xref: /freebsd/sys/kern/kern_lockf.c (revision 4f29da19bd44f0e99f021510460a81bf754c21d2)
1 /*-
2  * Copyright (c) 1982, 1986, 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from software contributed to Berkeley by
6  * Scooter Morris at Genentech Inc.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 4. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  *	@(#)ufs_lockf.c	8.3 (Berkeley) 1/6/94
33  */
34 
35 #include <sys/cdefs.h>
36 __FBSDID("$FreeBSD$");
37 
38 #include "opt_debug_lockf.h"
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/kernel.h>
43 #include <sys/limits.h>
44 #include <sys/lock.h>
45 #include <sys/mount.h>
46 #include <sys/mutex.h>
47 #include <sys/proc.h>
48 #include <sys/unistd.h>
49 #include <sys/vnode.h>
50 #include <sys/malloc.h>
51 #include <sys/fcntl.h>
52 #include <sys/lockf.h>
53 
54 /*
55  * This variable controls the maximum number of processes that will
56  * be checked in doing deadlock detection.
57  */
58 static int maxlockdepth = MAXDEPTH;
59 
60 #ifdef LOCKF_DEBUG
61 #include <sys/sysctl.h>
62 
63 #include <ufs/ufs/quota.h>
64 #include <ufs/ufs/inode.h>
65 
66 
67 static int	lockf_debug = 0;
68 SYSCTL_INT(_debug, OID_AUTO, lockf_debug, CTLFLAG_RW, &lockf_debug, 0, "");
69 #endif
70 
71 MALLOC_DEFINE(M_LOCKF, "lockf", "Byte-range locking structures");
72 
73 #define NOLOCKF (struct lockf *)0
74 #define SELF	0x1
75 #define OTHERS	0x2
76 static int	 lf_clearlock(struct lockf *);
77 static int	 lf_findoverlap(struct lockf *,
78 	    struct lockf *, int, struct lockf ***, struct lockf **);
79 static struct lockf *
80 	 lf_getblock(struct lockf *);
81 static int	 lf_getlock(struct lockf *, struct flock *);
82 static int	 lf_setlock(struct lockf *);
83 static void	 lf_split(struct lockf *, struct lockf *);
84 static void	 lf_wakelock(struct lockf *);
85 #ifdef LOCKF_DEBUG
86 static void	 lf_print(char *, struct lockf *);
87 static void	 lf_printlist(char *, struct lockf *);
88 #endif
89 
90 /*
91  * Advisory record locking support
92  */
93 int
94 lf_advlock(ap, head, size)
95 	struct vop_advlock_args /* {
96 		struct vnode *a_vp;
97 		caddr_t  a_id;
98 		int  a_op;
99 		struct flock *a_fl;
100 		int  a_flags;
101 	} */ *ap;
102 	struct lockf **head;
103 	u_quad_t size;
104 {
105 	register struct flock *fl = ap->a_fl;
106 	register struct lockf *lock;
107 	off_t start, end, oadd;
108 	int error;
109 
110 	mtx_lock(&Giant);
111 	/*
112 	 * Convert the flock structure into a start and end.
113 	 */
114 	switch (fl->l_whence) {
115 
116 	case SEEK_SET:
117 	case SEEK_CUR:
118 		/*
119 		 * Caller is responsible for adding any necessary offset
120 		 * when SEEK_CUR is used.
121 		 */
122 		start = fl->l_start;
123 		break;
124 
125 	case SEEK_END:
126 		if (size > OFF_MAX ||
127 		    (fl->l_start > 0 && size > OFF_MAX - fl->l_start)) {
128 			error = EOVERFLOW;
129 			goto out;
130 		}
131 		start = size + fl->l_start;
132 		break;
133 
134 	default:
135 		error = EINVAL;
136 		goto out;
137 	}
138 	if (start < 0) {
139 		error = EINVAL;
140 		goto out;
141 	}
142 	if (fl->l_len < 0) {
143 		if (start == 0) {
144 			error = EINVAL;
145 			goto out;
146 		}
147 		end = start - 1;
148 		start += fl->l_len;
149 		if (start < 0) {
150 			error = EINVAL;
151 			goto out;
152 		}
153 	} else if (fl->l_len == 0)
154 		end = -1;
155 	else {
156 		oadd = fl->l_len - 1;
157 		if (oadd > OFF_MAX - start) {
158 			error = EOVERFLOW;
159 			goto out;
160 		}
161 		end = start + oadd;
162 	}
163 	/*
164 	 * Avoid the common case of unlocking when inode has no locks.
165 	 */
166 	if (*head == (struct lockf *)0) {
167 		if (ap->a_op != F_SETLK) {
168 			fl->l_type = F_UNLCK;
169 			error = 0;
170 			goto out;
171 		}
172 	}
173 	/*
174 	 * Create the lockf structure
175 	 */
176 	MALLOC(lock, struct lockf *, sizeof *lock, M_LOCKF, M_WAITOK);
177 	lock->lf_start = start;
178 	lock->lf_end = end;
179 	lock->lf_id = ap->a_id;
180 	/*
181 	 * XXX The problem is that VTOI is ufs specific, so it will
182 	 * break LOCKF_DEBUG for all other FS's other than UFS because
183 	 * it casts the vnode->data ptr to struct inode *.
184 	 */
185 /*	lock->lf_inode = VTOI(ap->a_vp); */
186 	lock->lf_inode = (struct inode *)0;
187 	lock->lf_type = fl->l_type;
188 	lock->lf_head = head;
189 	lock->lf_next = (struct lockf *)0;
190 	TAILQ_INIT(&lock->lf_blkhd);
191 	lock->lf_flags = ap->a_flags;
192 	/*
193 	 * Do the requested operation.
194 	 */
195 	switch(ap->a_op) {
196 	case F_SETLK:
197 		error = lf_setlock(lock);
198 		goto out;
199 
200 	case F_UNLCK:
201 		error = lf_clearlock(lock);
202 		FREE(lock, M_LOCKF);
203 		goto out;
204 
205 	case F_GETLK:
206 		error = lf_getlock(lock, fl);
207 		FREE(lock, M_LOCKF);
208 		goto out;
209 
210 	default:
211 		free(lock, M_LOCKF);
212 		error = EINVAL;
213 		goto out;
214 	}
215 	/* NOTREACHED */
216 out:
217 	mtx_unlock(&Giant);
218 	return (error);
219 }
220 
221 /*
222  * Set a byte-range lock.
223  */
224 static int
225 lf_setlock(lock)
226 	register struct lockf *lock;
227 {
228 	register struct lockf *block;
229 	struct lockf **head = lock->lf_head;
230 	struct lockf **prev, *overlap, *ltmp;
231 	static char lockstr[] = "lockf";
232 	int ovcase, priority, needtolink, error;
233 
234 #ifdef LOCKF_DEBUG
235 	if (lockf_debug & 1)
236 		lf_print("lf_setlock", lock);
237 #endif /* LOCKF_DEBUG */
238 
239 	/*
240 	 * Set the priority
241 	 */
242 	priority = PLOCK;
243 	if (lock->lf_type == F_WRLCK)
244 		priority += 4;
245 	priority |= PCATCH;
246 	/*
247 	 * Scan lock list for this file looking for locks that would block us.
248 	 */
249 	while ((block = lf_getblock(lock))) {
250 		/*
251 		 * Free the structure and return if nonblocking.
252 		 */
253 		if ((lock->lf_flags & F_WAIT) == 0) {
254 			FREE(lock, M_LOCKF);
255 			return (EAGAIN);
256 		}
257 		/*
258 		 * We are blocked. Since flock style locks cover
259 		 * the whole file, there is no chance for deadlock.
260 		 * For byte-range locks we must check for deadlock.
261 		 *
262 		 * Deadlock detection is done by looking through the
263 		 * wait channels to see if there are any cycles that
264 		 * involve us. MAXDEPTH is set just to make sure we
265 		 * do not go off into neverland.
266 		 */
267 		if ((lock->lf_flags & F_POSIX) &&
268 		    (block->lf_flags & F_POSIX)) {
269 			register struct proc *wproc;
270 			struct thread *td;
271 			register struct lockf *waitblock;
272 			int i = 0;
273 
274 			/* The block is waiting on something */
275 			/* XXXKSE this is not complete under threads */
276 			wproc = (struct proc *)block->lf_id;
277 			mtx_lock_spin(&sched_lock);
278 			FOREACH_THREAD_IN_PROC(wproc, td) {
279 				while (td->td_wchan &&
280 				    (td->td_wmesg == lockstr) &&
281 				    (i++ < maxlockdepth)) {
282 					waitblock = (struct lockf *)td->td_wchan;
283 					/* Get the owner of the blocking lock */
284 					waitblock = waitblock->lf_next;
285 					if ((waitblock->lf_flags & F_POSIX) == 0)
286 						break;
287 					wproc = (struct proc *)waitblock->lf_id;
288 					if (wproc == (struct proc *)lock->lf_id) {
289 						mtx_unlock_spin(&sched_lock);
290 						free(lock, M_LOCKF);
291 						return (EDEADLK);
292 					}
293 				}
294 			}
295 			mtx_unlock_spin(&sched_lock);
296 		}
297 		/*
298 		 * For flock type locks, we must first remove
299 		 * any shared locks that we hold before we sleep
300 		 * waiting for an exclusive lock.
301 		 */
302 		if ((lock->lf_flags & F_FLOCK) &&
303 		    lock->lf_type == F_WRLCK) {
304 			lock->lf_type = F_UNLCK;
305 			(void) lf_clearlock(lock);
306 			lock->lf_type = F_WRLCK;
307 		}
308 		/*
309 		 * Add our lock to the blocked list and sleep until we're free.
310 		 * Remember who blocked us (for deadlock detection).
311 		 */
312 		lock->lf_next = block;
313 		TAILQ_INSERT_TAIL(&block->lf_blkhd, lock, lf_block);
314 #ifdef LOCKF_DEBUG
315 		if (lockf_debug & 1) {
316 			lf_print("lf_setlock: blocking on", block);
317 			lf_printlist("lf_setlock", block);
318 		}
319 #endif /* LOCKF_DEBUG */
320 		error = tsleep(lock, priority, lockstr, 0);
321 		/*
322 		 * We may have been awakened by a signal and/or by a
323 		 * debugger continuing us (in which cases we must remove
324 		 * ourselves from the blocked list) and/or by another
325 		 * process releasing a lock (in which case we have
326 		 * already been removed from the blocked list and our
327 		 * lf_next field set to NOLOCKF).
328 		 */
329 		if (lock->lf_next) {
330 			TAILQ_REMOVE(&lock->lf_next->lf_blkhd, lock, lf_block);
331 			lock->lf_next = NOLOCKF;
332 		}
333 		if (error) {
334 			free(lock, M_LOCKF);
335 			return (error);
336 		}
337 	}
338 	/*
339 	 * No blocks!!  Add the lock.  Note that we will
340 	 * downgrade or upgrade any overlapping locks this
341 	 * process already owns.
342 	 *
343 	 * Skip over locks owned by other processes.
344 	 * Handle any locks that overlap and are owned by ourselves.
345 	 */
346 	prev = head;
347 	block = *head;
348 	needtolink = 1;
349 	for (;;) {
350 		ovcase = lf_findoverlap(block, lock, SELF, &prev, &overlap);
351 		if (ovcase)
352 			block = overlap->lf_next;
353 		/*
354 		 * Six cases:
355 		 *	0) no overlap
356 		 *	1) overlap == lock
357 		 *	2) overlap contains lock
358 		 *	3) lock contains overlap
359 		 *	4) overlap starts before lock
360 		 *	5) overlap ends after lock
361 		 */
362 		switch (ovcase) {
363 		case 0: /* no overlap */
364 			if (needtolink) {
365 				*prev = lock;
366 				lock->lf_next = overlap;
367 			}
368 			break;
369 
370 		case 1: /* overlap == lock */
371 			/*
372 			 * If downgrading lock, others may be
373 			 * able to acquire it.
374 			 */
375 			if (lock->lf_type == F_RDLCK &&
376 			    overlap->lf_type == F_WRLCK)
377 				lf_wakelock(overlap);
378 			overlap->lf_type = lock->lf_type;
379 			FREE(lock, M_LOCKF);
380 			lock = overlap; /* for debug output below */
381 			break;
382 
383 		case 2: /* overlap contains lock */
384 			/*
385 			 * Check for common starting point and different types.
386 			 */
387 			if (overlap->lf_type == lock->lf_type) {
388 				free(lock, M_LOCKF);
389 				lock = overlap; /* for debug output below */
390 				break;
391 			}
392 			if (overlap->lf_start == lock->lf_start) {
393 				*prev = lock;
394 				lock->lf_next = overlap;
395 				overlap->lf_start = lock->lf_end + 1;
396 			} else
397 				lf_split(overlap, lock);
398 			lf_wakelock(overlap);
399 			break;
400 
401 		case 3: /* lock contains overlap */
402 			/*
403 			 * If downgrading lock, others may be able to
404 			 * acquire it, otherwise take the list.
405 			 */
406 			if (lock->lf_type == F_RDLCK &&
407 			    overlap->lf_type == F_WRLCK) {
408 				lf_wakelock(overlap);
409 			} else {
410 				while (!TAILQ_EMPTY(&overlap->lf_blkhd)) {
411 					ltmp = TAILQ_FIRST(&overlap->lf_blkhd);
412 					TAILQ_REMOVE(&overlap->lf_blkhd, ltmp,
413 					    lf_block);
414 					TAILQ_INSERT_TAIL(&lock->lf_blkhd,
415 					    ltmp, lf_block);
416 					ltmp->lf_next = lock;
417 				}
418 			}
419 			/*
420 			 * Add the new lock if necessary and delete the overlap.
421 			 */
422 			if (needtolink) {
423 				*prev = lock;
424 				lock->lf_next = overlap->lf_next;
425 				prev = &lock->lf_next;
426 				needtolink = 0;
427 			} else
428 				*prev = overlap->lf_next;
429 			free(overlap, M_LOCKF);
430 			continue;
431 
432 		case 4: /* overlap starts before lock */
433 			/*
434 			 * Add lock after overlap on the list.
435 			 */
436 			lock->lf_next = overlap->lf_next;
437 			overlap->lf_next = lock;
438 			overlap->lf_end = lock->lf_start - 1;
439 			prev = &lock->lf_next;
440 			lf_wakelock(overlap);
441 			needtolink = 0;
442 			continue;
443 
444 		case 5: /* overlap ends after lock */
445 			/*
446 			 * Add the new lock before overlap.
447 			 */
448 			if (needtolink) {
449 				*prev = lock;
450 				lock->lf_next = overlap;
451 			}
452 			overlap->lf_start = lock->lf_end + 1;
453 			lf_wakelock(overlap);
454 			break;
455 		}
456 		break;
457 	}
458 #ifdef LOCKF_DEBUG
459 	if (lockf_debug & 1) {
460 		lf_print("lf_setlock: got the lock", lock);
461 		lf_printlist("lf_setlock", lock);
462 	}
463 #endif /* LOCKF_DEBUG */
464 	return (0);
465 }
466 
467 /*
468  * Remove a byte-range lock on an inode.
469  *
470  * Generally, find the lock (or an overlap to that lock)
471  * and remove it (or shrink it), then wakeup anyone we can.
472  */
473 static int
474 lf_clearlock(unlock)
475 	register struct lockf *unlock;
476 {
477 	struct lockf **head = unlock->lf_head;
478 	register struct lockf *lf = *head;
479 	struct lockf *overlap, **prev;
480 	int ovcase;
481 
482 	if (lf == NOLOCKF)
483 		return (0);
484 #ifdef LOCKF_DEBUG
485 	if (unlock->lf_type != F_UNLCK)
486 		panic("lf_clearlock: bad type");
487 	if (lockf_debug & 1)
488 		lf_print("lf_clearlock", unlock);
489 #endif /* LOCKF_DEBUG */
490 	prev = head;
491 	while ((ovcase = lf_findoverlap(lf, unlock, SELF, &prev, &overlap))) {
492 		/*
493 		 * Wakeup the list of locks to be retried.
494 		 */
495 		lf_wakelock(overlap);
496 
497 		switch (ovcase) {
498 
499 		case 1: /* overlap == lock */
500 			*prev = overlap->lf_next;
501 			FREE(overlap, M_LOCKF);
502 			break;
503 
504 		case 2: /* overlap contains lock: split it */
505 			if (overlap->lf_start == unlock->lf_start) {
506 				overlap->lf_start = unlock->lf_end + 1;
507 				break;
508 			}
509 			lf_split(overlap, unlock);
510 			overlap->lf_next = unlock->lf_next;
511 			break;
512 
513 		case 3: /* lock contains overlap */
514 			*prev = overlap->lf_next;
515 			lf = overlap->lf_next;
516 			free(overlap, M_LOCKF);
517 			continue;
518 
519 		case 4: /* overlap starts before lock */
520 			overlap->lf_end = unlock->lf_start - 1;
521 			prev = &overlap->lf_next;
522 			lf = overlap->lf_next;
523 			continue;
524 
525 		case 5: /* overlap ends after lock */
526 			overlap->lf_start = unlock->lf_end + 1;
527 			break;
528 		}
529 		break;
530 	}
531 #ifdef LOCKF_DEBUG
532 	if (lockf_debug & 1)
533 		lf_printlist("lf_clearlock", unlock);
534 #endif /* LOCKF_DEBUG */
535 	return (0);
536 }
537 
538 /*
539  * Check whether there is a blocking lock,
540  * and if so return its process identifier.
541  */
542 static int
543 lf_getlock(lock, fl)
544 	register struct lockf *lock;
545 	register struct flock *fl;
546 {
547 	register struct lockf *block;
548 
549 #ifdef LOCKF_DEBUG
550 	if (lockf_debug & 1)
551 		lf_print("lf_getlock", lock);
552 #endif /* LOCKF_DEBUG */
553 
554 	if ((block = lf_getblock(lock))) {
555 		fl->l_type = block->lf_type;
556 		fl->l_whence = SEEK_SET;
557 		fl->l_start = block->lf_start;
558 		if (block->lf_end == -1)
559 			fl->l_len = 0;
560 		else
561 			fl->l_len = block->lf_end - block->lf_start + 1;
562 		if (block->lf_flags & F_POSIX)
563 			fl->l_pid = ((struct proc *)(block->lf_id))->p_pid;
564 		else
565 			fl->l_pid = -1;
566 	} else {
567 		fl->l_type = F_UNLCK;
568 	}
569 	return (0);
570 }
571 
572 /*
573  * Walk the list of locks for an inode and
574  * return the first blocking lock.
575  */
576 static struct lockf *
577 lf_getblock(lock)
578 	register struct lockf *lock;
579 {
580 	struct lockf **prev, *overlap, *lf = *(lock->lf_head);
581 	int ovcase;
582 
583 	prev = lock->lf_head;
584 	while ((ovcase = lf_findoverlap(lf, lock, OTHERS, &prev, &overlap))) {
585 		/*
586 		 * We've found an overlap, see if it blocks us
587 		 */
588 		if ((lock->lf_type == F_WRLCK || overlap->lf_type == F_WRLCK))
589 			return (overlap);
590 		/*
591 		 * Nope, point to the next one on the list and
592 		 * see if it blocks us
593 		 */
594 		lf = overlap->lf_next;
595 	}
596 	return (NOLOCKF);
597 }
598 
599 /*
600  * Walk the list of locks for an inode to
601  * find an overlapping lock (if any).
602  *
603  * NOTE: this returns only the FIRST overlapping lock.  There
604  *	 may be more than one.
605  */
606 static int
607 lf_findoverlap(lf, lock, type, prev, overlap)
608 	register struct lockf *lf;
609 	struct lockf *lock;
610 	int type;
611 	struct lockf ***prev;
612 	struct lockf **overlap;
613 {
614 	off_t start, end;
615 
616 	*overlap = lf;
617 	if (lf == NOLOCKF)
618 		return (0);
619 #ifdef LOCKF_DEBUG
620 	if (lockf_debug & 2)
621 		lf_print("lf_findoverlap: looking for overlap in", lock);
622 #endif /* LOCKF_DEBUG */
623 	start = lock->lf_start;
624 	end = lock->lf_end;
625 	while (lf != NOLOCKF) {
626 		if (((type & SELF) && lf->lf_id != lock->lf_id) ||
627 		    ((type & OTHERS) && lf->lf_id == lock->lf_id)) {
628 			*prev = &lf->lf_next;
629 			*overlap = lf = lf->lf_next;
630 			continue;
631 		}
632 #ifdef LOCKF_DEBUG
633 		if (lockf_debug & 2)
634 			lf_print("\tchecking", lf);
635 #endif /* LOCKF_DEBUG */
636 		/*
637 		 * OK, check for overlap
638 		 *
639 		 * Six cases:
640 		 *	0) no overlap
641 		 *	1) overlap == lock
642 		 *	2) overlap contains lock
643 		 *	3) lock contains overlap
644 		 *	4) overlap starts before lock
645 		 *	5) overlap ends after lock
646 		 */
647 		if ((lf->lf_end != -1 && start > lf->lf_end) ||
648 		    (end != -1 && lf->lf_start > end)) {
649 			/* Case 0 */
650 #ifdef LOCKF_DEBUG
651 			if (lockf_debug & 2)
652 				printf("no overlap\n");
653 #endif /* LOCKF_DEBUG */
654 			if ((type & SELF) && end != -1 && lf->lf_start > end)
655 				return (0);
656 			*prev = &lf->lf_next;
657 			*overlap = lf = lf->lf_next;
658 			continue;
659 		}
660 		if ((lf->lf_start == start) && (lf->lf_end == end)) {
661 			/* Case 1 */
662 #ifdef LOCKF_DEBUG
663 			if (lockf_debug & 2)
664 				printf("overlap == lock\n");
665 #endif /* LOCKF_DEBUG */
666 			return (1);
667 		}
668 		if ((lf->lf_start <= start) &&
669 		    (end != -1) &&
670 		    ((lf->lf_end >= end) || (lf->lf_end == -1))) {
671 			/* Case 2 */
672 #ifdef LOCKF_DEBUG
673 			if (lockf_debug & 2)
674 				printf("overlap contains lock\n");
675 #endif /* LOCKF_DEBUG */
676 			return (2);
677 		}
678 		if (start <= lf->lf_start &&
679 		           (end == -1 ||
680 			   (lf->lf_end != -1 && end >= lf->lf_end))) {
681 			/* Case 3 */
682 #ifdef LOCKF_DEBUG
683 			if (lockf_debug & 2)
684 				printf("lock contains overlap\n");
685 #endif /* LOCKF_DEBUG */
686 			return (3);
687 		}
688 		if ((lf->lf_start < start) &&
689 			((lf->lf_end >= start) || (lf->lf_end == -1))) {
690 			/* Case 4 */
691 #ifdef LOCKF_DEBUG
692 			if (lockf_debug & 2)
693 				printf("overlap starts before lock\n");
694 #endif /* LOCKF_DEBUG */
695 			return (4);
696 		}
697 		if ((lf->lf_start > start) &&
698 			(end != -1) &&
699 			((lf->lf_end > end) || (lf->lf_end == -1))) {
700 			/* Case 5 */
701 #ifdef LOCKF_DEBUG
702 			if (lockf_debug & 2)
703 				printf("overlap ends after lock\n");
704 #endif /* LOCKF_DEBUG */
705 			return (5);
706 		}
707 		panic("lf_findoverlap: default");
708 	}
709 	return (0);
710 }
711 
712 /*
713  * Split a lock and a contained region into
714  * two or three locks as necessary.
715  */
716 static void
717 lf_split(lock1, lock2)
718 	register struct lockf *lock1;
719 	register struct lockf *lock2;
720 {
721 	register struct lockf *splitlock;
722 
723 #ifdef LOCKF_DEBUG
724 	if (lockf_debug & 2) {
725 		lf_print("lf_split", lock1);
726 		lf_print("splitting from", lock2);
727 	}
728 #endif /* LOCKF_DEBUG */
729 	/*
730 	 * Check to see if spliting into only two pieces.
731 	 */
732 	if (lock1->lf_start == lock2->lf_start) {
733 		lock1->lf_start = lock2->lf_end + 1;
734 		lock2->lf_next = lock1;
735 		return;
736 	}
737 	if (lock1->lf_end == lock2->lf_end) {
738 		lock1->lf_end = lock2->lf_start - 1;
739 		lock2->lf_next = lock1->lf_next;
740 		lock1->lf_next = lock2;
741 		return;
742 	}
743 	/*
744 	 * Make a new lock consisting of the last part of
745 	 * the encompassing lock
746 	 */
747 	MALLOC(splitlock, struct lockf *, sizeof *splitlock, M_LOCKF, M_WAITOK);
748 	bcopy(lock1, splitlock, sizeof *splitlock);
749 	splitlock->lf_start = lock2->lf_end + 1;
750 	TAILQ_INIT(&splitlock->lf_blkhd);
751 	lock1->lf_end = lock2->lf_start - 1;
752 	/*
753 	 * OK, now link it in
754 	 */
755 	splitlock->lf_next = lock1->lf_next;
756 	lock2->lf_next = splitlock;
757 	lock1->lf_next = lock2;
758 }
759 
760 /*
761  * Wakeup a blocklist
762  */
763 static void
764 lf_wakelock(listhead)
765 	struct lockf *listhead;
766 {
767 	register struct lockf *wakelock;
768 
769 	while (!TAILQ_EMPTY(&listhead->lf_blkhd)) {
770 		wakelock = TAILQ_FIRST(&listhead->lf_blkhd);
771 		TAILQ_REMOVE(&listhead->lf_blkhd, wakelock, lf_block);
772 		wakelock->lf_next = NOLOCKF;
773 #ifdef LOCKF_DEBUG
774 		if (lockf_debug & 2)
775 			lf_print("lf_wakelock: awakening", wakelock);
776 #endif /* LOCKF_DEBUG */
777 		wakeup(wakelock);
778 	}
779 }
780 
781 #ifdef LOCKF_DEBUG
782 /*
783  * Print out a lock.
784  */
785 static void
786 lf_print(tag, lock)
787 	char *tag;
788 	register struct lockf *lock;
789 {
790 
791 	printf("%s: lock %p for ", tag, (void *)lock);
792 	if (lock->lf_flags & F_POSIX)
793 		printf("proc %ld", (long)((struct proc *)lock->lf_id)->p_pid);
794 	else
795 		printf("id %p", (void *)lock->lf_id);
796 	if (lock->lf_inode != (struct inode *)0)
797 		printf(" in ino %ju on dev <%s>, %s, start %jd, end %jd",
798 		    (uintmax_t)lock->lf_inode->i_number,
799 		    devtoname(lock->lf_inode->i_dev),
800 		    lock->lf_type == F_RDLCK ? "shared" :
801 		    lock->lf_type == F_WRLCK ? "exclusive" :
802 		    lock->lf_type == F_UNLCK ? "unlock" : "unknown",
803 		    (intmax_t)lock->lf_start, (intmax_t)lock->lf_end);
804 	else
805 		printf(" %s, start %jd, end %jd",
806 		    lock->lf_type == F_RDLCK ? "shared" :
807 		    lock->lf_type == F_WRLCK ? "exclusive" :
808 		    lock->lf_type == F_UNLCK ? "unlock" : "unknown",
809 		    (intmax_t)lock->lf_start, (intmax_t)lock->lf_end);
810 	if (!TAILQ_EMPTY(&lock->lf_blkhd))
811 		printf(" block %p\n", (void *)TAILQ_FIRST(&lock->lf_blkhd));
812 	else
813 		printf("\n");
814 }
815 
816 static void
817 lf_printlist(tag, lock)
818 	char *tag;
819 	struct lockf *lock;
820 {
821 	register struct lockf *lf, *blk;
822 
823 	if (lock->lf_inode == (struct inode *)0)
824 		return;
825 
826 	printf("%s: Lock list for ino %ju on dev <%s>:\n",
827 	    tag, (uintmax_t)lock->lf_inode->i_number,
828 	    devtoname(lock->lf_inode->i_dev));
829 	for (lf = lock->lf_inode->i_lockf; lf; lf = lf->lf_next) {
830 		printf("\tlock %p for ",(void *)lf);
831 		if (lf->lf_flags & F_POSIX)
832 			printf("proc %ld",
833 			    (long)((struct proc *)lf->lf_id)->p_pid);
834 		else
835 			printf("id %p", (void *)lf->lf_id);
836 		printf(", %s, start %jd, end %jd",
837 		    lf->lf_type == F_RDLCK ? "shared" :
838 		    lf->lf_type == F_WRLCK ? "exclusive" :
839 		    lf->lf_type == F_UNLCK ? "unlock" :
840 		    "unknown", (intmax_t)lf->lf_start, (intmax_t)lf->lf_end);
841 		TAILQ_FOREACH(blk, &lf->lf_blkhd, lf_block) {
842 			printf("\n\t\tlock request %p for ", (void *)blk);
843 			if (blk->lf_flags & F_POSIX)
844 				printf("proc %ld",
845 				    (long)((struct proc *)blk->lf_id)->p_pid);
846 			else
847 				printf("id %p", (void *)blk->lf_id);
848 			printf(", %s, start %jd, end %jd",
849 			    blk->lf_type == F_RDLCK ? "shared" :
850 			    blk->lf_type == F_WRLCK ? "exclusive" :
851 			    blk->lf_type == F_UNLCK ? "unlock" :
852 			    "unknown", (intmax_t)blk->lf_start,
853 			    (intmax_t)blk->lf_end);
854 			if (!TAILQ_EMPTY(&blk->lf_blkhd))
855 				panic("lf_printlist: bad list");
856 		}
857 		printf("\n");
858 	}
859 }
860 #endif /* LOCKF_DEBUG */
861