2279 init_once, NULL);
2280 return 0;
2281}
2282
2283core_initcall(filelock_init);