Home | History | Annotate | Line # | Download | only in kern
vfs_lockf.c revision 1.27
      1 /*	$NetBSD: vfs_lockf.c,v 1.27 2003/05/01 14:36:43 yamt 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. All advertising materials mentioning features or use of this software
     19  *    must display the following acknowledgement:
     20  *	This product includes software developed by the University of
     21  *	California, Berkeley and its contributors.
     22  * 4. Neither the name of the University nor the names of its contributors
     23  *    may be used to endorse or promote products derived from this software
     24  *    without specific prior written permission.
     25  *
     26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     36  * SUCH DAMAGE.
     37  *
     38  *	@(#)ufs_lockf.c	8.4 (Berkeley) 10/26/94
     39  */
     40 
     41 #include <sys/cdefs.h>
     42 __KERNEL_RCSID(0, "$NetBSD: vfs_lockf.c,v 1.27 2003/05/01 14:36:43 yamt Exp $");
     43 
     44 #include <sys/param.h>
     45 #include <sys/systm.h>
     46 #include <sys/kernel.h>
     47 #include <sys/file.h>
     48 #include <sys/proc.h>
     49 #include <sys/vnode.h>
     50 #include <sys/malloc.h>
     51 #include <sys/fcntl.h>
     52 #include <sys/lockf.h>
     53 
     54 MALLOC_DEFINE(M_LOCKF, "lockf", "Byte-range locking structures");
     55 
     56 /*
     57  * This variable controls the maximum number of processes that will
     58  * be checked in doing deadlock detection.
     59  */
     60 int maxlockdepth = MAXDEPTH;
     61 
     62 #ifdef LOCKF_DEBUG
     63 int	lockf_debug = 0;
     64 #endif
     65 
     66 #define NOLOCKF (struct lockf *)0
     67 #define SELF	0x1
     68 #define OTHERS	0x2
     69 
     70 static int lf_clearlock(struct lockf *, struct lockf **);
     71 static int lf_findoverlap(struct lockf *,
     72 	    struct lockf *, int, struct lockf ***, struct lockf **);
     73 static struct lockf *lf_getblock(struct lockf *);
     74 static int lf_getlock(struct lockf *, struct flock *);
     75 static int lf_setlock(struct lockf *, struct lockf **, struct simplelock *);
     76 static void lf_split(struct lockf *, struct lockf *, struct lockf **);
     77 static void lf_wakelock(struct lockf *);
     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  * Use pools for lock allocation.
     91  */
     92 
     93 /*
     94  * If there's a lot of lock contention on a single vnode, locking
     95  * schemes which allow for more paralleism would be needed.  Given how
     96  * infrequently byte-range locks are actually used in typical BSD
     97  * code, a more complex approach probably isn't worth it.
     98  */
     99 
    100 /*
    101  * Do an advisory lock operation.
    102  */
    103 int
    104 lf_advlock(struct vop_advlock_args *ap, struct lockf **head, off_t size)
    105 {
    106 	struct flock *fl = ap->a_fl;
    107 	struct lockf *lock = NULL;
    108 	struct lockf *sparelock;
    109 	struct vnode *vp = ap->a_vp;
    110 	off_t start, end;
    111 	int error;
    112 
    113 	/*
    114 	 * Convert the flock structure into a start and end.
    115 	 */
    116 	switch (fl->l_whence) {
    117 	case SEEK_SET:
    118 	case SEEK_CUR:
    119 		/*
    120 		 * Caller is responsible for adding any necessary offset
    121 		 * when SEEK_CUR is used.
    122 		 */
    123 		start = fl->l_start;
    124 		break;
    125 
    126 	case SEEK_END:
    127 		start = size + fl->l_start;
    128 		break;
    129 
    130 	default:
    131 		return (EINVAL);
    132 	}
    133 	if (start < 0)
    134 		return (EINVAL);
    135 
    136 	/*
    137 	 * allocate locks before acquire simple lock.
    138 	 * we need two locks in the worst case.
    139 	 */
    140 	switch (ap->a_op) {
    141 	case F_SETLK:
    142 	case F_UNLCK:
    143 		/*
    144 		 * XXX for F_UNLCK case, we can re-use lock.
    145 		 */
    146 		if ((fl->l_type & F_FLOCK) == 0) {
    147 			/*
    148 			 * byte-range lock might need one more lock.
    149 			 */
    150 			MALLOC(sparelock, struct lockf *, sizeof(*lock),
    151 			    M_LOCKF, M_WAITOK);
    152 			if (sparelock == NULL) {
    153 				error = ENOMEM;
    154 				goto quit;
    155 			}
    156 			break;
    157 		}
    158 		/* FALLTHROUGH */
    159 
    160 	case F_GETLK:
    161 		sparelock = NULL;
    162 		break;
    163 
    164 	default:
    165 		return (EINVAL);
    166 	}
    167 
    168 	MALLOC(lock, struct lockf *, sizeof(*lock), M_LOCKF, M_WAITOK);
    169 	if (lock == NULL) {
    170 		error = ENOMEM;
    171 		goto quit;
    172 	}
    173 
    174 	simple_lock(&vp->v_interlock);
    175 
    176 	/*
    177 	 * Avoid the common case of unlocking when inode has no locks.
    178 	 */
    179 	if (*head == (struct lockf *)0) {
    180 		if (ap->a_op != F_SETLK) {
    181 			fl->l_type = F_UNLCK;
    182 			error = 0;
    183 			goto quit_unlock;
    184 		}
    185 	}
    186 
    187 	if (fl->l_len == 0)
    188 		end = -1;
    189 	else
    190 		end = start + fl->l_len - 1;
    191 	/*
    192 	 * Create the lockf structure.
    193 	 */
    194 	lock->lf_start = start;
    195 	lock->lf_end = end;
    196 	/* XXX NJWLWP
    197 	 * I don't want to make the entire VFS universe use LWPs, because
    198 	 * they don't need them, for the most part. This is an exception,
    199 	 * and a kluge.
    200 	 */
    201 
    202 	lock->lf_head = head;
    203 	lock->lf_type = fl->l_type;
    204 	lock->lf_next = (struct lockf *)0;
    205 	TAILQ_INIT(&lock->lf_blkhd);
    206 	lock->lf_flags = ap->a_flags;
    207 	if (lock->lf_flags & F_POSIX) {
    208 		KASSERT(curproc == (struct proc *)ap->a_id);
    209 	}
    210 	lock->lf_id = (struct proc *)ap->a_id;
    211 	lock->lf_lwp = curlwp;
    212 
    213 	/*
    214 	 * Do the requested operation.
    215 	 */
    216 	switch (ap->a_op) {
    217 
    218 	case F_SETLK:
    219 		error = lf_setlock(lock, &sparelock, &vp->v_interlock);
    220 		lock = NULL; /* lf_setlock freed it */
    221 		break;
    222 
    223 	case F_UNLCK:
    224 		error = lf_clearlock(lock, &sparelock);
    225 		break;
    226 
    227 	case F_GETLK:
    228 		error = lf_getlock(lock, fl);
    229 		break;
    230 
    231 	default:
    232 		/* NOTREACHED */
    233 	}
    234 
    235 quit_unlock:
    236 	simple_unlock(&vp->v_interlock);
    237 quit:
    238 	if (lock)
    239 		FREE(lock, M_LOCKF);
    240 	if (sparelock)
    241 		FREE(sparelock, M_LOCKF);
    242 
    243 	return (error);
    244 }
    245 
    246 /*
    247  * Set a byte-range lock.
    248  */
    249 static int
    250 lf_setlock(struct lockf *lock, struct lockf **sparelock,
    251     struct simplelock *interlock)
    252 {
    253 	struct lockf *block;
    254 	struct lockf **head = lock->lf_head;
    255 	struct lockf **prev, *overlap, *ltmp;
    256 	static char lockstr[] = "lockf";
    257 	int ovcase, priority, needtolink, error;
    258 
    259 #ifdef LOCKF_DEBUG
    260 	if (lockf_debug & 1)
    261 		lf_print("lf_setlock", lock);
    262 #endif /* LOCKF_DEBUG */
    263 
    264 	/*
    265 	 * Set the priority
    266 	 */
    267 	priority = PLOCK;
    268 	if (lock->lf_type == F_WRLCK)
    269 		priority += 4;
    270 	priority |= PCATCH;
    271 	/*
    272 	 * Scan lock list for this file looking for locks that would block us.
    273 	 */
    274 	while ((block = lf_getblock(lock)) != NULL) {
    275 		/*
    276 		 * Free the structure and return if nonblocking.
    277 		 */
    278 		if ((lock->lf_flags & F_WAIT) == 0) {
    279 			FREE(lock, M_LOCKF);
    280 			return (EAGAIN);
    281 		}
    282 		/*
    283 		 * We are blocked. Since flock style locks cover
    284 		 * the whole file, there is no chance for deadlock.
    285 		 * For byte-range locks we must check for deadlock.
    286 		 *
    287 		 * Deadlock detection is done by looking through the
    288 		 * wait channels to see if there are any cycles that
    289 		 * involve us. MAXDEPTH is set just to make sure we
    290 		 * do not go off into neverneverland.
    291 		 */
    292 		if ((lock->lf_flags & F_POSIX) &&
    293 		    (block->lf_flags & F_POSIX)) {
    294 			struct lwp *wlwp;
    295 			struct lockf *waitblock;
    296 			int i = 0;
    297 
    298 			/*
    299 			 * The block is waiting on something.  if_lwp will be
    300 			 * 0 once the lock is granted, so we terminate the
    301 			 * loop if we find this.
    302 			 */
    303 			wlwp = block->lf_lwp;
    304 			while (wlwp && (i++ < maxlockdepth)) {
    305 				waitblock = (struct lockf *)wlwp->l_wchan;
    306 				/* Get the owner of the blocking lock */
    307 				waitblock = waitblock->lf_next;
    308 				if ((waitblock->lf_flags & F_POSIX) == 0)
    309 					break;
    310 				wlwp = waitblock->lf_lwp;
    311 				if (wlwp == lock->lf_lwp) {
    312 					free(lock, M_LOCKF);
    313 					return (EDEADLK);
    314 				}
    315 			}
    316 			/*
    317 			 * If we're still following a dependancy chain
    318 			 * after maxlockdepth iterations, assume we're in
    319 			 * a cycle to be safe.
    320 			 */
    321 			if (i >= maxlockdepth) {
    322 				free(lock, M_LOCKF);
    323 				return (EDEADLK);
    324 			}
    325 		}
    326 		/*
    327 		 * For flock type locks, we must first remove
    328 		 * any shared locks that we hold before we sleep
    329 		 * waiting for an exclusive lock.
    330 		 */
    331 		if ((lock->lf_flags & F_FLOCK) &&
    332 		    lock->lf_type == F_WRLCK) {
    333 			lock->lf_type = F_UNLCK;
    334 			(void) lf_clearlock(lock, NULL);
    335 			lock->lf_type = F_WRLCK;
    336 		}
    337 		/*
    338 		 * Add our lock to the blocked list and sleep until we're free.
    339 		 * Remember who blocked us (for deadlock detection).
    340 		 */
    341 		lock->lf_next = block;
    342 		TAILQ_INSERT_TAIL(&block->lf_blkhd, lock, lf_block);
    343 #ifdef LOCKF_DEBUG
    344 		if (lockf_debug & 1) {
    345 			lf_print("lf_setlock: blocking on", block);
    346 			lf_printlist("lf_setlock", block);
    347 		}
    348 #endif /* LOCKF_DEBUG */
    349 		error = ltsleep(lock, priority, lockstr, 0, interlock);
    350 
    351 		/*
    352 		 * We may have been awakened by a signal (in
    353 		 * which case we must remove ourselves from the
    354 		 * blocked list) and/or by another process
    355 		 * releasing a lock (in which case we have already
    356 		 * been removed from the blocked list and our
    357 		 * lf_next field set to NOLOCKF).
    358 		 */
    359 		if (lock->lf_next != NOLOCKF) {
    360 			TAILQ_REMOVE(&lock->lf_next->lf_blkhd, lock, lf_block);
    361 			lock->lf_next = NOLOCKF;
    362 		}
    363 		if (error) {
    364 			free(lock, M_LOCKF);
    365 			return (error);
    366 		}
    367 	}
    368 	/*
    369 	 * No blocks!!  Add the lock.  Note that we will
    370 	 * downgrade or upgrade any overlapping locks this
    371 	 * process already owns.
    372 	 *
    373 	 * Skip over locks owned by other processes.
    374 	 * Handle any locks that overlap and are owned by ourselves.
    375 	 */
    376 	lock->lf_lwp = 0;
    377 	prev = head;
    378 	block = *head;
    379 	needtolink = 1;
    380 	for (;;) {
    381 		ovcase = lf_findoverlap(block, lock, SELF, &prev, &overlap);
    382 		if (ovcase)
    383 			block = overlap->lf_next;
    384 		/*
    385 		 * Six cases:
    386 		 *	0) no overlap
    387 		 *	1) overlap == lock
    388 		 *	2) overlap contains lock
    389 		 *	3) lock contains overlap
    390 		 *	4) overlap starts before lock
    391 		 *	5) overlap ends after lock
    392 		 */
    393 		switch (ovcase) {
    394 		case 0: /* no overlap */
    395 			if (needtolink) {
    396 				*prev = lock;
    397 				lock->lf_next = overlap;
    398 			}
    399 			break;
    400 
    401 		case 1: /* overlap == lock */
    402 			/*
    403 			 * If downgrading lock, others may be
    404 			 * able to acquire it.
    405 			 */
    406 			if (lock->lf_type == F_RDLCK &&
    407 			    overlap->lf_type == F_WRLCK)
    408 				lf_wakelock(overlap);
    409 			overlap->lf_type = lock->lf_type;
    410 			FREE(lock, M_LOCKF);
    411 			lock = overlap; /* for debug output below */
    412 			break;
    413 
    414 		case 2: /* overlap contains lock */
    415 			/*
    416 			 * Check for common starting point and different types.
    417 			 */
    418 			if (overlap->lf_type == lock->lf_type) {
    419 				free(lock, M_LOCKF);
    420 				lock = overlap; /* for debug output below */
    421 				break;
    422 			}
    423 			if (overlap->lf_start == lock->lf_start) {
    424 				*prev = lock;
    425 				lock->lf_next = overlap;
    426 				overlap->lf_start = lock->lf_end + 1;
    427 			} else
    428 				lf_split(overlap, lock, sparelock);
    429 			lf_wakelock(overlap);
    430 			break;
    431 
    432 		case 3: /* lock contains overlap */
    433 			/*
    434 			 * If downgrading lock, others may be able to
    435 			 * acquire it, otherwise take the list.
    436 			 */
    437 			if (lock->lf_type == F_RDLCK &&
    438 			    overlap->lf_type == F_WRLCK) {
    439 				lf_wakelock(overlap);
    440 			} else {
    441 				while ((ltmp = TAILQ_FIRST(&overlap->lf_blkhd))) {
    442 					KASSERT(ltmp->lf_next == overlap);
    443 					TAILQ_REMOVE(&overlap->lf_blkhd, ltmp,
    444 					    lf_block);
    445 					ltmp->lf_next = lock;
    446 					TAILQ_INSERT_TAIL(&lock->lf_blkhd,
    447 					    ltmp, lf_block);
    448 				}
    449 			}
    450 			/*
    451 			 * Add the new lock if necessary and delete the overlap.
    452 			 */
    453 			if (needtolink) {
    454 				*prev = lock;
    455 				lock->lf_next = overlap->lf_next;
    456 				prev = &lock->lf_next;
    457 				needtolink = 0;
    458 			} else
    459 				*prev = overlap->lf_next;
    460 			free(overlap, M_LOCKF);
    461 			continue;
    462 
    463 		case 4: /* overlap starts before lock */
    464 			/*
    465 			 * Add lock after overlap on the list.
    466 			 */
    467 			lock->lf_next = overlap->lf_next;
    468 			overlap->lf_next = lock;
    469 			overlap->lf_end = lock->lf_start - 1;
    470 			prev = &lock->lf_next;
    471 			lf_wakelock(overlap);
    472 			needtolink = 0;
    473 			continue;
    474 
    475 		case 5: /* overlap ends after lock */
    476 			/*
    477 			 * Add the new lock before overlap.
    478 			 */
    479 			if (needtolink) {
    480 				*prev = lock;
    481 				lock->lf_next = overlap;
    482 			}
    483 			overlap->lf_start = lock->lf_end + 1;
    484 			lf_wakelock(overlap);
    485 			break;
    486 		}
    487 		break;
    488 	}
    489 #ifdef LOCKF_DEBUG
    490 	if (lockf_debug & 1) {
    491 		lf_print("lf_setlock: got the lock", lock);
    492 		lf_printlist("lf_setlock", lock);
    493 	}
    494 #endif /* LOCKF_DEBUG */
    495 	return (0);
    496 }
    497 
    498 /*
    499  * Remove a byte-range lock on an inode.
    500  *
    501  * Generally, find the lock (or an overlap to that lock)
    502  * and remove it (or shrink it), then wakeup anyone we can.
    503  */
    504 static int
    505 lf_clearlock(struct lockf *unlock, struct lockf **sparelock)
    506 {
    507 	struct lockf **head = unlock->lf_head;
    508 	struct lockf *lf = *head;
    509 	struct lockf *overlap, **prev;
    510 	int ovcase;
    511 
    512 	if (lf == NOLOCKF)
    513 		return (0);
    514 #ifdef LOCKF_DEBUG
    515 	if (unlock->lf_type != F_UNLCK)
    516 		panic("lf_clearlock: bad type");
    517 	if (lockf_debug & 1)
    518 		lf_print("lf_clearlock", unlock);
    519 #endif /* LOCKF_DEBUG */
    520 	prev = head;
    521 	while ((ovcase = lf_findoverlap(lf, unlock, SELF,
    522 					&prev, &overlap)) != 0) {
    523 		/*
    524 		 * Wakeup the list of locks to be retried.
    525 		 */
    526 		lf_wakelock(overlap);
    527 
    528 		switch (ovcase) {
    529 
    530 		case 1: /* overlap == lock */
    531 			*prev = overlap->lf_next;
    532 			FREE(overlap, M_LOCKF);
    533 			break;
    534 
    535 		case 2: /* overlap contains lock: split it */
    536 			if (overlap->lf_start == unlock->lf_start) {
    537 				overlap->lf_start = unlock->lf_end + 1;
    538 				break;
    539 			}
    540 			lf_split(overlap, unlock, sparelock);
    541 			overlap->lf_next = unlock->lf_next;
    542 			break;
    543 
    544 		case 3: /* lock contains overlap */
    545 			*prev = overlap->lf_next;
    546 			lf = overlap->lf_next;
    547 			free(overlap, M_LOCKF);
    548 			continue;
    549 
    550 		case 4: /* overlap starts before lock */
    551 			overlap->lf_end = unlock->lf_start - 1;
    552 			prev = &overlap->lf_next;
    553 			lf = overlap->lf_next;
    554 			continue;
    555 
    556 		case 5: /* overlap ends after lock */
    557 			overlap->lf_start = unlock->lf_end + 1;
    558 			break;
    559 		}
    560 		break;
    561 	}
    562 #ifdef LOCKF_DEBUG
    563 	if (lockf_debug & 1)
    564 		lf_printlist("lf_clearlock", unlock);
    565 #endif /* LOCKF_DEBUG */
    566 	return (0);
    567 }
    568 
    569 /*
    570  * Check whether there is a blocking lock,
    571  * and if so return its process identifier.
    572  */
    573 static int
    574 lf_getlock(struct lockf *lock, struct flock *fl)
    575 {
    576 	struct lockf *block;
    577 
    578 #ifdef LOCKF_DEBUG
    579 	if (lockf_debug & 1)
    580 		lf_print("lf_getlock", lock);
    581 #endif /* LOCKF_DEBUG */
    582 
    583 	if ((block = lf_getblock(lock)) != NULL) {
    584 		fl->l_type = block->lf_type;
    585 		fl->l_whence = SEEK_SET;
    586 		fl->l_start = block->lf_start;
    587 		if (block->lf_end == -1)
    588 			fl->l_len = 0;
    589 		else
    590 			fl->l_len = block->lf_end - block->lf_start + 1;
    591 		if (block->lf_flags & F_POSIX)
    592 			fl->l_pid = ((struct proc *)block->lf_id)->p_pid;
    593 		else
    594 			fl->l_pid = -1;
    595 	} else {
    596 		fl->l_type = F_UNLCK;
    597 	}
    598 	return (0);
    599 }
    600 
    601 /*
    602  * Walk the list of locks for an inode and
    603  * return the first blocking lock.
    604  */
    605 static struct lockf *
    606 lf_getblock(struct lockf *lock)
    607 {
    608 	struct lockf **prev, *overlap, *lf = *(lock->lf_head);
    609 
    610 	prev = lock->lf_head;
    611 	while (lf_findoverlap(lf, lock, OTHERS, &prev, &overlap) != 0) {
    612 		/*
    613 		 * We've found an overlap, see if it blocks us
    614 		 */
    615 		if ((lock->lf_type == F_WRLCK || overlap->lf_type == F_WRLCK))
    616 			return (overlap);
    617 		/*
    618 		 * Nope, point to the next one on the list and
    619 		 * see if it blocks us
    620 		 */
    621 		lf = overlap->lf_next;
    622 	}
    623 	return (NOLOCKF);
    624 }
    625 
    626 /*
    627  * Walk the list of locks for an inode to
    628  * find an overlapping lock (if any).
    629  *
    630  * NOTE: this returns only the FIRST overlapping lock.  There
    631  *	 may be more than one.
    632  */
    633 static int
    634 lf_findoverlap(struct lockf *lf, struct lockf *lock, int type,
    635     struct lockf ***prev, struct lockf **overlap)
    636 {
    637 	off_t start, end;
    638 
    639 	*overlap = lf;
    640 	if (lf == NOLOCKF)
    641 		return (0);
    642 #ifdef LOCKF_DEBUG
    643 	if (lockf_debug & 2)
    644 		lf_print("lf_findoverlap: looking for overlap in", lock);
    645 #endif /* LOCKF_DEBUG */
    646 	start = lock->lf_start;
    647 	end = lock->lf_end;
    648 	while (lf != NOLOCKF) {
    649 		if (((type == SELF) && lf->lf_id != lock->lf_id) ||
    650 		    ((type == OTHERS) && lf->lf_id == lock->lf_id)) {
    651 			*prev = &lf->lf_next;
    652 			*overlap = lf = lf->lf_next;
    653 			continue;
    654 		}
    655 #ifdef LOCKF_DEBUG
    656 		if (lockf_debug & 2)
    657 			lf_print("\tchecking", lf);
    658 #endif /* LOCKF_DEBUG */
    659 		/*
    660 		 * OK, check for overlap
    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 		if ((lf->lf_end != -1 && start > lf->lf_end) ||
    671 		    (end != -1 && lf->lf_start > end)) {
    672 			/* Case 0 */
    673 #ifdef LOCKF_DEBUG
    674 			if (lockf_debug & 2)
    675 				printf("no overlap\n");
    676 #endif /* LOCKF_DEBUG */
    677 			if ((type & SELF) && end != -1 && lf->lf_start > end)
    678 				return (0);
    679 			*prev = &lf->lf_next;
    680 			*overlap = lf = lf->lf_next;
    681 			continue;
    682 		}
    683 		if ((lf->lf_start == start) && (lf->lf_end == end)) {
    684 			/* Case 1 */
    685 #ifdef LOCKF_DEBUG
    686 			if (lockf_debug & 2)
    687 				printf("overlap == lock\n");
    688 #endif /* LOCKF_DEBUG */
    689 			return (1);
    690 		}
    691 		if ((lf->lf_start <= start) &&
    692 		    (end != -1) &&
    693 		    ((lf->lf_end >= end) || (lf->lf_end == -1))) {
    694 			/* Case 2 */
    695 #ifdef LOCKF_DEBUG
    696 			if (lockf_debug & 2)
    697 				printf("overlap contains lock\n");
    698 #endif /* LOCKF_DEBUG */
    699 			return (2);
    700 		}
    701 		if (start <= lf->lf_start &&
    702 		           (end == -1 ||
    703 			   (lf->lf_end != -1 && end >= lf->lf_end))) {
    704 			/* Case 3 */
    705 #ifdef LOCKF_DEBUG
    706 			if (lockf_debug & 2)
    707 				printf("lock contains overlap\n");
    708 #endif /* LOCKF_DEBUG */
    709 			return (3);
    710 		}
    711 		if ((lf->lf_start < start) &&
    712 			((lf->lf_end >= start) || (lf->lf_end == -1))) {
    713 			/* Case 4 */
    714 #ifdef LOCKF_DEBUG
    715 			if (lockf_debug & 2)
    716 				printf("overlap starts before lock\n");
    717 #endif /* LOCKF_DEBUG */
    718 			return (4);
    719 		}
    720 		if ((lf->lf_start > start) &&
    721 			(end != -1) &&
    722 			((lf->lf_end > end) || (lf->lf_end == -1))) {
    723 			/* Case 5 */
    724 #ifdef LOCKF_DEBUG
    725 			if (lockf_debug & 2)
    726 				printf("overlap ends after lock\n");
    727 #endif /* LOCKF_DEBUG */
    728 			return (5);
    729 		}
    730 		panic("lf_findoverlap: default");
    731 	}
    732 	return (0);
    733 }
    734 
    735 /*
    736  * Split a lock and a contained region into
    737  * two or three locks as necessary.
    738  */
    739 static void
    740 lf_split(struct lockf *lock1, struct lockf *lock2, struct lockf **sparelock)
    741 {
    742 	struct lockf *splitlock;
    743 
    744 #ifdef LOCKF_DEBUG
    745 	if (lockf_debug & 2) {
    746 		lf_print("lf_split", lock1);
    747 		lf_print("splitting from", lock2);
    748 	}
    749 #endif /* LOCKF_DEBUG */
    750 	/*
    751 	 * Check to see if spliting into only two pieces.
    752 	 */
    753 	if (lock1->lf_start == lock2->lf_start) {
    754 		lock1->lf_start = lock2->lf_end + 1;
    755 		lock2->lf_next = lock1;
    756 		return;
    757 	}
    758 	if (lock1->lf_end == lock2->lf_end) {
    759 		lock1->lf_end = lock2->lf_start - 1;
    760 		lock2->lf_next = lock1->lf_next;
    761 		lock1->lf_next = lock2;
    762 		return;
    763 	}
    764 	/*
    765 	 * Make a new lock consisting of the last part of
    766 	 * the encompassing lock
    767 	 */
    768 	splitlock = *sparelock;
    769 	*sparelock = NULL;
    770 	memcpy((caddr_t)splitlock, (caddr_t)lock1, sizeof(*splitlock));
    771 	splitlock->lf_start = lock2->lf_end + 1;
    772 	TAILQ_INIT(&splitlock->lf_blkhd);
    773 	lock1->lf_end = lock2->lf_start - 1;
    774 	/*
    775 	 * OK, now link it in
    776 	 */
    777 	splitlock->lf_next = lock1->lf_next;
    778 	lock2->lf_next = splitlock;
    779 	lock1->lf_next = lock2;
    780 }
    781 
    782 /*
    783  * Wakeup a blocklist
    784  */
    785 static void
    786 lf_wakelock(struct lockf *listhead)
    787 {
    788 	struct lockf *wakelock;
    789 
    790 	while ((wakelock = TAILQ_FIRST(&listhead->lf_blkhd))) {
    791 		KASSERT(wakelock->lf_next == listhead);
    792 		TAILQ_REMOVE(&listhead->lf_blkhd, wakelock, lf_block);
    793 		wakelock->lf_next = NOLOCKF;
    794 #ifdef LOCKF_DEBUG
    795 		if (lockf_debug & 2)
    796 			lf_print("lf_wakelock: awakening", wakelock);
    797 #endif
    798 		wakeup((caddr_t)wakelock);
    799 	}
    800 }
    801 
    802 #ifdef LOCKF_DEBUG
    803 /*
    804  * Print out a lock.
    805  */
    806 static void
    807 lf_print(char *tag, struct lockf *lock)
    808 {
    809 
    810 	printf("%s: lock %p for ", tag, lock);
    811 	if (lock->lf_flags & F_POSIX)
    812 		printf("proc %d", ((struct proc *)lock->lf_id)->p_pid);
    813 	else
    814 		printf("file 0x%p", (struct file *)lock->lf_id);
    815 	printf(" %s, start %qx, end %qx",
    816 		lock->lf_type == F_RDLCK ? "shared" :
    817 		lock->lf_type == F_WRLCK ? "exclusive" :
    818 		lock->lf_type == F_UNLCK ? "unlock" :
    819 		"unknown", lock->lf_start, lock->lf_end);
    820 	if (TAILQ_FIRST(&lock->lf_blkhd))
    821 		printf(" block %p\n", TAILQ_FIRST(&lock->lf_blkhd));
    822 	else
    823 		printf("\n");
    824 }
    825 
    826 static void
    827 lf_printlist(char *tag, struct lockf *lock)
    828 {
    829 	struct lockf *lf, *blk;
    830 
    831 	printf("%s: Lock list:\n", tag);
    832 	for (lf = *lock->lf_head; lf; lf = lf->lf_next) {
    833 		printf("\tlock %p for ", lf);
    834 		if (lf->lf_flags & F_POSIX)
    835 			printf("proc %d", ((struct proc *)lf->lf_id)->p_pid);
    836 		else
    837 			printf("file 0x%p", (struct file *)lf->lf_id);
    838 		printf(", %s, start %qx, end %qx",
    839 			lf->lf_type == F_RDLCK ? "shared" :
    840 			lf->lf_type == F_WRLCK ? "exclusive" :
    841 			lf->lf_type == F_UNLCK ? "unlock" :
    842 			"unknown", lf->lf_start, lf->lf_end);
    843 		TAILQ_FOREACH(blk, &lf->lf_blkhd, lf_block) {
    844 			if (blk->lf_flags & F_POSIX)
    845 				printf("proc %d",
    846 				    ((struct proc *)blk->lf_id)->p_pid);
    847 			else
    848 				printf("file 0x%p", (struct file *)blk->lf_id);
    849 			printf(", %s, start %qx, end %qx",
    850 				blk->lf_type == F_RDLCK ? "shared" :
    851 				blk->lf_type == F_WRLCK ? "exclusive" :
    852 				blk->lf_type == F_UNLCK ? "unlock" :
    853 				"unknown", blk->lf_start, blk->lf_end);
    854 			if (TAILQ_FIRST(&blk->lf_blkhd))
    855 				 panic("lf_printlist: bad list");
    856 		}
    857 		printf("\n");
    858 	}
    859 }
    860 #endif /* LOCKF_DEBUG */
    861