Home | History | Annotate | Line # | Download | only in kern
vfs_lockf.c revision 1.25
      1  1.25      yamt /*	$NetBSD: vfs_lockf.c,v 1.25 2003/05/01 13:06:59 yamt Exp $	*/
      2   1.5       cgd 
      3   1.1        ws /*
      4   1.4   mycroft  * Copyright (c) 1982, 1986, 1989, 1993
      5   1.4   mycroft  *	The Regents of the University of California.  All rights reserved.
      6   1.1        ws  *
      7   1.1        ws  * This code is derived from software contributed to Berkeley by
      8   1.1        ws  * Scooter Morris at Genentech Inc.
      9   1.1        ws  *
     10   1.1        ws  * Redistribution and use in source and binary forms, with or without
     11   1.1        ws  * modification, are permitted provided that the following conditions
     12   1.1        ws  * are met:
     13   1.1        ws  * 1. Redistributions of source code must retain the above copyright
     14   1.1        ws  *    notice, this list of conditions and the following disclaimer.
     15   1.1        ws  * 2. Redistributions in binary form must reproduce the above copyright
     16   1.1        ws  *    notice, this list of conditions and the following disclaimer in the
     17   1.1        ws  *    documentation and/or other materials provided with the distribution.
     18   1.1        ws  * 3. All advertising materials mentioning features or use of this software
     19   1.1        ws  *    must display the following acknowledgement:
     20   1.1        ws  *	This product includes software developed by the University of
     21   1.1        ws  *	California, Berkeley and its contributors.
     22   1.1        ws  * 4. Neither the name of the University nor the names of its contributors
     23   1.1        ws  *    may be used to endorse or promote products derived from this software
     24   1.1        ws  *    without specific prior written permission.
     25   1.1        ws  *
     26   1.1        ws  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     27   1.1        ws  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     28   1.1        ws  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     29   1.1        ws  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     30   1.1        ws  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     31   1.1        ws  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     32   1.1        ws  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     33   1.1        ws  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     34   1.1        ws  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     35   1.1        ws  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     36   1.1        ws  * SUCH DAMAGE.
     37   1.1        ws  *
     38  1.12      fvdl  *	@(#)ufs_lockf.c	8.4 (Berkeley) 10/26/94
     39   1.1        ws  */
     40  1.18     lukem 
     41  1.18     lukem #include <sys/cdefs.h>
     42  1.25      yamt __KERNEL_RCSID(0, "$NetBSD: vfs_lockf.c,v 1.25 2003/05/01 13:06:59 yamt Exp $");
     43   1.1        ws 
     44   1.1        ws #include <sys/param.h>
     45   1.1        ws #include <sys/systm.h>
     46   1.1        ws #include <sys/kernel.h>
     47   1.1        ws #include <sys/file.h>
     48   1.1        ws #include <sys/proc.h>
     49   1.1        ws #include <sys/vnode.h>
     50   1.1        ws #include <sys/malloc.h>
     51   1.1        ws #include <sys/fcntl.h>
     52   1.1        ws #include <sys/lockf.h>
     53  1.22   thorpej 
     54  1.22   thorpej MALLOC_DEFINE(M_LOCKF, "lockf", "Byte-range locking structures");
     55   1.1        ws 
     56   1.1        ws /*
     57   1.6   mycroft  * This variable controls the maximum number of processes that will
     58   1.6   mycroft  * be checked in doing deadlock detection.
     59   1.6   mycroft  */
     60   1.6   mycroft int maxlockdepth = MAXDEPTH;
     61   1.6   mycroft 
     62   1.6   mycroft #ifdef LOCKF_DEBUG
     63   1.6   mycroft int	lockf_debug = 0;
     64   1.6   mycroft #endif
     65   1.6   mycroft 
     66   1.6   mycroft #define NOLOCKF (struct lockf *)0
     67   1.6   mycroft #define SELF	0x1
     68   1.6   mycroft #define OTHERS	0x2
     69   1.6   mycroft 
     70  1.25      yamt static int lf_clearlock(struct lockf *);
     71  1.25      yamt static int lf_findoverlap(struct lockf *,
     72  1.25      yamt 	    struct lockf *, int, struct lockf ***, struct lockf **);
     73  1.25      yamt static struct lockf *lf_getblock(struct lockf *);
     74  1.25      yamt static int lf_getlock(struct lockf *, struct flock *);
     75  1.25      yamt static int lf_setlock(struct lockf *);
     76  1.25      yamt static void lf_split(struct lockf *, struct lockf *);
     77  1.25      yamt static void lf_wakelock(struct lockf *);
     78  1.24      yamt 
     79  1.24      yamt #ifdef LOCKF_DEBUG
     80  1.25      yamt static void lf_print(char *, struct lockf *);
     81  1.25      yamt static void lf_printlist(char *, struct lockf *);
     82  1.24      yamt #endif
     83  1.24      yamt 
     84   1.6   mycroft /*
     85  1.16  sommerfe  * XXX TODO
     86  1.16  sommerfe  * Misc cleanups: "caddr_t id" should be visible in the API as a
     87  1.16  sommerfe  * "struct proc *".
     88  1.16  sommerfe  * (This requires rototilling all VFS's which support advisory locking).
     89  1.16  sommerfe  *
     90  1.16  sommerfe  * Use pools for lock allocation.
     91  1.16  sommerfe  */
     92  1.16  sommerfe 
     93  1.16  sommerfe /*
     94  1.16  sommerfe  * XXXSMP TODO: Using either (a) a global lock, or (b) the vnode's
     95  1.16  sommerfe  * interlock should be sufficient; (b) requires a change to the API
     96  1.16  sommerfe  * because the vnode isn't visible here.
     97  1.16  sommerfe  *
     98  1.16  sommerfe  * If there's a lot of lock contention on a single vnode, locking
     99  1.16  sommerfe  * schemes which allow for more paralleism would be needed.  Given how
    100  1.16  sommerfe  * infrequently byte-range locks are actually used in typical BSD
    101  1.16  sommerfe  * code, a more complex approach probably isn't worth it.
    102  1.16  sommerfe  */
    103  1.16  sommerfe 
    104  1.16  sommerfe /*
    105   1.4   mycroft  * Do an advisory lock operation.
    106   1.1        ws  */
    107   1.4   mycroft int
    108  1.25      yamt lf_advlock(struct vop_advlock_args *ap, struct lockf **head, off_t size)
    109   1.1        ws {
    110  1.17  jdolecek 	struct flock *fl = ap->a_fl;
    111  1.15  augustss 	struct lockf *lock;
    112   1.1        ws 	off_t start, end;
    113   1.1        ws 	int error;
    114   1.1        ws 
    115   1.1        ws 	/*
    116   1.1        ws 	 * Convert the flock structure into a start and end.
    117   1.1        ws 	 */
    118   1.1        ws 	switch (fl->l_whence) {
    119   1.1        ws 	case SEEK_SET:
    120   1.1        ws 	case SEEK_CUR:
    121   1.1        ws 		/*
    122   1.1        ws 		 * Caller is responsible for adding any necessary offset
    123   1.1        ws 		 * when SEEK_CUR is used.
    124   1.1        ws 		 */
    125   1.1        ws 		start = fl->l_start;
    126   1.1        ws 		break;
    127   1.1        ws 
    128   1.1        ws 	case SEEK_END:
    129   1.1        ws 		start = size + fl->l_start;
    130   1.1        ws 		break;
    131   1.1        ws 
    132   1.1        ws 	default:
    133   1.1        ws 		return (EINVAL);
    134   1.1        ws 	}
    135   1.1        ws 	if (start < 0)
    136   1.1        ws 		return (EINVAL);
    137  1.10    kleink 
    138  1.10    kleink 	/*
    139  1.10    kleink 	 * Avoid the common case of unlocking when inode has no locks.
    140  1.10    kleink 	 */
    141  1.10    kleink 	if (*head == (struct lockf *)0) {
    142  1.17  jdolecek 		if (ap->a_op != F_SETLK) {
    143  1.10    kleink 			fl->l_type = F_UNLCK;
    144  1.10    kleink 			return (0);
    145  1.10    kleink 		}
    146  1.10    kleink 	}
    147  1.10    kleink 
    148   1.1        ws 	if (fl->l_len == 0)
    149   1.1        ws 		end = -1;
    150   1.1        ws 	else
    151   1.1        ws 		end = start + fl->l_len - 1;
    152   1.1        ws 	/*
    153   1.4   mycroft 	 * Create the lockf structure.
    154   1.1        ws 	 */
    155  1.13     perry 	MALLOC(lock, struct lockf *, sizeof(*lock), M_LOCKF, M_WAITOK);
    156   1.1        ws 	lock->lf_start = start;
    157   1.1        ws 	lock->lf_end = end;
    158  1.21   thorpej 	/* XXX NJWLWP
    159  1.21   thorpej 	 * I don't want to make the entire VFS universe use LWPs, because
    160  1.21   thorpej 	 * they don't need them, for the most part. This is an exception,
    161  1.21   thorpej 	 * and a kluge.
    162  1.21   thorpej 	 */
    163  1.21   thorpej 
    164   1.1        ws 	lock->lf_head = head;
    165   1.1        ws 	lock->lf_type = fl->l_type;
    166   1.1        ws 	lock->lf_next = (struct lockf *)0;
    167  1.12      fvdl 	TAILQ_INIT(&lock->lf_blkhd);
    168  1.17  jdolecek 	lock->lf_flags = ap->a_flags;
    169  1.21   thorpej 	if (lock->lf_flags & F_POSIX) {
    170  1.21   thorpej 		KASSERT(curproc == (struct proc *)ap->a_id);
    171  1.21   thorpej 	}
    172  1.23   mycroft 	lock->lf_id = (struct proc *)ap->a_id;
    173  1.23   mycroft 	lock->lf_lwp = curlwp;
    174  1.21   thorpej 
    175   1.1        ws 	/*
    176   1.1        ws 	 * Do the requested operation.
    177   1.1        ws 	 */
    178  1.17  jdolecek 	switch (ap->a_op) {
    179   1.4   mycroft 
    180   1.1        ws 	case F_SETLK:
    181   1.1        ws 		return (lf_setlock(lock));
    182   1.1        ws 
    183   1.1        ws 	case F_UNLCK:
    184   1.1        ws 		error = lf_clearlock(lock);
    185   1.1        ws 		FREE(lock, M_LOCKF);
    186   1.1        ws 		return (error);
    187   1.1        ws 
    188   1.1        ws 	case F_GETLK:
    189   1.1        ws 		error = lf_getlock(lock, fl);
    190   1.1        ws 		FREE(lock, M_LOCKF);
    191   1.1        ws 		return (error);
    192   1.4   mycroft 
    193   1.1        ws 	default:
    194   1.4   mycroft 		FREE(lock, M_LOCKF);
    195   1.1        ws 		return (EINVAL);
    196   1.1        ws 	}
    197   1.1        ws 	/* NOTREACHED */
    198   1.1        ws }
    199   1.1        ws 
    200   1.1        ws /*
    201   1.1        ws  * Set a byte-range lock.
    202   1.1        ws  */
    203  1.24      yamt static int
    204  1.25      yamt lf_setlock(struct lockf *lock)
    205   1.1        ws {
    206  1.15  augustss 	struct lockf *block;
    207   1.1        ws 	struct lockf **head = lock->lf_head;
    208   1.1        ws 	struct lockf **prev, *overlap, *ltmp;
    209   1.1        ws 	static char lockstr[] = "lockf";
    210   1.1        ws 	int ovcase, priority, needtolink, error;
    211   1.1        ws 
    212   1.1        ws #ifdef LOCKF_DEBUG
    213   1.1        ws 	if (lockf_debug & 1)
    214   1.1        ws 		lf_print("lf_setlock", lock);
    215   1.1        ws #endif /* LOCKF_DEBUG */
    216   1.1        ws 
    217   1.1        ws 	/*
    218   1.1        ws 	 * Set the priority
    219   1.1        ws 	 */
    220   1.1        ws 	priority = PLOCK;
    221   1.1        ws 	if (lock->lf_type == F_WRLCK)
    222   1.1        ws 		priority += 4;
    223   1.1        ws 	priority |= PCATCH;
    224   1.1        ws 	/*
    225   1.1        ws 	 * Scan lock list for this file looking for locks that would block us.
    226   1.1        ws 	 */
    227   1.7  christos 	while ((block = lf_getblock(lock)) != NULL) {
    228   1.1        ws 		/*
    229   1.1        ws 		 * Free the structure and return if nonblocking.
    230   1.1        ws 		 */
    231   1.1        ws 		if ((lock->lf_flags & F_WAIT) == 0) {
    232   1.1        ws 			FREE(lock, M_LOCKF);
    233   1.1        ws 			return (EAGAIN);
    234   1.1        ws 		}
    235   1.1        ws 		/*
    236   1.1        ws 		 * We are blocked. Since flock style locks cover
    237   1.1        ws 		 * the whole file, there is no chance for deadlock.
    238   1.1        ws 		 * For byte-range locks we must check for deadlock.
    239   1.1        ws 		 *
    240   1.1        ws 		 * Deadlock detection is done by looking through the
    241   1.1        ws 		 * wait channels to see if there are any cycles that
    242   1.1        ws 		 * involve us. MAXDEPTH is set just to make sure we
    243  1.16  sommerfe 		 * do not go off into neverneverland.
    244   1.1        ws 		 */
    245   1.1        ws 		if ((lock->lf_flags & F_POSIX) &&
    246   1.1        ws 		    (block->lf_flags & F_POSIX)) {
    247  1.21   thorpej 			struct lwp *wlwp;
    248  1.15  augustss 			struct lockf *waitblock;
    249   1.1        ws 			int i = 0;
    250   1.1        ws 
    251  1.23   mycroft 			/*
    252  1.23   mycroft 			 * The block is waiting on something.  if_lwp will be
    253  1.23   mycroft 			 * 0 once the lock is granted, so we terminate the
    254  1.23   mycroft 			 * loop if we find this.
    255  1.23   mycroft 			 */
    256  1.23   mycroft 			wlwp = block->lf_lwp;
    257  1.23   mycroft 			while (wlwp && (i++ < maxlockdepth)) {
    258  1.21   thorpej 				waitblock = (struct lockf *)wlwp->l_wchan;
    259   1.1        ws 				/* Get the owner of the blocking lock */
    260   1.1        ws 				waitblock = waitblock->lf_next;
    261   1.1        ws 				if ((waitblock->lf_flags & F_POSIX) == 0)
    262   1.1        ws 					break;
    263  1.23   mycroft 				wlwp = waitblock->lf_lwp;
    264  1.23   mycroft 				if (wlwp == lock->lf_lwp) {
    265   1.1        ws 					free(lock, M_LOCKF);
    266   1.1        ws 					return (EDEADLK);
    267   1.1        ws 				}
    268   1.1        ws 			}
    269  1.16  sommerfe 			/*
    270  1.16  sommerfe 			 * If we're still following a dependancy chain
    271  1.16  sommerfe 			 * after maxlockdepth iterations, assume we're in
    272  1.16  sommerfe 			 * a cycle to be safe.
    273  1.16  sommerfe 			 */
    274  1.16  sommerfe 			if (i >= maxlockdepth) {
    275  1.16  sommerfe 				free(lock, M_LOCKF);
    276  1.16  sommerfe 				return (EDEADLK);
    277  1.16  sommerfe 			}
    278   1.1        ws 		}
    279   1.1        ws 		/*
    280   1.1        ws 		 * For flock type locks, we must first remove
    281   1.1        ws 		 * any shared locks that we hold before we sleep
    282   1.1        ws 		 * waiting for an exclusive lock.
    283   1.1        ws 		 */
    284   1.1        ws 		if ((lock->lf_flags & F_FLOCK) &&
    285   1.1        ws 		    lock->lf_type == F_WRLCK) {
    286   1.1        ws 			lock->lf_type = F_UNLCK;
    287   1.1        ws 			(void) lf_clearlock(lock);
    288   1.1        ws 			lock->lf_type = F_WRLCK;
    289   1.1        ws 		}
    290   1.1        ws 		/*
    291   1.1        ws 		 * Add our lock to the blocked list and sleep until we're free.
    292   1.1        ws 		 * Remember who blocked us (for deadlock detection).
    293   1.1        ws 		 */
    294   1.1        ws 		lock->lf_next = block;
    295  1.12      fvdl 		TAILQ_INSERT_TAIL(&block->lf_blkhd, lock, lf_block);
    296   1.1        ws #ifdef LOCKF_DEBUG
    297   1.1        ws 		if (lockf_debug & 1) {
    298   1.1        ws 			lf_print("lf_setlock: blocking on", block);
    299   1.1        ws 			lf_printlist("lf_setlock", block);
    300   1.1        ws 		}
    301   1.1        ws #endif /* LOCKF_DEBUG */
    302   1.7  christos 		error = tsleep((caddr_t)lock, priority, lockstr, 0);
    303  1.16  sommerfe 
    304  1.16  sommerfe 		/*
    305  1.16  sommerfe 		 * We may have been awakened by a signal (in
    306  1.16  sommerfe 		 * which case we must remove ourselves from the
    307  1.16  sommerfe 		 * blocked list) and/or by another process
    308  1.16  sommerfe 		 * releasing a lock (in which case we have already
    309  1.16  sommerfe 		 * been removed from the blocked list and our
    310  1.16  sommerfe 		 * lf_next field set to NOLOCKF).
    311  1.16  sommerfe 		 */
    312  1.16  sommerfe 		if (lock->lf_next != NOLOCKF) {
    313  1.16  sommerfe 			TAILQ_REMOVE(&lock->lf_next->lf_blkhd, lock, lf_block);
    314  1.16  sommerfe 			lock->lf_next = NOLOCKF;
    315  1.16  sommerfe 		}
    316   1.7  christos 		if (error) {
    317   1.4   mycroft 			free(lock, M_LOCKF);
    318   1.4   mycroft 			return (error);
    319   1.1        ws 		}
    320   1.1        ws 	}
    321   1.1        ws 	/*
    322   1.1        ws 	 * No blocks!!  Add the lock.  Note that we will
    323   1.1        ws 	 * downgrade or upgrade any overlapping locks this
    324   1.1        ws 	 * process already owns.
    325   1.1        ws 	 *
    326   1.1        ws 	 * Skip over locks owned by other processes.
    327   1.1        ws 	 * Handle any locks that overlap and are owned by ourselves.
    328   1.1        ws 	 */
    329  1.23   mycroft 	lock->lf_lwp = 0;
    330   1.1        ws 	prev = head;
    331   1.1        ws 	block = *head;
    332   1.1        ws 	needtolink = 1;
    333   1.1        ws 	for (;;) {
    334   1.7  christos 		ovcase = lf_findoverlap(block, lock, SELF, &prev, &overlap);
    335   1.7  christos 		if (ovcase)
    336   1.1        ws 			block = overlap->lf_next;
    337   1.1        ws 		/*
    338   1.1        ws 		 * Six cases:
    339   1.1        ws 		 *	0) no overlap
    340   1.1        ws 		 *	1) overlap == lock
    341   1.1        ws 		 *	2) overlap contains lock
    342   1.1        ws 		 *	3) lock contains overlap
    343   1.1        ws 		 *	4) overlap starts before lock
    344   1.1        ws 		 *	5) overlap ends after lock
    345   1.1        ws 		 */
    346   1.1        ws 		switch (ovcase) {
    347   1.1        ws 		case 0: /* no overlap */
    348   1.1        ws 			if (needtolink) {
    349   1.1        ws 				*prev = lock;
    350   1.1        ws 				lock->lf_next = overlap;
    351   1.1        ws 			}
    352   1.1        ws 			break;
    353   1.1        ws 
    354   1.1        ws 		case 1: /* overlap == lock */
    355   1.1        ws 			/*
    356   1.1        ws 			 * If downgrading lock, others may be
    357   1.1        ws 			 * able to acquire it.
    358   1.1        ws 			 */
    359   1.1        ws 			if (lock->lf_type == F_RDLCK &&
    360   1.1        ws 			    overlap->lf_type == F_WRLCK)
    361   1.1        ws 				lf_wakelock(overlap);
    362   1.1        ws 			overlap->lf_type = lock->lf_type;
    363   1.1        ws 			FREE(lock, M_LOCKF);
    364   1.1        ws 			lock = overlap; /* for debug output below */
    365   1.1        ws 			break;
    366   1.1        ws 
    367   1.1        ws 		case 2: /* overlap contains lock */
    368   1.1        ws 			/*
    369   1.1        ws 			 * Check for common starting point and different types.
    370   1.1        ws 			 */
    371   1.1        ws 			if (overlap->lf_type == lock->lf_type) {
    372   1.1        ws 				free(lock, M_LOCKF);
    373   1.1        ws 				lock = overlap; /* for debug output below */
    374   1.1        ws 				break;
    375   1.1        ws 			}
    376   1.1        ws 			if (overlap->lf_start == lock->lf_start) {
    377   1.1        ws 				*prev = lock;
    378   1.1        ws 				lock->lf_next = overlap;
    379   1.1        ws 				overlap->lf_start = lock->lf_end + 1;
    380   1.1        ws 			} else
    381   1.1        ws 				lf_split(overlap, lock);
    382   1.1        ws 			lf_wakelock(overlap);
    383   1.1        ws 			break;
    384   1.1        ws 
    385   1.1        ws 		case 3: /* lock contains overlap */
    386   1.1        ws 			/*
    387   1.1        ws 			 * If downgrading lock, others may be able to
    388   1.1        ws 			 * acquire it, otherwise take the list.
    389   1.1        ws 			 */
    390   1.1        ws 			if (lock->lf_type == F_RDLCK &&
    391   1.1        ws 			    overlap->lf_type == F_WRLCK) {
    392   1.1        ws 				lf_wakelock(overlap);
    393   1.1        ws 			} else {
    394  1.19      matt 				while ((ltmp = TAILQ_FIRST(&overlap->lf_blkhd))) {
    395  1.16  sommerfe 					KASSERT(ltmp->lf_next == overlap);
    396  1.12      fvdl 					TAILQ_REMOVE(&overlap->lf_blkhd, ltmp,
    397  1.12      fvdl 					    lf_block);
    398  1.16  sommerfe 					ltmp->lf_next = lock;
    399  1.12      fvdl 					TAILQ_INSERT_TAIL(&lock->lf_blkhd,
    400  1.12      fvdl 					    ltmp, lf_block);
    401  1.12      fvdl 				}
    402   1.1        ws 			}
    403   1.1        ws 			/*
    404   1.1        ws 			 * Add the new lock if necessary and delete the overlap.
    405   1.1        ws 			 */
    406   1.1        ws 			if (needtolink) {
    407   1.1        ws 				*prev = lock;
    408   1.1        ws 				lock->lf_next = overlap->lf_next;
    409   1.1        ws 				prev = &lock->lf_next;
    410   1.1        ws 				needtolink = 0;
    411   1.1        ws 			} else
    412   1.1        ws 				*prev = overlap->lf_next;
    413   1.1        ws 			free(overlap, M_LOCKF);
    414   1.1        ws 			continue;
    415   1.1        ws 
    416   1.1        ws 		case 4: /* overlap starts before lock */
    417   1.1        ws 			/*
    418   1.1        ws 			 * Add lock after overlap on the list.
    419   1.1        ws 			 */
    420   1.1        ws 			lock->lf_next = overlap->lf_next;
    421   1.1        ws 			overlap->lf_next = lock;
    422   1.1        ws 			overlap->lf_end = lock->lf_start - 1;
    423   1.1        ws 			prev = &lock->lf_next;
    424   1.1        ws 			lf_wakelock(overlap);
    425   1.1        ws 			needtolink = 0;
    426   1.1        ws 			continue;
    427   1.1        ws 
    428   1.1        ws 		case 5: /* overlap ends after lock */
    429   1.1        ws 			/*
    430   1.1        ws 			 * Add the new lock before overlap.
    431   1.1        ws 			 */
    432   1.1        ws 			if (needtolink) {
    433   1.1        ws 				*prev = lock;
    434   1.1        ws 				lock->lf_next = overlap;
    435   1.1        ws 			}
    436   1.1        ws 			overlap->lf_start = lock->lf_end + 1;
    437   1.1        ws 			lf_wakelock(overlap);
    438   1.1        ws 			break;
    439   1.1        ws 		}
    440   1.1        ws 		break;
    441   1.1        ws 	}
    442   1.1        ws #ifdef LOCKF_DEBUG
    443   1.1        ws 	if (lockf_debug & 1) {
    444   1.1        ws 		lf_print("lf_setlock: got the lock", lock);
    445   1.1        ws 		lf_printlist("lf_setlock", lock);
    446   1.1        ws 	}
    447   1.1        ws #endif /* LOCKF_DEBUG */
    448   1.1        ws 	return (0);
    449   1.1        ws }
    450   1.1        ws 
    451   1.1        ws /*
    452   1.1        ws  * Remove a byte-range lock on an inode.
    453   1.1        ws  *
    454   1.1        ws  * Generally, find the lock (or an overlap to that lock)
    455   1.1        ws  * and remove it (or shrink it), then wakeup anyone we can.
    456   1.1        ws  */
    457  1.24      yamt static int
    458  1.25      yamt lf_clearlock(struct lockf *unlock)
    459   1.1        ws {
    460   1.1        ws 	struct lockf **head = unlock->lf_head;
    461  1.15  augustss 	struct lockf *lf = *head;
    462   1.1        ws 	struct lockf *overlap, **prev;
    463   1.1        ws 	int ovcase;
    464   1.1        ws 
    465   1.1        ws 	if (lf == NOLOCKF)
    466   1.1        ws 		return (0);
    467   1.1        ws #ifdef LOCKF_DEBUG
    468   1.1        ws 	if (unlock->lf_type != F_UNLCK)
    469   1.1        ws 		panic("lf_clearlock: bad type");
    470   1.1        ws 	if (lockf_debug & 1)
    471   1.1        ws 		lf_print("lf_clearlock", unlock);
    472   1.1        ws #endif /* LOCKF_DEBUG */
    473   1.1        ws 	prev = head;
    474   1.7  christos 	while ((ovcase = lf_findoverlap(lf, unlock, SELF,
    475   1.7  christos 					&prev, &overlap)) != 0) {
    476   1.1        ws 		/*
    477   1.1        ws 		 * Wakeup the list of locks to be retried.
    478   1.1        ws 		 */
    479   1.1        ws 		lf_wakelock(overlap);
    480   1.1        ws 
    481   1.1        ws 		switch (ovcase) {
    482   1.1        ws 
    483   1.1        ws 		case 1: /* overlap == lock */
    484   1.1        ws 			*prev = overlap->lf_next;
    485   1.1        ws 			FREE(overlap, M_LOCKF);
    486   1.1        ws 			break;
    487   1.1        ws 
    488   1.1        ws 		case 2: /* overlap contains lock: split it */
    489   1.1        ws 			if (overlap->lf_start == unlock->lf_start) {
    490   1.1        ws 				overlap->lf_start = unlock->lf_end + 1;
    491   1.1        ws 				break;
    492   1.1        ws 			}
    493   1.1        ws 			lf_split(overlap, unlock);
    494   1.1        ws 			overlap->lf_next = unlock->lf_next;
    495   1.1        ws 			break;
    496   1.1        ws 
    497   1.1        ws 		case 3: /* lock contains overlap */
    498   1.1        ws 			*prev = overlap->lf_next;
    499   1.1        ws 			lf = overlap->lf_next;
    500   1.1        ws 			free(overlap, M_LOCKF);
    501   1.1        ws 			continue;
    502   1.1        ws 
    503   1.1        ws 		case 4: /* overlap starts before lock */
    504   1.1        ws 			overlap->lf_end = unlock->lf_start - 1;
    505   1.1        ws 			prev = &overlap->lf_next;
    506   1.1        ws 			lf = overlap->lf_next;
    507   1.1        ws 			continue;
    508   1.1        ws 
    509   1.1        ws 		case 5: /* overlap ends after lock */
    510   1.1        ws 			overlap->lf_start = unlock->lf_end + 1;
    511   1.1        ws 			break;
    512   1.1        ws 		}
    513   1.1        ws 		break;
    514   1.1        ws 	}
    515   1.1        ws #ifdef LOCKF_DEBUG
    516   1.1        ws 	if (lockf_debug & 1)
    517   1.1        ws 		lf_printlist("lf_clearlock", unlock);
    518   1.1        ws #endif /* LOCKF_DEBUG */
    519   1.1        ws 	return (0);
    520   1.1        ws }
    521   1.1        ws 
    522   1.1        ws /*
    523   1.1        ws  * Check whether there is a blocking lock,
    524   1.1        ws  * and if so return its process identifier.
    525   1.1        ws  */
    526  1.24      yamt static int
    527  1.25      yamt lf_getlock(struct lockf *lock, struct flock *fl)
    528   1.1        ws {
    529  1.15  augustss 	struct lockf *block;
    530   1.1        ws 
    531   1.1        ws #ifdef LOCKF_DEBUG
    532   1.1        ws 	if (lockf_debug & 1)
    533   1.1        ws 		lf_print("lf_getlock", lock);
    534   1.1        ws #endif /* LOCKF_DEBUG */
    535   1.1        ws 
    536   1.7  christos 	if ((block = lf_getblock(lock)) != NULL) {
    537   1.1        ws 		fl->l_type = block->lf_type;
    538   1.1        ws 		fl->l_whence = SEEK_SET;
    539   1.1        ws 		fl->l_start = block->lf_start;
    540   1.1        ws 		if (block->lf_end == -1)
    541   1.1        ws 			fl->l_len = 0;
    542   1.1        ws 		else
    543   1.1        ws 			fl->l_len = block->lf_end - block->lf_start + 1;
    544   1.1        ws 		if (block->lf_flags & F_POSIX)
    545  1.23   mycroft 			fl->l_pid = ((struct proc *)block->lf_id)->p_pid;
    546   1.1        ws 		else
    547   1.1        ws 			fl->l_pid = -1;
    548   1.1        ws 	} else {
    549   1.1        ws 		fl->l_type = F_UNLCK;
    550   1.1        ws 	}
    551   1.1        ws 	return (0);
    552   1.1        ws }
    553   1.1        ws 
    554   1.1        ws /*
    555   1.1        ws  * Walk the list of locks for an inode and
    556   1.1        ws  * return the first blocking lock.
    557   1.1        ws  */
    558  1.24      yamt static struct lockf *
    559  1.25      yamt lf_getblock(struct lockf *lock)
    560   1.1        ws {
    561   1.1        ws 	struct lockf **prev, *overlap, *lf = *(lock->lf_head);
    562   1.1        ws 
    563   1.1        ws 	prev = lock->lf_head;
    564  1.20    simonb 	while (lf_findoverlap(lf, lock, OTHERS, &prev, &overlap) != 0) {
    565   1.1        ws 		/*
    566   1.1        ws 		 * We've found an overlap, see if it blocks us
    567   1.1        ws 		 */
    568   1.1        ws 		if ((lock->lf_type == F_WRLCK || overlap->lf_type == F_WRLCK))
    569   1.1        ws 			return (overlap);
    570   1.1        ws 		/*
    571   1.1        ws 		 * Nope, point to the next one on the list and
    572   1.1        ws 		 * see if it blocks us
    573   1.1        ws 		 */
    574   1.1        ws 		lf = overlap->lf_next;
    575   1.1        ws 	}
    576   1.1        ws 	return (NOLOCKF);
    577   1.1        ws }
    578   1.1        ws 
    579   1.1        ws /*
    580   1.1        ws  * Walk the list of locks for an inode to
    581   1.1        ws  * find an overlapping lock (if any).
    582   1.1        ws  *
    583   1.1        ws  * NOTE: this returns only the FIRST overlapping lock.  There
    584   1.1        ws  *	 may be more than one.
    585   1.1        ws  */
    586  1.24      yamt static int
    587  1.25      yamt lf_findoverlap(struct lockf *lf, struct lockf *lock, int type,
    588  1.25      yamt     struct lockf ***prev, struct lockf **overlap)
    589   1.1        ws {
    590   1.1        ws 	off_t start, end;
    591   1.1        ws 
    592   1.1        ws 	*overlap = lf;
    593   1.1        ws 	if (lf == NOLOCKF)
    594   1.1        ws 		return (0);
    595   1.1        ws #ifdef LOCKF_DEBUG
    596   1.1        ws 	if (lockf_debug & 2)
    597   1.1        ws 		lf_print("lf_findoverlap: looking for overlap in", lock);
    598   1.1        ws #endif /* LOCKF_DEBUG */
    599   1.1        ws 	start = lock->lf_start;
    600   1.1        ws 	end = lock->lf_end;
    601   1.1        ws 	while (lf != NOLOCKF) {
    602  1.23   mycroft 		if (((type == SELF) && lf->lf_id != lock->lf_id) ||
    603  1.23   mycroft 		    ((type == OTHERS) && lf->lf_id == lock->lf_id)) {
    604   1.1        ws 			*prev = &lf->lf_next;
    605   1.1        ws 			*overlap = lf = lf->lf_next;
    606   1.1        ws 			continue;
    607   1.1        ws 		}
    608   1.1        ws #ifdef LOCKF_DEBUG
    609   1.1        ws 		if (lockf_debug & 2)
    610   1.1        ws 			lf_print("\tchecking", lf);
    611   1.1        ws #endif /* LOCKF_DEBUG */
    612   1.1        ws 		/*
    613   1.1        ws 		 * OK, check for overlap
    614   1.1        ws 		 *
    615   1.1        ws 		 * Six cases:
    616   1.1        ws 		 *	0) no overlap
    617   1.1        ws 		 *	1) overlap == lock
    618   1.1        ws 		 *	2) overlap contains lock
    619   1.1        ws 		 *	3) lock contains overlap
    620   1.1        ws 		 *	4) overlap starts before lock
    621   1.1        ws 		 *	5) overlap ends after lock
    622   1.1        ws 		 */
    623   1.1        ws 		if ((lf->lf_end != -1 && start > lf->lf_end) ||
    624   1.1        ws 		    (end != -1 && lf->lf_start > end)) {
    625   1.1        ws 			/* Case 0 */
    626   1.1        ws #ifdef LOCKF_DEBUG
    627   1.1        ws 			if (lockf_debug & 2)
    628   1.9  christos 				printf("no overlap\n");
    629   1.1        ws #endif /* LOCKF_DEBUG */
    630   1.1        ws 			if ((type & SELF) && end != -1 && lf->lf_start > end)
    631   1.1        ws 				return (0);
    632   1.1        ws 			*prev = &lf->lf_next;
    633   1.1        ws 			*overlap = lf = lf->lf_next;
    634   1.1        ws 			continue;
    635   1.1        ws 		}
    636   1.1        ws 		if ((lf->lf_start == start) && (lf->lf_end == end)) {
    637   1.1        ws 			/* Case 1 */
    638   1.1        ws #ifdef LOCKF_DEBUG
    639   1.1        ws 			if (lockf_debug & 2)
    640   1.9  christos 				printf("overlap == lock\n");
    641   1.1        ws #endif /* LOCKF_DEBUG */
    642   1.1        ws 			return (1);
    643   1.1        ws 		}
    644   1.1        ws 		if ((lf->lf_start <= start) &&
    645   1.1        ws 		    (end != -1) &&
    646   1.1        ws 		    ((lf->lf_end >= end) || (lf->lf_end == -1))) {
    647   1.1        ws 			/* Case 2 */
    648   1.1        ws #ifdef LOCKF_DEBUG
    649   1.1        ws 			if (lockf_debug & 2)
    650   1.9  christos 				printf("overlap contains lock\n");
    651   1.1        ws #endif /* LOCKF_DEBUG */
    652   1.1        ws 			return (2);
    653   1.1        ws 		}
    654   1.1        ws 		if (start <= lf->lf_start &&
    655   1.4   mycroft 		           (end == -1 ||
    656   1.1        ws 			   (lf->lf_end != -1 && end >= lf->lf_end))) {
    657   1.1        ws 			/* Case 3 */
    658   1.1        ws #ifdef LOCKF_DEBUG
    659   1.1        ws 			if (lockf_debug & 2)
    660   1.9  christos 				printf("lock contains overlap\n");
    661   1.1        ws #endif /* LOCKF_DEBUG */
    662   1.1        ws 			return (3);
    663   1.1        ws 		}
    664   1.1        ws 		if ((lf->lf_start < start) &&
    665   1.1        ws 			((lf->lf_end >= start) || (lf->lf_end == -1))) {
    666   1.1        ws 			/* Case 4 */
    667   1.1        ws #ifdef LOCKF_DEBUG
    668   1.1        ws 			if (lockf_debug & 2)
    669   1.9  christos 				printf("overlap starts before lock\n");
    670   1.1        ws #endif /* LOCKF_DEBUG */
    671   1.1        ws 			return (4);
    672   1.1        ws 		}
    673   1.1        ws 		if ((lf->lf_start > start) &&
    674   1.1        ws 			(end != -1) &&
    675   1.1        ws 			((lf->lf_end > end) || (lf->lf_end == -1))) {
    676   1.1        ws 			/* Case 5 */
    677   1.1        ws #ifdef LOCKF_DEBUG
    678   1.1        ws 			if (lockf_debug & 2)
    679   1.9  christos 				printf("overlap ends after lock\n");
    680   1.1        ws #endif /* LOCKF_DEBUG */
    681   1.1        ws 			return (5);
    682   1.1        ws 		}
    683   1.1        ws 		panic("lf_findoverlap: default");
    684   1.1        ws 	}
    685   1.1        ws 	return (0);
    686   1.1        ws }
    687   1.1        ws 
    688   1.1        ws /*
    689   1.1        ws  * Split a lock and a contained region into
    690   1.1        ws  * two or three locks as necessary.
    691   1.1        ws  */
    692  1.24      yamt static void
    693  1.25      yamt lf_split(struct lockf *lock1, struct lockf *lock2)
    694   1.1        ws {
    695  1.15  augustss 	struct lockf *splitlock;
    696   1.1        ws 
    697   1.1        ws #ifdef LOCKF_DEBUG
    698   1.1        ws 	if (lockf_debug & 2) {
    699   1.1        ws 		lf_print("lf_split", lock1);
    700   1.1        ws 		lf_print("splitting from", lock2);
    701   1.1        ws 	}
    702   1.1        ws #endif /* LOCKF_DEBUG */
    703   1.1        ws 	/*
    704   1.1        ws 	 * Check to see if spliting into only two pieces.
    705   1.1        ws 	 */
    706   1.1        ws 	if (lock1->lf_start == lock2->lf_start) {
    707   1.1        ws 		lock1->lf_start = lock2->lf_end + 1;
    708   1.1        ws 		lock2->lf_next = lock1;
    709   1.1        ws 		return;
    710   1.1        ws 	}
    711   1.1        ws 	if (lock1->lf_end == lock2->lf_end) {
    712   1.1        ws 		lock1->lf_end = lock2->lf_start - 1;
    713   1.1        ws 		lock2->lf_next = lock1->lf_next;
    714   1.1        ws 		lock1->lf_next = lock2;
    715   1.1        ws 		return;
    716   1.1        ws 	}
    717   1.1        ws 	/*
    718   1.1        ws 	 * Make a new lock consisting of the last part of
    719   1.1        ws 	 * the encompassing lock
    720   1.1        ws 	 */
    721  1.13     perry 	MALLOC(splitlock, struct lockf *, sizeof(*splitlock), M_LOCKF, M_WAITOK);
    722  1.14     perry 	memcpy((caddr_t)splitlock, (caddr_t)lock1, sizeof(*splitlock));
    723   1.1        ws 	splitlock->lf_start = lock2->lf_end + 1;
    724  1.12      fvdl 	TAILQ_INIT(&splitlock->lf_blkhd);
    725   1.1        ws 	lock1->lf_end = lock2->lf_start - 1;
    726   1.1        ws 	/*
    727   1.1        ws 	 * OK, now link it in
    728   1.1        ws 	 */
    729   1.1        ws 	splitlock->lf_next = lock1->lf_next;
    730   1.1        ws 	lock2->lf_next = splitlock;
    731   1.1        ws 	lock1->lf_next = lock2;
    732   1.1        ws }
    733   1.1        ws 
    734   1.1        ws /*
    735   1.1        ws  * Wakeup a blocklist
    736   1.1        ws  */
    737  1.24      yamt static void
    738  1.25      yamt lf_wakelock(struct lockf *listhead)
    739   1.1        ws {
    740  1.15  augustss 	struct lockf *wakelock;
    741   1.1        ws 
    742  1.19      matt 	while ((wakelock = TAILQ_FIRST(&listhead->lf_blkhd))) {
    743  1.16  sommerfe 		KASSERT(wakelock->lf_next == listhead);
    744  1.12      fvdl 		TAILQ_REMOVE(&listhead->lf_blkhd, wakelock, lf_block);
    745   1.1        ws 		wakelock->lf_next = NOLOCKF;
    746   1.1        ws #ifdef LOCKF_DEBUG
    747   1.1        ws 		if (lockf_debug & 2)
    748   1.1        ws 			lf_print("lf_wakelock: awakening", wakelock);
    749  1.12      fvdl #endif
    750   1.1        ws 		wakeup((caddr_t)wakelock);
    751   1.1        ws 	}
    752   1.1        ws }
    753   1.1        ws 
    754   1.1        ws #ifdef LOCKF_DEBUG
    755   1.1        ws /*
    756   1.1        ws  * Print out a lock.
    757   1.1        ws  */
    758  1.24      yamt static void
    759  1.25      yamt lf_print(char *tag, struct lockf *lock)
    760   1.1        ws {
    761   1.1        ws 
    762   1.9  christos 	printf("%s: lock %p for ", tag, lock);
    763   1.1        ws 	if (lock->lf_flags & F_POSIX)
    764  1.23   mycroft 		printf("proc %d", ((struct proc *)lock->lf_id)->p_pid);
    765   1.1        ws 	else
    766  1.23   mycroft 		printf("file 0x%p", (struct file *)lock->lf_id);
    767  1.11       jtk 	printf(" %s, start %qx, end %qx",
    768   1.1        ws 		lock->lf_type == F_RDLCK ? "shared" :
    769   1.1        ws 		lock->lf_type == F_WRLCK ? "exclusive" :
    770   1.1        ws 		lock->lf_type == F_UNLCK ? "unlock" :
    771   1.1        ws 		"unknown", lock->lf_start, lock->lf_end);
    772  1.19      matt 	if (TAILQ_FIRST(&lock->lf_blkhd))
    773  1.19      matt 		printf(" block %p\n", TAILQ_FIRST(&lock->lf_blkhd));
    774   1.1        ws 	else
    775   1.9  christos 		printf("\n");
    776   1.1        ws }
    777   1.1        ws 
    778  1.24      yamt static void
    779  1.25      yamt lf_printlist(char *tag, struct lockf *lock)
    780   1.1        ws {
    781  1.15  augustss 	struct lockf *lf, *blk;
    782   1.1        ws 
    783  1.11       jtk 	printf("%s: Lock list:\n", tag);
    784  1.12      fvdl 	for (lf = *lock->lf_head; lf; lf = lf->lf_next) {
    785   1.9  christos 		printf("\tlock %p for ", lf);
    786   1.1        ws 		if (lf->lf_flags & F_POSIX)
    787  1.23   mycroft 			printf("proc %d", ((struct proc *)lf->lf_id)->p_pid);
    788   1.1        ws 		else
    789  1.23   mycroft 			printf("file 0x%p", (struct file *)lf->lf_id);
    790  1.11       jtk 		printf(", %s, start %qx, end %qx",
    791   1.1        ws 			lf->lf_type == F_RDLCK ? "shared" :
    792   1.1        ws 			lf->lf_type == F_WRLCK ? "exclusive" :
    793   1.1        ws 			lf->lf_type == F_UNLCK ? "unlock" :
    794   1.1        ws 			"unknown", lf->lf_start, lf->lf_end);
    795  1.19      matt 		TAILQ_FOREACH(blk, &lf->lf_blkhd, lf_block) {
    796  1.12      fvdl 			if (blk->lf_flags & F_POSIX)
    797  1.12      fvdl 				printf("proc %d",
    798  1.23   mycroft 				    ((struct proc *)blk->lf_id)->p_pid);
    799  1.12      fvdl 			else
    800  1.23   mycroft 				printf("file 0x%p", (struct file *)blk->lf_id);
    801  1.12      fvdl 			printf(", %s, start %qx, end %qx",
    802  1.12      fvdl 				blk->lf_type == F_RDLCK ? "shared" :
    803  1.12      fvdl 				blk->lf_type == F_WRLCK ? "exclusive" :
    804  1.12      fvdl 				blk->lf_type == F_UNLCK ? "unlock" :
    805  1.12      fvdl 				"unknown", blk->lf_start, blk->lf_end);
    806  1.19      matt 			if (TAILQ_FIRST(&blk->lf_blkhd))
    807  1.12      fvdl 				 panic("lf_printlist: bad list");
    808  1.12      fvdl 		}
    809  1.12      fvdl 		printf("\n");
    810   1.1        ws 	}
    811   1.1        ws }
    812   1.1        ws #endif /* LOCKF_DEBUG */
    813