xref: /freebsd/sys/kern/kern_lockf.c (revision 7660b554bc59a07be0431c17e0e33815818baa69)
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  * 3. All advertising materials mentioning features or use of this software
17  *    must display the following acknowledgement:
18  *	This product includes software developed by the University of
19  *	California, Berkeley and its contributors.
20  * 4. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)ufs_lockf.c	8.3 (Berkeley) 1/6/94
37  */
38 
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41 
42 #include "opt_debug_lockf.h"
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/kernel.h>
47 #include <sys/limits.h>
48 #include <sys/lock.h>
49 #include <sys/mount.h>
50 #include <sys/mutex.h>
51 #include <sys/proc.h>
52 #include <sys/unistd.h>
53 #include <sys/vnode.h>
54 #include <sys/malloc.h>
55 #include <sys/fcntl.h>
56 #include <sys/lockf.h>
57 
58 /*
59  * This variable controls the maximum number of processes that will
60  * be checked in doing deadlock detection.
61  */
62 static int maxlockdepth = MAXDEPTH;
63 
64 #ifdef LOCKF_DEBUG
65 #include <sys/sysctl.h>
66 
67 #include <ufs/ufs/quota.h>
68 #include <ufs/ufs/inode.h>
69 
70 
71 static int	lockf_debug = 0;
72 SYSCTL_INT(_debug, OID_AUTO, lockf_debug, CTLFLAG_RW, &lockf_debug, 0, "");
73 #endif
74 
75 MALLOC_DEFINE(M_LOCKF, "lockf", "Byte-range locking structures");
76 
77 #define NOLOCKF (struct lockf *)0
78 #define SELF	0x1
79 #define OTHERS	0x2
80 static int	 lf_clearlock(struct lockf *);
81 static int	 lf_findoverlap(struct lockf *,
82 	    struct lockf *, int, struct lockf ***, struct lockf **);
83 static struct lockf *
84 	 lf_getblock(struct lockf *);
85 static int	 lf_getlock(struct lockf *, struct flock *);
86 static int	 lf_setlock(struct lockf *);
87 static void	 lf_split(struct lockf *, struct lockf *);
88 static void	 lf_wakelock(struct lockf *);
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 	/*
111 	 * Convert the flock structure into a start and end.
112 	 */
113 	switch (fl->l_whence) {
114 
115 	case SEEK_SET:
116 	case SEEK_CUR:
117 		/*
118 		 * Caller is responsible for adding any necessary offset
119 		 * when SEEK_CUR is used.
120 		 */
121 		start = fl->l_start;
122 		break;
123 
124 	case SEEK_END:
125 		if (size > OFF_MAX ||
126 		    (fl->l_start > 0 && size > OFF_MAX - fl->l_start))
127 			return (EOVERFLOW);
128 		start = size + fl->l_start;
129 		break;
130 
131 	default:
132 		return (EINVAL);
133 	}
134 	if (start < 0)
135 		return (EINVAL);
136 	if (fl->l_len < 0) {
137 		if (start == 0)
138 			return (EINVAL);
139 		end = start - 1;
140 		start += fl->l_len;
141 		if (start < 0)
142 			return (EINVAL);
143 	} else if (fl->l_len == 0)
144 		end = -1;
145 	else {
146 		oadd = fl->l_len - 1;
147 		if (oadd > OFF_MAX - start)
148 			return (EOVERFLOW);
149 		end = start + oadd;
150 	}
151 	/*
152 	 * Avoid the common case of unlocking when inode has no locks.
153 	 */
154 	if (*head == (struct lockf *)0) {
155 		if (ap->a_op != F_SETLK) {
156 			fl->l_type = F_UNLCK;
157 			return (0);
158 		}
159 	}
160 	/*
161 	 * Create the lockf structure
162 	 */
163 	MALLOC(lock, struct lockf *, sizeof *lock, M_LOCKF, M_WAITOK);
164 	lock->lf_start = start;
165 	lock->lf_end = end;
166 	lock->lf_id = ap->a_id;
167 	/*
168 	 * XXX The problem is that VTOI is ufs specific, so it will
169 	 * break LOCKF_DEBUG for all other FS's other than UFS because
170 	 * it casts the vnode->data ptr to struct inode *.
171 	 */
172 /*	lock->lf_inode = VTOI(ap->a_vp); */
173 	lock->lf_inode = (struct inode *)0;
174 	lock->lf_type = fl->l_type;
175 	lock->lf_head = head;
176 	lock->lf_next = (struct lockf *)0;
177 	TAILQ_INIT(&lock->lf_blkhd);
178 	lock->lf_flags = ap->a_flags;
179 	/*
180 	 * Do the requested operation.
181 	 */
182 	switch(ap->a_op) {
183 	case F_SETLK:
184 		return (lf_setlock(lock));
185 
186 	case F_UNLCK:
187 		error = lf_clearlock(lock);
188 		FREE(lock, M_LOCKF);
189 		return (error);
190 
191 	case F_GETLK:
192 		error = lf_getlock(lock, fl);
193 		FREE(lock, M_LOCKF);
194 		return (error);
195 
196 	default:
197 		free(lock, M_LOCKF);
198 		return (EINVAL);
199 	}
200 	/* NOTREACHED */
201 }
202 
203 /*
204  * Set a byte-range lock.
205  */
206 static int
207 lf_setlock(lock)
208 	register struct lockf *lock;
209 {
210 	register struct lockf *block;
211 	struct lockf **head = lock->lf_head;
212 	struct lockf **prev, *overlap, *ltmp;
213 	static char lockstr[] = "lockf";
214 	int ovcase, priority, needtolink, error;
215 
216 #ifdef LOCKF_DEBUG
217 	if (lockf_debug & 1)
218 		lf_print("lf_setlock", lock);
219 #endif /* LOCKF_DEBUG */
220 
221 	/*
222 	 * Set the priority
223 	 */
224 	priority = PLOCK;
225 	if (lock->lf_type == F_WRLCK)
226 		priority += 4;
227 	priority |= PCATCH;
228 	/*
229 	 * Scan lock list for this file looking for locks that would block us.
230 	 */
231 	while ((block = lf_getblock(lock))) {
232 		/*
233 		 * Free the structure and return if nonblocking.
234 		 */
235 		if ((lock->lf_flags & F_WAIT) == 0) {
236 			FREE(lock, M_LOCKF);
237 			return (EAGAIN);
238 		}
239 		/*
240 		 * We are blocked. Since flock style locks cover
241 		 * the whole file, there is no chance for deadlock.
242 		 * For byte-range locks we must check for deadlock.
243 		 *
244 		 * Deadlock detection is done by looking through the
245 		 * wait channels to see if there are any cycles that
246 		 * involve us. MAXDEPTH is set just to make sure we
247 		 * do not go off into neverland.
248 		 */
249 		if ((lock->lf_flags & F_POSIX) &&
250 		    (block->lf_flags & F_POSIX)) {
251 			register struct proc *wproc;
252 			struct thread *td;
253 			register struct lockf *waitblock;
254 			int i = 0;
255 
256 			/* The block is waiting on something */
257 			/* XXXKSE this is not complete under threads */
258 			wproc = (struct proc *)block->lf_id;
259 			mtx_lock_spin(&sched_lock);
260 			FOREACH_THREAD_IN_PROC(wproc, td) {
261 				while (td->td_wchan &&
262 				    (td->td_wmesg == lockstr) &&
263 				    (i++ < maxlockdepth)) {
264 					waitblock = (struct lockf *)td->td_wchan;
265 					/* Get the owner of the blocking lock */
266 					waitblock = waitblock->lf_next;
267 					if ((waitblock->lf_flags & F_POSIX) == 0)
268 						break;
269 					wproc = (struct proc *)waitblock->lf_id;
270 					if (wproc == (struct proc *)lock->lf_id) {
271 						mtx_unlock_spin(&sched_lock);
272 						free(lock, M_LOCKF);
273 						return (EDEADLK);
274 					}
275 				}
276 			}
277 			mtx_unlock_spin(&sched_lock);
278 		}
279 		/*
280 		 * For flock type locks, we must first remove
281 		 * any shared locks that we hold before we sleep
282 		 * waiting for an exclusive lock.
283 		 */
284 		if ((lock->lf_flags & F_FLOCK) &&
285 		    lock->lf_type == F_WRLCK) {
286 			lock->lf_type = F_UNLCK;
287 			(void) lf_clearlock(lock);
288 			lock->lf_type = F_WRLCK;
289 		}
290 		/*
291 		 * Add our lock to the blocked list and sleep until we're free.
292 		 * Remember who blocked us (for deadlock detection).
293 		 */
294 		lock->lf_next = block;
295 		TAILQ_INSERT_TAIL(&block->lf_blkhd, lock, lf_block);
296 #ifdef LOCKF_DEBUG
297 		if (lockf_debug & 1) {
298 			lf_print("lf_setlock: blocking on", block);
299 			lf_printlist("lf_setlock", block);
300 		}
301 #endif /* LOCKF_DEBUG */
302 		error = tsleep(lock, priority, lockstr, 0);
303 		/*
304 		 * We may have been awakened by a signal and/or by a
305 		 * debugger continuing us (in which cases we must remove
306 		 * ourselves from the blocked list) and/or by another
307 		 * process releasing a lock (in which case we have
308 		 * already been removed from the blocked list and our
309 		 * lf_next field set to NOLOCKF).
310 		 */
311 		if (lock->lf_next) {
312 			TAILQ_REMOVE(&lock->lf_next->lf_blkhd, lock, lf_block);
313 			lock->lf_next = NOLOCKF;
314 		}
315 		if (error) {
316 			free(lock, M_LOCKF);
317 			return (error);
318 		}
319 	}
320 	/*
321 	 * No blocks!!  Add the lock.  Note that we will
322 	 * downgrade or upgrade any overlapping locks this
323 	 * process already owns.
324 	 *
325 	 * Skip over locks owned by other processes.
326 	 * Handle any locks that overlap and are owned by ourselves.
327 	 */
328 	prev = head;
329 	block = *head;
330 	needtolink = 1;
331 	for (;;) {
332 		ovcase = lf_findoverlap(block, lock, SELF, &prev, &overlap);
333 		if (ovcase)
334 			block = overlap->lf_next;
335 		/*
336 		 * Six cases:
337 		 *	0) no overlap
338 		 *	1) overlap == lock
339 		 *	2) overlap contains lock
340 		 *	3) lock contains overlap
341 		 *	4) overlap starts before lock
342 		 *	5) overlap ends after lock
343 		 */
344 		switch (ovcase) {
345 		case 0: /* no overlap */
346 			if (needtolink) {
347 				*prev = lock;
348 				lock->lf_next = overlap;
349 			}
350 			break;
351 
352 		case 1: /* overlap == lock */
353 			/*
354 			 * If downgrading lock, others may be
355 			 * able to acquire it.
356 			 */
357 			if (lock->lf_type == F_RDLCK &&
358 			    overlap->lf_type == F_WRLCK)
359 				lf_wakelock(overlap);
360 			overlap->lf_type = lock->lf_type;
361 			FREE(lock, M_LOCKF);
362 			lock = overlap; /* for debug output below */
363 			break;
364 
365 		case 2: /* overlap contains lock */
366 			/*
367 			 * Check for common starting point and different types.
368 			 */
369 			if (overlap->lf_type == lock->lf_type) {
370 				free(lock, M_LOCKF);
371 				lock = overlap; /* for debug output below */
372 				break;
373 			}
374 			if (overlap->lf_start == lock->lf_start) {
375 				*prev = lock;
376 				lock->lf_next = overlap;
377 				overlap->lf_start = lock->lf_end + 1;
378 			} else
379 				lf_split(overlap, lock);
380 			lf_wakelock(overlap);
381 			break;
382 
383 		case 3: /* lock contains overlap */
384 			/*
385 			 * If downgrading lock, others may be able to
386 			 * acquire it, otherwise take the list.
387 			 */
388 			if (lock->lf_type == F_RDLCK &&
389 			    overlap->lf_type == F_WRLCK) {
390 				lf_wakelock(overlap);
391 			} else {
392 				while (!TAILQ_EMPTY(&overlap->lf_blkhd)) {
393 					ltmp = TAILQ_FIRST(&overlap->lf_blkhd);
394 					TAILQ_REMOVE(&overlap->lf_blkhd, ltmp,
395 					    lf_block);
396 					TAILQ_INSERT_TAIL(&lock->lf_blkhd,
397 					    ltmp, lf_block);
398 					ltmp->lf_next = lock;
399 				}
400 			}
401 			/*
402 			 * Add the new lock if necessary and delete the overlap.
403 			 */
404 			if (needtolink) {
405 				*prev = lock;
406 				lock->lf_next = overlap->lf_next;
407 				prev = &lock->lf_next;
408 				needtolink = 0;
409 			} else
410 				*prev = overlap->lf_next;
411 			free(overlap, M_LOCKF);
412 			continue;
413 
414 		case 4: /* overlap starts before lock */
415 			/*
416 			 * Add lock after overlap on the list.
417 			 */
418 			lock->lf_next = overlap->lf_next;
419 			overlap->lf_next = lock;
420 			overlap->lf_end = lock->lf_start - 1;
421 			prev = &lock->lf_next;
422 			lf_wakelock(overlap);
423 			needtolink = 0;
424 			continue;
425 
426 		case 5: /* overlap ends after lock */
427 			/*
428 			 * Add the new lock before overlap.
429 			 */
430 			if (needtolink) {
431 				*prev = lock;
432 				lock->lf_next = overlap;
433 			}
434 			overlap->lf_start = lock->lf_end + 1;
435 			lf_wakelock(overlap);
436 			break;
437 		}
438 		break;
439 	}
440 #ifdef LOCKF_DEBUG
441 	if (lockf_debug & 1) {
442 		lf_print("lf_setlock: got the lock", lock);
443 		lf_printlist("lf_setlock", lock);
444 	}
445 #endif /* LOCKF_DEBUG */
446 	return (0);
447 }
448 
449 /*
450  * Remove a byte-range lock on an inode.
451  *
452  * Generally, find the lock (or an overlap to that lock)
453  * and remove it (or shrink it), then wakeup anyone we can.
454  */
455 static int
456 lf_clearlock(unlock)
457 	register struct lockf *unlock;
458 {
459 	struct lockf **head = unlock->lf_head;
460 	register struct lockf *lf = *head;
461 	struct lockf *overlap, **prev;
462 	int ovcase;
463 
464 	if (lf == NOLOCKF)
465 		return (0);
466 #ifdef LOCKF_DEBUG
467 	if (unlock->lf_type != F_UNLCK)
468 		panic("lf_clearlock: bad type");
469 	if (lockf_debug & 1)
470 		lf_print("lf_clearlock", unlock);
471 #endif /* LOCKF_DEBUG */
472 	prev = head;
473 	while ((ovcase = lf_findoverlap(lf, unlock, SELF, &prev, &overlap))) {
474 		/*
475 		 * Wakeup the list of locks to be retried.
476 		 */
477 		lf_wakelock(overlap);
478 
479 		switch (ovcase) {
480 
481 		case 1: /* overlap == lock */
482 			*prev = overlap->lf_next;
483 			FREE(overlap, M_LOCKF);
484 			break;
485 
486 		case 2: /* overlap contains lock: split it */
487 			if (overlap->lf_start == unlock->lf_start) {
488 				overlap->lf_start = unlock->lf_end + 1;
489 				break;
490 			}
491 			lf_split(overlap, unlock);
492 			overlap->lf_next = unlock->lf_next;
493 			break;
494 
495 		case 3: /* lock contains overlap */
496 			*prev = overlap->lf_next;
497 			lf = overlap->lf_next;
498 			free(overlap, M_LOCKF);
499 			continue;
500 
501 		case 4: /* overlap starts before lock */
502 			overlap->lf_end = unlock->lf_start - 1;
503 			prev = &overlap->lf_next;
504 			lf = overlap->lf_next;
505 			continue;
506 
507 		case 5: /* overlap ends after lock */
508 			overlap->lf_start = unlock->lf_end + 1;
509 			break;
510 		}
511 		break;
512 	}
513 #ifdef LOCKF_DEBUG
514 	if (lockf_debug & 1)
515 		lf_printlist("lf_clearlock", unlock);
516 #endif /* LOCKF_DEBUG */
517 	return (0);
518 }
519 
520 /*
521  * Check whether there is a blocking lock,
522  * and if so return its process identifier.
523  */
524 static int
525 lf_getlock(lock, fl)
526 	register struct lockf *lock;
527 	register struct flock *fl;
528 {
529 	register struct lockf *block;
530 
531 #ifdef LOCKF_DEBUG
532 	if (lockf_debug & 1)
533 		lf_print("lf_getlock", lock);
534 #endif /* LOCKF_DEBUG */
535 
536 	if ((block = lf_getblock(lock))) {
537 		fl->l_type = block->lf_type;
538 		fl->l_whence = SEEK_SET;
539 		fl->l_start = block->lf_start;
540 		if (block->lf_end == -1)
541 			fl->l_len = 0;
542 		else
543 			fl->l_len = block->lf_end - block->lf_start + 1;
544 		if (block->lf_flags & F_POSIX)
545 			fl->l_pid = ((struct proc *)(block->lf_id))->p_pid;
546 		else
547 			fl->l_pid = -1;
548 	} else {
549 		fl->l_type = F_UNLCK;
550 	}
551 	return (0);
552 }
553 
554 /*
555  * Walk the list of locks for an inode and
556  * return the first blocking lock.
557  */
558 static struct lockf *
559 lf_getblock(lock)
560 	register struct lockf *lock;
561 {
562 	struct lockf **prev, *overlap, *lf = *(lock->lf_head);
563 	int ovcase;
564 
565 	prev = lock->lf_head;
566 	while ((ovcase = lf_findoverlap(lf, lock, OTHERS, &prev, &overlap))) {
567 		/*
568 		 * We've found an overlap, see if it blocks us
569 		 */
570 		if ((lock->lf_type == F_WRLCK || overlap->lf_type == F_WRLCK))
571 			return (overlap);
572 		/*
573 		 * Nope, point to the next one on the list and
574 		 * see if it blocks us
575 		 */
576 		lf = overlap->lf_next;
577 	}
578 	return (NOLOCKF);
579 }
580 
581 /*
582  * Walk the list of locks for an inode to
583  * find an overlapping lock (if any).
584  *
585  * NOTE: this returns only the FIRST overlapping lock.  There
586  *	 may be more than one.
587  */
588 static int
589 lf_findoverlap(lf, lock, type, prev, overlap)
590 	register struct lockf *lf;
591 	struct lockf *lock;
592 	int type;
593 	struct lockf ***prev;
594 	struct lockf **overlap;
595 {
596 	off_t start, end;
597 
598 	*overlap = lf;
599 	if (lf == NOLOCKF)
600 		return (0);
601 #ifdef LOCKF_DEBUG
602 	if (lockf_debug & 2)
603 		lf_print("lf_findoverlap: looking for overlap in", lock);
604 #endif /* LOCKF_DEBUG */
605 	start = lock->lf_start;
606 	end = lock->lf_end;
607 	while (lf != NOLOCKF) {
608 		if (((type & SELF) && lf->lf_id != lock->lf_id) ||
609 		    ((type & OTHERS) && lf->lf_id == lock->lf_id)) {
610 			*prev = &lf->lf_next;
611 			*overlap = lf = lf->lf_next;
612 			continue;
613 		}
614 #ifdef LOCKF_DEBUG
615 		if (lockf_debug & 2)
616 			lf_print("\tchecking", lf);
617 #endif /* LOCKF_DEBUG */
618 		/*
619 		 * OK, check for overlap
620 		 *
621 		 * Six cases:
622 		 *	0) no overlap
623 		 *	1) overlap == lock
624 		 *	2) overlap contains lock
625 		 *	3) lock contains overlap
626 		 *	4) overlap starts before lock
627 		 *	5) overlap ends after lock
628 		 */
629 		if ((lf->lf_end != -1 && start > lf->lf_end) ||
630 		    (end != -1 && lf->lf_start > end)) {
631 			/* Case 0 */
632 #ifdef LOCKF_DEBUG
633 			if (lockf_debug & 2)
634 				printf("no overlap\n");
635 #endif /* LOCKF_DEBUG */
636 			if ((type & SELF) && end != -1 && lf->lf_start > end)
637 				return (0);
638 			*prev = &lf->lf_next;
639 			*overlap = lf = lf->lf_next;
640 			continue;
641 		}
642 		if ((lf->lf_start == start) && (lf->lf_end == end)) {
643 			/* Case 1 */
644 #ifdef LOCKF_DEBUG
645 			if (lockf_debug & 2)
646 				printf("overlap == lock\n");
647 #endif /* LOCKF_DEBUG */
648 			return (1);
649 		}
650 		if ((lf->lf_start <= start) &&
651 		    (end != -1) &&
652 		    ((lf->lf_end >= end) || (lf->lf_end == -1))) {
653 			/* Case 2 */
654 #ifdef LOCKF_DEBUG
655 			if (lockf_debug & 2)
656 				printf("overlap contains lock\n");
657 #endif /* LOCKF_DEBUG */
658 			return (2);
659 		}
660 		if (start <= lf->lf_start &&
661 		           (end == -1 ||
662 			   (lf->lf_end != -1 && end >= lf->lf_end))) {
663 			/* Case 3 */
664 #ifdef LOCKF_DEBUG
665 			if (lockf_debug & 2)
666 				printf("lock contains overlap\n");
667 #endif /* LOCKF_DEBUG */
668 			return (3);
669 		}
670 		if ((lf->lf_start < start) &&
671 			((lf->lf_end >= start) || (lf->lf_end == -1))) {
672 			/* Case 4 */
673 #ifdef LOCKF_DEBUG
674 			if (lockf_debug & 2)
675 				printf("overlap starts before lock\n");
676 #endif /* LOCKF_DEBUG */
677 			return (4);
678 		}
679 		if ((lf->lf_start > start) &&
680 			(end != -1) &&
681 			((lf->lf_end > end) || (lf->lf_end == -1))) {
682 			/* Case 5 */
683 #ifdef LOCKF_DEBUG
684 			if (lockf_debug & 2)
685 				printf("overlap ends after lock\n");
686 #endif /* LOCKF_DEBUG */
687 			return (5);
688 		}
689 		panic("lf_findoverlap: default");
690 	}
691 	return (0);
692 }
693 
694 /*
695  * Split a lock and a contained region into
696  * two or three locks as necessary.
697  */
698 static void
699 lf_split(lock1, lock2)
700 	register struct lockf *lock1;
701 	register struct lockf *lock2;
702 {
703 	register struct lockf *splitlock;
704 
705 #ifdef LOCKF_DEBUG
706 	if (lockf_debug & 2) {
707 		lf_print("lf_split", lock1);
708 		lf_print("splitting from", lock2);
709 	}
710 #endif /* LOCKF_DEBUG */
711 	/*
712 	 * Check to see if spliting into only two pieces.
713 	 */
714 	if (lock1->lf_start == lock2->lf_start) {
715 		lock1->lf_start = lock2->lf_end + 1;
716 		lock2->lf_next = lock1;
717 		return;
718 	}
719 	if (lock1->lf_end == lock2->lf_end) {
720 		lock1->lf_end = lock2->lf_start - 1;
721 		lock2->lf_next = lock1->lf_next;
722 		lock1->lf_next = lock2;
723 		return;
724 	}
725 	/*
726 	 * Make a new lock consisting of the last part of
727 	 * the encompassing lock
728 	 */
729 	MALLOC(splitlock, struct lockf *, sizeof *splitlock, M_LOCKF, M_WAITOK);
730 	bcopy(lock1, splitlock, sizeof *splitlock);
731 	splitlock->lf_start = lock2->lf_end + 1;
732 	TAILQ_INIT(&splitlock->lf_blkhd);
733 	lock1->lf_end = lock2->lf_start - 1;
734 	/*
735 	 * OK, now link it in
736 	 */
737 	splitlock->lf_next = lock1->lf_next;
738 	lock2->lf_next = splitlock;
739 	lock1->lf_next = lock2;
740 }
741 
742 /*
743  * Wakeup a blocklist
744  */
745 static void
746 lf_wakelock(listhead)
747 	struct lockf *listhead;
748 {
749 	register struct lockf *wakelock;
750 
751 	while (!TAILQ_EMPTY(&listhead->lf_blkhd)) {
752 		wakelock = TAILQ_FIRST(&listhead->lf_blkhd);
753 		TAILQ_REMOVE(&listhead->lf_blkhd, wakelock, lf_block);
754 		wakelock->lf_next = NOLOCKF;
755 #ifdef LOCKF_DEBUG
756 		if (lockf_debug & 2)
757 			lf_print("lf_wakelock: awakening", wakelock);
758 #endif /* LOCKF_DEBUG */
759 		wakeup(wakelock);
760 	}
761 }
762 
763 #ifdef LOCKF_DEBUG
764 /*
765  * Print out a lock.
766  */
767 void
768 lf_print(tag, lock)
769 	char *tag;
770 	register struct lockf *lock;
771 {
772 
773 	printf("%s: lock %p for ", tag, (void *)lock);
774 	if (lock->lf_flags & F_POSIX)
775 		printf("proc %ld", (long)((struct proc *)lock->lf_id)->p_pid);
776 	else
777 		printf("id %p", (void *)lock->lf_id);
778 	if (lock->lf_inode != (struct inode *)0)
779 		printf(" in ino %ju on dev <%d, %d>, %s, start %jd, end %jd",
780 		    (uintmax_t)lock->lf_inode->i_number,
781 		    major(lock->lf_inode->i_dev),
782 		    minor(lock->lf_inode->i_dev),
783 		    lock->lf_type == F_RDLCK ? "shared" :
784 		    lock->lf_type == F_WRLCK ? "exclusive" :
785 		    lock->lf_type == F_UNLCK ? "unlock" : "unknown",
786 		    (intmax_t)lock->lf_start, (intmax_t)lock->lf_end);
787 	else
788 		printf(" %s, start %jd, end %jd",
789 		    lock->lf_type == F_RDLCK ? "shared" :
790 		    lock->lf_type == F_WRLCK ? "exclusive" :
791 		    lock->lf_type == F_UNLCK ? "unlock" : "unknown",
792 		    (intmax_t)lock->lf_start, (intmax_t)lock->lf_end);
793 	if (!TAILQ_EMPTY(&lock->lf_blkhd))
794 		printf(" block %p\n", (void *)TAILQ_FIRST(&lock->lf_blkhd));
795 	else
796 		printf("\n");
797 }
798 
799 void
800 lf_printlist(tag, lock)
801 	char *tag;
802 	struct lockf *lock;
803 {
804 	register struct lockf *lf, *blk;
805 
806 	if (lock->lf_inode == (struct inode *)0)
807 		return;
808 
809 	printf("%s: Lock list for ino %ju on dev <%d, %d>:\n",
810 	    tag, (uintmax_t)lock->lf_inode->i_number,
811 	    major(lock->lf_inode->i_dev),
812 	    minor(lock->lf_inode->i_dev));
813 	for (lf = lock->lf_inode->i_lockf; lf; lf = lf->lf_next) {
814 		printf("\tlock %p for ",(void *)lf);
815 		if (lf->lf_flags & F_POSIX)
816 			printf("proc %ld",
817 			    (long)((struct proc *)lf->lf_id)->p_pid);
818 		else
819 			printf("id %p", (void *)lf->lf_id);
820 		printf(", %s, start %jd, end %jd",
821 		    lf->lf_type == F_RDLCK ? "shared" :
822 		    lf->lf_type == F_WRLCK ? "exclusive" :
823 		    lf->lf_type == F_UNLCK ? "unlock" :
824 		    "unknown", (intmax_t)lf->lf_start, (intmax_t)lf->lf_end);
825 		TAILQ_FOREACH(blk, &lf->lf_blkhd, lf_block) {
826 			printf("\n\t\tlock request %p for ", (void *)blk);
827 			if (blk->lf_flags & F_POSIX)
828 				printf("proc %ld",
829 				    (long)((struct proc *)blk->lf_id)->p_pid);
830 			else
831 				printf("id %p", (void *)blk->lf_id);
832 			printf(", %s, start %jd, end %jd",
833 			    blk->lf_type == F_RDLCK ? "shared" :
834 			    blk->lf_type == F_WRLCK ? "exclusive" :
835 			    blk->lf_type == F_UNLCK ? "unlock" :
836 			    "unknown", (intmax_t)blk->lf_start,
837 			    (intmax_t)blk->lf_end);
838 			if (!TAILQ_EMPTY(&blk->lf_blkhd))
839 				panic("lf_printlist: bad list");
840 		}
841 		printf("\n");
842 	}
843 }
844 #endif /* LOCKF_DEBUG */
845