Home | History | Annotate | Line # | Download | only in kern
vfs_lockf.c revision 1.27
      1  1.27      yamt /*	$NetBSD: vfs_lockf.c,v 1.27 2003/05/01 14:36:43 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  1.27      yamt __KERNEL_RCSID(0, "$NetBSD: vfs_lockf.c,v 1.27 2003/05/01 14:36:43 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  1.27      yamt static int lf_clearlock(struct lockf *, 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  1.27      yamt static int lf_setlock(struct lockf *, struct lockf **, struct simplelock *);
     76  1.27      yamt static void lf_split(struct lockf *, 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  * If there's a lot of lock contention on a single vnode, locking
     95  1.16  sommerfe  * schemes which allow for more paralleism would be needed.  Given how
     96  1.16  sommerfe  * infrequently byte-range locks are actually used in typical BSD
     97  1.16  sommerfe  * code, a more complex approach probably isn't worth it.
     98  1.16  sommerfe  */
     99  1.16  sommerfe 
    100  1.16  sommerfe /*
    101   1.4   mycroft  * Do an advisory lock operation.
    102   1.1        ws  */
    103   1.4   mycroft int
    104  1.25      yamt lf_advlock(struct vop_advlock_args *ap, struct lockf **head, off_t size)
    105   1.1        ws {
    106  1.17  jdolecek 	struct flock *fl = ap->a_fl;
    107  1.27      yamt 	struct lockf *lock = NULL;
    108  1.27      yamt 	struct lockf *sparelock;
    109  1.27      yamt 	struct vnode *vp = ap->a_vp;
    110   1.1        ws 	off_t start, end;
    111   1.1        ws 	int error;
    112   1.1        ws 
    113   1.1        ws 	/*
    114   1.1        ws 	 * Convert the flock structure into a start and end.
    115   1.1        ws 	 */
    116   1.1        ws 	switch (fl->l_whence) {
    117   1.1        ws 	case SEEK_SET:
    118   1.1        ws 	case SEEK_CUR:
    119   1.1        ws 		/*
    120   1.1        ws 		 * Caller is responsible for adding any necessary offset
    121   1.1        ws 		 * when SEEK_CUR is used.
    122   1.1        ws 		 */
    123   1.1        ws 		start = fl->l_start;
    124   1.1        ws 		break;
    125   1.1        ws 
    126   1.1        ws 	case SEEK_END:
    127   1.1        ws 		start = size + fl->l_start;
    128   1.1        ws 		break;
    129   1.1        ws 
    130   1.1        ws 	default:
    131   1.1        ws 		return (EINVAL);
    132   1.1        ws 	}
    133   1.1        ws 	if (start < 0)
    134   1.1        ws 		return (EINVAL);
    135  1.10    kleink 
    136  1.10    kleink 	/*
    137  1.27      yamt 	 * allocate locks before acquire simple lock.
    138  1.27      yamt 	 * we need two locks in the worst case.
    139  1.27      yamt 	 */
    140  1.27      yamt 	switch (ap->a_op) {
    141  1.27      yamt 	case F_SETLK:
    142  1.27      yamt 	case F_UNLCK:
    143  1.27      yamt 		/*
    144  1.27      yamt 		 * XXX for F_UNLCK case, we can re-use lock.
    145  1.27      yamt 		 */
    146  1.27      yamt 		if ((fl->l_type & F_FLOCK) == 0) {
    147  1.27      yamt 			/*
    148  1.27      yamt 			 * byte-range lock might need one more lock.
    149  1.27      yamt 			 */
    150  1.27      yamt 			MALLOC(sparelock, struct lockf *, sizeof(*lock),
    151  1.27      yamt 			    M_LOCKF, M_WAITOK);
    152  1.27      yamt 			if (sparelock == NULL) {
    153  1.27      yamt 				error = ENOMEM;
    154  1.27      yamt 				goto quit;
    155  1.27      yamt 			}
    156  1.27      yamt 			break;
    157  1.27      yamt 		}
    158  1.27      yamt 		/* FALLTHROUGH */
    159  1.27      yamt 
    160  1.27      yamt 	case F_GETLK:
    161  1.27      yamt 		sparelock = NULL;
    162  1.27      yamt 		break;
    163  1.27      yamt 
    164  1.27      yamt 	default:
    165  1.27      yamt 		return (EINVAL);
    166  1.27      yamt 	}
    167  1.27      yamt 
    168  1.27      yamt 	MALLOC(lock, struct lockf *, sizeof(*lock), M_LOCKF, M_WAITOK);
    169  1.27      yamt 	if (lock == NULL) {
    170  1.27      yamt 		error = ENOMEM;
    171  1.27      yamt 		goto quit;
    172  1.27      yamt 	}
    173  1.27      yamt 
    174  1.27      yamt 	simple_lock(&vp->v_interlock);
    175  1.27      yamt 
    176  1.27      yamt 	/*
    177  1.10    kleink 	 * Avoid the common case of unlocking when inode has no locks.
    178  1.10    kleink 	 */
    179  1.10    kleink 	if (*head == (struct lockf *)0) {
    180  1.17  jdolecek 		if (ap->a_op != F_SETLK) {
    181  1.10    kleink 			fl->l_type = F_UNLCK;
    182  1.27      yamt 			error = 0;
    183  1.27      yamt 			goto quit_unlock;
    184  1.10    kleink 		}
    185  1.10    kleink 	}
    186  1.10    kleink 
    187   1.1        ws 	if (fl->l_len == 0)
    188   1.1        ws 		end = -1;
    189   1.1        ws 	else
    190   1.1        ws 		end = start + fl->l_len - 1;
    191   1.1        ws 	/*
    192   1.4   mycroft 	 * Create the lockf structure.
    193   1.1        ws 	 */
    194   1.1        ws 	lock->lf_start = start;
    195   1.1        ws 	lock->lf_end = end;
    196  1.21   thorpej 	/* XXX NJWLWP
    197  1.21   thorpej 	 * I don't want to make the entire VFS universe use LWPs, because
    198  1.21   thorpej 	 * they don't need them, for the most part. This is an exception,
    199  1.21   thorpej 	 * and a kluge.
    200  1.21   thorpej 	 */
    201  1.21   thorpej 
    202   1.1        ws 	lock->lf_head = head;
    203   1.1        ws 	lock->lf_type = fl->l_type;
    204   1.1        ws 	lock->lf_next = (struct lockf *)0;
    205  1.12      fvdl 	TAILQ_INIT(&lock->lf_blkhd);
    206  1.17  jdolecek 	lock->lf_flags = ap->a_flags;
    207  1.21   thorpej 	if (lock->lf_flags & F_POSIX) {
    208  1.21   thorpej 		KASSERT(curproc == (struct proc *)ap->a_id);
    209  1.21   thorpej 	}
    210  1.23   mycroft 	lock->lf_id = (struct proc *)ap->a_id;
    211  1.23   mycroft 	lock->lf_lwp = curlwp;
    212  1.21   thorpej 
    213   1.1        ws 	/*
    214   1.1        ws 	 * Do the requested operation.
    215   1.1        ws 	 */
    216  1.17  jdolecek 	switch (ap->a_op) {
    217   1.4   mycroft 
    218   1.1        ws 	case F_SETLK:
    219  1.27      yamt 		error = lf_setlock(lock, &sparelock, &vp->v_interlock);
    220  1.27      yamt 		lock = NULL; /* lf_setlock freed it */
    221  1.27      yamt 		break;
    222   1.1        ws 
    223   1.1        ws 	case F_UNLCK:
    224  1.27      yamt 		error = lf_clearlock(lock, &sparelock);
    225  1.27      yamt 		break;
    226   1.1        ws 
    227   1.1        ws 	case F_GETLK:
    228   1.1        ws 		error = lf_getlock(lock, fl);
    229  1.27      yamt 		break;
    230   1.4   mycroft 
    231   1.1        ws 	default:
    232  1.27      yamt 		/* NOTREACHED */
    233  1.27      yamt 	}
    234  1.27      yamt 
    235  1.27      yamt quit_unlock:
    236  1.27      yamt 	simple_unlock(&vp->v_interlock);
    237  1.27      yamt quit:
    238  1.27      yamt 	if (lock)
    239   1.4   mycroft 		FREE(lock, M_LOCKF);
    240  1.27      yamt 	if (sparelock)
    241  1.27      yamt 		FREE(sparelock, M_LOCKF);
    242  1.27      yamt 
    243  1.27      yamt 	return (error);
    244   1.1        ws }
    245   1.1        ws 
    246   1.1        ws /*
    247   1.1        ws  * Set a byte-range lock.
    248   1.1        ws  */
    249  1.24      yamt static int
    250  1.27      yamt lf_setlock(struct lockf *lock, struct lockf **sparelock,
    251  1.27      yamt     struct simplelock *interlock)
    252   1.1        ws {
    253  1.15  augustss 	struct lockf *block;
    254   1.1        ws 	struct lockf **head = lock->lf_head;
    255   1.1        ws 	struct lockf **prev, *overlap, *ltmp;
    256   1.1        ws 	static char lockstr[] = "lockf";
    257   1.1        ws 	int ovcase, priority, needtolink, error;
    258   1.1        ws 
    259   1.1        ws #ifdef LOCKF_DEBUG
    260   1.1        ws 	if (lockf_debug & 1)
    261   1.1        ws 		lf_print("lf_setlock", lock);
    262   1.1        ws #endif /* LOCKF_DEBUG */
    263   1.1        ws 
    264   1.1        ws 	/*
    265   1.1        ws 	 * Set the priority
    266   1.1        ws 	 */
    267   1.1        ws 	priority = PLOCK;
    268   1.1        ws 	if (lock->lf_type == F_WRLCK)
    269   1.1        ws 		priority += 4;
    270   1.1        ws 	priority |= PCATCH;
    271   1.1        ws 	/*
    272   1.1        ws 	 * Scan lock list for this file looking for locks that would block us.
    273   1.1        ws 	 */
    274   1.7  christos 	while ((block = lf_getblock(lock)) != NULL) {
    275   1.1        ws 		/*
    276   1.1        ws 		 * Free the structure and return if nonblocking.
    277   1.1        ws 		 */
    278   1.1        ws 		if ((lock->lf_flags & F_WAIT) == 0) {
    279   1.1        ws 			FREE(lock, M_LOCKF);
    280   1.1        ws 			return (EAGAIN);
    281   1.1        ws 		}
    282   1.1        ws 		/*
    283   1.1        ws 		 * We are blocked. Since flock style locks cover
    284   1.1        ws 		 * the whole file, there is no chance for deadlock.
    285   1.1        ws 		 * For byte-range locks we must check for deadlock.
    286   1.1        ws 		 *
    287   1.1        ws 		 * Deadlock detection is done by looking through the
    288   1.1        ws 		 * wait channels to see if there are any cycles that
    289   1.1        ws 		 * involve us. MAXDEPTH is set just to make sure we
    290  1.16  sommerfe 		 * do not go off into neverneverland.
    291   1.1        ws 		 */
    292   1.1        ws 		if ((lock->lf_flags & F_POSIX) &&
    293   1.1        ws 		    (block->lf_flags & F_POSIX)) {
    294  1.21   thorpej 			struct lwp *wlwp;
    295  1.15  augustss 			struct lockf *waitblock;
    296   1.1        ws 			int i = 0;
    297   1.1        ws 
    298  1.23   mycroft 			/*
    299  1.23   mycroft 			 * The block is waiting on something.  if_lwp will be
    300  1.23   mycroft 			 * 0 once the lock is granted, so we terminate the
    301  1.23   mycroft 			 * loop if we find this.
    302  1.23   mycroft 			 */
    303  1.23   mycroft 			wlwp = block->lf_lwp;
    304  1.23   mycroft 			while (wlwp && (i++ < maxlockdepth)) {
    305  1.21   thorpej 				waitblock = (struct lockf *)wlwp->l_wchan;
    306   1.1        ws 				/* Get the owner of the blocking lock */
    307   1.1        ws 				waitblock = waitblock->lf_next;
    308   1.1        ws 				if ((waitblock->lf_flags & F_POSIX) == 0)
    309   1.1        ws 					break;
    310  1.23   mycroft 				wlwp = waitblock->lf_lwp;
    311  1.23   mycroft 				if (wlwp == lock->lf_lwp) {
    312   1.1        ws 					free(lock, M_LOCKF);
    313   1.1        ws 					return (EDEADLK);
    314   1.1        ws 				}
    315   1.1        ws 			}
    316  1.16  sommerfe 			/*
    317  1.16  sommerfe 			 * If we're still following a dependancy chain
    318  1.16  sommerfe 			 * after maxlockdepth iterations, assume we're in
    319  1.16  sommerfe 			 * a cycle to be safe.
    320  1.16  sommerfe 			 */
    321  1.16  sommerfe 			if (i >= maxlockdepth) {
    322  1.16  sommerfe 				free(lock, M_LOCKF);
    323  1.16  sommerfe 				return (EDEADLK);
    324  1.16  sommerfe 			}
    325   1.1        ws 		}
    326   1.1        ws 		/*
    327   1.1        ws 		 * For flock type locks, we must first remove
    328   1.1        ws 		 * any shared locks that we hold before we sleep
    329   1.1        ws 		 * waiting for an exclusive lock.
    330   1.1        ws 		 */
    331   1.1        ws 		if ((lock->lf_flags & F_FLOCK) &&
    332   1.1        ws 		    lock->lf_type == F_WRLCK) {
    333   1.1        ws 			lock->lf_type = F_UNLCK;
    334  1.27      yamt 			(void) lf_clearlock(lock, NULL);
    335   1.1        ws 			lock->lf_type = F_WRLCK;
    336   1.1        ws 		}
    337   1.1        ws 		/*
    338   1.1        ws 		 * Add our lock to the blocked list and sleep until we're free.
    339   1.1        ws 		 * Remember who blocked us (for deadlock detection).
    340   1.1        ws 		 */
    341   1.1        ws 		lock->lf_next = block;
    342  1.12      fvdl 		TAILQ_INSERT_TAIL(&block->lf_blkhd, lock, lf_block);
    343   1.1        ws #ifdef LOCKF_DEBUG
    344   1.1        ws 		if (lockf_debug & 1) {
    345   1.1        ws 			lf_print("lf_setlock: blocking on", block);
    346   1.1        ws 			lf_printlist("lf_setlock", block);
    347   1.1        ws 		}
    348   1.1        ws #endif /* LOCKF_DEBUG */
    349  1.27      yamt 		error = ltsleep(lock, priority, lockstr, 0, interlock);
    350  1.16  sommerfe 
    351  1.16  sommerfe 		/*
    352  1.16  sommerfe 		 * We may have been awakened by a signal (in
    353  1.16  sommerfe 		 * which case we must remove ourselves from the
    354  1.16  sommerfe 		 * blocked list) and/or by another process
    355  1.16  sommerfe 		 * releasing a lock (in which case we have already
    356  1.16  sommerfe 		 * been removed from the blocked list and our
    357  1.16  sommerfe 		 * lf_next field set to NOLOCKF).
    358  1.16  sommerfe 		 */
    359  1.16  sommerfe 		if (lock->lf_next != NOLOCKF) {
    360  1.16  sommerfe 			TAILQ_REMOVE(&lock->lf_next->lf_blkhd, lock, lf_block);
    361  1.16  sommerfe 			lock->lf_next = NOLOCKF;
    362  1.16  sommerfe 		}
    363   1.7  christos 		if (error) {
    364   1.4   mycroft 			free(lock, M_LOCKF);
    365   1.4   mycroft 			return (error);
    366   1.1        ws 		}
    367   1.1        ws 	}
    368   1.1        ws 	/*
    369   1.1        ws 	 * No blocks!!  Add the lock.  Note that we will
    370   1.1        ws 	 * downgrade or upgrade any overlapping locks this
    371   1.1        ws 	 * process already owns.
    372   1.1        ws 	 *
    373   1.1        ws 	 * Skip over locks owned by other processes.
    374   1.1        ws 	 * Handle any locks that overlap and are owned by ourselves.
    375   1.1        ws 	 */
    376  1.23   mycroft 	lock->lf_lwp = 0;
    377   1.1        ws 	prev = head;
    378   1.1        ws 	block = *head;
    379   1.1        ws 	needtolink = 1;
    380   1.1        ws 	for (;;) {
    381   1.7  christos 		ovcase = lf_findoverlap(block, lock, SELF, &prev, &overlap);
    382   1.7  christos 		if (ovcase)
    383   1.1        ws 			block = overlap->lf_next;
    384   1.1        ws 		/*
    385   1.1        ws 		 * Six cases:
    386   1.1        ws 		 *	0) no overlap
    387   1.1        ws 		 *	1) overlap == lock
    388   1.1        ws 		 *	2) overlap contains lock
    389   1.1        ws 		 *	3) lock contains overlap
    390   1.1        ws 		 *	4) overlap starts before lock
    391   1.1        ws 		 *	5) overlap ends after lock
    392   1.1        ws 		 */
    393   1.1        ws 		switch (ovcase) {
    394   1.1        ws 		case 0: /* no overlap */
    395   1.1        ws 			if (needtolink) {
    396   1.1        ws 				*prev = lock;
    397   1.1        ws 				lock->lf_next = overlap;
    398   1.1        ws 			}
    399   1.1        ws 			break;
    400   1.1        ws 
    401   1.1        ws 		case 1: /* overlap == lock */
    402   1.1        ws 			/*
    403   1.1        ws 			 * If downgrading lock, others may be
    404   1.1        ws 			 * able to acquire it.
    405   1.1        ws 			 */
    406   1.1        ws 			if (lock->lf_type == F_RDLCK &&
    407   1.1        ws 			    overlap->lf_type == F_WRLCK)
    408   1.1        ws 				lf_wakelock(overlap);
    409   1.1        ws 			overlap->lf_type = lock->lf_type;
    410   1.1        ws 			FREE(lock, M_LOCKF);
    411   1.1        ws 			lock = overlap; /* for debug output below */
    412   1.1        ws 			break;
    413   1.1        ws 
    414   1.1        ws 		case 2: /* overlap contains lock */
    415   1.1        ws 			/*
    416   1.1        ws 			 * Check for common starting point and different types.
    417   1.1        ws 			 */
    418   1.1        ws 			if (overlap->lf_type == lock->lf_type) {
    419   1.1        ws 				free(lock, M_LOCKF);
    420   1.1        ws 				lock = overlap; /* for debug output below */
    421   1.1        ws 				break;
    422   1.1        ws 			}
    423   1.1        ws 			if (overlap->lf_start == lock->lf_start) {
    424   1.1        ws 				*prev = lock;
    425   1.1        ws 				lock->lf_next = overlap;
    426   1.1        ws 				overlap->lf_start = lock->lf_end + 1;
    427   1.1        ws 			} else
    428  1.27      yamt 				lf_split(overlap, lock, sparelock);
    429   1.1        ws 			lf_wakelock(overlap);
    430   1.1        ws 			break;
    431   1.1        ws 
    432   1.1        ws 		case 3: /* lock contains overlap */
    433   1.1        ws 			/*
    434   1.1        ws 			 * If downgrading lock, others may be able to
    435   1.1        ws 			 * acquire it, otherwise take the list.
    436   1.1        ws 			 */
    437   1.1        ws 			if (lock->lf_type == F_RDLCK &&
    438   1.1        ws 			    overlap->lf_type == F_WRLCK) {
    439   1.1        ws 				lf_wakelock(overlap);
    440   1.1        ws 			} else {
    441  1.19      matt 				while ((ltmp = TAILQ_FIRST(&overlap->lf_blkhd))) {
    442  1.16  sommerfe 					KASSERT(ltmp->lf_next == overlap);
    443  1.12      fvdl 					TAILQ_REMOVE(&overlap->lf_blkhd, ltmp,
    444  1.12      fvdl 					    lf_block);
    445  1.16  sommerfe 					ltmp->lf_next = lock;
    446  1.12      fvdl 					TAILQ_INSERT_TAIL(&lock->lf_blkhd,
    447  1.12      fvdl 					    ltmp, lf_block);
    448  1.12      fvdl 				}
    449   1.1        ws 			}
    450   1.1        ws 			/*
    451   1.1        ws 			 * Add the new lock if necessary and delete the overlap.
    452   1.1        ws 			 */
    453   1.1        ws 			if (needtolink) {
    454   1.1        ws 				*prev = lock;
    455   1.1        ws 				lock->lf_next = overlap->lf_next;
    456   1.1        ws 				prev = &lock->lf_next;
    457   1.1        ws 				needtolink = 0;
    458   1.1        ws 			} else
    459   1.1        ws 				*prev = overlap->lf_next;
    460   1.1        ws 			free(overlap, M_LOCKF);
    461   1.1        ws 			continue;
    462   1.1        ws 
    463   1.1        ws 		case 4: /* overlap starts before lock */
    464   1.1        ws 			/*
    465   1.1        ws 			 * Add lock after overlap on the list.
    466   1.1        ws 			 */
    467   1.1        ws 			lock->lf_next = overlap->lf_next;
    468   1.1        ws 			overlap->lf_next = lock;
    469   1.1        ws 			overlap->lf_end = lock->lf_start - 1;
    470   1.1        ws 			prev = &lock->lf_next;
    471   1.1        ws 			lf_wakelock(overlap);
    472   1.1        ws 			needtolink = 0;
    473   1.1        ws 			continue;
    474   1.1        ws 
    475   1.1        ws 		case 5: /* overlap ends after lock */
    476   1.1        ws 			/*
    477   1.1        ws 			 * Add the new lock before overlap.
    478   1.1        ws 			 */
    479   1.1        ws 			if (needtolink) {
    480   1.1        ws 				*prev = lock;
    481   1.1        ws 				lock->lf_next = overlap;
    482   1.1        ws 			}
    483   1.1        ws 			overlap->lf_start = lock->lf_end + 1;
    484   1.1        ws 			lf_wakelock(overlap);
    485   1.1        ws 			break;
    486   1.1        ws 		}
    487   1.1        ws 		break;
    488   1.1        ws 	}
    489   1.1        ws #ifdef LOCKF_DEBUG
    490   1.1        ws 	if (lockf_debug & 1) {
    491   1.1        ws 		lf_print("lf_setlock: got the lock", lock);
    492   1.1        ws 		lf_printlist("lf_setlock", lock);
    493   1.1        ws 	}
    494   1.1        ws #endif /* LOCKF_DEBUG */
    495   1.1        ws 	return (0);
    496   1.1        ws }
    497   1.1        ws 
    498   1.1        ws /*
    499   1.1        ws  * Remove a byte-range lock on an inode.
    500   1.1        ws  *
    501   1.1        ws  * Generally, find the lock (or an overlap to that lock)
    502   1.1        ws  * and remove it (or shrink it), then wakeup anyone we can.
    503   1.1        ws  */
    504  1.24      yamt static int
    505  1.27      yamt lf_clearlock(struct lockf *unlock, struct lockf **sparelock)
    506   1.1        ws {
    507   1.1        ws 	struct lockf **head = unlock->lf_head;
    508  1.15  augustss 	struct lockf *lf = *head;
    509   1.1        ws 	struct lockf *overlap, **prev;
    510   1.1        ws 	int ovcase;
    511   1.1        ws 
    512   1.1        ws 	if (lf == NOLOCKF)
    513   1.1        ws 		return (0);
    514   1.1        ws #ifdef LOCKF_DEBUG
    515   1.1        ws 	if (unlock->lf_type != F_UNLCK)
    516   1.1        ws 		panic("lf_clearlock: bad type");
    517   1.1        ws 	if (lockf_debug & 1)
    518   1.1        ws 		lf_print("lf_clearlock", unlock);
    519   1.1        ws #endif /* LOCKF_DEBUG */
    520   1.1        ws 	prev = head;
    521   1.7  christos 	while ((ovcase = lf_findoverlap(lf, unlock, SELF,
    522   1.7  christos 					&prev, &overlap)) != 0) {
    523   1.1        ws 		/*
    524   1.1        ws 		 * Wakeup the list of locks to be retried.
    525   1.1        ws 		 */
    526   1.1        ws 		lf_wakelock(overlap);
    527   1.1        ws 
    528   1.1        ws 		switch (ovcase) {
    529   1.1        ws 
    530   1.1        ws 		case 1: /* overlap == lock */
    531   1.1        ws 			*prev = overlap->lf_next;
    532   1.1        ws 			FREE(overlap, M_LOCKF);
    533   1.1        ws 			break;
    534   1.1        ws 
    535   1.1        ws 		case 2: /* overlap contains lock: split it */
    536   1.1        ws 			if (overlap->lf_start == unlock->lf_start) {
    537   1.1        ws 				overlap->lf_start = unlock->lf_end + 1;
    538   1.1        ws 				break;
    539   1.1        ws 			}
    540  1.27      yamt 			lf_split(overlap, unlock, sparelock);
    541   1.1        ws 			overlap->lf_next = unlock->lf_next;
    542   1.1        ws 			break;
    543   1.1        ws 
    544   1.1        ws 		case 3: /* lock contains overlap */
    545   1.1        ws 			*prev = overlap->lf_next;
    546   1.1        ws 			lf = overlap->lf_next;
    547   1.1        ws 			free(overlap, M_LOCKF);
    548   1.1        ws 			continue;
    549   1.1        ws 
    550   1.1        ws 		case 4: /* overlap starts before lock */
    551   1.1        ws 			overlap->lf_end = unlock->lf_start - 1;
    552   1.1        ws 			prev = &overlap->lf_next;
    553   1.1        ws 			lf = overlap->lf_next;
    554   1.1        ws 			continue;
    555   1.1        ws 
    556   1.1        ws 		case 5: /* overlap ends after lock */
    557   1.1        ws 			overlap->lf_start = unlock->lf_end + 1;
    558   1.1        ws 			break;
    559   1.1        ws 		}
    560   1.1        ws 		break;
    561   1.1        ws 	}
    562   1.1        ws #ifdef LOCKF_DEBUG
    563   1.1        ws 	if (lockf_debug & 1)
    564   1.1        ws 		lf_printlist("lf_clearlock", unlock);
    565   1.1        ws #endif /* LOCKF_DEBUG */
    566   1.1        ws 	return (0);
    567   1.1        ws }
    568   1.1        ws 
    569   1.1        ws /*
    570   1.1        ws  * Check whether there is a blocking lock,
    571   1.1        ws  * and if so return its process identifier.
    572   1.1        ws  */
    573  1.24      yamt static int
    574  1.25      yamt lf_getlock(struct lockf *lock, struct flock *fl)
    575   1.1        ws {
    576  1.15  augustss 	struct lockf *block;
    577   1.1        ws 
    578   1.1        ws #ifdef LOCKF_DEBUG
    579   1.1        ws 	if (lockf_debug & 1)
    580   1.1        ws 		lf_print("lf_getlock", lock);
    581   1.1        ws #endif /* LOCKF_DEBUG */
    582   1.1        ws 
    583   1.7  christos 	if ((block = lf_getblock(lock)) != NULL) {
    584   1.1        ws 		fl->l_type = block->lf_type;
    585   1.1        ws 		fl->l_whence = SEEK_SET;
    586   1.1        ws 		fl->l_start = block->lf_start;
    587   1.1        ws 		if (block->lf_end == -1)
    588   1.1        ws 			fl->l_len = 0;
    589   1.1        ws 		else
    590   1.1        ws 			fl->l_len = block->lf_end - block->lf_start + 1;
    591   1.1        ws 		if (block->lf_flags & F_POSIX)
    592  1.23   mycroft 			fl->l_pid = ((struct proc *)block->lf_id)->p_pid;
    593   1.1        ws 		else
    594   1.1        ws 			fl->l_pid = -1;
    595   1.1        ws 	} else {
    596   1.1        ws 		fl->l_type = F_UNLCK;
    597   1.1        ws 	}
    598   1.1        ws 	return (0);
    599   1.1        ws }
    600   1.1        ws 
    601   1.1        ws /*
    602   1.1        ws  * Walk the list of locks for an inode and
    603   1.1        ws  * return the first blocking lock.
    604   1.1        ws  */
    605  1.24      yamt static struct lockf *
    606  1.25      yamt lf_getblock(struct lockf *lock)
    607   1.1        ws {
    608   1.1        ws 	struct lockf **prev, *overlap, *lf = *(lock->lf_head);
    609   1.1        ws 
    610   1.1        ws 	prev = lock->lf_head;
    611  1.20    simonb 	while (lf_findoverlap(lf, lock, OTHERS, &prev, &overlap) != 0) {
    612   1.1        ws 		/*
    613   1.1        ws 		 * We've found an overlap, see if it blocks us
    614   1.1        ws 		 */
    615   1.1        ws 		if ((lock->lf_type == F_WRLCK || overlap->lf_type == F_WRLCK))
    616   1.1        ws 			return (overlap);
    617   1.1        ws 		/*
    618   1.1        ws 		 * Nope, point to the next one on the list and
    619   1.1        ws 		 * see if it blocks us
    620   1.1        ws 		 */
    621   1.1        ws 		lf = overlap->lf_next;
    622   1.1        ws 	}
    623   1.1        ws 	return (NOLOCKF);
    624   1.1        ws }
    625   1.1        ws 
    626   1.1        ws /*
    627   1.1        ws  * Walk the list of locks for an inode to
    628   1.1        ws  * find an overlapping lock (if any).
    629   1.1        ws  *
    630   1.1        ws  * NOTE: this returns only the FIRST overlapping lock.  There
    631   1.1        ws  *	 may be more than one.
    632   1.1        ws  */
    633  1.24      yamt static int
    634  1.25      yamt lf_findoverlap(struct lockf *lf, struct lockf *lock, int type,
    635  1.25      yamt     struct lockf ***prev, struct lockf **overlap)
    636   1.1        ws {
    637   1.1        ws 	off_t start, end;
    638   1.1        ws 
    639   1.1        ws 	*overlap = lf;
    640   1.1        ws 	if (lf == NOLOCKF)
    641   1.1        ws 		return (0);
    642   1.1        ws #ifdef LOCKF_DEBUG
    643   1.1        ws 	if (lockf_debug & 2)
    644   1.1        ws 		lf_print("lf_findoverlap: looking for overlap in", lock);
    645   1.1        ws #endif /* LOCKF_DEBUG */
    646   1.1        ws 	start = lock->lf_start;
    647   1.1        ws 	end = lock->lf_end;
    648   1.1        ws 	while (lf != NOLOCKF) {
    649  1.23   mycroft 		if (((type == SELF) && lf->lf_id != lock->lf_id) ||
    650  1.23   mycroft 		    ((type == OTHERS) && lf->lf_id == lock->lf_id)) {
    651   1.1        ws 			*prev = &lf->lf_next;
    652   1.1        ws 			*overlap = lf = lf->lf_next;
    653   1.1        ws 			continue;
    654   1.1        ws 		}
    655   1.1        ws #ifdef LOCKF_DEBUG
    656   1.1        ws 		if (lockf_debug & 2)
    657   1.1        ws 			lf_print("\tchecking", lf);
    658   1.1        ws #endif /* LOCKF_DEBUG */
    659   1.1        ws 		/*
    660   1.1        ws 		 * OK, check for overlap
    661   1.1        ws 		 *
    662   1.1        ws 		 * Six cases:
    663   1.1        ws 		 *	0) no overlap
    664   1.1        ws 		 *	1) overlap == lock
    665   1.1        ws 		 *	2) overlap contains lock
    666   1.1        ws 		 *	3) lock contains overlap
    667   1.1        ws 		 *	4) overlap starts before lock
    668   1.1        ws 		 *	5) overlap ends after lock
    669   1.1        ws 		 */
    670   1.1        ws 		if ((lf->lf_end != -1 && start > lf->lf_end) ||
    671   1.1        ws 		    (end != -1 && lf->lf_start > end)) {
    672   1.1        ws 			/* Case 0 */
    673   1.1        ws #ifdef LOCKF_DEBUG
    674   1.1        ws 			if (lockf_debug & 2)
    675   1.9  christos 				printf("no overlap\n");
    676   1.1        ws #endif /* LOCKF_DEBUG */
    677   1.1        ws 			if ((type & SELF) && end != -1 && lf->lf_start > end)
    678   1.1        ws 				return (0);
    679   1.1        ws 			*prev = &lf->lf_next;
    680   1.1        ws 			*overlap = lf = lf->lf_next;
    681   1.1        ws 			continue;
    682   1.1        ws 		}
    683   1.1        ws 		if ((lf->lf_start == start) && (lf->lf_end == end)) {
    684   1.1        ws 			/* Case 1 */
    685   1.1        ws #ifdef LOCKF_DEBUG
    686   1.1        ws 			if (lockf_debug & 2)
    687   1.9  christos 				printf("overlap == lock\n");
    688   1.1        ws #endif /* LOCKF_DEBUG */
    689   1.1        ws 			return (1);
    690   1.1        ws 		}
    691   1.1        ws 		if ((lf->lf_start <= start) &&
    692   1.1        ws 		    (end != -1) &&
    693   1.1        ws 		    ((lf->lf_end >= end) || (lf->lf_end == -1))) {
    694   1.1        ws 			/* Case 2 */
    695   1.1        ws #ifdef LOCKF_DEBUG
    696   1.1        ws 			if (lockf_debug & 2)
    697   1.9  christos 				printf("overlap contains lock\n");
    698   1.1        ws #endif /* LOCKF_DEBUG */
    699   1.1        ws 			return (2);
    700   1.1        ws 		}
    701   1.1        ws 		if (start <= lf->lf_start &&
    702   1.4   mycroft 		           (end == -1 ||
    703   1.1        ws 			   (lf->lf_end != -1 && end >= lf->lf_end))) {
    704   1.1        ws 			/* Case 3 */
    705   1.1        ws #ifdef LOCKF_DEBUG
    706   1.1        ws 			if (lockf_debug & 2)
    707   1.9  christos 				printf("lock contains overlap\n");
    708   1.1        ws #endif /* LOCKF_DEBUG */
    709   1.1        ws 			return (3);
    710   1.1        ws 		}
    711   1.1        ws 		if ((lf->lf_start < start) &&
    712   1.1        ws 			((lf->lf_end >= start) || (lf->lf_end == -1))) {
    713   1.1        ws 			/* Case 4 */
    714   1.1        ws #ifdef LOCKF_DEBUG
    715   1.1        ws 			if (lockf_debug & 2)
    716   1.9  christos 				printf("overlap starts before lock\n");
    717   1.1        ws #endif /* LOCKF_DEBUG */
    718   1.1        ws 			return (4);
    719   1.1        ws 		}
    720   1.1        ws 		if ((lf->lf_start > start) &&
    721   1.1        ws 			(end != -1) &&
    722   1.1        ws 			((lf->lf_end > end) || (lf->lf_end == -1))) {
    723   1.1        ws 			/* Case 5 */
    724   1.1        ws #ifdef LOCKF_DEBUG
    725   1.1        ws 			if (lockf_debug & 2)
    726   1.9  christos 				printf("overlap ends after lock\n");
    727   1.1        ws #endif /* LOCKF_DEBUG */
    728   1.1        ws 			return (5);
    729   1.1        ws 		}
    730   1.1        ws 		panic("lf_findoverlap: default");
    731   1.1        ws 	}
    732   1.1        ws 	return (0);
    733   1.1        ws }
    734   1.1        ws 
    735   1.1        ws /*
    736   1.1        ws  * Split a lock and a contained region into
    737   1.1        ws  * two or three locks as necessary.
    738   1.1        ws  */
    739  1.24      yamt static void
    740  1.27      yamt lf_split(struct lockf *lock1, struct lockf *lock2, struct lockf **sparelock)
    741   1.1        ws {
    742  1.15  augustss 	struct lockf *splitlock;
    743   1.1        ws 
    744   1.1        ws #ifdef LOCKF_DEBUG
    745   1.1        ws 	if (lockf_debug & 2) {
    746   1.1        ws 		lf_print("lf_split", lock1);
    747   1.1        ws 		lf_print("splitting from", lock2);
    748   1.1        ws 	}
    749   1.1        ws #endif /* LOCKF_DEBUG */
    750   1.1        ws 	/*
    751   1.1        ws 	 * Check to see if spliting into only two pieces.
    752   1.1        ws 	 */
    753   1.1        ws 	if (lock1->lf_start == lock2->lf_start) {
    754   1.1        ws 		lock1->lf_start = lock2->lf_end + 1;
    755   1.1        ws 		lock2->lf_next = lock1;
    756   1.1        ws 		return;
    757   1.1        ws 	}
    758   1.1        ws 	if (lock1->lf_end == lock2->lf_end) {
    759   1.1        ws 		lock1->lf_end = lock2->lf_start - 1;
    760   1.1        ws 		lock2->lf_next = lock1->lf_next;
    761   1.1        ws 		lock1->lf_next = lock2;
    762   1.1        ws 		return;
    763   1.1        ws 	}
    764   1.1        ws 	/*
    765   1.1        ws 	 * Make a new lock consisting of the last part of
    766   1.1        ws 	 * the encompassing lock
    767   1.1        ws 	 */
    768  1.27      yamt 	splitlock = *sparelock;
    769  1.27      yamt 	*sparelock = NULL;
    770  1.14     perry 	memcpy((caddr_t)splitlock, (caddr_t)lock1, sizeof(*splitlock));
    771   1.1        ws 	splitlock->lf_start = lock2->lf_end + 1;
    772  1.12      fvdl 	TAILQ_INIT(&splitlock->lf_blkhd);
    773   1.1        ws 	lock1->lf_end = lock2->lf_start - 1;
    774   1.1        ws 	/*
    775   1.1        ws 	 * OK, now link it in
    776   1.1        ws 	 */
    777   1.1        ws 	splitlock->lf_next = lock1->lf_next;
    778   1.1        ws 	lock2->lf_next = splitlock;
    779   1.1        ws 	lock1->lf_next = lock2;
    780   1.1        ws }
    781   1.1        ws 
    782   1.1        ws /*
    783   1.1        ws  * Wakeup a blocklist
    784   1.1        ws  */
    785  1.24      yamt static void
    786  1.25      yamt lf_wakelock(struct lockf *listhead)
    787   1.1        ws {
    788  1.15  augustss 	struct lockf *wakelock;
    789   1.1        ws 
    790  1.19      matt 	while ((wakelock = TAILQ_FIRST(&listhead->lf_blkhd))) {
    791  1.16  sommerfe 		KASSERT(wakelock->lf_next == listhead);
    792  1.12      fvdl 		TAILQ_REMOVE(&listhead->lf_blkhd, wakelock, lf_block);
    793   1.1        ws 		wakelock->lf_next = NOLOCKF;
    794   1.1        ws #ifdef LOCKF_DEBUG
    795   1.1        ws 		if (lockf_debug & 2)
    796   1.1        ws 			lf_print("lf_wakelock: awakening", wakelock);
    797  1.12      fvdl #endif
    798   1.1        ws 		wakeup((caddr_t)wakelock);
    799   1.1        ws 	}
    800   1.1        ws }
    801   1.1        ws 
    802   1.1        ws #ifdef LOCKF_DEBUG
    803   1.1        ws /*
    804   1.1        ws  * Print out a lock.
    805   1.1        ws  */
    806  1.24      yamt static void
    807  1.25      yamt lf_print(char *tag, struct lockf *lock)
    808   1.1        ws {
    809   1.1        ws 
    810   1.9  christos 	printf("%s: lock %p for ", tag, lock);
    811   1.1        ws 	if (lock->lf_flags & F_POSIX)
    812  1.23   mycroft 		printf("proc %d", ((struct proc *)lock->lf_id)->p_pid);
    813   1.1        ws 	else
    814  1.23   mycroft 		printf("file 0x%p", (struct file *)lock->lf_id);
    815  1.11       jtk 	printf(" %s, start %qx, end %qx",
    816   1.1        ws 		lock->lf_type == F_RDLCK ? "shared" :
    817   1.1        ws 		lock->lf_type == F_WRLCK ? "exclusive" :
    818   1.1        ws 		lock->lf_type == F_UNLCK ? "unlock" :
    819   1.1        ws 		"unknown", lock->lf_start, lock->lf_end);
    820  1.19      matt 	if (TAILQ_FIRST(&lock->lf_blkhd))
    821  1.19      matt 		printf(" block %p\n", TAILQ_FIRST(&lock->lf_blkhd));
    822   1.1        ws 	else
    823   1.9  christos 		printf("\n");
    824   1.1        ws }
    825   1.1        ws 
    826  1.24      yamt static void
    827  1.25      yamt lf_printlist(char *tag, struct lockf *lock)
    828   1.1        ws {
    829  1.15  augustss 	struct lockf *lf, *blk;
    830   1.1        ws 
    831  1.11       jtk 	printf("%s: Lock list:\n", tag);
    832  1.12      fvdl 	for (lf = *lock->lf_head; lf; lf = lf->lf_next) {
    833   1.9  christos 		printf("\tlock %p for ", lf);
    834   1.1        ws 		if (lf->lf_flags & F_POSIX)
    835  1.23   mycroft 			printf("proc %d", ((struct proc *)lf->lf_id)->p_pid);
    836   1.1        ws 		else
    837  1.23   mycroft 			printf("file 0x%p", (struct file *)lf->lf_id);
    838  1.11       jtk 		printf(", %s, start %qx, end %qx",
    839   1.1        ws 			lf->lf_type == F_RDLCK ? "shared" :
    840   1.1        ws 			lf->lf_type == F_WRLCK ? "exclusive" :
    841   1.1        ws 			lf->lf_type == F_UNLCK ? "unlock" :
    842   1.1        ws 			"unknown", lf->lf_start, lf->lf_end);
    843  1.19      matt 		TAILQ_FOREACH(blk, &lf->lf_blkhd, lf_block) {
    844  1.12      fvdl 			if (blk->lf_flags & F_POSIX)
    845  1.12      fvdl 				printf("proc %d",
    846  1.23   mycroft 				    ((struct proc *)blk->lf_id)->p_pid);
    847  1.12      fvdl 			else
    848  1.23   mycroft 				printf("file 0x%p", (struct file *)blk->lf_id);
    849  1.12      fvdl 			printf(", %s, start %qx, end %qx",
    850  1.12      fvdl 				blk->lf_type == F_RDLCK ? "shared" :
    851  1.12      fvdl 				blk->lf_type == F_WRLCK ? "exclusive" :
    852  1.12      fvdl 				blk->lf_type == F_UNLCK ? "unlock" :
    853  1.12      fvdl 				"unknown", blk->lf_start, blk->lf_end);
    854  1.19      matt 			if (TAILQ_FIRST(&blk->lf_blkhd))
    855  1.12      fvdl 				 panic("lf_printlist: bad list");
    856  1.12      fvdl 		}
    857  1.12      fvdl 		printf("\n");
    858   1.1        ws 	}
    859   1.1        ws }
    860   1.1        ws #endif /* LOCKF_DEBUG */
    861