![[first]](../icons/first.png){kind=icon}
![[previous]](../icons/n_left.png){kind=icon}
![[next]](../icons/n_right.png){kind=icon}
![[last]](../icons/last.png){kind=icon}
![[top]](../icons/n_top.png){kind=icon}
![[bottom]](../icons/bottom.png){kind=icon}
![[index]](../icons/index.png){kind=icon}
![[help]](../icons/help.png){kind=icon}
*/
1 /*
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 - F_FLOCK and F_POSIX.
32 *
33 * F_POSIX locks are created with calls to fcntl() and lockf() through the
34 * fcntl() system call. They have the semantics described above.
35 *
36 * F_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 * F_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 * F_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 (begemot@bgm.rosprint.net), 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 'linux/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
80 #include <linux/malloc.h>
81 #include <linux/sched.h>
82 #include <linux/kernel.h>
83 #include <linux/errno.h>
84 #include <linux/stat.h>
85 #include <linux/fcntl.h>
86
87 #include <asm/segment.h>
88
89 #define OFFSET_MAX ((off_t)0x7fffffff) /* FIXME: move elsewhere? */
90
91 static int flock_make_lock(struct file *filp, struct file_lock *fl,
92 unsigned int cmd);
93 static int posix_make_lock(struct file *filp, struct file_lock *fl,
94 struct flock *l);
95 static int flock_locks_conflict(struct file_lock *caller_fl,
96 struct file_lock *sys_fl);
97 static int posix_locks_conflict(struct file_lock *caller_fl,
98 struct file_lock *sys_fl);
99 static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl);
100 static int flock_lock_file(struct file *filp, struct file_lock *caller,
101 unsigned int wait);
102 static int posix_lock_file(struct file *filp, struct file_lock *caller,
103 unsigned int wait);
104 static int locks_deadlock(struct task_struct *my_task,
105 struct task_struct *blocked_task);
106 static int locks_overlap(struct file_lock *fl1, struct file_lock *fl2);
107
108 static struct file_lock *locks_alloc_lock(struct file_lock *fl);
109 static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl);
110 static void locks_delete_lock(struct file_lock **fl, unsigned int wait);
111 static char *lock_get_status(struct file_lock *fl, char *p, int id, char *pfx);
112
113 static struct file_lock *file_lock_table = NULL;
114
115 /* Free lock not inserted in any queue */
116 static inline void locks_free_lock(struct file_lock *fl)
/* ![[next]](../icons/right.png)
![[first]](../icons/n_first.png)
![[top]](../icons/top.png)
*/
117 {
118 kfree(fl);
119 return;
120 }
121
122 /* Add lock fl to the blocked list pointed to by block.
123 * We search to the end of the existing list and insert the the new
124 * struct. This ensures processes will be woken up in the order they
125 * blocked.
126 * NOTE: nowhere does the documentation insist that processes be woken
127 * up in this order, but it seems like the reasonable thing to do.
128 * If the blocked list gets long then this search could get expensive,
129 * in which case we could consider waking the processes up in reverse
130 * order, or making the blocked list a doubly linked circular list.
131 *
132 * This functions are called only from one place (flock_lock_file)
133 * so they are inlined now. -- Dmitry Gorodchanin 02/09/96.
134 */
135
136 static inline void locks_insert_block(struct file_lock *bfl,
/* ![[previous]](../icons/left.png)
*/
137 struct file_lock *fl)
138 {
139 while (bfl->fl_block != NULL) {
140 bfl = bfl->fl_block;
141 }
142
143 bfl->fl_block = fl;
144 fl->fl_block = NULL;
145
146 return;
147 }
148
149 static inline void locks_delete_block(struct file_lock *bfl,
/* */
150 struct file_lock *fl)
151 {
152 struct file_lock *tfl;
153
154 while ((tfl = bfl->fl_block) != NULL) {
155 if (tfl == fl) {
156 bfl->fl_block = fl->fl_block;
157 fl->fl_block = NULL;
158 return;
159 }
160 bfl = tfl;
161 }
162 return;
163 }
164
165 /* flock() system call entry point. Apply a FLOCK style lock to
166 * an open file descriptor.
167 */
168 asmlinkage int sys_flock(unsigned int fd, unsigned int cmd)
/* */
169 {
170 struct file_lock file_lock;
171 struct file *filp;
172
173 if ((fd >= NR_OPEN) || !(filp = current->files->fd[fd]))
174 return (-EBADF);
175
176 if (!flock_make_lock(filp, &file_lock, cmd))
177 return (-EINVAL);
178
179 if ((file_lock.fl_type != F_UNLCK) && !(filp->f_mode & 3))
180 return (-EBADF);
181
182 return (flock_lock_file(filp, &file_lock, cmd & LOCK_UN ? 0 : cmd & LOCK_NB ? 0 : 1));
183 }
184
185 /* Report the first existing lock that would conflict with l.
186 * This implements the F_GETLK command of fcntl().
187 */
188 int fcntl_getlk(unsigned int fd, struct flock *l)
/* */
189 {
190 int error;
191 struct flock flock;
192 struct file *filp;
193 struct file_lock *fl,file_lock;
194
195 if ((fd >= NR_OPEN) || !(filp = current->files->fd[fd]))
196 return (-EBADF);
197 error = verify_area(VERIFY_WRITE, l, sizeof(*l));
198 if (error)
199 return (error);
200
201 memcpy_fromfs(&flock, l, sizeof(flock));
202 if ((flock.l_type == F_UNLCK) || (flock.l_type == F_EXLCK) ||
203 (flock.l_type == F_SHLCK))
204 return (-EINVAL);
205
206 if (!filp->f_inode || !posix_make_lock(filp, &file_lock, &flock))
207 return (-EINVAL);
208
209 for (fl = filp->f_inode->i_flock; fl != NULL; fl = fl->fl_next) {
210 if (posix_locks_conflict(&file_lock, fl)) {
211 flock.l_pid = fl->fl_owner->pid;
212 flock.l_start = fl->fl_start;
213 flock.l_len = fl->fl_end == OFFSET_MAX ? 0 :
214 fl->fl_end - fl->fl_start + 1;
215 flock.l_whence = 0;
216 flock.l_type = fl->fl_type;
217 memcpy_tofs(l, &flock, sizeof(flock));
218 return (0);
219 }
220 }
221
222 flock.l_type = F_UNLCK; /* no conflict found */
223 memcpy_tofs(l, &flock, sizeof(flock));
224 return (0);
225 }
226
227 /* Apply the lock described by l to an open file descriptor.
228 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
229 * It also emulates flock() in a pretty broken way for older C
230 * libraries.
231 */
232 int fcntl_setlk(unsigned int fd, unsigned int cmd, struct flock *l)
/* */
233 {
234 int error;
235 struct file *filp;
236 struct file_lock file_lock;
237 struct flock flock;
238 struct inode *inode;
239
240 /*
241 * Get arguments and validate them ...
242 */
243
244 if ((fd >= NR_OPEN) || !(filp = current->files->fd[fd]))
245 return (-EBADF);
246
247 error = verify_area(VERIFY_READ, l, sizeof(*l));
248 if (error)
249 return (error);
250
251 if (!(inode = filp->f_inode))
252 return (-EINVAL);
253
254 /* Don't allow mandatory locks on files that may be memory mapped
255 * and shared.
256 */
257 if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID && inode->i_mmap) {
258 struct vm_area_struct *vma = inode->i_mmap;
259 do {
260 if (vma->vm_flags & VM_MAYSHARE)
261 return (-EAGAIN);
262 vma = vma->vm_next_share;
263 } while (vma != inode->i_mmap);
264 }
265
266 memcpy_fromfs(&flock, l, sizeof(flock));
267 if (!posix_make_lock(filp, &file_lock, &flock))
268 return (-EINVAL);
269
270 switch (flock.l_type) {
271 case F_RDLCK :
272 if (!(filp->f_mode & 1))
273 return (-EBADF);
274 break;
275 case F_WRLCK :
276 if (!(filp->f_mode & 2))
277 return (-EBADF);
278 break;
279 case F_SHLCK :
280 case F_EXLCK :
281 #if 1
282 /* warn a bit for now, but don't overdo it */
283 {
284 static int count = 0;
285 if (count < 5) {
286 count++;
287 printk(KERN_WARNING
288 "fcntl_setlk() called by process %d with broken flock() emulation\n",
289 current->pid);
290 }
291 }
292 #endif
293 if (!(filp->f_mode & 3))
294 return (-EBADF);
295 break;
296 case F_UNLCK :
297 break;
298 }
299
300 return (posix_lock_file(filp, &file_lock, cmd == F_SETLKW));
301 }
302
303 /* This function is called when the file is closed.
304 */
305 void locks_remove_locks(struct task_struct *task, struct file *filp)
/* */
306 {
307 struct file_lock *fl;
308 struct file_lock **before;
309
310 /* For POSIX locks we free all locks on this file for the given task.
311 * For FLOCK we only free locks on this *open* file if it is the last
312 * close on that file.
313 */
314 before = &filp->f_inode->i_flock;
315 while ((fl = *before) != NULL) {
316 if (((fl->fl_flags & F_POSIX) && (fl->fl_owner == task)) ||
317 ((fl->fl_flags & F_FLOCK) && (fl->fl_file == filp) &&
318 (filp->f_count == 1)))
319 locks_delete_lock(before, 0);
320 else
321 before = &fl->fl_next;
322 }
323
324 return;
325 }
326
327 int locks_verify_locked(struct inode *inode)
/* */
328 {
329 /* Candidates for mandatory locking have the setgid bit set
330 * but no group execute bit - an otherwise meaningless combination.
331 */
332 if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID)
333 return (locks_mandatory_locked(inode));
334 return (0);
335 }
336
337 int locks_mandatory_locked(struct inode *inode)
/* */
338 {
339 struct file_lock *fl;
340
341 /* Search the lock list for this inode for any POSIX locks.
342 */
343 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
344 if ((fl->fl_flags & F_POSIX) && (fl->fl_owner != current))
345 return (-EAGAIN);
346 }
347 return (0);
348 }
349
350 int locks_verify_area(int read_write, struct inode *inode, struct file *filp,
/* */
351 unsigned int offset, unsigned int count)
352 {
353 /* Candidates for mandatory locking have the setgid bit set
354 * but no group execute bit - an otherwise meaningless combination.
355 */
356 if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID)
357 return (locks_mandatory_area(read_write, inode, filp, offset,
358 count));
359 return (0);
360 }
361
362 int locks_mandatory_area(int read_write, struct inode *inode,
/* */
363 struct file *filp, unsigned int offset,
364 unsigned int count)
365 {
366 struct file_lock *fl;
367
368 repeat:
369 /*
370 * Search the lock list for this inode for locks that conflict with
371 * the proposed read/write.
372 */
373 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
374 if ((fl->fl_flags & F_FLOCK) ||
375 ((fl->fl_flags & F_POSIX) && (fl->fl_owner == current)))
376 continue;
377 if (fl->fl_end < offset ||
378 fl->fl_start >= offset + count)
379 continue;
380 /*
381 * Block for writes against a "read" lock, and both reads and
382 * writes against a "write" lock.
383 */
384 if (read_write == FLOCK_VERIFY_WRITE ||
385 fl->fl_type == F_WRLCK) {
386 if (filp && (filp->f_flags & O_NONBLOCK))
387 return (-EAGAIN);
388 if (current->signal & ~current->blocked)
389 return (-ERESTARTSYS);
390 if (locks_deadlock(current, fl->fl_owner))
391 return (-EDEADLOCK);
392 interruptible_sleep_on(&fl->fl_wait);
393 if (current->signal & ~current->blocked)
394 return (-ERESTARTSYS);
395 /*
396 * If we've been sleeping someone might have changed
397 * the permissions behind our back.
398 */
399 if ((inode->i_mode & (S_ISGID | S_IXGRP)) != S_ISGID)
400 break;
401 goto repeat;
402 }
403 }
404 return (0);
405 }
406
407 /* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
408 * style lock.
409 */
410 static int posix_make_lock(struct file *filp, struct file_lock *fl,
/* */
411 struct flock *l)
412 {
413 off_t start;
414
415 fl->fl_flags = F_POSIX;
416
417 switch (l->l_type) {
418 case F_RDLCK :
419 case F_WRLCK :
420 case F_UNLCK :
421 fl->fl_type = l->l_type;
422 break;
423 case F_SHLCK :
424 fl->fl_type = F_RDLCK;
425 fl->fl_flags |= F_BROKEN;
426 break;
427 case F_EXLCK :
428 fl->fl_type = F_WRLCK;
429 fl->fl_flags |= F_BROKEN;
430 break;
431 default :
432 return (0);
433 }
434
435 switch (l->l_whence) {
436 case 0 : /*SEEK_SET*/
437 start = 0;
438 break;
439 case 1 : /*SEEK_CUR*/
440 start = filp->f_pos;
441 break;
442 case 2 : /*SEEK_END*/
443 start = filp->f_inode->i_size;
444 break;
445 default :
446 return (0);
447 }
448
449 if (((start += l->l_start) < 0) || (l->l_len < 0))
450 return (0);
451 fl->fl_start = start; /* we record the absolute position */
452 if ((l->l_len == 0) || ((fl->fl_end = start + l->l_len - 1) < 0))
453 fl->fl_end = OFFSET_MAX;
454
455 fl->fl_file = filp;
456 fl->fl_owner = current;
457 fl->fl_wait = NULL; /* just for cleanliness */
458
459 return (1);
460 }
461
462 /* Verify a call to flock() and fill in a file_lock structure with
463 * an appropriate FLOCK lock.
464 */
465 static int flock_make_lock(struct file *filp, struct file_lock *fl,
/* */
466 unsigned int cmd)
467 {
468 if (!filp->f_inode) /* just in case */
469 return (0);
470
471 switch (cmd & ~LOCK_NB) {
472 case LOCK_SH :
473 fl->fl_type = F_RDLCK;
474 break;
475 case LOCK_EX :
476 fl->fl_type = F_WRLCK;
477 break;
478 case LOCK_UN :
479 fl->fl_type = F_UNLCK;
480 break;
481 default :
482 return (0);
483 }
484
485 fl->fl_flags = F_FLOCK;
486 fl->fl_start = 0;
487 fl->fl_end = OFFSET_MAX;
488 fl->fl_file = filp;
489 fl->fl_owner = current;
490 fl->fl_wait = NULL; /* just for cleanliness */
491
492 return (1);
493 }
494
495 /* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
496 * checking before calling the locks_conflict().
497 */
498 static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
/* */
499 {
500 /* POSIX locks owned by the same process do not conflict with
501 * each other.
502 */
503 if ((sys_fl->fl_flags & F_POSIX) &&
504 (caller_fl->fl_owner == sys_fl->fl_owner))
505 return (0);
506
507 return (locks_conflict(caller_fl, sys_fl));
508 }
509
510 /* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
511 * checking before calling the locks_conflict().
512 */
513 static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
/* */
514 {
515 /* FLOCK locks referring to the same filp do not conflict with
516 * each other.
517 */
518 if ((sys_fl->fl_flags & F_FLOCK) &&
519 (caller_fl->fl_file == sys_fl->fl_file))
520 return (0);
521
522 return (locks_conflict(caller_fl, sys_fl));
523 }
524
525 /* Determine if lock sys_fl blocks lock caller_fl. Common functionality
526 * checks for overlapping locks and shared/exclusive status.
527 */
528 static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
/* */
529 {
530 if (!locks_overlap(caller_fl, sys_fl))
531 return (0);
532
533 switch (caller_fl->fl_type) {
534 case F_RDLCK :
535 return (sys_fl->fl_type == F_WRLCK);
536
537 case F_WRLCK :
538 return (1);
539
540 default:
541 printk("locks_conflict(): impossible lock type - %d\n",
542 caller_fl->fl_type);
543 break;
544 }
545 return (0); /* This should never happen */
546 }
547
548 /* Check if two locks overlap each other.
549 */
550 static int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
/* */
551 {
552 return ((fl1->fl_end >= fl2->fl_start) &&
553 (fl2->fl_end >= fl1->fl_start));
554 }
555
556 /* This function tests for deadlock condition before putting a process to
557 * sleep. The detection scheme is no longer recursive. Recursive was neat,
558 * but dangerous - we risked stack corruption if the lock data was bad, or
559 * if the recursion was too deep for any other reason.
560 *
561 * We rely on the fact that a task can only be on one lock's wait queue
562 * at a time. When we find blocked_task on a wait queue we can re-search
563 * with blocked_task equal to that queue's owner, until either blocked_task
564 * isn't found, or blocked_task is found on a queue owned by my_task.
565 */
566 static int locks_deadlock(struct task_struct *my_task,
/* */
567 struct task_struct *blocked_task)
568 {
569 struct wait_queue *dlock_wait;
570 struct file_lock *fl;
571
572 next_task:
573 if (my_task == blocked_task)
574 return (1);
575 for (fl = file_lock_table; fl != NULL; fl = fl->fl_nextlink) {
576 if (fl->fl_owner == NULL || fl->fl_wait == NULL)
577 continue;
578 dlock_wait = fl->fl_wait;
579 do {
580 if (dlock_wait->task == blocked_task) {
581 if (fl->fl_owner == my_task) {
582 return (1);
583 }
584 blocked_task = fl->fl_owner;
585 goto next_task;
586 }
587 dlock_wait = dlock_wait->next;
588 } while (dlock_wait != fl->fl_wait);
589 }
590 return (0);
591 }
592
593 /* Try to create a FLOCK lock on filp. We rely on FLOCK locks being sorted
594 * first in an inode's lock list, and always insert new locks at the head
595 * of the list.
596 */
597 static int flock_lock_file(struct file *filp, struct file_lock *caller,
/* */
598 unsigned int wait)
599 {
600 struct file_lock *fl;
601 struct file_lock *new_fl;
602 struct file_lock **before;
603 int change = 0;
604
605 /* This a compact little algorithm based on us always placing FLOCK
606 * locks at the front of the list.
607 */
608 before = &filp->f_inode->i_flock;
609 while ((fl = *before) && (fl->fl_flags & F_FLOCK)) {
610 if (caller->fl_file == fl->fl_file) {
611 if (caller->fl_type == fl->fl_type)
612 return (0);
613 change = 1;
614 break;
615 }
616 before = &fl->fl_next;
617 }
618 /* change means that we are changing the type of an existing lock, or
619 * or else unlocking it.
620 */
621 if (change)
622 locks_delete_lock(before, caller->fl_type != F_UNLCK);
623 if (caller->fl_type == F_UNLCK)
624 return (0);
625 if ((new_fl = locks_alloc_lock(caller)) == NULL)
626 return (-ENOLCK);
627 repeat:
628 for (fl = filp->f_inode->i_flock; fl != NULL; fl = fl->fl_next) {
629 if (!flock_locks_conflict(new_fl, fl))
630 continue;
631
632 if (wait) {
633 if (current->signal & ~current->blocked) {
634 /* Note: new_fl is not in any queue at this
635 * point, so we must use locks_free_lock()
636 * instead of locks_delete_lock()
637 * Dmitry Gorodchanin 09/02/96.
638 */
639 locks_free_lock(new_fl);
640 return (-ERESTARTSYS);
641 }
642 /* Try to avoid deadlocks due to pathological programs that
643 * mix calls to flock() and fcntl(). Return EAGAIN, because
644 * EDEADLOCK isn't a documented return value for flock().
645 */
646 if (locks_deadlock(new_fl->fl_owner, fl->fl_owner)) {
647 locks_free_lock(new_fl);
648 return (-EAGAIN);
649 }
650 locks_insert_block(fl, new_fl);
651 interruptible_sleep_on(&new_fl->fl_wait);
652 wake_up(&new_fl->fl_wait);
653 if (current->signal & ~current->blocked) {
654 /* If we are here, than we were awakened
655 * by a signal, so new_fl is still in the
656 * block queue of fl. We need to remove
657 * new_fl and then free it.
658 * Dmitry Gorodchanin 09/02/96.
659 */
660 locks_delete_block(fl, new_fl);
661 locks_free_lock(new_fl);
662 return (-ERESTARTSYS);
663 }
664 goto repeat;
665 }
666
667 locks_free_lock(new_fl);
668 return (-EAGAIN);
669 }
670 locks_insert_lock(&filp->f_inode->i_flock, new_fl);
671 return (0);
672 }
673
674 /* Add a POSIX style lock to a file.
675 * We merge adjacent locks whenever possible. POSIX locks come after FLOCK
676 * locks in the list and are sorted by owner task, then by starting address
677 *
678 * Kai Petzke writes:
679 * To make freeing a lock much faster, we keep a pointer to the lock before the
680 * actual one. But the real gain of the new coding was, that lock_it() and
681 * unlock_it() became one function.
682 *
683 * To all purists: Yes, I use a few goto's. Just pass on to the next function.
684 */
685
686 static int posix_lock_file(struct file *filp, struct file_lock *caller,
/* */
687 unsigned int wait)
688 {
689 struct file_lock *fl;
690 struct file_lock *new_fl;
691 struct file_lock *left = NULL;
692 struct file_lock *right = NULL;
693 struct file_lock **before;
694 int added = 0;
695
696 if (caller->fl_type != F_UNLCK) {
697 repeat:
698 for (fl = filp->f_inode->i_flock; fl != NULL; fl = fl->fl_next) {
699 if (!posix_locks_conflict(caller, fl))
700 continue;
701 if (wait) {
702 if (current->signal & ~current->blocked)
703 return (-ERESTARTSYS);
704 if (locks_deadlock(caller->fl_owner, fl->fl_owner))
705 return (-EDEADLOCK);
706 interruptible_sleep_on(&fl->fl_wait);
707 if (current->signal & ~current->blocked)
708 return (-ERESTARTSYS);
709 goto repeat;
710 }
711 return (-EAGAIN);
712 }
713 }
714 /*
715 * Find the first old lock with the same owner as the new lock.
716 */
717
718 before = &filp->f_inode->i_flock;
719
720 /* First skip FLOCK locks and locks owned by other processes.
721 */
722 while ((fl = *before) && ((fl->fl_flags & F_FLOCK) ||
723 (caller->fl_owner != fl->fl_owner))) {
724 before = &fl->fl_next;
725 }
726
727
728 /* Process locks with this owner.
729 */
730 while ((fl = *before) && (caller->fl_owner == fl->fl_owner)) {
731 /* Detect adjacent or overlapping regions (if same lock type)
732 */
733 if (caller->fl_type == fl->fl_type) {
734 if (fl->fl_end < caller->fl_start - 1)
735 goto next_lock;
736 /* If the next lock in the list has entirely bigger
737 * addresses than the new one, insert the lock here.
738 */
739 if (fl->fl_start > caller->fl_end + 1)
740 break;
741
742 /* If we come here, the new and old lock are of the
743 * same type and adjacent or overlapping. Make one
744 * lock yielding from the lower start address of both
745 * locks to the higher end address.
746 */
747 if (fl->fl_start > caller->fl_start)
748 fl->fl_start = caller->fl_start;
749 else
750 caller->fl_start = fl->fl_start;
751 if (fl->fl_end < caller->fl_end)
752 fl->fl_end = caller->fl_end;
753 else
754 caller->fl_end = fl->fl_end;
755 if (added) {
756 locks_delete_lock(before, 0);
757 continue;
758 }
759 caller = fl;
760 added = 1;
761 }
762 else {
763 /* Processing for different lock types is a bit
764 * more complex.
765 */
766 if (fl->fl_end < caller->fl_start)
767 goto next_lock;
768 if (fl->fl_start > caller->fl_end)
769 break;
770 if (caller->fl_type == F_UNLCK)
771 added = 1;
772 if (fl->fl_start < caller->fl_start)
773 left = fl;
774 /* If the next lock in the list has a higher end
775 * address than the new one, insert the new one here.
776 */
777 if (fl->fl_end > caller->fl_end) {
778 right = fl;
779 break;
780 }
781 if (fl->fl_start >= caller->fl_start) {
782 /* The new lock completely replaces an old
783 * one (This may happen several times).
784 */
785 if (added) {
786 locks_delete_lock(before, 0);
787 continue;
788 }
789 /* Replace the old lock with the new one.
790 * Wake up anybody waiting for the old one,
791 * as the change in lock type might satisfy
792 * their needs.
793 */
794 wake_up(&fl->fl_wait);
795 fl->fl_start = caller->fl_start;
796 fl->fl_end = caller->fl_end;
797 fl->fl_type = caller->fl_type;
798 caller = fl;
799 added = 1;
800 }
801 }
802 /* Go on to next lock.
803 */
804 next_lock:
805 before = &fl->fl_next;
806 }
807
808 if (!added) {
809 if (caller->fl_type == F_UNLCK)
810 return (0);
811 if ((new_fl = locks_alloc_lock(caller)) == NULL)
812 return (-ENOLCK);
813 locks_insert_lock(before, new_fl);
814 }
815 if (right) {
816 if (left == right) {
817 /* The new lock breaks the old one in two pieces, so we
818 * have to allocate one more lock (in this case, even
819 * F_UNLCK may fail!).
820 */
821 if ((left = locks_alloc_lock(right)) == NULL) {
822 if (!added)
823 locks_delete_lock(before, 0);
824 return (-ENOLCK);
825 }
826 locks_insert_lock(before, left);
827 }
828 right->fl_start = caller->fl_end + 1;
829 }
830 if (left)
831 left->fl_end = caller->fl_start - 1;
832 return (0);
833 }
834
835 /* Allocate memory for a new lock and initialize its fields from
836 * fl. The lock is not inserted into any lists until locks_insert_lock()
837 * or locks_insert_block() are called.
838 */
839
840 static struct file_lock *locks_alloc_lock(struct file_lock *fl)
/* */
841 {
842 struct file_lock *tmp;
843
844 /* Okay, let's make a new file_lock structure... */
845 if ((tmp = (struct file_lock *)kmalloc(sizeof(struct file_lock),
846 GFP_ATOMIC)) == NULL)
847 return (tmp);
848
849 tmp->fl_nextlink = NULL;
850 tmp->fl_prevlink = NULL;
851 tmp->fl_next = NULL;
852 tmp->fl_block = NULL;
853 tmp->fl_flags = fl->fl_flags;
854 tmp->fl_owner = fl->fl_owner;
855 tmp->fl_file = fl->fl_file;
856 tmp->fl_wait = NULL;
857 tmp->fl_type = fl->fl_type;
858 tmp->fl_start = fl->fl_start;
859 tmp->fl_end = fl->fl_end;
860
861 return (tmp);
862 }
863
864 /* Insert file lock fl into an inode's lock list at the position indicated
865 * by pos. At the same time add the lock to the global file lock list.
866 */
867
868 static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
/* */
869 {
870 fl->fl_nextlink = file_lock_table;
871 fl->fl_prevlink = NULL;
872 if (file_lock_table != NULL)
873 file_lock_table->fl_prevlink = fl;
874 file_lock_table = fl;
875 fl->fl_next = *pos; /* insert into file's list */
876 *pos = fl;
877
878 return;
879 }
880
881 /* Delete a lock and free it.
882 * First remove our lock from the lock lists. Then remove all the blocked
883 * locks from our blocked list, waking up the processes that own them. If
884 * told to wait, then sleep on each of these lock's wait queues. Each
885 * blocked process will wake up and immediately wake up its own wait queue
886 * allowing us to be scheduled again. Lastly, wake up our own wait queue
887 * before freeing the file_lock structure.
888 */
889
890 static void locks_delete_lock(struct file_lock **fl_p, unsigned int wait)
/* */
891 {
892 struct file_lock *fl;
893 struct file_lock *pfl;
894 struct file_lock *nfl;
895
896 fl = *fl_p;
897 *fl_p = fl->fl_next;
898 pfl = fl->fl_prevlink;
899 nfl = fl->fl_nextlink;
900
901 if (nfl != NULL)
902 nfl->fl_prevlink = pfl;
903
904 if (pfl != NULL)
905 pfl->fl_nextlink = nfl;
906 else
907 file_lock_table = nfl;
908
909 while ((nfl = fl->fl_block) != NULL) {
910 fl->fl_block = nfl->fl_block;
911 nfl->fl_block = NULL;
912 wake_up(&nfl->fl_wait);
913 if (wait)
914 sleep_on(&nfl->fl_wait);
915 }
916
917 wake_up(&fl->fl_wait);
918 kfree(fl);
919
920 return;
921 }
922
923
924 static char *lock_get_status(struct file_lock *fl, char *p, int id, char *pfx)
/* */
925 {
926 struct wait_queue *wt;
927
928 p += sprintf(p, "%d:%s ", id, pfx);
929 if (fl->fl_flags & F_POSIX) {
930 p += sprintf(p, "%s %s ",
931 (fl->fl_flags & F_BROKEN) ? "BROKEN" : "POSIX ",
932 ((fl->fl_file->f_inode->i_mode & (S_IXGRP | S_ISGID))
933 == S_ISGID) ? "MANDATORY" : "ADVISORY ");
934 }
935 else {
936 p += sprintf(p, "FLOCK ADVISORY ");
937 }
938 p += sprintf(p, "%s ", (fl->fl_type == F_RDLCK) ? "READ " : "WRITE");
939 p += sprintf(p, "%d %04x:%ld %ld %ld ",
940 fl->fl_owner->pid, fl->fl_file->f_inode->i_dev,
941 fl->fl_file->f_inode->i_ino, fl->fl_start,
942 fl->fl_end);
943 p += sprintf(p, "%08lx %08lx %08lx %08lx %08lx\n%d:%s",
944 (long)fl, (long)fl->fl_prevlink, (long)fl->fl_nextlink,
945 (long)fl->fl_next, (long)fl->fl_block, id, pfx);
946 if ((wt = fl->fl_wait) != NULL) {
947 do {
948 p += sprintf(p, " %d", wt->task->pid);
949 wt = wt->next;
950 } while (wt != fl->fl_wait);
951 }
952 p += sprintf(p, "\n");
953 return (p);
954 }
955
956 int get_locks_status(char *buf)
/* ![[last]](../icons/n_last.png)
*/
957 {
958 struct file_lock *fl;
959 struct file_lock *bfl;
960 char *p;
961 int i;
962
963 p = buf;
964 for (fl = file_lock_table, i = 1; fl != NULL; fl = fl->fl_nextlink, i++) {
965 p = lock_get_status(fl, p, i, "");
966 for (bfl = fl; bfl->fl_block != NULL; bfl = bfl->fl_block)
967 p = lock_get_status(bfl->fl_block, p, i, " ->");
968 }
969 return (p - buf);
970 }
971
![[bottom]](../icons/n_bottom.png){kind=icon}
*/