xref: /netbsd-src/sys/kern/vfs_lockf.c (revision 20e85ad185ab16980f1219a557c42e057edb42ea)
1 /*	$NetBSD: vfs_lockf.c,v 1.39 2005/03/25 22:48:23 christos 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 1.39 2005/03/25 22:48:23 christos 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 
50 POOL_INIT(lockfpool, sizeof(struct lockf), 0, 0, 0, "lockfpl",
51     &pool_allocator_nointr);
52 
53 /*
54  * This variable controls the maximum number of processes that will
55  * be checked in doing deadlock detection.
56  */
57 int maxlockdepth = MAXDEPTH;
58 
59 #ifdef LOCKF_DEBUG
60 int	lockf_debug = 0;
61 #endif
62 
63 #define NOLOCKF (struct lockf *)0
64 #define SELF	0x1
65 #define OTHERS	0x2
66 
67 static int lf_clearlock(struct lockf *, struct lockf **);
68 static int lf_findoverlap(struct lockf *,
69 	    struct lockf *, int, struct lockf ***, struct lockf **);
70 static struct lockf *lf_getblock(struct lockf *);
71 static int lf_getlock(struct lockf *, struct flock *);
72 static int lf_setlock(struct lockf *, struct lockf **, struct simplelock *);
73 static void lf_split(struct lockf *, struct lockf *, struct lockf **);
74 static void lf_wakelock(struct lockf *);
75 static struct lockf *lf_alloc(uid_t, int);
76 static void lf_free(struct lockf *);
77 
78 
79 #ifdef LOCKF_DEBUG
80 static void lf_print(char *, struct lockf *);
81 static void lf_printlist(char *, struct lockf *);
82 #endif
83 
84 /*
85  * XXX TODO
86  * Misc cleanups: "caddr_t id" should be visible in the API as a
87  * "struct proc *".
88  * (This requires rototilling all VFS's which support advisory locking).
89  */
90 
91 /*
92  * If there's a lot of lock contention on a single vnode, locking
93  * schemes which allow for more paralleism would be needed.  Given how
94  * infrequently byte-range locks are actually used in typical BSD
95  * code, a more complex approach probably isn't worth it.
96  */
97 
98 /*
99  * We enforce a limit on locks by uid, so that a single user cannot
100  * run the kernel out of memory.  For now, the limit is pretty coarse.
101  * There is no limit on root.
102  *
103  * Splitting a lock will always succeed, regardless of current allocations.
104  * If you're slightly above the limit, we still have to permit an allocation
105  * so that the unlock can succeed.  If the unlocking causes too many splits,
106  * however, you're totally cutoff.
107  */
108 int maxlocksperuid = 1024;
109 
110 /*
111  * 3 options for allowfail.
112  * 0 - always allocate.  1 - cutoff at limit.  2 - cutoff at double limit.
113  */
114 struct lockf *
115 lf_alloc(uid_t uid, int allowfail)
116 {
117 	struct uidinfo *uip;
118 	struct lockf *lock;
119 
120 	uip = uid_find(uid);
121 	if (uid && allowfail && uip->ui_lockcnt >
122 	    (allowfail == 1 ? maxlocksperuid : (maxlocksperuid * 2)))
123 		return (NULL);
124 	uip->ui_lockcnt++;
125 	lock = pool_get(&lockfpool, PR_WAITOK);
126 	lock->lf_uid = uid;
127 	return (lock);
128 }
129 
130 void
131 lf_free(struct lockf *lock)
132 {
133 	struct uidinfo *uip;
134 
135 	uip = uid_find(lock->lf_uid);
136 	uip->ui_lockcnt--;
137 	pool_put(&lockfpool, lock);
138 }
139 
140 /*
141  * Do an advisory lock operation.
142  */
143 int
144 lf_advlock(struct vop_advlock_args *ap, struct lockf **head, off_t size)
145 {
146 	struct proc *p = curproc;
147 	struct flock *fl = ap->a_fl;
148 	struct lockf *lock = NULL;
149 	struct lockf *sparelock;
150 	struct simplelock *interlock = &ap->a_vp->v_interlock;
151 	off_t start, end;
152 	int error = 0;
153 
154 	/*
155 	 * Convert the flock structure into a start and end.
156 	 */
157 	switch (fl->l_whence) {
158 	case SEEK_SET:
159 	case SEEK_CUR:
160 		/*
161 		 * Caller is responsible for adding any necessary offset
162 		 * when SEEK_CUR is used.
163 		 */
164 		start = fl->l_start;
165 		break;
166 
167 	case SEEK_END:
168 		start = size + fl->l_start;
169 		break;
170 
171 	default:
172 		return EINVAL;
173 	}
174 	if (start < 0)
175 		return EINVAL;
176 
177 	/*
178 	 * allocate locks before acquire simple lock.
179 	 * we need two locks in the worst case.
180 	 */
181 	switch (ap->a_op) {
182 	case F_SETLK:
183 	case F_UNLCK:
184 		/*
185 		 * XXX for F_UNLCK case, we can re-use lock.
186 		 */
187 		if ((fl->l_type & F_FLOCK) == 0) {
188 			/*
189 			 * byte-range lock might need one more lock.
190 			 */
191 			sparelock = lf_alloc(p->p_ucred->cr_uid, 0);
192 			if (sparelock == NULL) {
193 				error = ENOMEM;
194 				goto quit;
195 			}
196 			break;
197 		}
198 		/* FALLTHROUGH */
199 
200 	case F_GETLK:
201 		sparelock = NULL;
202 		break;
203 
204 	default:
205 		return EINVAL;
206 	}
207 
208 	lock = lf_alloc(p->p_ucred->cr_uid, ap->a_op != F_UNLCK ? 1 : 2);
209 	if (lock == NULL) {
210 		error = ENOMEM;
211 		goto quit;
212 	}
213 
214 	simple_lock(interlock);
215 
216 	/*
217 	 * Avoid the common case of unlocking when inode has no locks.
218 	 */
219 	if (*head == (struct lockf *)0) {
220 		if (ap->a_op != F_SETLK) {
221 			fl->l_type = F_UNLCK;
222 			error = 0;
223 			goto quit_unlock;
224 		}
225 	}
226 
227 	if (fl->l_len == 0)
228 		end = -1;
229 	else
230 		end = start + fl->l_len - 1;
231 	/*
232 	 * Create the lockf structure.
233 	 */
234 	lock->lf_start = start;
235 	lock->lf_end = end;
236 	/* XXX NJWLWP
237 	 * I don't want to make the entire VFS universe use LWPs, because
238 	 * they don't need them, for the most part. This is an exception,
239 	 * and a kluge.
240 	 */
241 
242 	lock->lf_head = head;
243 	lock->lf_type = fl->l_type;
244 	lock->lf_next = (struct lockf *)0;
245 	TAILQ_INIT(&lock->lf_blkhd);
246 	lock->lf_flags = ap->a_flags;
247 	if (lock->lf_flags & F_POSIX) {
248 		KASSERT(curproc == (struct proc *)ap->a_id);
249 	}
250 	lock->lf_id = (struct proc *)ap->a_id;
251 	lock->lf_lwp = curlwp;
252 
253 	/*
254 	 * Do the requested operation.
255 	 */
256 	switch (ap->a_op) {
257 
258 	case F_SETLK:
259 		error = lf_setlock(lock, &sparelock, interlock);
260 		lock = NULL; /* lf_setlock freed it */
261 		break;
262 
263 	case F_UNLCK:
264 		error = lf_clearlock(lock, &sparelock);
265 		break;
266 
267 	case F_GETLK:
268 		error = lf_getlock(lock, fl);
269 		break;
270 
271 	default:
272 		break;
273 		/* NOTREACHED */
274 	}
275 
276 quit_unlock:
277 	simple_unlock(interlock);
278 quit:
279 	if (lock)
280 		lf_free(lock);
281 	if (sparelock)
282 		lf_free(sparelock);
283 
284 	return error;
285 }
286 
287 /*
288  * Set a byte-range lock.
289  */
290 static int
291 lf_setlock(struct lockf *lock, struct lockf **sparelock,
292     struct simplelock *interlock)
293 {
294 	struct lockf *block;
295 	struct lockf **head = lock->lf_head;
296 	struct lockf **prev, *overlap, *ltmp;
297 	static char lockstr[] = "lockf";
298 	int ovcase, priority, needtolink, error;
299 
300 #ifdef LOCKF_DEBUG
301 	if (lockf_debug & 1)
302 		lf_print("lf_setlock", lock);
303 #endif /* LOCKF_DEBUG */
304 
305 	/*
306 	 * Set the priority
307 	 */
308 	priority = PLOCK;
309 	if (lock->lf_type == F_WRLCK)
310 		priority += 4;
311 	priority |= PCATCH;
312 	/*
313 	 * Scan lock list for this file looking for locks that would block us.
314 	 */
315 	while ((block = lf_getblock(lock)) != NULL) {
316 		/*
317 		 * Free the structure and return if nonblocking.
318 		 */
319 		if ((lock->lf_flags & F_WAIT) == 0) {
320 			lf_free(lock);
321 			return EAGAIN;
322 		}
323 		/*
324 		 * We are blocked. Since flock style locks cover
325 		 * the whole file, there is no chance for deadlock.
326 		 * For byte-range locks we must check for deadlock.
327 		 *
328 		 * Deadlock detection is done by looking through the
329 		 * wait channels to see if there are any cycles that
330 		 * involve us. MAXDEPTH is set just to make sure we
331 		 * do not go off into neverneverland.
332 		 */
333 		if ((lock->lf_flags & F_POSIX) &&
334 		    (block->lf_flags & F_POSIX)) {
335 			struct lwp *wlwp;
336 			struct lockf *waitblock;
337 			int i = 0;
338 
339 			/*
340 			 * The block is waiting on something.  if_lwp will be
341 			 * 0 once the lock is granted, so we terminate the
342 			 * loop if we find this.
343 			 */
344 			wlwp = block->lf_lwp;
345 			while (wlwp && (i++ < maxlockdepth)) {
346 				waitblock = (struct lockf *)wlwp->l_wchan;
347 				/* Get the owner of the blocking lock */
348 				waitblock = waitblock->lf_next;
349 				if ((waitblock->lf_flags & F_POSIX) == 0)
350 					break;
351 				wlwp = waitblock->lf_lwp;
352 				if (wlwp == lock->lf_lwp) {
353 					lf_free(lock);
354 					return EDEADLK;
355 				}
356 			}
357 			/*
358 			 * If we're still following a dependency chain
359 			 * after maxlockdepth iterations, assume we're in
360 			 * a cycle to be safe.
361 			 */
362 			if (i >= maxlockdepth) {
363 				lf_free(lock);
364 				return EDEADLK;
365 			}
366 		}
367 		/*
368 		 * For flock type locks, we must first remove
369 		 * any shared locks that we hold before we sleep
370 		 * waiting for an exclusive lock.
371 		 */
372 		if ((lock->lf_flags & F_FLOCK) &&
373 		    lock->lf_type == F_WRLCK) {
374 			lock->lf_type = F_UNLCK;
375 			(void) lf_clearlock(lock, NULL);
376 			lock->lf_type = F_WRLCK;
377 		}
378 		/*
379 		 * Add our lock to the blocked list and sleep until we're free.
380 		 * Remember who blocked us (for deadlock detection).
381 		 */
382 		lock->lf_next = block;
383 		TAILQ_INSERT_TAIL(&block->lf_blkhd, lock, lf_block);
384 #ifdef LOCKF_DEBUG
385 		if (lockf_debug & 1) {
386 			lf_print("lf_setlock: blocking on", block);
387 			lf_printlist("lf_setlock", block);
388 		}
389 #endif /* LOCKF_DEBUG */
390 		error = ltsleep(lock, priority, lockstr, 0, interlock);
391 
392 		/*
393 		 * We may have been awakened by a signal (in
394 		 * which case we must remove ourselves from the
395 		 * blocked list) and/or by another process
396 		 * releasing a lock (in which case we have already
397 		 * been removed from the blocked list and our
398 		 * lf_next field set to NOLOCKF).
399 		 */
400 		if (lock->lf_next != NOLOCKF) {
401 			TAILQ_REMOVE(&lock->lf_next->lf_blkhd, lock, lf_block);
402 			lock->lf_next = NOLOCKF;
403 		}
404 		if (error) {
405 			lf_free(lock);
406 			return error;
407 		}
408 	}
409 	/*
410 	 * No blocks!!  Add the lock.  Note that we will
411 	 * downgrade or upgrade any overlapping locks this
412 	 * process already owns.
413 	 *
414 	 * Skip over locks owned by other processes.
415 	 * Handle any locks that overlap and are owned by ourselves.
416 	 */
417 	lock->lf_lwp = 0;
418 	prev = head;
419 	block = *head;
420 	needtolink = 1;
421 	for (;;) {
422 		ovcase = lf_findoverlap(block, lock, SELF, &prev, &overlap);
423 		if (ovcase)
424 			block = overlap->lf_next;
425 		/*
426 		 * Six cases:
427 		 *	0) no overlap
428 		 *	1) overlap == lock
429 		 *	2) overlap contains lock
430 		 *	3) lock contains overlap
431 		 *	4) overlap starts before lock
432 		 *	5) overlap ends after lock
433 		 */
434 		switch (ovcase) {
435 		case 0: /* no overlap */
436 			if (needtolink) {
437 				*prev = lock;
438 				lock->lf_next = overlap;
439 			}
440 			break;
441 
442 		case 1: /* overlap == lock */
443 			/*
444 			 * If downgrading lock, others may be
445 			 * able to acquire it.
446 			 */
447 			if (lock->lf_type == F_RDLCK &&
448 			    overlap->lf_type == F_WRLCK)
449 				lf_wakelock(overlap);
450 			overlap->lf_type = lock->lf_type;
451 			lf_free(lock);
452 			lock = overlap; /* for debug output below */
453 			break;
454 
455 		case 2: /* overlap contains lock */
456 			/*
457 			 * Check for common starting point and different types.
458 			 */
459 			if (overlap->lf_type == lock->lf_type) {
460 				lf_free(lock);
461 				lock = overlap; /* for debug output below */
462 				break;
463 			}
464 			if (overlap->lf_start == lock->lf_start) {
465 				*prev = lock;
466 				lock->lf_next = overlap;
467 				overlap->lf_start = lock->lf_end + 1;
468 			} else
469 				lf_split(overlap, lock, sparelock);
470 			lf_wakelock(overlap);
471 			break;
472 
473 		case 3: /* lock contains overlap */
474 			/*
475 			 * If downgrading lock, others may be able to
476 			 * acquire it, otherwise take the list.
477 			 */
478 			if (lock->lf_type == F_RDLCK &&
479 			    overlap->lf_type == F_WRLCK) {
480 				lf_wakelock(overlap);
481 			} else {
482 				while ((ltmp = TAILQ_FIRST(&overlap->lf_blkhd))) {
483 					KASSERT(ltmp->lf_next == overlap);
484 					TAILQ_REMOVE(&overlap->lf_blkhd, ltmp,
485 					    lf_block);
486 					ltmp->lf_next = lock;
487 					TAILQ_INSERT_TAIL(&lock->lf_blkhd,
488 					    ltmp, lf_block);
489 				}
490 			}
491 			/*
492 			 * Add the new lock if necessary and delete the overlap.
493 			 */
494 			if (needtolink) {
495 				*prev = lock;
496 				lock->lf_next = overlap->lf_next;
497 				prev = &lock->lf_next;
498 				needtolink = 0;
499 			} else
500 				*prev = overlap->lf_next;
501 			lf_free(overlap);
502 			continue;
503 
504 		case 4: /* overlap starts before lock */
505 			/*
506 			 * Add lock after overlap on the list.
507 			 */
508 			lock->lf_next = overlap->lf_next;
509 			overlap->lf_next = lock;
510 			overlap->lf_end = lock->lf_start - 1;
511 			prev = &lock->lf_next;
512 			lf_wakelock(overlap);
513 			needtolink = 0;
514 			continue;
515 
516 		case 5: /* overlap ends after lock */
517 			/*
518 			 * Add the new lock before overlap.
519 			 */
520 			if (needtolink) {
521 				*prev = lock;
522 				lock->lf_next = overlap;
523 			}
524 			overlap->lf_start = lock->lf_end + 1;
525 			lf_wakelock(overlap);
526 			break;
527 		}
528 		break;
529 	}
530 #ifdef LOCKF_DEBUG
531 	if (lockf_debug & 1) {
532 		lf_print("lf_setlock: got the lock", lock);
533 		lf_printlist("lf_setlock", lock);
534 	}
535 #endif /* LOCKF_DEBUG */
536 	return 0;
537 }
538 
539 /*
540  * Remove a byte-range lock on an inode.
541  *
542  * Generally, find the lock (or an overlap to that lock)
543  * and remove it (or shrink it), then wakeup anyone we can.
544  */
545 static int
546 lf_clearlock(struct lockf *unlock, struct lockf **sparelock)
547 {
548 	struct lockf **head = unlock->lf_head;
549 	struct lockf *lf = *head;
550 	struct lockf *overlap, **prev;
551 	int ovcase;
552 
553 	if (lf == NOLOCKF)
554 		return 0;
555 #ifdef LOCKF_DEBUG
556 	if (unlock->lf_type != F_UNLCK)
557 		panic("lf_clearlock: bad type");
558 	if (lockf_debug & 1)
559 		lf_print("lf_clearlock", unlock);
560 #endif /* LOCKF_DEBUG */
561 	prev = head;
562 	while ((ovcase = lf_findoverlap(lf, unlock, SELF,
563 					&prev, &overlap)) != 0) {
564 		/*
565 		 * Wakeup the list of locks to be retried.
566 		 */
567 		lf_wakelock(overlap);
568 
569 		switch (ovcase) {
570 
571 		case 1: /* overlap == lock */
572 			*prev = overlap->lf_next;
573 			lf_free(overlap);
574 			break;
575 
576 		case 2: /* overlap contains lock: split it */
577 			if (overlap->lf_start == unlock->lf_start) {
578 				overlap->lf_start = unlock->lf_end + 1;
579 				break;
580 			}
581 			lf_split(overlap, unlock, sparelock);
582 			overlap->lf_next = unlock->lf_next;
583 			break;
584 
585 		case 3: /* lock contains overlap */
586 			*prev = overlap->lf_next;
587 			lf = overlap->lf_next;
588 			lf_free(overlap);
589 			continue;
590 
591 		case 4: /* overlap starts before lock */
592 			overlap->lf_end = unlock->lf_start - 1;
593 			prev = &overlap->lf_next;
594 			lf = overlap->lf_next;
595 			continue;
596 
597 		case 5: /* overlap ends after lock */
598 			overlap->lf_start = unlock->lf_end + 1;
599 			break;
600 		}
601 		break;
602 	}
603 #ifdef LOCKF_DEBUG
604 	if (lockf_debug & 1)
605 		lf_printlist("lf_clearlock", unlock);
606 #endif /* LOCKF_DEBUG */
607 	return 0;
608 }
609 
610 /*
611  * Check whether there is a blocking lock,
612  * and if so return its process identifier.
613  */
614 static int
615 lf_getlock(struct lockf *lock, struct flock *fl)
616 {
617 	struct lockf *block;
618 
619 #ifdef LOCKF_DEBUG
620 	if (lockf_debug & 1)
621 		lf_print("lf_getlock", lock);
622 #endif /* LOCKF_DEBUG */
623 
624 	if ((block = lf_getblock(lock)) != NULL) {
625 		fl->l_type = block->lf_type;
626 		fl->l_whence = SEEK_SET;
627 		fl->l_start = block->lf_start;
628 		if (block->lf_end == -1)
629 			fl->l_len = 0;
630 		else
631 			fl->l_len = block->lf_end - block->lf_start + 1;
632 		if (block->lf_flags & F_POSIX)
633 			fl->l_pid = ((struct proc *)block->lf_id)->p_pid;
634 		else
635 			fl->l_pid = -1;
636 	} else {
637 		fl->l_type = F_UNLCK;
638 	}
639 	return 0;
640 }
641 
642 /*
643  * Walk the list of locks for an inode and
644  * return the first blocking lock.
645  */
646 static struct lockf *
647 lf_getblock(struct lockf *lock)
648 {
649 	struct lockf **prev, *overlap, *lf = *(lock->lf_head);
650 
651 	prev = lock->lf_head;
652 	while (lf_findoverlap(lf, lock, OTHERS, &prev, &overlap) != 0) {
653 		/*
654 		 * We've found an overlap, see if it blocks us
655 		 */
656 		if ((lock->lf_type == F_WRLCK || overlap->lf_type == F_WRLCK))
657 			return overlap;
658 		/*
659 		 * Nope, point to the next one on the list and
660 		 * see if it blocks us
661 		 */
662 		lf = overlap->lf_next;
663 	}
664 	return NOLOCKF;
665 }
666 
667 /*
668  * Walk the list of locks for an inode to
669  * find an overlapping lock (if any).
670  *
671  * NOTE: this returns only the FIRST overlapping lock.  There
672  *	 may be more than one.
673  */
674 static int
675 lf_findoverlap(struct lockf *lf, struct lockf *lock, int type,
676     struct lockf ***prev, struct lockf **overlap)
677 {
678 	off_t start, end;
679 
680 	*overlap = lf;
681 	if (lf == NOLOCKF)
682 		return 0;
683 #ifdef LOCKF_DEBUG
684 	if (lockf_debug & 2)
685 		lf_print("lf_findoverlap: looking for overlap in", lock);
686 #endif /* LOCKF_DEBUG */
687 	start = lock->lf_start;
688 	end = lock->lf_end;
689 	while (lf != NOLOCKF) {
690 		if (((type == SELF) && lf->lf_id != lock->lf_id) ||
691 		    ((type == OTHERS) && lf->lf_id == lock->lf_id)) {
692 			*prev = &lf->lf_next;
693 			*overlap = lf = lf->lf_next;
694 			continue;
695 		}
696 #ifdef LOCKF_DEBUG
697 		if (lockf_debug & 2)
698 			lf_print("\tchecking", lf);
699 #endif /* LOCKF_DEBUG */
700 		/*
701 		 * OK, check for overlap
702 		 *
703 		 * Six cases:
704 		 *	0) no overlap
705 		 *	1) overlap == lock
706 		 *	2) overlap contains lock
707 		 *	3) lock contains overlap
708 		 *	4) overlap starts before lock
709 		 *	5) overlap ends after lock
710 		 */
711 		if ((lf->lf_end != -1 && start > lf->lf_end) ||
712 		    (end != -1 && lf->lf_start > end)) {
713 			/* Case 0 */
714 #ifdef LOCKF_DEBUG
715 			if (lockf_debug & 2)
716 				printf("no overlap\n");
717 #endif /* LOCKF_DEBUG */
718 			if ((type & SELF) && end != -1 && lf->lf_start > end)
719 				return 0;
720 			*prev = &lf->lf_next;
721 			*overlap = lf = lf->lf_next;
722 			continue;
723 		}
724 		if ((lf->lf_start == start) && (lf->lf_end == end)) {
725 			/* Case 1 */
726 #ifdef LOCKF_DEBUG
727 			if (lockf_debug & 2)
728 				printf("overlap == lock\n");
729 #endif /* LOCKF_DEBUG */
730 			return 1;
731 		}
732 		if ((lf->lf_start <= start) &&
733 		    (end != -1) &&
734 		    ((lf->lf_end >= end) || (lf->lf_end == -1))) {
735 			/* Case 2 */
736 #ifdef LOCKF_DEBUG
737 			if (lockf_debug & 2)
738 				printf("overlap contains lock\n");
739 #endif /* LOCKF_DEBUG */
740 			return 2;
741 		}
742 		if (start <= lf->lf_start &&
743 		           (end == -1 ||
744 			   (lf->lf_end != -1 && end >= lf->lf_end))) {
745 			/* Case 3 */
746 #ifdef LOCKF_DEBUG
747 			if (lockf_debug & 2)
748 				printf("lock contains overlap\n");
749 #endif /* LOCKF_DEBUG */
750 			return 3;
751 		}
752 		if ((lf->lf_start < start) &&
753 			((lf->lf_end >= start) || (lf->lf_end == -1))) {
754 			/* Case 4 */
755 #ifdef LOCKF_DEBUG
756 			if (lockf_debug & 2)
757 				printf("overlap starts before lock\n");
758 #endif /* LOCKF_DEBUG */
759 			return 4;
760 		}
761 		if ((lf->lf_start > start) &&
762 			(end != -1) &&
763 			((lf->lf_end > end) || (lf->lf_end == -1))) {
764 			/* Case 5 */
765 #ifdef LOCKF_DEBUG
766 			if (lockf_debug & 2)
767 				printf("overlap ends after lock\n");
768 #endif /* LOCKF_DEBUG */
769 			return 5;
770 		}
771 		panic("lf_findoverlap: default");
772 	}
773 	return 0;
774 }
775 
776 /*
777  * Split a lock and a contained region into
778  * two or three locks as necessary.
779  */
780 static void
781 lf_split(struct lockf *lock1, struct lockf *lock2, struct lockf **sparelock)
782 {
783 	struct lockf *splitlock;
784 
785 #ifdef LOCKF_DEBUG
786 	if (lockf_debug & 2) {
787 		lf_print("lf_split", lock1);
788 		lf_print("splitting from", lock2);
789 	}
790 #endif /* LOCKF_DEBUG */
791 	/*
792 	 * Check to see if spliting into only two pieces.
793 	 */
794 	if (lock1->lf_start == lock2->lf_start) {
795 		lock1->lf_start = lock2->lf_end + 1;
796 		lock2->lf_next = lock1;
797 		return;
798 	}
799 	if (lock1->lf_end == lock2->lf_end) {
800 		lock1->lf_end = lock2->lf_start - 1;
801 		lock2->lf_next = lock1->lf_next;
802 		lock1->lf_next = lock2;
803 		return;
804 	}
805 	/*
806 	 * Make a new lock consisting of the last part of
807 	 * the encompassing lock
808 	 */
809 	splitlock = *sparelock;
810 	*sparelock = NULL;
811 	memcpy(splitlock, lock1, sizeof(*splitlock));
812 	splitlock->lf_start = lock2->lf_end + 1;
813 	TAILQ_INIT(&splitlock->lf_blkhd);
814 	lock1->lf_end = lock2->lf_start - 1;
815 	/*
816 	 * OK, now link it in
817 	 */
818 	splitlock->lf_next = lock1->lf_next;
819 	lock2->lf_next = splitlock;
820 	lock1->lf_next = lock2;
821 }
822 
823 /*
824  * Wakeup a blocklist
825  */
826 static void
827 lf_wakelock(struct lockf *listhead)
828 {
829 	struct lockf *wakelock;
830 
831 	while ((wakelock = TAILQ_FIRST(&listhead->lf_blkhd))) {
832 		KASSERT(wakelock->lf_next == listhead);
833 		TAILQ_REMOVE(&listhead->lf_blkhd, wakelock, lf_block);
834 		wakelock->lf_next = NOLOCKF;
835 #ifdef LOCKF_DEBUG
836 		if (lockf_debug & 2)
837 			lf_print("lf_wakelock: awakening", wakelock);
838 #endif
839 		wakeup(wakelock);
840 	}
841 }
842 
843 #ifdef LOCKF_DEBUG
844 /*
845  * Print out a lock.
846  */
847 static void
848 lf_print(char *tag, struct lockf *lock)
849 {
850 
851 	printf("%s: lock %p for ", tag, lock);
852 	if (lock->lf_flags & F_POSIX)
853 		printf("proc %d", ((struct proc *)lock->lf_id)->p_pid);
854 	else
855 		printf("file 0x%p", (struct file *)lock->lf_id);
856 	printf(" %s, start %qx, end %qx",
857 		lock->lf_type == F_RDLCK ? "shared" :
858 		lock->lf_type == F_WRLCK ? "exclusive" :
859 		lock->lf_type == F_UNLCK ? "unlock" :
860 		"unknown", lock->lf_start, lock->lf_end);
861 	if (TAILQ_FIRST(&lock->lf_blkhd))
862 		printf(" block %p\n", TAILQ_FIRST(&lock->lf_blkhd));
863 	else
864 		printf("\n");
865 }
866 
867 static void
868 lf_printlist(char *tag, struct lockf *lock)
869 {
870 	struct lockf *lf, *blk;
871 
872 	printf("%s: Lock list:\n", tag);
873 	for (lf = *lock->lf_head; lf; lf = lf->lf_next) {
874 		printf("\tlock %p for ", lf);
875 		if (lf->lf_flags & F_POSIX)
876 			printf("proc %d", ((struct proc *)lf->lf_id)->p_pid);
877 		else
878 			printf("file 0x%p", (struct file *)lf->lf_id);
879 		printf(", %s, start %qx, end %qx",
880 			lf->lf_type == F_RDLCK ? "shared" :
881 			lf->lf_type == F_WRLCK ? "exclusive" :
882 			lf->lf_type == F_UNLCK ? "unlock" :
883 			"unknown", lf->lf_start, lf->lf_end);
884 		TAILQ_FOREACH(blk, &lf->lf_blkhd, lf_block) {
885 			if (blk->lf_flags & F_POSIX)
886 				printf("proc %d",
887 				    ((struct proc *)blk->lf_id)->p_pid);
888 			else
889 				printf("file 0x%p", (struct file *)blk->lf_id);
890 			printf(", %s, start %qx, end %qx",
891 				blk->lf_type == F_RDLCK ? "shared" :
892 				blk->lf_type == F_WRLCK ? "exclusive" :
893 				blk->lf_type == F_UNLCK ? "unlock" :
894 				"unknown", blk->lf_start, blk->lf_end);
895 			if (TAILQ_FIRST(&blk->lf_blkhd))
896 				 panic("lf_printlist: bad list");
897 		}
898 		printf("\n");
899 	}
900 }
901 #endif /* LOCKF_DEBUG */
902