blob: 11dbf08651b70861d31530ef73436dc2e3beffc1 [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>
Al Viro9f3acc32008-04-24 07:44:08 -0400119#include <linux/fdtable.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120#include <linux/fs.h>
121#include <linux/init.h>
122#include <linux/module.h>
123#include <linux/security.h>
124#include <linux/slab.h>
125#include <linux/smp_lock.h>
126#include <linux/syscalls.h>
127#include <linux/time.h>
Dipankar Sarma4fb3a532005-09-16 19:28:13 -0700128#include <linux/rcupdate.h>
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000129#include <linux/pid_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131#include <asm/uaccess.h>
132
133#define IS_POSIX(fl) (fl->fl_flags & FL_POSIX)
134#define IS_FLOCK(fl) (fl->fl_flags & FL_FLOCK)
135#define IS_LEASE(fl) (fl->fl_flags & FL_LEASE)
136
137int leases_enable = 1;
138int lease_break_time = 45;
139
140#define for_each_lock(inode, lockp) \
141 for (lockp = &inode->i_flock; *lockp != NULL; lockp = &(*lockp)->fl_next)
142
Christoph Hellwig26bcbf92006-03-20 13:44:40 -0500143static LIST_HEAD(file_lock_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144static LIST_HEAD(blocked_list);
145
Christoph Lametere18b8902006-12-06 20:33:20 -0800146static struct kmem_cache *filelock_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147
148/* Allocate an empty lock structure. */
149static struct file_lock *locks_alloc_lock(void)
150{
Christoph Lametere94b1762006-12-06 20:33:17 -0800151 return kmem_cache_alloc(filelock_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152}
153
Trond Myklebust47831f32006-03-20 13:44:05 -0500154static void locks_release_private(struct file_lock *fl)
155{
156 if (fl->fl_ops) {
157 if (fl->fl_ops->fl_release_private)
158 fl->fl_ops->fl_release_private(fl);
159 fl->fl_ops = NULL;
160 }
161 if (fl->fl_lmops) {
162 if (fl->fl_lmops->fl_release_private)
163 fl->fl_lmops->fl_release_private(fl);
164 fl->fl_lmops = NULL;
165 }
166
167}
168
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169/* Free a lock which is not in use. */
Matt Mackall33443c42006-01-08 01:05:22 -0800170static void locks_free_lock(struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171{
Miklos Szeredi5ce29642006-03-31 02:30:29 -0800172 BUG_ON(waitqueue_active(&fl->fl_wait));
173 BUG_ON(!list_empty(&fl->fl_block));
174 BUG_ON(!list_empty(&fl->fl_link));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175
Trond Myklebust47831f32006-03-20 13:44:05 -0500176 locks_release_private(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177 kmem_cache_free(filelock_cache, fl);
178}
179
180void locks_init_lock(struct file_lock *fl)
181{
182 INIT_LIST_HEAD(&fl->fl_link);
183 INIT_LIST_HEAD(&fl->fl_block);
184 init_waitqueue_head(&fl->fl_wait);
185 fl->fl_next = NULL;
186 fl->fl_fasync = NULL;
187 fl->fl_owner = NULL;
188 fl->fl_pid = 0;
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000189 fl->fl_nspid = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 fl->fl_file = NULL;
191 fl->fl_flags = 0;
192 fl->fl_type = 0;
193 fl->fl_start = fl->fl_end = 0;
194 fl->fl_ops = NULL;
195 fl->fl_lmops = NULL;
196}
197
198EXPORT_SYMBOL(locks_init_lock);
199
200/*
201 * Initialises the fields of the file lock which are invariant for
202 * free file_locks.
203 */
Christoph Lameter4ba9b9d2007-10-16 23:25:51 -0700204static void init_once(struct kmem_cache *cache, void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205{
206 struct file_lock *lock = (struct file_lock *) foo;
207
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 locks_init_lock(lock);
209}
210
Trond Myklebust47831f32006-03-20 13:44:05 -0500211static void locks_copy_private(struct file_lock *new, struct file_lock *fl)
212{
213 if (fl->fl_ops) {
214 if (fl->fl_ops->fl_copy_lock)
215 fl->fl_ops->fl_copy_lock(new, fl);
216 new->fl_ops = fl->fl_ops;
217 }
218 if (fl->fl_lmops) {
219 if (fl->fl_lmops->fl_copy_lock)
220 fl->fl_lmops->fl_copy_lock(new, fl);
221 new->fl_lmops = fl->fl_lmops;
222 }
223}
224
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225/*
226 * Initialize a new lock from an existing file_lock structure.
227 */
J. Bruce Fields1a747ee2008-04-24 10:08:22 -0400228void __locks_copy_lock(struct file_lock *new, const struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229{
230 new->fl_owner = fl->fl_owner;
231 new->fl_pid = fl->fl_pid;
Trond Myklebust09969052006-03-20 13:44:38 -0500232 new->fl_file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 new->fl_flags = fl->fl_flags;
234 new->fl_type = fl->fl_type;
235 new->fl_start = fl->fl_start;
236 new->fl_end = fl->fl_end;
Trond Myklebust09969052006-03-20 13:44:38 -0500237 new->fl_ops = NULL;
238 new->fl_lmops = NULL;
239}
Roland Dreier3dd7b712008-04-25 15:32:51 -0700240EXPORT_SYMBOL(__locks_copy_lock);
Trond Myklebust09969052006-03-20 13:44:38 -0500241
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 */
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500464static struct file_lock *lease_alloc(struct file *filp, int type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465{
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)
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500470 return ERR_PTR(error);
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);
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500475 return ERR_PTR(error);
Trond Myklebust75dff552006-05-07 23:02:42 -0400476 }
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500477 return fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478}
479
480/* Check if two locks overlap each other.
481 */
482static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
483{
484 return ((fl1->fl_end >= fl2->fl_start) &&
485 (fl2->fl_end >= fl1->fl_start));
486}
487
488/*
489 * Check whether two locks have the same owner.
490 */
Matt Mackall33443c42006-01-08 01:05:22 -0800491static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492{
493 if (fl1->fl_lmops && fl1->fl_lmops->fl_compare_owner)
494 return fl2->fl_lmops == fl1->fl_lmops &&
495 fl1->fl_lmops->fl_compare_owner(fl1, fl2);
496 return fl1->fl_owner == fl2->fl_owner;
497}
498
499/* Remove waiter from blocker's block list.
500 * When blocker ends up pointing to itself then the list is empty.
501 */
Matt Mackall33443c42006-01-08 01:05:22 -0800502static void __locks_delete_block(struct file_lock *waiter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503{
504 list_del_init(&waiter->fl_block);
505 list_del_init(&waiter->fl_link);
506 waiter->fl_next = NULL;
507}
508
509/*
510 */
511static void locks_delete_block(struct file_lock *waiter)
512{
513 lock_kernel();
514 __locks_delete_block(waiter);
515 unlock_kernel();
516}
517
518/* Insert waiter into blocker's block list.
519 * We use a circular list so that processes can be easily woken up in
520 * the order they blocked. The documentation doesn't require this but
521 * it seems like the reasonable thing to do.
522 */
523static void locks_insert_block(struct file_lock *blocker,
524 struct file_lock *waiter)
525{
J. Bruce Fields6dc0fe82006-03-26 01:37:24 -0800526 BUG_ON(!list_empty(&waiter->fl_block));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 list_add_tail(&waiter->fl_block, &blocker->fl_block);
528 waiter->fl_next = blocker;
529 if (IS_POSIX(blocker))
530 list_add(&waiter->fl_link, &blocked_list);
531}
532
533/* Wake up processes blocked waiting for blocker.
534 * If told to wait then schedule the processes until the block list
535 * is empty, otherwise empty the block list ourselves.
536 */
537static void locks_wake_up_blocks(struct file_lock *blocker)
538{
539 while (!list_empty(&blocker->fl_block)) {
Pavel Emelyanovf0c1cd02007-09-19 16:44:07 +0400540 struct file_lock *waiter;
541
542 waiter = list_first_entry(&blocker->fl_block,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543 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
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000559 fl->fl_nspid = get_pid(task_tgid(current));
560
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 /* insert into file's list */
562 fl->fl_next = *pos;
563 *pos = fl;
564
565 if (fl->fl_ops && fl->fl_ops->fl_insert)
566 fl->fl_ops->fl_insert(fl);
567}
568
569/*
570 * Delete a lock and then free it.
571 * Wake up processes that are blocked waiting for this lock,
572 * notify the FS that the lock has been cleared and
573 * finally free the lock.
574 */
575static void locks_delete_lock(struct file_lock **thisfl_p)
576{
577 struct file_lock *fl = *thisfl_p;
578
579 *thisfl_p = fl->fl_next;
580 fl->fl_next = NULL;
581 list_del_init(&fl->fl_link);
582
583 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
584 if (fl->fl_fasync != NULL) {
585 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
586 fl->fl_fasync = NULL;
587 }
588
589 if (fl->fl_ops && fl->fl_ops->fl_remove)
590 fl->fl_ops->fl_remove(fl);
591
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000592 if (fl->fl_nspid) {
593 put_pid(fl->fl_nspid);
594 fl->fl_nspid = NULL;
595 }
596
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 locks_wake_up_blocks(fl);
598 locks_free_lock(fl);
599}
600
601/* Determine if lock sys_fl blocks lock caller_fl. Common functionality
602 * checks for shared/exclusive status of overlapping locks.
603 */
604static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
605{
606 if (sys_fl->fl_type == F_WRLCK)
607 return 1;
608 if (caller_fl->fl_type == F_WRLCK)
609 return 1;
610 return 0;
611}
612
613/* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
614 * checking before calling the locks_conflict().
615 */
616static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
617{
618 /* POSIX locks owned by the same process do not conflict with
619 * each other.
620 */
621 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
622 return (0);
623
624 /* Check whether they overlap */
625 if (!locks_overlap(caller_fl, sys_fl))
626 return 0;
627
628 return (locks_conflict(caller_fl, sys_fl));
629}
630
631/* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
632 * checking before calling the locks_conflict().
633 */
634static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
635{
636 /* FLOCK locks referring to the same filp do not conflict with
637 * each other.
638 */
639 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
640 return (0);
641 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
642 return 0;
643
644 return (locks_conflict(caller_fl, sys_fl));
645}
646
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400647void
Marc Eshel9d6a8c52007-02-21 00:55:18 -0500648posix_test_lock(struct file *filp, struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649{
650 struct file_lock *cfl;
651
652 lock_kernel();
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800653 for (cfl = filp->f_path.dentry->d_inode->i_flock; cfl; cfl = cfl->fl_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 if (!IS_POSIX(cfl))
655 continue;
J. Bruce Fieldsb842e242007-05-10 19:02:07 -0400656 if (posix_locks_conflict(fl, cfl))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 break;
658 }
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000659 if (cfl) {
Marc Eshel9d6a8c52007-02-21 00:55:18 -0500660 __locks_copy_lock(fl, cfl);
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000661 if (cfl->fl_nspid)
Pavel Emelyanov6c5f3e72008-02-08 04:19:20 -0800662 fl->fl_pid = pid_vnr(cfl->fl_nspid);
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000663 } else
J. Bruce Fields129a84d2007-05-10 18:38:43 -0400664 fl->fl_type = F_UNLCK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 unlock_kernel();
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400666 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668EXPORT_SYMBOL(posix_test_lock);
669
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400670/*
671 * Deadlock detection:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400673 * We attempt to detect deadlocks that are due purely to posix file
674 * locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400676 * We assume that a task can be waiting for at most one lock at a time.
677 * So for any acquired lock, the process holding that lock may be
678 * waiting on at most one other lock. That lock in turns may be held by
679 * someone waiting for at most one other lock. Given a requested lock
680 * caller_fl which is about to wait for a conflicting lock block_fl, we
681 * follow this chain of waiters to ensure we are not about to create a
682 * cycle.
J. Bruce Fields97855b42007-10-30 11:20:02 -0400683 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400684 * Since we do this before we ever put a process to sleep on a lock, we
685 * are ensured that there is never a cycle; that is what guarantees that
686 * the while() loop in posix_locks_deadlock() eventually completes.
687 *
688 * Note: the above assumption may not be true when handling lock
689 * requests from a broken NFS client. It may also fail in the presence
690 * of tasks (such as posix threads) sharing the same open file table.
691 *
692 * To handle those cases, we just bail out after a few iterations.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 */
J. Bruce Fields97855b42007-10-30 11:20:02 -0400694
695#define MAX_DEADLK_ITERATIONS 10
696
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400697/* Find a lock that the owner of the given block_fl is blocking on. */
698static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
699{
700 struct file_lock *fl;
701
702 list_for_each_entry(fl, &blocked_list, fl_link) {
703 if (posix_same_owner(fl, block_fl))
704 return fl->fl_next;
705 }
706 return NULL;
707}
708
Adrian Bunkb0904e12006-06-23 02:05:13 -0700709static int posix_locks_deadlock(struct file_lock *caller_fl,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 struct file_lock *block_fl)
711{
J. Bruce Fields97855b42007-10-30 11:20:02 -0400712 int i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400714 while ((block_fl = what_owner_is_waiting_for(block_fl))) {
715 if (i++ > MAX_DEADLK_ITERATIONS)
716 return 0;
717 if (posix_same_owner(caller_fl, block_fl))
718 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 }
720 return 0;
721}
722
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723/* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
J. Bruce Fields02888f42007-09-12 15:45:07 -0400724 * after any leases, but before any posix locks.
Trond Myklebustf475ae92006-06-29 16:38:32 -0400725 *
726 * Note that if called with an FL_EXISTS argument, the caller may determine
727 * whether or not a lock was successfully freed by testing the return
728 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 */
Trond Myklebust993dfa82006-03-31 02:30:55 -0800730static int flock_lock_file(struct file *filp, struct file_lock *request)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731{
Trond Myklebust993dfa82006-03-31 02:30:55 -0800732 struct file_lock *new_fl = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 struct file_lock **before;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800734 struct inode * inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 int error = 0;
736 int found = 0;
737
738 lock_kernel();
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400739 if (request->fl_flags & FL_ACCESS)
740 goto find_conflict;
Pavel Emelyanov84d535a2007-09-11 16:38:13 +0400741
742 if (request->fl_type != F_UNLCK) {
743 error = -ENOMEM;
744 new_fl = locks_alloc_lock();
745 if (new_fl == NULL)
746 goto out;
747 error = 0;
748 }
749
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 for_each_lock(inode, before) {
751 struct file_lock *fl = *before;
752 if (IS_POSIX(fl))
753 break;
754 if (IS_LEASE(fl))
755 continue;
756 if (filp != fl->fl_file)
757 continue;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800758 if (request->fl_type == fl->fl_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 goto out;
760 found = 1;
761 locks_delete_lock(before);
762 break;
763 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764
Trond Myklebustf475ae92006-06-29 16:38:32 -0400765 if (request->fl_type == F_UNLCK) {
766 if ((request->fl_flags & FL_EXISTS) && !found)
767 error = -ENOENT;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800768 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -0400769 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770
771 /*
772 * If a higher-priority process was blocked on the old file lock,
773 * give it the opportunity to lock the file.
774 */
775 if (found)
Linus Torvaldsc3921ab2008-05-11 16:04:48 -0700776 cond_resched_bkl();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400778find_conflict:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 for_each_lock(inode, before) {
780 struct file_lock *fl = *before;
781 if (IS_POSIX(fl))
782 break;
783 if (IS_LEASE(fl))
784 continue;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800785 if (!flock_locks_conflict(request, fl))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 continue;
787 error = -EAGAIN;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800788 if (request->fl_flags & FL_SLEEP)
789 locks_insert_block(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 goto out;
791 }
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400792 if (request->fl_flags & FL_ACCESS)
793 goto out;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800794 locks_copy_lock(new_fl, request);
Pavel Emelyanov0e2f6db2007-09-11 15:24:01 -0700795 locks_insert_lock(before, new_fl);
Trond Myklebust993dfa82006-03-31 02:30:55 -0800796 new_fl = NULL;
Kirill Korotaev9cedc192006-06-14 17:59:35 +0400797 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798
799out:
800 unlock_kernel();
Trond Myklebust993dfa82006-03-31 02:30:55 -0800801 if (new_fl)
802 locks_free_lock(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 return error;
804}
805
Marc Eshel150b3932007-01-18 16:15:35 -0500806static int __posix_lock_file(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807{
808 struct file_lock *fl;
Miklos Szeredi39005d02006-06-23 02:05:10 -0700809 struct file_lock *new_fl = NULL;
810 struct file_lock *new_fl2 = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 struct file_lock *left = NULL;
812 struct file_lock *right = NULL;
813 struct file_lock **before;
814 int error, added = 0;
815
816 /*
817 * We may need two file_lock structures for this operation,
818 * so we get them in advance to avoid races.
Miklos Szeredi39005d02006-06-23 02:05:10 -0700819 *
820 * In some cases we can be sure, that no new locks will be needed
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 */
Miklos Szeredi39005d02006-06-23 02:05:10 -0700822 if (!(request->fl_flags & FL_ACCESS) &&
823 (request->fl_type != F_UNLCK ||
824 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
825 new_fl = locks_alloc_lock();
826 new_fl2 = locks_alloc_lock();
827 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828
829 lock_kernel();
830 if (request->fl_type != F_UNLCK) {
831 for_each_lock(inode, before) {
J. Bruce Fields526985b2006-11-14 16:54:36 -0500832 fl = *before;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 if (!IS_POSIX(fl))
834 continue;
835 if (!posix_locks_conflict(request, fl))
836 continue;
Andy Adamson5842add2006-03-26 01:37:26 -0800837 if (conflock)
J. Bruce Fields1a747ee2008-04-24 10:08:22 -0400838 __locks_copy_lock(conflock, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839 error = -EAGAIN;
840 if (!(request->fl_flags & FL_SLEEP))
841 goto out;
842 error = -EDEADLK;
843 if (posix_locks_deadlock(request, fl))
844 goto out;
845 error = -EAGAIN;
846 locks_insert_block(fl, request);
847 goto out;
848 }
849 }
850
851 /* If we're just looking for a conflict, we're done. */
852 error = 0;
853 if (request->fl_flags & FL_ACCESS)
854 goto out;
855
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 * Find the first old lock with the same owner as the new lock.
858 */
859
860 before = &inode->i_flock;
861
862 /* First skip locks owned by other processes. */
863 while ((fl = *before) && (!IS_POSIX(fl) ||
864 !posix_same_owner(request, fl))) {
865 before = &fl->fl_next;
866 }
867
868 /* Process locks with this owner. */
869 while ((fl = *before) && posix_same_owner(request, fl)) {
870 /* Detect adjacent or overlapping regions (if same lock type)
871 */
872 if (request->fl_type == fl->fl_type) {
Olaf Kirch449231d2005-08-25 16:25:35 -0700873 /* In all comparisons of start vs end, use
874 * "start - 1" rather than "end + 1". If end
875 * is OFFSET_MAX, end + 1 will become negative.
876 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 if (fl->fl_end < request->fl_start - 1)
878 goto next_lock;
879 /* If the next lock in the list has entirely bigger
880 * addresses than the new one, insert the lock here.
881 */
Olaf Kirch449231d2005-08-25 16:25:35 -0700882 if (fl->fl_start - 1 > request->fl_end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 break;
884
885 /* If we come here, the new and old lock are of the
886 * same type and adjacent or overlapping. Make one
887 * lock yielding from the lower start address of both
888 * locks to the higher end address.
889 */
890 if (fl->fl_start > request->fl_start)
891 fl->fl_start = request->fl_start;
892 else
893 request->fl_start = fl->fl_start;
894 if (fl->fl_end < request->fl_end)
895 fl->fl_end = request->fl_end;
896 else
897 request->fl_end = fl->fl_end;
898 if (added) {
899 locks_delete_lock(before);
900 continue;
901 }
902 request = fl;
903 added = 1;
904 }
905 else {
906 /* Processing for different lock types is a bit
907 * more complex.
908 */
909 if (fl->fl_end < request->fl_start)
910 goto next_lock;
911 if (fl->fl_start > request->fl_end)
912 break;
913 if (request->fl_type == F_UNLCK)
914 added = 1;
915 if (fl->fl_start < request->fl_start)
916 left = fl;
917 /* If the next lock in the list has a higher end
918 * address than the new one, insert the new one here.
919 */
920 if (fl->fl_end > request->fl_end) {
921 right = fl;
922 break;
923 }
924 if (fl->fl_start >= request->fl_start) {
925 /* The new lock completely replaces an old
926 * one (This may happen several times).
927 */
928 if (added) {
929 locks_delete_lock(before);
930 continue;
931 }
932 /* Replace the old lock with the new one.
933 * Wake up anybody waiting for the old one,
934 * as the change in lock type might satisfy
935 * their needs.
936 */
937 locks_wake_up_blocks(fl);
938 fl->fl_start = request->fl_start;
939 fl->fl_end = request->fl_end;
940 fl->fl_type = request->fl_type;
Trond Myklebust47831f32006-03-20 13:44:05 -0500941 locks_release_private(fl);
942 locks_copy_private(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943 request = fl;
944 added = 1;
945 }
946 }
947 /* Go on to next lock.
948 */
949 next_lock:
950 before = &fl->fl_next;
951 }
952
Miklos Szeredi0d9a4902006-06-23 02:05:09 -0700953 /*
954 * The above code only modifies existing locks in case of
955 * merging or replacing. If new lock(s) need to be inserted
956 * all modifications are done bellow this, so it's safe yet to
957 * bail out.
958 */
959 error = -ENOLCK; /* "no luck" */
960 if (right && left == right && !new_fl2)
961 goto out;
962
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 error = 0;
964 if (!added) {
Trond Myklebustf475ae92006-06-29 16:38:32 -0400965 if (request->fl_type == F_UNLCK) {
966 if (request->fl_flags & FL_EXISTS)
967 error = -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -0400969 }
Miklos Szeredi0d9a4902006-06-23 02:05:09 -0700970
971 if (!new_fl) {
972 error = -ENOLCK;
973 goto out;
974 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 locks_copy_lock(new_fl, request);
976 locks_insert_lock(before, new_fl);
977 new_fl = NULL;
978 }
979 if (right) {
980 if (left == right) {
981 /* The new lock breaks the old one in two pieces,
982 * so we have to use the second new lock.
983 */
984 left = new_fl2;
985 new_fl2 = NULL;
986 locks_copy_lock(left, right);
987 locks_insert_lock(before, left);
988 }
989 right->fl_start = request->fl_end + 1;
990 locks_wake_up_blocks(right);
991 }
992 if (left) {
993 left->fl_end = request->fl_start - 1;
994 locks_wake_up_blocks(left);
995 }
996 out:
997 unlock_kernel();
998 /*
999 * Free any unused locks.
1000 */
1001 if (new_fl)
1002 locks_free_lock(new_fl);
1003 if (new_fl2)
1004 locks_free_lock(new_fl2);
1005 return error;
1006}
1007
1008/**
1009 * posix_lock_file - Apply a POSIX-style lock to a file
1010 * @filp: The file to apply the lock to
1011 * @fl: The lock to be applied
Marc Eshel150b3932007-01-18 16:15:35 -05001012 * @conflock: Place to return a copy of the conflicting lock, if found.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 *
1014 * Add a POSIX style lock to a file.
1015 * We merge adjacent & overlapping locks whenever possible.
1016 * POSIX locks are sorted by owner task, then by starting address
Trond Myklebustf475ae92006-06-29 16:38:32 -04001017 *
1018 * Note that if called with an FL_EXISTS argument, the caller may determine
1019 * whether or not a lock was successfully freed by testing the return
1020 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 */
Marc Eshel150b3932007-01-18 16:15:35 -05001022int posix_lock_file(struct file *filp, struct file_lock *fl,
Andy Adamson5842add2006-03-26 01:37:26 -08001023 struct file_lock *conflock)
1024{
Marc Eshel150b3932007-01-18 16:15:35 -05001025 return __posix_lock_file(filp->f_path.dentry->d_inode, fl, conflock);
Andy Adamson5842add2006-03-26 01:37:26 -08001026}
Marc Eshel150b3932007-01-18 16:15:35 -05001027EXPORT_SYMBOL(posix_lock_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028
1029/**
1030 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1031 * @filp: The file to apply the lock to
1032 * @fl: The lock to be applied
1033 *
1034 * Add a POSIX style lock to a file.
1035 * We merge adjacent & overlapping locks whenever possible.
1036 * POSIX locks are sorted by owner task, then by starting address
1037 */
1038int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1039{
1040 int error;
1041 might_sleep ();
1042 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001043 error = posix_lock_file(filp, fl, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1045 break;
1046 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1047 if (!error)
1048 continue;
1049
1050 locks_delete_block(fl);
1051 break;
1052 }
1053 return error;
1054}
1055EXPORT_SYMBOL(posix_lock_file_wait);
1056
1057/**
1058 * locks_mandatory_locked - Check for an active lock
1059 * @inode: the file to check
1060 *
1061 * Searches the inode's list of locks to find any POSIX locks which conflict.
1062 * This function is called from locks_verify_locked() only.
1063 */
1064int locks_mandatory_locked(struct inode *inode)
1065{
1066 fl_owner_t owner = current->files;
1067 struct file_lock *fl;
1068
1069 /*
1070 * Search the lock list for this inode for any POSIX locks.
1071 */
1072 lock_kernel();
1073 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1074 if (!IS_POSIX(fl))
1075 continue;
1076 if (fl->fl_owner != owner)
1077 break;
1078 }
1079 unlock_kernel();
1080 return fl ? -EAGAIN : 0;
1081}
1082
1083/**
1084 * locks_mandatory_area - Check for a conflicting lock
1085 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1086 * for shared
1087 * @inode: the file to check
1088 * @filp: how the file was opened (if it was)
1089 * @offset: start of area to check
1090 * @count: length of area to check
1091 *
1092 * Searches the inode's list of locks to find any POSIX locks which conflict.
1093 * This function is called from rw_verify_area() and
1094 * locks_verify_truncate().
1095 */
1096int locks_mandatory_area(int read_write, struct inode *inode,
1097 struct file *filp, loff_t offset,
1098 size_t count)
1099{
1100 struct file_lock fl;
1101 int error;
1102
1103 locks_init_lock(&fl);
1104 fl.fl_owner = current->files;
1105 fl.fl_pid = current->tgid;
1106 fl.fl_file = filp;
1107 fl.fl_flags = FL_POSIX | FL_ACCESS;
1108 if (filp && !(filp->f_flags & O_NONBLOCK))
1109 fl.fl_flags |= FL_SLEEP;
1110 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1111 fl.fl_start = offset;
1112 fl.fl_end = offset + count - 1;
1113
1114 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001115 error = __posix_lock_file(inode, &fl, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 if (error != -EAGAIN)
1117 break;
1118 if (!(fl.fl_flags & FL_SLEEP))
1119 break;
1120 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1121 if (!error) {
1122 /*
1123 * If we've been sleeping someone might have
1124 * changed the permissions behind our back.
1125 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07001126 if (__mandatory_lock(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 continue;
1128 }
1129
1130 locks_delete_block(&fl);
1131 break;
1132 }
1133
1134 return error;
1135}
1136
1137EXPORT_SYMBOL(locks_mandatory_area);
1138
1139/* We already had a lease on this file; just change its type */
1140int lease_modify(struct file_lock **before, int arg)
1141{
1142 struct file_lock *fl = *before;
1143 int error = assign_type(fl, arg);
1144
1145 if (error)
1146 return error;
1147 locks_wake_up_blocks(fl);
1148 if (arg == F_UNLCK)
1149 locks_delete_lock(before);
1150 return 0;
1151}
1152
1153EXPORT_SYMBOL(lease_modify);
1154
1155static void time_out_leases(struct inode *inode)
1156{
1157 struct file_lock **before;
1158 struct file_lock *fl;
1159
1160 before = &inode->i_flock;
1161 while ((fl = *before) && IS_LEASE(fl) && (fl->fl_type & F_INPROGRESS)) {
1162 if ((fl->fl_break_time == 0)
1163 || time_before(jiffies, fl->fl_break_time)) {
1164 before = &fl->fl_next;
1165 continue;
1166 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167 lease_modify(before, fl->fl_type & ~F_INPROGRESS);
1168 if (fl == *before) /* lease_modify may have freed fl */
1169 before = &fl->fl_next;
1170 }
1171}
1172
1173/**
1174 * __break_lease - revoke all outstanding leases on file
1175 * @inode: the inode of the file to return
1176 * @mode: the open mode (read or write)
1177 *
david m. richter87250dd2007-05-09 16:10:27 -04001178 * break_lease (inlined for speed) has checked there already is at least
1179 * some kind of lock (maybe a lease) on this file. Leases are broken on
1180 * a call to open() or truncate(). This function can sleep unless you
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181 * specified %O_NONBLOCK to your open().
1182 */
1183int __break_lease(struct inode *inode, unsigned int mode)
1184{
1185 int error = 0, future;
1186 struct file_lock *new_fl, *flock;
1187 struct file_lock *fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 unsigned long break_time;
1189 int i_have_this_lease = 0;
1190
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -05001191 new_fl = lease_alloc(NULL, mode & FMODE_WRITE ? F_WRLCK : F_RDLCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192
1193 lock_kernel();
1194
1195 time_out_leases(inode);
1196
1197 flock = inode->i_flock;
1198 if ((flock == NULL) || !IS_LEASE(flock))
1199 goto out;
1200
1201 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1202 if (fl->fl_owner == current->files)
1203 i_have_this_lease = 1;
1204
1205 if (mode & FMODE_WRITE) {
1206 /* If we want write access, we have to revoke any lease. */
1207 future = F_UNLCK | F_INPROGRESS;
1208 } else if (flock->fl_type & F_INPROGRESS) {
1209 /* If the lease is already being broken, we just leave it */
1210 future = flock->fl_type;
1211 } else if (flock->fl_type & F_WRLCK) {
1212 /* Downgrade the exclusive lease to a read-only lease. */
1213 future = F_RDLCK | F_INPROGRESS;
1214 } else {
1215 /* the existing lease was read-only, so we can read too. */
1216 goto out;
1217 }
1218
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -05001219 if (IS_ERR(new_fl) && !i_have_this_lease
1220 && ((mode & O_NONBLOCK) == 0)) {
1221 error = PTR_ERR(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 goto out;
1223 }
1224
1225 break_time = 0;
1226 if (lease_break_time > 0) {
1227 break_time = jiffies + lease_break_time * HZ;
1228 if (break_time == 0)
1229 break_time++; /* so that 0 means no break time */
1230 }
1231
1232 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1233 if (fl->fl_type != future) {
1234 fl->fl_type = future;
1235 fl->fl_break_time = break_time;
1236 /* lease must have lmops break callback */
1237 fl->fl_lmops->fl_break(fl);
1238 }
1239 }
1240
1241 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1242 error = -EWOULDBLOCK;
1243 goto out;
1244 }
1245
1246restart:
1247 break_time = flock->fl_break_time;
1248 if (break_time != 0) {
1249 break_time -= jiffies;
1250 if (break_time == 0)
1251 break_time++;
1252 }
Matthew Wilcox4321e012008-01-14 21:28:30 -07001253 locks_insert_block(flock, new_fl);
1254 error = wait_event_interruptible_timeout(new_fl->fl_wait,
1255 !new_fl->fl_next, break_time);
1256 __locks_delete_block(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 if (error >= 0) {
1258 if (error == 0)
1259 time_out_leases(inode);
1260 /* Wait for the next lease that has not been broken yet */
1261 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1262 flock = flock->fl_next) {
1263 if (flock->fl_type & F_INPROGRESS)
1264 goto restart;
1265 }
1266 error = 0;
1267 }
1268
1269out:
1270 unlock_kernel();
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -05001271 if (!IS_ERR(new_fl))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272 locks_free_lock(new_fl);
1273 return error;
1274}
1275
1276EXPORT_SYMBOL(__break_lease);
1277
1278/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07001279 * lease_get_mtime - get the last modified time of an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 * @inode: the inode
1281 * @time: pointer to a timespec which will contain the last modified time
1282 *
1283 * This is to force NFS clients to flush their caches for files with
1284 * exclusive leases. The justification is that if someone has an
Randy Dunlapa6b91912008-03-19 17:01:00 -07001285 * exclusive lease, then they could be modifying it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 */
1287void lease_get_mtime(struct inode *inode, struct timespec *time)
1288{
1289 struct file_lock *flock = inode->i_flock;
1290 if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK))
1291 *time = current_fs_time(inode->i_sb);
1292 else
1293 *time = inode->i_mtime;
1294}
1295
1296EXPORT_SYMBOL(lease_get_mtime);
1297
1298/**
1299 * fcntl_getlease - Enquire what lease is currently active
1300 * @filp: the file
1301 *
1302 * The value returned by this function will be one of
1303 * (if no lease break is pending):
1304 *
1305 * %F_RDLCK to indicate a shared lease is held.
1306 *
1307 * %F_WRLCK to indicate an exclusive lease is held.
1308 *
1309 * %F_UNLCK to indicate no lease is held.
1310 *
1311 * (if a lease break is pending):
1312 *
1313 * %F_RDLCK to indicate an exclusive lease needs to be
1314 * changed to a shared lease (or removed).
1315 *
1316 * %F_UNLCK to indicate the lease needs to be removed.
1317 *
1318 * XXX: sfr & willy disagree over whether F_INPROGRESS
1319 * should be returned to userspace.
1320 */
1321int fcntl_getlease(struct file *filp)
1322{
1323 struct file_lock *fl;
1324 int type = F_UNLCK;
1325
1326 lock_kernel();
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001327 time_out_leases(filp->f_path.dentry->d_inode);
1328 for (fl = filp->f_path.dentry->d_inode->i_flock; fl && IS_LEASE(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329 fl = fl->fl_next) {
1330 if (fl->fl_file == filp) {
1331 type = fl->fl_type & ~F_INPROGRESS;
1332 break;
1333 }
1334 }
1335 unlock_kernel();
1336 return type;
1337}
1338
1339/**
Christoph Hellwig0af1a452007-07-31 00:39:22 -07001340 * generic_setlease - sets a lease on an open file
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 * @filp: file pointer
1342 * @arg: type of lease to obtain
1343 * @flp: input - file_lock to use, output - file_lock inserted
1344 *
1345 * The (input) flp->fl_lmops->fl_break function is required
1346 * by break_lease().
1347 *
1348 * Called with kernel lock held.
1349 */
Christoph Hellwig0af1a452007-07-31 00:39:22 -07001350int generic_setlease(struct file *filp, long arg, struct file_lock **flp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351{
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001352 struct file_lock *fl, **before, **my_before = NULL, *lease;
Pavel Emelyanov85c59582007-09-20 12:45:02 +04001353 struct file_lock *new_fl = NULL;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001354 struct dentry *dentry = filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 struct inode *inode = dentry->d_inode;
1356 int error, rdlease_count = 0, wrlease_count = 0;
1357
J. Bruce Fields6d5e8b02007-05-31 17:03:46 -04001358 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1359 return -EACCES;
1360 if (!S_ISREG(inode->i_mode))
1361 return -EINVAL;
1362 error = security_file_lock(filp, arg);
1363 if (error)
1364 return error;
1365
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366 time_out_leases(inode);
1367
J. Bruce Fieldsd2ab0b0c2007-06-30 12:40:32 -04001368 BUG_ON(!(*flp)->fl_lmops->fl_break);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001370 lease = *flp;
1371
David M. Richter288b2fd2008-04-23 16:29:00 -04001372 if (arg != F_UNLCK) {
David M. Richter19085552008-04-23 16:29:01 -04001373 error = -ENOMEM;
1374 new_fl = locks_alloc_lock();
1375 if (new_fl == NULL)
1376 goto out;
1377
David M. Richter288b2fd2008-04-23 16:29:00 -04001378 error = -EAGAIN;
1379 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1380 goto out;
1381 if ((arg == F_WRLCK)
1382 && ((atomic_read(&dentry->d_count) > 1)
1383 || (atomic_read(&inode->i_count) > 1)))
1384 goto out;
David M. Richter288b2fd2008-04-23 16:29:00 -04001385 }
Pavel Emelyanov85c59582007-09-20 12:45:02 +04001386
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 /*
1388 * At this point, we know that if there is an exclusive
1389 * lease on this file, then we hold it on this filp
1390 * (otherwise our open of this file would have blocked).
1391 * And if we are trying to acquire an exclusive lease,
1392 * then the file is not open by anyone (including us)
1393 * except for this filp.
1394 */
1395 for (before = &inode->i_flock;
1396 ((fl = *before) != NULL) && IS_LEASE(fl);
1397 before = &fl->fl_next) {
1398 if (lease->fl_lmops->fl_mylease(fl, lease))
1399 my_before = before;
1400 else if (fl->fl_type == (F_INPROGRESS | F_UNLCK))
1401 /*
1402 * Someone is in the process of opening this
1403 * file for writing so we may not take an
1404 * exclusive lease on it.
1405 */
1406 wrlease_count++;
1407 else
1408 rdlease_count++;
1409 }
1410
David M. Richter5fcc60c2008-04-23 16:28:59 -04001411 error = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 if ((arg == F_RDLCK && (wrlease_count > 0)) ||
1413 (arg == F_WRLCK && ((rdlease_count + wrlease_count) > 0)))
1414 goto out;
1415
1416 if (my_before != NULL) {
Trond Myklebust75dff552006-05-07 23:02:42 -04001417 *flp = *my_before;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418 error = lease->fl_lmops->fl_change(my_before, arg);
1419 goto out;
1420 }
1421
1422 error = 0;
1423 if (arg == F_UNLCK)
1424 goto out;
1425
1426 error = -EINVAL;
1427 if (!leases_enable)
1428 goto out;
1429
Pavel Emelyanov85c59582007-09-20 12:45:02 +04001430 locks_copy_lock(new_fl, lease);
1431 locks_insert_lock(before, new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432
Pavel Emelyanov85c59582007-09-20 12:45:02 +04001433 *flp = new_fl;
1434 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436out:
Pavel Emelyanov85c59582007-09-20 12:45:02 +04001437 if (new_fl != NULL)
1438 locks_free_lock(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439 return error;
1440}
Christoph Hellwig0af1a452007-07-31 00:39:22 -07001441EXPORT_SYMBOL(generic_setlease);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442
1443 /**
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001444 * vfs_setlease - sets a lease on an open file
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445 * @filp: file pointer
1446 * @arg: type of lease to obtain
1447 * @lease: file_lock to use
1448 *
1449 * Call this to establish a lease on the file.
J. Bruce Fieldsf9ffed22006-11-14 15:51:40 -05001450 * The (*lease)->fl_lmops->fl_break operation must be set; if not,
1451 * break_lease will oops!
1452 *
1453 * This will call the filesystem's setlease file method, if
1454 * defined. Note that there is no getlease method; instead, the
1455 * filesystem setlease method should call back to setlease() to
1456 * add a lease to the inode's lease list, where fcntl_getlease() can
1457 * find it. Since fcntl_getlease() only reports whether the current
1458 * task holds a lease, a cluster filesystem need only do this for
1459 * leases held by processes on this node.
1460 *
1461 * There is also no break_lease method; filesystems that
1462 * handle their own leases shoud break leases themselves from the
1463 * filesystem's open, create, and (on truncate) setattr methods.
1464 *
1465 * Warning: the only current setlease methods exist only to disable
1466 * leases in certain cases. More vfs changes may be required to
1467 * allow a full filesystem lease implementation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 */
1469
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001470int vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472 int error;
1473
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 lock_kernel();
J. Bruce Fieldsf9ffed22006-11-14 15:51:40 -05001475 if (filp->f_op && filp->f_op->setlease)
1476 error = filp->f_op->setlease(filp, arg, lease);
1477 else
Christoph Hellwig0af1a452007-07-31 00:39:22 -07001478 error = generic_setlease(filp, arg, lease);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 unlock_kernel();
1480
1481 return error;
1482}
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001483EXPORT_SYMBOL_GPL(vfs_setlease);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484
1485/**
1486 * fcntl_setlease - sets a lease on an open file
1487 * @fd: open file descriptor
1488 * @filp: file pointer
1489 * @arg: type of lease to obtain
1490 *
1491 * Call this fcntl to establish a lease on the file.
1492 * Note that you also need to call %F_SETSIG to
1493 * receive a signal when the lease is broken.
1494 */
1495int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1496{
1497 struct file_lock fl, *flp = &fl;
David M. Richter9d91cdc2008-04-23 16:29:02 -04001498 struct inode *inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 int error;
1500
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501 locks_init_lock(&fl);
1502 error = lease_init(filp, arg, &fl);
1503 if (error)
1504 return error;
1505
1506 lock_kernel();
1507
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001508 error = vfs_setlease(filp, arg, &flp);
J. Bruce Fieldsdc15ae12005-11-10 19:08:00 -05001509 if (error || arg == F_UNLCK)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 goto out_unlock;
1511
1512 error = fasync_helper(fd, filp, 1, &flp->fl_fasync);
1513 if (error < 0) {
J. Bruce Fields6d5e8b02007-05-31 17:03:46 -04001514 /* remove lease just inserted by setlease */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515 flp->fl_type = F_UNLCK | F_INPROGRESS;
J. Bruce Fields6d5e8b02007-05-31 17:03:46 -04001516 flp->fl_break_time = jiffies - 10;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517 time_out_leases(inode);
1518 goto out_unlock;
1519 }
1520
Eric W. Biederman609d7fa2006-10-02 02:17:15 -07001521 error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522out_unlock:
1523 unlock_kernel();
1524 return error;
1525}
1526
1527/**
1528 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1529 * @filp: The file to apply the lock to
1530 * @fl: The lock to be applied
1531 *
1532 * Add a FLOCK style lock to a file.
1533 */
1534int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1535{
1536 int error;
1537 might_sleep();
1538 for (;;) {
1539 error = flock_lock_file(filp, fl);
1540 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1541 break;
1542 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1543 if (!error)
1544 continue;
1545
1546 locks_delete_block(fl);
1547 break;
1548 }
1549 return error;
1550}
1551
1552EXPORT_SYMBOL(flock_lock_file_wait);
1553
1554/**
1555 * sys_flock: - flock() system call.
1556 * @fd: the file descriptor to lock.
1557 * @cmd: the type of lock to apply.
1558 *
1559 * Apply a %FL_FLOCK style lock to an open file descriptor.
1560 * The @cmd can be one of
1561 *
1562 * %LOCK_SH -- a shared lock.
1563 *
1564 * %LOCK_EX -- an exclusive lock.
1565 *
1566 * %LOCK_UN -- remove an existing lock.
1567 *
1568 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1569 *
1570 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1571 * processes read and write access respectively.
1572 */
1573asmlinkage long sys_flock(unsigned int fd, unsigned int cmd)
1574{
1575 struct file *filp;
1576 struct file_lock *lock;
1577 int can_sleep, unlock;
1578 int error;
1579
1580 error = -EBADF;
1581 filp = fget(fd);
1582 if (!filp)
1583 goto out;
1584
1585 can_sleep = !(cmd & LOCK_NB);
1586 cmd &= ~LOCK_NB;
1587 unlock = (cmd == LOCK_UN);
1588
1589 if (!unlock && !(cmd & LOCK_MAND) && !(filp->f_mode & 3))
1590 goto out_putf;
1591
1592 error = flock_make_lock(filp, &lock, cmd);
1593 if (error)
1594 goto out_putf;
1595 if (can_sleep)
1596 lock->fl_flags |= FL_SLEEP;
1597
1598 error = security_file_lock(filp, cmd);
1599 if (error)
1600 goto out_free;
1601
1602 if (filp->f_op && filp->f_op->flock)
1603 error = filp->f_op->flock(filp,
1604 (can_sleep) ? F_SETLKW : F_SETLK,
1605 lock);
1606 else
1607 error = flock_lock_file_wait(filp, lock);
1608
1609 out_free:
Trond Myklebust993dfa82006-03-31 02:30:55 -08001610 locks_free_lock(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611
1612 out_putf:
1613 fput(filp);
1614 out:
1615 return error;
1616}
1617
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001618/**
1619 * vfs_test_lock - test file byte range lock
1620 * @filp: The file to test lock for
J. Bruce Fields6924c552007-05-11 16:22:50 -04001621 * @fl: The lock to test; also used to hold result
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001622 *
1623 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
1624 * setting conf->fl_type to something other than F_UNLCK.
1625 */
1626int vfs_test_lock(struct file *filp, struct file_lock *fl)
1627{
1628 if (filp->f_op && filp->f_op->lock)
1629 return filp->f_op->lock(filp, F_GETLK, fl);
1630 posix_test_lock(filp, fl);
1631 return 0;
1632}
1633EXPORT_SYMBOL_GPL(vfs_test_lock);
1634
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001635static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
1636{
1637 flock->l_pid = fl->fl_pid;
1638#if BITS_PER_LONG == 32
1639 /*
1640 * Make sure we can represent the posix lock via
1641 * legacy 32bit flock.
1642 */
1643 if (fl->fl_start > OFFT_OFFSET_MAX)
1644 return -EOVERFLOW;
1645 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
1646 return -EOVERFLOW;
1647#endif
1648 flock->l_start = fl->fl_start;
1649 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1650 fl->fl_end - fl->fl_start + 1;
1651 flock->l_whence = 0;
J. Bruce Fields129a84d2007-05-10 18:38:43 -04001652 flock->l_type = fl->fl_type;
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001653 return 0;
1654}
1655
1656#if BITS_PER_LONG == 32
1657static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
1658{
1659 flock->l_pid = fl->fl_pid;
1660 flock->l_start = fl->fl_start;
1661 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1662 fl->fl_end - fl->fl_start + 1;
1663 flock->l_whence = 0;
1664 flock->l_type = fl->fl_type;
1665}
1666#endif
1667
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668/* Report the first existing lock that would conflict with l.
1669 * This implements the F_GETLK command of fcntl().
1670 */
1671int fcntl_getlk(struct file *filp, struct flock __user *l)
1672{
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001673 struct file_lock file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674 struct flock flock;
1675 int error;
1676
1677 error = -EFAULT;
1678 if (copy_from_user(&flock, l, sizeof(flock)))
1679 goto out;
1680 error = -EINVAL;
1681 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1682 goto out;
1683
1684 error = flock_to_posix_lock(filp, &file_lock, &flock);
1685 if (error)
1686 goto out;
1687
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001688 error = vfs_test_lock(filp, &file_lock);
1689 if (error)
1690 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001692 flock.l_type = file_lock.fl_type;
1693 if (file_lock.fl_type != F_UNLCK) {
1694 error = posix_lock_to_flock(&flock, &file_lock);
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001695 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697 }
1698 error = -EFAULT;
1699 if (!copy_to_user(l, &flock, sizeof(flock)))
1700 error = 0;
1701out:
1702 return error;
1703}
1704
Marc Eshel7723ec92007-01-18 15:08:55 -05001705/**
1706 * vfs_lock_file - file byte range lock
1707 * @filp: The file to apply the lock to
1708 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
1709 * @fl: The lock to be applied
Marc Eshel150b3932007-01-18 16:15:35 -05001710 * @conf: Place to return a copy of the conflicting lock, if found.
1711 *
1712 * A caller that doesn't care about the conflicting lock may pass NULL
1713 * as the final argument.
1714 *
1715 * If the filesystem defines a private ->lock() method, then @conf will
1716 * be left unchanged; so a caller that cares should initialize it to
1717 * some acceptable default.
Marc Eshel2beb6612006-12-05 23:31:28 -05001718 *
1719 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
1720 * locks, the ->lock() interface may return asynchronously, before the lock has
1721 * been granted or denied by the underlying filesystem, if (and only if)
1722 * fl_grant is set. Callers expecting ->lock() to return asynchronously
1723 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
1724 * the request is for a blocking lock. When ->lock() does return asynchronously,
1725 * it must return -EINPROGRESS, and call ->fl_grant() when the lock
1726 * request completes.
1727 * If the request is for non-blocking lock the file system should return
1728 * -EINPROGRESS then try to get the lock and call the callback routine with
1729 * the result. If the request timed out the callback routine will return a
1730 * nonzero return code and the file system should release the lock. The file
1731 * system is also responsible to keep a corresponding posix lock when it
1732 * grants a lock so the VFS can find out which locks are locally held and do
1733 * the correct lock cleanup when required.
1734 * The underlying filesystem must not drop the kernel lock or call
1735 * ->fl_grant() before returning to the caller with a -EINPROGRESS
1736 * return code.
Marc Eshel7723ec92007-01-18 15:08:55 -05001737 */
Marc Eshel150b3932007-01-18 16:15:35 -05001738int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
Marc Eshel7723ec92007-01-18 15:08:55 -05001739{
1740 if (filp->f_op && filp->f_op->lock)
1741 return filp->f_op->lock(filp, cmd, fl);
1742 else
Marc Eshel150b3932007-01-18 16:15:35 -05001743 return posix_lock_file(filp, fl, conf);
Marc Eshel7723ec92007-01-18 15:08:55 -05001744}
1745EXPORT_SYMBOL_GPL(vfs_lock_file);
1746
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747/* Apply the lock described by l to an open file descriptor.
1748 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1749 */
Peter Staubachc2936212005-07-27 11:45:09 -07001750int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1751 struct flock __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752{
1753 struct file_lock *file_lock = locks_alloc_lock();
1754 struct flock flock;
1755 struct inode *inode;
Al Viro0b2bac22008-05-06 13:58:34 -04001756 struct file *f;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757 int error;
1758
1759 if (file_lock == NULL)
1760 return -ENOLCK;
1761
1762 /*
1763 * This might block, so we do it before checking the inode.
1764 */
1765 error = -EFAULT;
1766 if (copy_from_user(&flock, l, sizeof(flock)))
1767 goto out;
1768
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001769 inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770
1771 /* Don't allow mandatory locks on files that may be memory mapped
1772 * and shared.
1773 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07001774 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775 error = -EAGAIN;
1776 goto out;
1777 }
1778
Peter Staubachc2936212005-07-27 11:45:09 -07001779again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780 error = flock_to_posix_lock(filp, file_lock, &flock);
1781 if (error)
1782 goto out;
1783 if (cmd == F_SETLKW) {
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
J. Bruce Fields19e729a2008-04-14 15:03:02 -04001808 if (filp->f_op && filp->f_op->lock != NULL)
1809 error = filp->f_op->lock(filp, cmd, file_lock);
1810 else {
1811 for (;;) {
1812 error = posix_lock_file(filp, file_lock, NULL);
1813 if (error != -EAGAIN || cmd == F_SETLK)
1814 break;
1815 error = wait_event_interruptible(file_lock->fl_wait,
1816 !file_lock->fl_next);
1817 if (!error)
1818 continue;
Marc Eshel7723ec92007-01-18 15:08:55 -05001819
J. Bruce Fields19e729a2008-04-14 15:03:02 -04001820 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 */
Al Viro0b2bac22008-05-06 13:58:34 -04001829 /*
1830 * we need that spin_lock here - it prevents reordering between
1831 * update of inode->i_flock and check for it done in close().
1832 * rcu_read_lock() wouldn't do.
1833 */
1834 spin_lock(&current->files->file_lock);
1835 f = fcheck(fd);
1836 spin_unlock(&current->files->file_lock);
1837 if (!error && f != filp && flock.l_type != F_UNLCK) {
Peter Staubachc2936212005-07-27 11:45:09 -07001838 flock.l_type = F_UNLCK;
1839 goto again;
1840 }
1841
1842out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843 locks_free_lock(file_lock);
1844 return error;
1845}
1846
1847#if BITS_PER_LONG == 32
1848/* Report the first existing lock that would conflict with l.
1849 * This implements the F_GETLK command of fcntl().
1850 */
1851int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1852{
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001853 struct file_lock file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854 struct flock64 flock;
1855 int error;
1856
1857 error = -EFAULT;
1858 if (copy_from_user(&flock, l, sizeof(flock)))
1859 goto out;
1860 error = -EINVAL;
1861 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1862 goto out;
1863
1864 error = flock64_to_posix_lock(filp, &file_lock, &flock);
1865 if (error)
1866 goto out;
1867
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001868 error = vfs_test_lock(filp, &file_lock);
1869 if (error)
1870 goto out;
1871
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001872 flock.l_type = file_lock.fl_type;
1873 if (file_lock.fl_type != F_UNLCK)
1874 posix_lock_to_flock64(&flock, &file_lock);
1875
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876 error = -EFAULT;
1877 if (!copy_to_user(l, &flock, sizeof(flock)))
1878 error = 0;
1879
1880out:
1881 return error;
1882}
1883
1884/* Apply the lock described by l to an open file descriptor.
1885 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1886 */
Peter Staubachc2936212005-07-27 11:45:09 -07001887int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
1888 struct flock64 __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889{
1890 struct file_lock *file_lock = locks_alloc_lock();
1891 struct flock64 flock;
1892 struct inode *inode;
Al Viro0b2bac22008-05-06 13:58:34 -04001893 struct file *f;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894 int error;
1895
1896 if (file_lock == NULL)
1897 return -ENOLCK;
1898
1899 /*
1900 * This might block, so we do it before checking the inode.
1901 */
1902 error = -EFAULT;
1903 if (copy_from_user(&flock, l, sizeof(flock)))
1904 goto out;
1905
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001906 inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907
1908 /* Don't allow mandatory locks on files that may be memory mapped
1909 * and shared.
1910 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07001911 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912 error = -EAGAIN;
1913 goto out;
1914 }
1915
Peter Staubachc2936212005-07-27 11:45:09 -07001916again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917 error = flock64_to_posix_lock(filp, file_lock, &flock);
1918 if (error)
1919 goto out;
1920 if (cmd == F_SETLKW64) {
1921 file_lock->fl_flags |= FL_SLEEP;
1922 }
1923
1924 error = -EBADF;
1925 switch (flock.l_type) {
1926 case F_RDLCK:
1927 if (!(filp->f_mode & FMODE_READ))
1928 goto out;
1929 break;
1930 case F_WRLCK:
1931 if (!(filp->f_mode & FMODE_WRITE))
1932 goto out;
1933 break;
1934 case F_UNLCK:
1935 break;
1936 default:
1937 error = -EINVAL;
1938 goto out;
1939 }
1940
1941 error = security_file_lock(filp, file_lock->fl_type);
1942 if (error)
1943 goto out;
1944
J. Bruce Fields19e729a2008-04-14 15:03:02 -04001945 if (filp->f_op && filp->f_op->lock != NULL)
1946 error = filp->f_op->lock(filp, cmd, file_lock);
1947 else {
1948 for (;;) {
1949 error = posix_lock_file(filp, file_lock, NULL);
1950 if (error != -EAGAIN || cmd == F_SETLK64)
1951 break;
1952 error = wait_event_interruptible(file_lock->fl_wait,
1953 !file_lock->fl_next);
1954 if (!error)
1955 continue;
Marc Eshel7723ec92007-01-18 15:08:55 -05001956
J. Bruce Fields19e729a2008-04-14 15:03:02 -04001957 locks_delete_block(file_lock);
1958 break;
1959 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960 }
1961
Peter Staubachc2936212005-07-27 11:45:09 -07001962 /*
1963 * Attempt to detect a close/fcntl race and recover by
1964 * releasing the lock that was just acquired.
1965 */
Al Viro0b2bac22008-05-06 13:58:34 -04001966 spin_lock(&current->files->file_lock);
1967 f = fcheck(fd);
1968 spin_unlock(&current->files->file_lock);
1969 if (!error && f != filp && flock.l_type != F_UNLCK) {
Peter Staubachc2936212005-07-27 11:45:09 -07001970 flock.l_type = F_UNLCK;
1971 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972 }
1973
1974out:
1975 locks_free_lock(file_lock);
1976 return error;
1977}
1978#endif /* BITS_PER_LONG == 32 */
1979
1980/*
1981 * This function is called when the file is being removed
1982 * from the task's fd array. POSIX locks belonging to this task
1983 * are deleted at this time.
1984 */
1985void locks_remove_posix(struct file *filp, fl_owner_t owner)
1986{
Miklos Szerediff7b86b2006-06-23 02:05:11 -07001987 struct file_lock lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988
1989 /*
1990 * If there are no locks held on this file, we don't need to call
1991 * posix_lock_file(). Another process could be setting a lock on this
1992 * file at the same time, but we wouldn't remove that lock anyway.
1993 */
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001994 if (!filp->f_path.dentry->d_inode->i_flock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995 return;
1996
1997 lock.fl_type = F_UNLCK;
Miklos Szeredi75e1fcc2006-06-23 02:05:12 -07001998 lock.fl_flags = FL_POSIX | FL_CLOSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999 lock.fl_start = 0;
2000 lock.fl_end = OFFSET_MAX;
2001 lock.fl_owner = owner;
2002 lock.fl_pid = current->tgid;
2003 lock.fl_file = filp;
2004 lock.fl_ops = NULL;
2005 lock.fl_lmops = NULL;
2006
Marc Eshel150b3932007-01-18 16:15:35 -05002007 vfs_lock_file(filp, F_SETLK, &lock, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009 if (lock.fl_ops && lock.fl_ops->fl_release_private)
2010 lock.fl_ops->fl_release_private(&lock);
2011}
2012
2013EXPORT_SYMBOL(locks_remove_posix);
2014
2015/*
2016 * This function is called on the last close of an open file.
2017 */
2018void locks_remove_flock(struct file *filp)
2019{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08002020 struct inode * inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021 struct file_lock *fl;
2022 struct file_lock **before;
2023
2024 if (!inode->i_flock)
2025 return;
2026
2027 if (filp->f_op && filp->f_op->flock) {
2028 struct file_lock fl = {
2029 .fl_pid = current->tgid,
2030 .fl_file = filp,
2031 .fl_flags = FL_FLOCK,
2032 .fl_type = F_UNLCK,
2033 .fl_end = OFFSET_MAX,
2034 };
2035 filp->f_op->flock(filp, F_SETLKW, &fl);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00002036 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2037 fl.fl_ops->fl_release_private(&fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038 }
2039
2040 lock_kernel();
2041 before = &inode->i_flock;
2042
2043 while ((fl = *before) != NULL) {
2044 if (fl->fl_file == filp) {
Peter Staubachc2936212005-07-27 11:45:09 -07002045 if (IS_FLOCK(fl)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046 locks_delete_lock(before);
2047 continue;
2048 }
2049 if (IS_LEASE(fl)) {
2050 lease_modify(before, F_UNLCK);
2051 continue;
2052 }
2053 /* What? */
2054 BUG();
2055 }
2056 before = &fl->fl_next;
2057 }
2058 unlock_kernel();
2059}
2060
2061/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062 * posix_unblock_lock - stop waiting for a file lock
2063 * @filp: how the file was opened
2064 * @waiter: the lock which was waiting
2065 *
2066 * lockd needs to block waiting for locks.
2067 */
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002068int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069posix_unblock_lock(struct file *filp, struct file_lock *waiter)
2070{
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002071 int status = 0;
2072
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073 lock_kernel();
J. Bruce Fields5996a292006-01-03 09:55:44 +01002074 if (waiter->fl_next)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075 __locks_delete_block(waiter);
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002076 else
2077 status = -ENOENT;
J. Bruce Fields5996a292006-01-03 09:55:44 +01002078 unlock_kernel();
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002079 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080}
2081
2082EXPORT_SYMBOL(posix_unblock_lock);
2083
Marc Eshel9b9d2ab2007-01-18 17:52:58 -05002084/**
2085 * vfs_cancel_lock - file byte range unblock lock
2086 * @filp: The file to apply the unblock to
2087 * @fl: The lock to be unblocked
2088 *
2089 * Used by lock managers to cancel blocked requests
2090 */
2091int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2092{
2093 if (filp->f_op && filp->f_op->lock)
2094 return filp->f_op->lock(filp, F_CANCELLK, fl);
2095 return 0;
2096}
2097
2098EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2099
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002100#ifdef CONFIG_PROC_FS
2101#include <linux/seq_file.h>
2102
2103static void lock_get_status(struct seq_file *f, struct file_lock *fl,
2104 int id, char *pfx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105{
2106 struct inode *inode = NULL;
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002107 unsigned int fl_pid;
2108
2109 if (fl->fl_nspid)
Pavel Emelyanov6c5f3e72008-02-08 04:19:20 -08002110 fl_pid = pid_vnr(fl->fl_nspid);
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002111 else
2112 fl_pid = fl->fl_pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113
2114 if (fl->fl_file != NULL)
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08002115 inode = fl->fl_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002117 seq_printf(f, "%d:%s ", id, pfx);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118 if (IS_POSIX(fl)) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002119 seq_printf(f, "%6s %s ",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002120 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2121 (inode == NULL) ? "*NOINODE*" :
Pavel Emelyanova16877c2007-10-01 14:41:11 -07002122 mandatory_lock(inode) ? "MANDATORY" : "ADVISORY ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123 } else if (IS_FLOCK(fl)) {
2124 if (fl->fl_type & LOCK_MAND) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002125 seq_printf(f, "FLOCK MSNFS ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002127 seq_printf(f, "FLOCK ADVISORY ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002128 }
2129 } else if (IS_LEASE(fl)) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002130 seq_printf(f, "LEASE ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131 if (fl->fl_type & F_INPROGRESS)
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002132 seq_printf(f, "BREAKING ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133 else if (fl->fl_file)
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002134 seq_printf(f, "ACTIVE ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135 else
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002136 seq_printf(f, "BREAKER ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002138 seq_printf(f, "UNKNOWN UNKNOWN ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002139 }
2140 if (fl->fl_type & LOCK_MAND) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002141 seq_printf(f, "%s ",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142 (fl->fl_type & LOCK_READ)
2143 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2144 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2145 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002146 seq_printf(f, "%s ",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147 (fl->fl_type & F_INPROGRESS)
2148 ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ "
2149 : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ ");
2150 }
2151 if (inode) {
2152#ifdef WE_CAN_BREAK_LSLK_NOW
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002153 seq_printf(f, "%d %s:%ld ", fl_pid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154 inode->i_sb->s_id, inode->i_ino);
2155#else
2156 /* userspace relies on this representation of dev_t ;-( */
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002157 seq_printf(f, "%d %02x:%02x:%ld ", fl_pid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158 MAJOR(inode->i_sb->s_dev),
2159 MINOR(inode->i_sb->s_dev), inode->i_ino);
2160#endif
2161 } else {
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002162 seq_printf(f, "%d <none>:0 ", fl_pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002163 }
2164 if (IS_POSIX(fl)) {
2165 if (fl->fl_end == OFFSET_MAX)
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002166 seq_printf(f, "%Ld EOF\n", fl->fl_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002167 else
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002168 seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002170 seq_printf(f, "0 EOF\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171 }
2172}
2173
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002174static int locks_show(struct seq_file *f, void *v)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175{
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002176 struct file_lock *fl, *bfl;
2177
2178 fl = list_entry(v, struct file_lock, fl_link);
2179
2180 lock_get_status(f, fl, (long)f->private, "");
2181
2182 list_for_each_entry(bfl, &fl->fl_block, fl_block)
2183 lock_get_status(f, bfl, (long)f->private, " ->");
2184
2185 f->private++;
2186 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187}
2188
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002189static void *locks_start(struct seq_file *f, loff_t *pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002190{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002191 lock_kernel();
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002192 f->private = (void *)1;
2193 return seq_list_start(&file_lock_list, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002194}
2195
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002196static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2197{
2198 return seq_list_next(v, &file_lock_list, pos);
2199}
2200
2201static void locks_stop(struct seq_file *f, void *v)
2202{
2203 unlock_kernel();
2204}
2205
2206struct seq_operations locks_seq_operations = {
2207 .start = locks_start,
2208 .next = locks_next,
2209 .stop = locks_stop,
2210 .show = locks_show,
2211};
2212#endif
2213
Linus Torvalds1da177e2005-04-16 15:20:36 -07002214/**
2215 * lock_may_read - checks that the region is free of locks
2216 * @inode: the inode that is being read
2217 * @start: the first byte to read
2218 * @len: the number of bytes to read
2219 *
2220 * Emulates Windows locking requirements. Whole-file
2221 * mandatory locks (share modes) can prohibit a read and
2222 * byte-range POSIX locks can prohibit a read if they overlap.
2223 *
2224 * N.B. this function is only ever called
2225 * from knfsd and ownership of locks is never checked.
2226 */
2227int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2228{
2229 struct file_lock *fl;
2230 int result = 1;
2231 lock_kernel();
2232 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2233 if (IS_POSIX(fl)) {
2234 if (fl->fl_type == F_RDLCK)
2235 continue;
2236 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2237 continue;
2238 } else if (IS_FLOCK(fl)) {
2239 if (!(fl->fl_type & LOCK_MAND))
2240 continue;
2241 if (fl->fl_type & LOCK_READ)
2242 continue;
2243 } else
2244 continue;
2245 result = 0;
2246 break;
2247 }
2248 unlock_kernel();
2249 return result;
2250}
2251
2252EXPORT_SYMBOL(lock_may_read);
2253
2254/**
2255 * lock_may_write - checks that the region is free of locks
2256 * @inode: the inode that is being written
2257 * @start: the first byte to write
2258 * @len: the number of bytes to write
2259 *
2260 * Emulates Windows locking requirements. Whole-file
2261 * mandatory locks (share modes) can prohibit a write and
2262 * byte-range POSIX locks can prohibit a write if they overlap.
2263 *
2264 * N.B. this function is only ever called
2265 * from knfsd and ownership of locks is never checked.
2266 */
2267int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2268{
2269 struct file_lock *fl;
2270 int result = 1;
2271 lock_kernel();
2272 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2273 if (IS_POSIX(fl)) {
2274 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2275 continue;
2276 } else if (IS_FLOCK(fl)) {
2277 if (!(fl->fl_type & LOCK_MAND))
2278 continue;
2279 if (fl->fl_type & LOCK_WRITE)
2280 continue;
2281 } else
2282 continue;
2283 result = 0;
2284 break;
2285 }
2286 unlock_kernel();
2287 return result;
2288}
2289
2290EXPORT_SYMBOL(lock_may_write);
2291
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292static int __init filelock_init(void)
2293{
2294 filelock_cache = kmem_cache_create("file_lock_cache",
2295 sizeof(struct file_lock), 0, SLAB_PANIC,
Paul Mundt20c2df82007-07-20 10:11:58 +09002296 init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002297 return 0;
2298}
2299
2300core_initcall(filelock_init);