xref: /netbsd-src/sys/kern/vfs_lockf.c (revision cac8e449158efc7261bebc8657cbb0125a2cfdde)
1 /*	$NetBSD: vfs_lockf.c,v 1.66 2008/08/01 07:11:24 skrll Exp $	*/
2 
3 /*
4  * Copyright (c) 1982, 1986, 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Scooter Morris at Genentech Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)ufs_lockf.c	8.4 (Berkeley) 10/26/94
35  */
36 
37 #include <sys/cdefs.h>
38 __KERNEL_RCSID(0, "$NetBSD: vfs_lockf.c,v 1.66 2008/08/01 07:11:24 skrll Exp $");
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/kernel.h>
43 #include <sys/file.h>
44 #include <sys/proc.h>
45 #include <sys/vnode.h>
46 #include <sys/pool.h>
47 #include <sys/fcntl.h>
48 #include <sys/lockf.h>
49 #include <sys/atomic.h>
50 #include <sys/kauth.h>
51 
52 /*
53  * The lockf structure is a kernel structure which contains the information
54  * associated with a byte range lock.  The lockf structures are linked into
55  * the vnode structure.  Locks are sorted by the starting byte of the lock for
56  * efficiency.
57  *
58  * lf_next is used for two purposes, depending on whether the lock is
59  * being held, or is in conflict with an existing lock.  If this lock
60  * is held, it indicates the next lock on the same vnode.
61  * For pending locks, if lock->lf_next is non-NULL, then lock->lf_block
62  * must be queued on the lf_blkhd TAILQ of lock->lf_next.
63  */
64 
65 TAILQ_HEAD(locklist, lockf);
66 
67 struct lockf {
68 	kcondvar_t lf_cv;	 /* Signalling */
69 	short	lf_flags;	 /* Lock semantics: F_POSIX, F_FLOCK, F_WAIT */
70 	short	lf_type;	 /* Lock type: F_RDLCK, F_WRLCK */
71 	off_t	lf_start;	 /* The byte # of the start of the lock */
72 	off_t	lf_end;		 /* The byte # of the end of the lock (-1=EOF)*/
73 	void	*lf_id;		 /* process or file description holding lock */
74 	struct	lockf **lf_head; /* Back pointer to the head of lockf list */
75 	struct	lockf *lf_next;	 /* Next lock on this vnode, or blocking lock */
76 	struct  locklist lf_blkhd; /* List of requests blocked on this lock */
77 	TAILQ_ENTRY(lockf) lf_block;/* A request waiting for a lock */
78 	uid_t	lf_uid;		 /* User ID responsible */
79 };
80 
81 /* Maximum length of sleep chains to traverse to try and detect deadlock. */
82 #define MAXDEPTH 50
83 
84 static pool_cache_t lockf_cache;
85 static kmutex_t *lockf_lock;
86 static char lockstr[] = "lockf";
87 
88 /*
89  * This variable controls the maximum number of processes that will
90  * be checked in doing deadlock detection.
91  */
92 int maxlockdepth = MAXDEPTH;
93 
94 #ifdef LOCKF_DEBUG
95 int	lockf_debug = 0;
96 #endif
97 
98 #define SELF	0x1
99 #define OTHERS	0x2
100 
101 /*
102  * XXX TODO
103  * Misc cleanups: "void *id" should be visible in the API as a
104  * "struct proc *".
105  * (This requires rototilling all VFS's which support advisory locking).
106  */
107 
108 /*
109  * If there's a lot of lock contention on a single vnode, locking
110  * schemes which allow for more paralleism would be needed.  Given how
111  * infrequently byte-range locks are actually used in typical BSD
112  * code, a more complex approach probably isn't worth it.
113  */
114 
115 /*
116  * We enforce a limit on locks by uid, so that a single user cannot
117  * run the kernel out of memory.  For now, the limit is pretty coarse.
118  * There is no limit on root.
119  *
120  * Splitting a lock will always succeed, regardless of current allocations.
121  * If you're slightly above the limit, we still have to permit an allocation
122  * so that the unlock can succeed.  If the unlocking causes too many splits,
123  * however, you're totally cutoff.
124  */
125 int maxlocksperuid = 1024;
126 
127 #ifdef LOCKF_DEBUG
128 /*
129  * Print out a lock.
130  */
131 static void
132 lf_print(const char *tag, struct lockf *lock)
133 {
134 
135 	printf("%s: lock %p for ", tag, lock);
136 	if (lock->lf_flags & F_POSIX)
137 		printf("proc %d", ((struct proc *)lock->lf_id)->p_pid);
138 	else
139 		printf("file %p", (struct file *)lock->lf_id);
140 	printf(" %s, start %qx, end %qx",
141 		lock->lf_type == F_RDLCK ? "shared" :
142 		lock->lf_type == F_WRLCK ? "exclusive" :
143 		lock->lf_type == F_UNLCK ? "unlock" :
144 		"unknown", lock->lf_start, lock->lf_end);
145 	if (TAILQ_FIRST(&lock->lf_blkhd))
146 		printf(" block %p\n", TAILQ_FIRST(&lock->lf_blkhd));
147 	else
148 		printf("\n");
149 }
150 
151 static void
152 lf_printlist(const char *tag, struct lockf *lock)
153 {
154 	struct lockf *lf, *blk;
155 
156 	printf("%s: Lock list:\n", tag);
157 	for (lf = *lock->lf_head; lf; lf = lf->lf_next) {
158 		printf("\tlock %p for ", lf);
159 		if (lf->lf_flags & F_POSIX)
160 			printf("proc %d", ((struct proc *)lf->lf_id)->p_pid);
161 		else
162 			printf("file %p", (struct file *)lf->lf_id);
163 		printf(", %s, start %qx, end %qx",
164 			lf->lf_type == F_RDLCK ? "shared" :
165 			lf->lf_type == F_WRLCK ? "exclusive" :
166 			lf->lf_type == F_UNLCK ? "unlock" :
167 			"unknown", lf->lf_start, lf->lf_end);
168 		TAILQ_FOREACH(blk, &lf->lf_blkhd, lf_block) {
169 			if (blk->lf_flags & F_POSIX)
170 				printf("; proc %d",
171 				    ((struct proc *)blk->lf_id)->p_pid);
172 			else
173 				printf("; file %p", (struct file *)blk->lf_id);
174 			printf(", %s, start %qx, end %qx",
175 				blk->lf_type == F_RDLCK ? "shared" :
176 				blk->lf_type == F_WRLCK ? "exclusive" :
177 				blk->lf_type == F_UNLCK ? "unlock" :
178 				"unknown", blk->lf_start, blk->lf_end);
179 			if (TAILQ_FIRST(&blk->lf_blkhd))
180 				 panic("lf_printlist: bad list");
181 		}
182 		printf("\n");
183 	}
184 }
185 #endif /* LOCKF_DEBUG */
186 
187 /*
188  * 3 options for allowfail.
189  * 0 - always allocate.  1 - cutoff at limit.  2 - cutoff at double limit.
190  */
191 static struct lockf *
192 lf_alloc(uid_t uid, int allowfail)
193 {
194 	struct uidinfo *uip;
195 	struct lockf *lock;
196 	u_long lcnt;
197 
198 	uip = uid_find(uid);
199 	lcnt = atomic_inc_ulong_nv(&uip->ui_lockcnt);
200 	if (uid && allowfail && lcnt >
201 	    (allowfail == 1 ? maxlocksperuid : (maxlocksperuid * 2))) {
202 		atomic_dec_ulong(&uip->ui_lockcnt);
203 		return NULL;
204 	}
205 
206 	lock = pool_cache_get(lockf_cache, PR_WAITOK);
207 	lock->lf_uid = uid;
208 	return lock;
209 }
210 
211 static void
212 lf_free(struct lockf *lock)
213 {
214 	struct uidinfo *uip;
215 
216 	uip = uid_find(lock->lf_uid);
217 	atomic_dec_ulong(&uip->ui_lockcnt);
218 	pool_cache_put(lockf_cache, lock);
219 }
220 
221 static int
222 lf_ctor(void *arg, void *obj, int flag)
223 {
224 	struct lockf *lock;
225 
226 	lock = obj;
227 	cv_init(&lock->lf_cv, lockstr);
228 
229 	return 0;
230 }
231 
232 static void
233 lf_dtor(void *arg, void *obj)
234 {
235 	struct lockf *lock;
236 
237 	lock = obj;
238 	cv_destroy(&lock->lf_cv);
239 }
240 
241 /*
242  * Walk the list of locks for an inode to
243  * find an overlapping lock (if any).
244  *
245  * NOTE: this returns only the FIRST overlapping lock.  There
246  *	 may be more than one.
247  */
248 static int
249 lf_findoverlap(struct lockf *lf, struct lockf *lock, int type,
250     struct lockf ***prev, struct lockf **overlap)
251 {
252 	off_t start, end;
253 
254 	*overlap = lf;
255 	if (lf == NULL)
256 		return 0;
257 #ifdef LOCKF_DEBUG
258 	if (lockf_debug & 2)
259 		lf_print("lf_findoverlap: looking for overlap in", lock);
260 #endif /* LOCKF_DEBUG */
261 	start = lock->lf_start;
262 	end = lock->lf_end;
263 	while (lf != NULL) {
264 		if (((type == SELF) && lf->lf_id != lock->lf_id) ||
265 		    ((type == OTHERS) && lf->lf_id == lock->lf_id)) {
266 			*prev = &lf->lf_next;
267 			*overlap = lf = lf->lf_next;
268 			continue;
269 		}
270 #ifdef LOCKF_DEBUG
271 		if (lockf_debug & 2)
272 			lf_print("\tchecking", lf);
273 #endif /* LOCKF_DEBUG */
274 		/*
275 		 * OK, check for overlap
276 		 *
277 		 * Six cases:
278 		 *	0) no overlap
279 		 *	1) overlap == lock
280 		 *	2) overlap contains lock
281 		 *	3) lock contains overlap
282 		 *	4) overlap starts before lock
283 		 *	5) overlap ends after lock
284 		 */
285 		if ((lf->lf_end != -1 && start > lf->lf_end) ||
286 		    (end != -1 && lf->lf_start > end)) {
287 			/* Case 0 */
288 #ifdef LOCKF_DEBUG
289 			if (lockf_debug & 2)
290 				printf("no overlap\n");
291 #endif /* LOCKF_DEBUG */
292 			if ((type & SELF) && end != -1 && lf->lf_start > end)
293 				return 0;
294 			*prev = &lf->lf_next;
295 			*overlap = lf = lf->lf_next;
296 			continue;
297 		}
298 		if ((lf->lf_start == start) && (lf->lf_end == end)) {
299 			/* Case 1 */
300 #ifdef LOCKF_DEBUG
301 			if (lockf_debug & 2)
302 				printf("overlap == lock\n");
303 #endif /* LOCKF_DEBUG */
304 			return 1;
305 		}
306 		if ((lf->lf_start <= start) &&
307 		    (end != -1) &&
308 		    ((lf->lf_end >= end) || (lf->lf_end == -1))) {
309 			/* Case 2 */
310 #ifdef LOCKF_DEBUG
311 			if (lockf_debug & 2)
312 				printf("overlap contains lock\n");
313 #endif /* LOCKF_DEBUG */
314 			return 2;
315 		}
316 		if (start <= lf->lf_start &&
317 		           (end == -1 ||
318 			   (lf->lf_end != -1 && end >= lf->lf_end))) {
319 			/* Case 3 */
320 #ifdef LOCKF_DEBUG
321 			if (lockf_debug & 2)
322 				printf("lock contains overlap\n");
323 #endif /* LOCKF_DEBUG */
324 			return 3;
325 		}
326 		if ((lf->lf_start < start) &&
327 			((lf->lf_end >= start) || (lf->lf_end == -1))) {
328 			/* Case 4 */
329 #ifdef LOCKF_DEBUG
330 			if (lockf_debug & 2)
331 				printf("overlap starts before lock\n");
332 #endif /* LOCKF_DEBUG */
333 			return 4;
334 		}
335 		if ((lf->lf_start > start) &&
336 			(end != -1) &&
337 			((lf->lf_end > end) || (lf->lf_end == -1))) {
338 			/* Case 5 */
339 #ifdef LOCKF_DEBUG
340 			if (lockf_debug & 2)
341 				printf("overlap ends after lock\n");
342 #endif /* LOCKF_DEBUG */
343 			return 5;
344 		}
345 		panic("lf_findoverlap: default");
346 	}
347 	return 0;
348 }
349 
350 /*
351  * Split a lock and a contained region into
352  * two or three locks as necessary.
353  */
354 static void
355 lf_split(struct lockf *lock1, struct lockf *lock2, struct lockf **sparelock)
356 {
357 	struct lockf *splitlock;
358 
359 #ifdef LOCKF_DEBUG
360 	if (lockf_debug & 2) {
361 		lf_print("lf_split", lock1);
362 		lf_print("splitting from", lock2);
363 	}
364 #endif /* LOCKF_DEBUG */
365 	/*
366 	 * Check to see if spliting into only two pieces.
367 	 */
368 	if (lock1->lf_start == lock2->lf_start) {
369 		lock1->lf_start = lock2->lf_end + 1;
370 		lock2->lf_next = lock1;
371 		return;
372 	}
373 	if (lock1->lf_end == lock2->lf_end) {
374 		lock1->lf_end = lock2->lf_start - 1;
375 		lock2->lf_next = lock1->lf_next;
376 		lock1->lf_next = lock2;
377 		return;
378 	}
379 	/*
380 	 * Make a new lock consisting of the last part of
381 	 * the encompassing lock
382 	 */
383 	splitlock = *sparelock;
384 	*sparelock = NULL;
385 	memcpy(splitlock, lock1, sizeof(*splitlock));
386 	splitlock->lf_start = lock2->lf_end + 1;
387 	TAILQ_INIT(&splitlock->lf_blkhd);
388 	lock1->lf_end = lock2->lf_start - 1;
389 	/*
390 	 * OK, now link it in
391 	 */
392 	splitlock->lf_next = lock1->lf_next;
393 	lock2->lf_next = splitlock;
394 	lock1->lf_next = lock2;
395 }
396 
397 /*
398  * Wakeup a blocklist
399  */
400 static void
401 lf_wakelock(struct lockf *listhead)
402 {
403 	struct lockf *wakelock;
404 
405 	while ((wakelock = TAILQ_FIRST(&listhead->lf_blkhd))) {
406 		KASSERT(wakelock->lf_next == listhead);
407 		TAILQ_REMOVE(&listhead->lf_blkhd, wakelock, lf_block);
408 		wakelock->lf_next = NULL;
409 #ifdef LOCKF_DEBUG
410 		if (lockf_debug & 2)
411 			lf_print("lf_wakelock: awakening", wakelock);
412 #endif
413 		cv_broadcast(&wakelock->lf_cv);
414 	}
415 }
416 
417 /*
418  * Remove a byte-range lock on an inode.
419  *
420  * Generally, find the lock (or an overlap to that lock)
421  * and remove it (or shrink it), then wakeup anyone we can.
422  */
423 static int
424 lf_clearlock(struct lockf *unlock, struct lockf **sparelock)
425 {
426 	struct lockf **head = unlock->lf_head;
427 	struct lockf *lf = *head;
428 	struct lockf *overlap, **prev;
429 	int ovcase;
430 
431 	if (lf == NULL)
432 		return 0;
433 #ifdef LOCKF_DEBUG
434 	if (unlock->lf_type != F_UNLCK)
435 		panic("lf_clearlock: bad type");
436 	if (lockf_debug & 1)
437 		lf_print("lf_clearlock", unlock);
438 #endif /* LOCKF_DEBUG */
439 	prev = head;
440 	while ((ovcase = lf_findoverlap(lf, unlock, SELF,
441 	    &prev, &overlap)) != 0) {
442 		/*
443 		 * Wakeup the list of locks to be retried.
444 		 */
445 		lf_wakelock(overlap);
446 
447 		switch (ovcase) {
448 
449 		case 1: /* overlap == lock */
450 			*prev = overlap->lf_next;
451 			lf_free(overlap);
452 			break;
453 
454 		case 2: /* overlap contains lock: split it */
455 			if (overlap->lf_start == unlock->lf_start) {
456 				overlap->lf_start = unlock->lf_end + 1;
457 				break;
458 			}
459 			lf_split(overlap, unlock, sparelock);
460 			overlap->lf_next = unlock->lf_next;
461 			break;
462 
463 		case 3: /* lock contains overlap */
464 			*prev = overlap->lf_next;
465 			lf = overlap->lf_next;
466 			lf_free(overlap);
467 			continue;
468 
469 		case 4: /* overlap starts before lock */
470 			overlap->lf_end = unlock->lf_start - 1;
471 			prev = &overlap->lf_next;
472 			lf = overlap->lf_next;
473 			continue;
474 
475 		case 5: /* overlap ends after lock */
476 			overlap->lf_start = unlock->lf_end + 1;
477 			break;
478 		}
479 		break;
480 	}
481 #ifdef LOCKF_DEBUG
482 	if (lockf_debug & 1)
483 		lf_printlist("lf_clearlock", unlock);
484 #endif /* LOCKF_DEBUG */
485 	return 0;
486 }
487 
488 /*
489  * Walk the list of locks for an inode and
490  * return the first blocking lock.
491  */
492 static struct lockf *
493 lf_getblock(struct lockf *lock)
494 {
495 	struct lockf **prev, *overlap, *lf = *(lock->lf_head);
496 
497 	prev = lock->lf_head;
498 	while (lf_findoverlap(lf, lock, OTHERS, &prev, &overlap) != 0) {
499 		/*
500 		 * We've found an overlap, see if it blocks us
501 		 */
502 		if ((lock->lf_type == F_WRLCK || overlap->lf_type == F_WRLCK))
503 			return overlap;
504 		/*
505 		 * Nope, point to the next one on the list and
506 		 * see if it blocks us
507 		 */
508 		lf = overlap->lf_next;
509 	}
510 	return NULL;
511 }
512 
513 /*
514  * Set a byte-range lock.
515  */
516 static int
517 lf_setlock(struct lockf *lock, struct lockf **sparelock,
518     kmutex_t *interlock)
519 {
520 	struct lockf *block;
521 	struct lockf **head = lock->lf_head;
522 	struct lockf **prev, *overlap, *ltmp;
523 	int ovcase, needtolink, error;
524 
525 #ifdef LOCKF_DEBUG
526 	if (lockf_debug & 1)
527 		lf_print("lf_setlock", lock);
528 #endif /* LOCKF_DEBUG */
529 
530 	/*
531 	 * Scan lock list for this file looking for locks that would block us.
532 	 */
533 	while ((block = lf_getblock(lock)) != NULL) {
534 		/*
535 		 * Free the structure and return if nonblocking.
536 		 */
537 		if ((lock->lf_flags & F_WAIT) == 0) {
538 			lf_free(lock);
539 			return EAGAIN;
540 		}
541 		/*
542 		 * We are blocked. Since flock style locks cover
543 		 * the whole file, there is no chance for deadlock.
544 		 * For byte-range locks we must check for deadlock.
545 		 *
546 		 * Deadlock detection is done by looking through the
547 		 * wait channels to see if there are any cycles that
548 		 * involve us. MAXDEPTH is set just to make sure we
549 		 * do not go off into neverneverland.
550 		 */
551 		if ((lock->lf_flags & F_POSIX) &&
552 		    (block->lf_flags & F_POSIX)) {
553 			struct lwp *wlwp;
554 			volatile const struct lockf *waitblock;
555 			int i = 0;
556 			struct proc *p;
557 
558 			p = (struct proc *)block->lf_id;
559 			KASSERT(p != NULL);
560 			while (i++ < maxlockdepth) {
561 				mutex_enter(p->p_lock);
562 				if (p->p_nlwps > 1) {
563 					mutex_exit(p->p_lock);
564 					break;
565 				}
566 				wlwp = LIST_FIRST(&p->p_lwps);
567 				lwp_lock(wlwp);
568 				if (wlwp->l_wchan == NULL ||
569 				    wlwp->l_wmesg != lockstr) {
570 					lwp_unlock(wlwp);
571 					mutex_exit(p->p_lock);
572 					break;
573 				}
574 				waitblock = wlwp->l_wchan;
575 				lwp_unlock(wlwp);
576 				mutex_exit(p->p_lock);
577 				if (waitblock == NULL) {
578 					/*
579 					 * this lwp just got up but
580 					 * not returned from ltsleep yet.
581 					 */
582 					break;
583 				}
584 				/* Get the owner of the blocking lock */
585 				waitblock = waitblock->lf_next;
586 				if ((waitblock->lf_flags & F_POSIX) == 0)
587 					break;
588 				p = (struct proc *)waitblock->lf_id;
589 				if (p == curproc) {
590 					lf_free(lock);
591 					return EDEADLK;
592 				}
593 			}
594 			/*
595 			 * If we're still following a dependency chain
596 			 * after maxlockdepth iterations, assume we're in
597 			 * a cycle to be safe.
598 			 */
599 			if (i >= maxlockdepth) {
600 				lf_free(lock);
601 				return EDEADLK;
602 			}
603 		}
604 		/*
605 		 * For flock type locks, we must first remove
606 		 * any shared locks that we hold before we sleep
607 		 * waiting for an exclusive lock.
608 		 */
609 		if ((lock->lf_flags & F_FLOCK) &&
610 		    lock->lf_type == F_WRLCK) {
611 			lock->lf_type = F_UNLCK;
612 			(void) lf_clearlock(lock, NULL);
613 			lock->lf_type = F_WRLCK;
614 		}
615 		/*
616 		 * Add our lock to the blocked list and sleep until we're free.
617 		 * Remember who blocked us (for deadlock detection).
618 		 */
619 		lock->lf_next = block;
620 		TAILQ_INSERT_TAIL(&block->lf_blkhd, lock, lf_block);
621 #ifdef LOCKF_DEBUG
622 		if (lockf_debug & 1) {
623 			lf_print("lf_setlock: blocking on", block);
624 			lf_printlist("lf_setlock", block);
625 		}
626 #endif /* LOCKF_DEBUG */
627 		error = cv_wait_sig(&lock->lf_cv, interlock);
628 
629 		/*
630 		 * We may have been awoken by a signal (in
631 		 * which case we must remove ourselves from the
632 		 * blocked list) and/or by another process
633 		 * releasing a lock (in which case we have already
634 		 * been removed from the blocked list and our
635 		 * lf_next field set to NULL).
636 		 */
637 		if (lock->lf_next != NULL) {
638 			TAILQ_REMOVE(&lock->lf_next->lf_blkhd, lock, lf_block);
639 			lock->lf_next = NULL;
640 		}
641 		if (error) {
642 			lf_free(lock);
643 			return error;
644 		}
645 	}
646 	/*
647 	 * No blocks!!  Add the lock.  Note that we will
648 	 * downgrade or upgrade any overlapping locks this
649 	 * process already owns.
650 	 *
651 	 * Skip over locks owned by other processes.
652 	 * Handle any locks that overlap and are owned by ourselves.
653 	 */
654 	prev = head;
655 	block = *head;
656 	needtolink = 1;
657 	for (;;) {
658 		ovcase = lf_findoverlap(block, lock, SELF, &prev, &overlap);
659 		if (ovcase)
660 			block = overlap->lf_next;
661 		/*
662 		 * Six cases:
663 		 *	0) no overlap
664 		 *	1) overlap == lock
665 		 *	2) overlap contains lock
666 		 *	3) lock contains overlap
667 		 *	4) overlap starts before lock
668 		 *	5) overlap ends after lock
669 		 */
670 		switch (ovcase) {
671 		case 0: /* no overlap */
672 			if (needtolink) {
673 				*prev = lock;
674 				lock->lf_next = overlap;
675 			}
676 			break;
677 
678 		case 1: /* overlap == lock */
679 			/*
680 			 * If downgrading lock, others may be
681 			 * able to acquire it.
682 			 */
683 			if (lock->lf_type == F_RDLCK &&
684 			    overlap->lf_type == F_WRLCK)
685 				lf_wakelock(overlap);
686 			overlap->lf_type = lock->lf_type;
687 			lf_free(lock);
688 			lock = overlap; /* for debug output below */
689 			break;
690 
691 		case 2: /* overlap contains lock */
692 			/*
693 			 * Check for common starting point and different types.
694 			 */
695 			if (overlap->lf_type == lock->lf_type) {
696 				lf_free(lock);
697 				lock = overlap; /* for debug output below */
698 				break;
699 			}
700 			if (overlap->lf_start == lock->lf_start) {
701 				*prev = lock;
702 				lock->lf_next = overlap;
703 				overlap->lf_start = lock->lf_end + 1;
704 			} else
705 				lf_split(overlap, lock, sparelock);
706 			lf_wakelock(overlap);
707 			break;
708 
709 		case 3: /* lock contains overlap */
710 			/*
711 			 * If downgrading lock, others may be able to
712 			 * acquire it, otherwise take the list.
713 			 */
714 			if (lock->lf_type == F_RDLCK &&
715 			    overlap->lf_type == F_WRLCK) {
716 				lf_wakelock(overlap);
717 			} else {
718 				while ((ltmp = TAILQ_FIRST(&overlap->lf_blkhd))) {
719 					KASSERT(ltmp->lf_next == overlap);
720 					TAILQ_REMOVE(&overlap->lf_blkhd, ltmp,
721 					    lf_block);
722 					ltmp->lf_next = lock;
723 					TAILQ_INSERT_TAIL(&lock->lf_blkhd,
724 					    ltmp, lf_block);
725 				}
726 			}
727 			/*
728 			 * Add the new lock if necessary and delete the overlap.
729 			 */
730 			if (needtolink) {
731 				*prev = lock;
732 				lock->lf_next = overlap->lf_next;
733 				prev = &lock->lf_next;
734 				needtolink = 0;
735 			} else
736 				*prev = overlap->lf_next;
737 			lf_free(overlap);
738 			continue;
739 
740 		case 4: /* overlap starts before lock */
741 			/*
742 			 * Add lock after overlap on the list.
743 			 */
744 			lock->lf_next = overlap->lf_next;
745 			overlap->lf_next = lock;
746 			overlap->lf_end = lock->lf_start - 1;
747 			prev = &lock->lf_next;
748 			lf_wakelock(overlap);
749 			needtolink = 0;
750 			continue;
751 
752 		case 5: /* overlap ends after lock */
753 			/*
754 			 * Add the new lock before overlap.
755 			 */
756 			if (needtolink) {
757 				*prev = lock;
758 				lock->lf_next = overlap;
759 			}
760 			overlap->lf_start = lock->lf_end + 1;
761 			lf_wakelock(overlap);
762 			break;
763 		}
764 		break;
765 	}
766 #ifdef LOCKF_DEBUG
767 	if (lockf_debug & 1) {
768 		lf_print("lf_setlock: got the lock", lock);
769 		lf_printlist("lf_setlock", lock);
770 	}
771 #endif /* LOCKF_DEBUG */
772 	return 0;
773 }
774 
775 /*
776  * Check whether there is a blocking lock,
777  * and if so return its process identifier.
778  */
779 static int
780 lf_getlock(struct lockf *lock, struct flock *fl)
781 {
782 	struct lockf *block;
783 
784 #ifdef LOCKF_DEBUG
785 	if (lockf_debug & 1)
786 		lf_print("lf_getlock", lock);
787 #endif /* LOCKF_DEBUG */
788 
789 	if ((block = lf_getblock(lock)) != NULL) {
790 		fl->l_type = block->lf_type;
791 		fl->l_whence = SEEK_SET;
792 		fl->l_start = block->lf_start;
793 		if (block->lf_end == -1)
794 			fl->l_len = 0;
795 		else
796 			fl->l_len = block->lf_end - block->lf_start + 1;
797 		if (block->lf_flags & F_POSIX)
798 			fl->l_pid = ((struct proc *)block->lf_id)->p_pid;
799 		else
800 			fl->l_pid = -1;
801 	} else {
802 		fl->l_type = F_UNLCK;
803 	}
804 	return 0;
805 }
806 
807 /*
808  * Do an advisory lock operation.
809  */
810 int
811 lf_advlock(struct vop_advlock_args *ap, struct lockf **head, off_t size)
812 {
813 	struct lwp *l = curlwp;
814 	struct flock *fl = ap->a_fl;
815 	struct lockf *lock = NULL;
816 	struct lockf *sparelock;
817 	kmutex_t *interlock = lockf_lock;
818 	off_t start, end;
819 	int error = 0;
820 
821 	/*
822 	 * Convert the flock structure into a start and end.
823 	 */
824 	switch (fl->l_whence) {
825 	case SEEK_SET:
826 	case SEEK_CUR:
827 		/*
828 		 * Caller is responsible for adding any necessary offset
829 		 * when SEEK_CUR is used.
830 		 */
831 		start = fl->l_start;
832 		break;
833 
834 	case SEEK_END:
835 		start = size + fl->l_start;
836 		break;
837 
838 	default:
839 		return EINVAL;
840 	}
841 	if (start < 0)
842 		return EINVAL;
843 
844 	/*
845 	 * Allocate locks before acquiring the interlock.  We need two
846 	 * locks in the worst case.
847 	 */
848 	switch (ap->a_op) {
849 	case F_SETLK:
850 	case F_UNLCK:
851 		/*
852 		 * XXX For F_UNLCK case, we can re-use the lock.
853 		 */
854 		if ((ap->a_flags & F_FLOCK) == 0) {
855 			/*
856 			 * Byte-range lock might need one more lock.
857 			 */
858 			sparelock = lf_alloc(kauth_cred_geteuid(l->l_cred), 0);
859 			if (sparelock == NULL) {
860 				error = ENOMEM;
861 				goto quit;
862 			}
863 			break;
864 		}
865 		/* FALLTHROUGH */
866 
867 	case F_GETLK:
868 		sparelock = NULL;
869 		break;
870 
871 	default:
872 		return EINVAL;
873 	}
874 
875 	lock = lf_alloc(kauth_cred_geteuid(l->l_cred),
876 	    ap->a_op != F_UNLCK ? 1 : 2);
877 	if (lock == NULL) {
878 		error = ENOMEM;
879 		goto quit;
880 	}
881 
882 	mutex_enter(interlock);
883 
884 	/*
885 	 * Avoid the common case of unlocking when inode has no locks.
886 	 */
887 	if (*head == (struct lockf *)0) {
888 		if (ap->a_op != F_SETLK) {
889 			fl->l_type = F_UNLCK;
890 			error = 0;
891 			goto quit_unlock;
892 		}
893 	}
894 
895 	if (fl->l_len == 0)
896 		end = -1;
897 	else
898 		end = start + fl->l_len - 1;
899 	/*
900 	 * Create the lockf structure.
901 	 */
902 	lock->lf_start = start;
903 	lock->lf_end = end;
904 	lock->lf_head = head;
905 	lock->lf_type = fl->l_type;
906 	lock->lf_next = (struct lockf *)0;
907 	TAILQ_INIT(&lock->lf_blkhd);
908 	lock->lf_flags = ap->a_flags;
909 	if (lock->lf_flags & F_POSIX) {
910 		KASSERT(curproc == (struct proc *)ap->a_id);
911 	}
912 	lock->lf_id = (struct proc *)ap->a_id;
913 
914 	/*
915 	 * Do the requested operation.
916 	 */
917 	switch (ap->a_op) {
918 
919 	case F_SETLK:
920 		error = lf_setlock(lock, &sparelock, interlock);
921 		lock = NULL; /* lf_setlock freed it */
922 		break;
923 
924 	case F_UNLCK:
925 		error = lf_clearlock(lock, &sparelock);
926 		break;
927 
928 	case F_GETLK:
929 		error = lf_getlock(lock, fl);
930 		break;
931 
932 	default:
933 		break;
934 		/* NOTREACHED */
935 	}
936 
937 quit_unlock:
938 	mutex_exit(interlock);
939 quit:
940 	if (lock)
941 		lf_free(lock);
942 	if (sparelock)
943 		lf_free(sparelock);
944 
945 	return error;
946 }
947 
948 /*
949  * Initialize subsystem.   XXX We use a global lock.  This could be the
950  * vnode interlock, but the deadlock detection code may need to inspect
951  * locks belonging to other files.
952  */
953 void
954 lf_init(void)
955 {
956 
957 	lockf_cache = pool_cache_init(sizeof(struct lockf), 0, 0, 0, "lockf",
958  	    NULL, IPL_NONE, lf_ctor, lf_dtor, NULL);
959         lockf_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
960 }
961