Home | History | Annotate | Line # | Download | only in lfs
lfs_inode.c revision 1.50.2.1
      1  1.50.2.1   nathanw /*	$NetBSD: lfs_inode.c,v 1.50.2.1 2001/06/21 20:10:10 nathanw Exp $	*/
      2       1.2       cgd 
      3      1.18  perseant /*-
      4      1.44  perseant  * Copyright (c) 1999, 2000 The NetBSD Foundation, Inc.
      5      1.18  perseant  * All rights reserved.
      6      1.18  perseant  *
      7      1.18  perseant  * This code is derived from software contributed to The NetBSD Foundation
      8      1.18  perseant  * by Konrad E. Schroder <perseant (at) hhhh.org>.
      9      1.18  perseant  *
     10      1.18  perseant  * Redistribution and use in source and binary forms, with or without
     11      1.18  perseant  * modification, are permitted provided that the following conditions
     12      1.18  perseant  * are met:
     13      1.18  perseant  * 1. Redistributions of source code must retain the above copyright
     14      1.18  perseant  *    notice, this list of conditions and the following disclaimer.
     15      1.18  perseant  * 2. Redistributions in binary form must reproduce the above copyright
     16      1.18  perseant  *    notice, this list of conditions and the following disclaimer in the
     17      1.18  perseant  *    documentation and/or other materials provided with the distribution.
     18      1.18  perseant  * 3. All advertising materials mentioning features or use of this software
     19      1.18  perseant  *    must display the following acknowledgement:
     20      1.18  perseant  *      This product includes software developed by the NetBSD
     21      1.18  perseant  *      Foundation, Inc. and its contributors.
     22      1.18  perseant  * 4. Neither the name of The NetBSD Foundation nor the names of its
     23      1.18  perseant  *    contributors may be used to endorse or promote products derived
     24      1.18  perseant  *    from this software without specific prior written permission.
     25      1.18  perseant  *
     26      1.18  perseant  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     27      1.18  perseant  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     28      1.18  perseant  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     29      1.18  perseant  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     30      1.18  perseant  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     31      1.18  perseant  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     32      1.18  perseant  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     33      1.18  perseant  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     34      1.18  perseant  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     35      1.18  perseant  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     36      1.18  perseant  * POSSIBILITY OF SUCH DAMAGE.
     37      1.18  perseant  */
     38       1.1   mycroft /*
     39       1.1   mycroft  * Copyright (c) 1986, 1989, 1991, 1993
     40       1.1   mycroft  *	The Regents of the University of California.  All rights reserved.
     41       1.1   mycroft  *
     42       1.1   mycroft  * Redistribution and use in source and binary forms, with or without
     43       1.1   mycroft  * modification, are permitted provided that the following conditions
     44       1.1   mycroft  * are met:
     45       1.1   mycroft  * 1. Redistributions of source code must retain the above copyright
     46       1.1   mycroft  *    notice, this list of conditions and the following disclaimer.
     47       1.1   mycroft  * 2. Redistributions in binary form must reproduce the above copyright
     48       1.1   mycroft  *    notice, this list of conditions and the following disclaimer in the
     49       1.1   mycroft  *    documentation and/or other materials provided with the distribution.
     50       1.1   mycroft  * 3. All advertising materials mentioning features or use of this software
     51       1.1   mycroft  *    must display the following acknowledgement:
     52       1.1   mycroft  *	This product includes software developed by the University of
     53       1.1   mycroft  *	California, Berkeley and its contributors.
     54       1.1   mycroft  * 4. Neither the name of the University nor the names of its contributors
     55       1.1   mycroft  *    may be used to endorse or promote products derived from this software
     56       1.1   mycroft  *    without specific prior written permission.
     57       1.1   mycroft  *
     58       1.1   mycroft  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     59       1.1   mycroft  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     60       1.1   mycroft  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     61       1.1   mycroft  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     62       1.1   mycroft  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     63       1.1   mycroft  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     64       1.1   mycroft  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     65       1.1   mycroft  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     66       1.1   mycroft  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     67       1.1   mycroft  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     68       1.1   mycroft  * SUCH DAMAGE.
     69       1.1   mycroft  *
     70      1.12      fvdl  *	@(#)lfs_inode.c	8.9 (Berkeley) 5/8/95
     71       1.1   mycroft  */
     72      1.13    scottr 
     73  1.50.2.1   nathanw #if defined(_KERNEL_OPT)
     74      1.13    scottr #include "opt_quota.h"
     75      1.14    scottr #endif
     76       1.1   mycroft 
     77       1.1   mycroft #include <sys/param.h>
     78       1.1   mycroft #include <sys/systm.h>
     79       1.1   mycroft #include <sys/mount.h>
     80       1.1   mycroft #include <sys/proc.h>
     81       1.1   mycroft #include <sys/file.h>
     82       1.1   mycroft #include <sys/buf.h>
     83       1.1   mycroft #include <sys/vnode.h>
     84       1.1   mycroft #include <sys/kernel.h>
     85       1.1   mycroft #include <sys/malloc.h>
     86      1.38  perseant #include <sys/trace.h>
     87      1.38  perseant #include <sys/resourcevar.h>
     88       1.1   mycroft 
     89       1.1   mycroft #include <ufs/ufs/quota.h>
     90       1.1   mycroft #include <ufs/ufs/inode.h>
     91       1.1   mycroft #include <ufs/ufs/ufsmount.h>
     92       1.1   mycroft #include <ufs/ufs/ufs_extern.h>
     93       1.1   mycroft 
     94       1.1   mycroft #include <ufs/lfs/lfs.h>
     95       1.1   mycroft #include <ufs/lfs/lfs_extern.h>
     96       1.1   mycroft 
     97      1.35  perseant extern int locked_queue_count;
     98      1.35  perseant extern long locked_queue_bytes;
     99      1.21  perseant 
    100      1.38  perseant static int lfs_update_seguse(struct lfs *, long, size_t);
    101      1.38  perseant static int lfs_indirtrunc (struct inode *, ufs_daddr_t, ufs_daddr_t,
    102      1.48  perseant 			   ufs_daddr_t, int, long *, long *, long *, size_t *,
    103      1.48  perseant 			   struct proc *);
    104      1.38  perseant static int lfs_blkfree (struct lfs *, daddr_t, size_t, long *, size_t *);
    105      1.38  perseant static int lfs_vtruncbuf(struct vnode *, daddr_t, int, int);
    106      1.38  perseant 
    107       1.1   mycroft /* Search a block for a specific dinode. */
    108       1.1   mycroft struct dinode *
    109      1.30  perseant lfs_ifind(fs, ino, bp)
    110       1.1   mycroft 	struct lfs *fs;
    111       1.1   mycroft 	ino_t ino;
    112      1.30  perseant 	struct buf *bp;
    113       1.1   mycroft {
    114      1.32  augustss 	int cnt;
    115      1.32  augustss 	struct dinode *dip = (struct dinode *)bp->b_data;
    116      1.32  augustss 	struct dinode *ldip;
    117      1.18  perseant 
    118       1.1   mycroft 	for (cnt = INOPB(fs), ldip = dip + (cnt - 1); cnt--; --ldip)
    119       1.1   mycroft 		if (ldip->di_inumber == ino)
    120       1.1   mycroft 			return (ldip);
    121      1.18  perseant 
    122      1.40  perseant 	printf("offset is 0x%x (seg %d)\n", fs->lfs_offset,
    123      1.40  perseant 	       datosn(fs,fs->lfs_offset));
    124      1.40  perseant 	printf("block is 0x%x (seg %d)\n", bp->b_blkno,
    125      1.40  perseant 	       datosn(fs,bp->b_blkno));
    126       1.1   mycroft 	panic("lfs_ifind: dinode %u not found", ino);
    127       1.1   mycroft 	/* NOTREACHED */
    128       1.1   mycroft }
    129       1.1   mycroft 
    130       1.1   mycroft int
    131       1.4  christos lfs_update(v)
    132       1.4  christos 	void *v;
    133       1.4  christos {
    134       1.1   mycroft 	struct vop_update_args /* {
    135      1.18  perseant 				  struct vnode *a_vp;
    136      1.18  perseant 				  struct timespec *a_access;
    137      1.18  perseant 				  struct timespec *a_modify;
    138      1.36  perseant 				  int a_flags;
    139      1.18  perseant 				  } */ *ap = v;
    140       1.1   mycroft 	struct inode *ip;
    141      1.12      fvdl 	struct vnode *vp = ap->a_vp;
    142      1.42  perseant 	int oflag;
    143       1.6   mycroft 	struct timespec ts;
    144      1.25  perseant 	struct lfs *fs = VFSTOUFS(vp->v_mount)->um_lfs;
    145      1.18  perseant 
    146      1.12      fvdl 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
    147       1.1   mycroft 		return (0);
    148      1.12      fvdl 	ip = VTOI(vp);
    149      1.18  perseant 
    150      1.18  perseant 	/*
    151      1.18  perseant 	 * If we are called from vinvalbuf, and the file's blocks have
    152      1.18  perseant 	 * already been scheduled for writing, but the writes have not
    153      1.18  perseant 	 * yet completed, lfs_vflush will not be called, and vinvalbuf
    154      1.18  perseant 	 * will cause a panic.  So, we must wait until any pending write
    155      1.36  perseant 	 * for our inode completes, if we are called with UPDATE_WAIT set.
    156      1.18  perseant 	 */
    157      1.36  perseant 	while((ap->a_flags & (UPDATE_WAIT|UPDATE_DIROP)) == UPDATE_WAIT &&
    158      1.36  perseant 	    WRITEINPROG(vp)) {
    159      1.23  perseant #ifdef DEBUG_LFS
    160      1.38  perseant 		printf("lfs_update: sleeping on inode %d (in-progress)\n",
    161      1.38  perseant 		       ip->i_number);
    162      1.23  perseant #endif
    163      1.18  perseant 		tsleep(vp, (PRIBIO+1), "lfs_update", 0);
    164      1.18  perseant 	}
    165      1.18  perseant 	oflag = ip->i_flag;
    166       1.6   mycroft 	TIMEVAL_TO_TIMESPEC(&time, &ts);
    167      1.18  perseant 	LFS_ITIMES(ip,
    168      1.18  perseant 		   ap->a_access ? ap->a_access : &ts,
    169      1.18  perseant 		   ap->a_modify ? ap->a_modify : &ts, &ts);
    170      1.37  perseant 	if ((ip->i_flag & (IN_MODIFIED | IN_ACCESSED | IN_CLEANING)) == 0) {
    171       1.1   mycroft 		return (0);
    172      1.18  perseant 	}
    173      1.18  perseant 
    174       1.1   mycroft 	/* If sync, push back the vnode and any dirty blocks it may have. */
    175      1.36  perseant 	if((ap->a_flags & (UPDATE_WAIT|UPDATE_DIROP))==UPDATE_WAIT) {
    176      1.25  perseant 		/* Avoid flushing VDIROP. */
    177      1.25  perseant 		++fs->lfs_diropwait;
    178      1.25  perseant 		while(vp->v_flag & VDIROP) {
    179      1.25  perseant #ifdef DEBUG_LFS
    180      1.38  perseant 			printf("lfs_update: sleeping on inode %d (dirops)\n",
    181      1.38  perseant 			       ip->i_number);
    182      1.38  perseant 			printf("lfs_update: vflags 0x%lx, iflags 0x%x\n",
    183      1.38  perseant 				vp->v_flag, ip->i_flag);
    184      1.25  perseant #endif
    185      1.38  perseant 			if(fs->lfs_dirops == 0)
    186      1.38  perseant 				lfs_flush_fs(fs, SEGM_SYNC);
    187      1.25  perseant 			else
    188      1.38  perseant 				tsleep(&fs->lfs_writer, PRIBIO+1, "lfs_fsync",
    189      1.38  perseant 				       0);
    190      1.25  perseant 			/* XXX KS - by falling out here, are we writing the vn
    191      1.25  perseant 			twice? */
    192      1.25  perseant 		}
    193      1.25  perseant 		--fs->lfs_diropwait;
    194      1.25  perseant 		return lfs_vflush(vp);
    195      1.25  perseant         }
    196      1.25  perseant 	return 0;
    197       1.1   mycroft }
    198       1.1   mycroft 
    199      1.38  perseant #define	SINGLE	0	/* index of single indirect block */
    200      1.38  perseant #define	DOUBLE	1	/* index of double indirect block */
    201      1.38  perseant #define	TRIPLE	2	/* index of triple indirect block */
    202       1.1   mycroft /*
    203      1.38  perseant  * Truncate the inode oip to at most length size, freeing the
    204      1.38  perseant  * disk blocks.
    205       1.1   mycroft  */
    206      1.44  perseant /* VOP_BWRITE 1 + NIADDR + VOP_BALLOC == 2 + 2*NIADDR times */
    207       1.1   mycroft int
    208       1.4  christos lfs_truncate(v)
    209       1.4  christos 	void *v;
    210       1.4  christos {
    211       1.1   mycroft 	struct vop_truncate_args /* {
    212       1.1   mycroft 		struct vnode *a_vp;
    213       1.1   mycroft 		off_t a_length;
    214       1.1   mycroft 		int a_flags;
    215       1.1   mycroft 		struct ucred *a_cred;
    216       1.1   mycroft 		struct proc *a_p;
    217       1.4  christos 	} */ *ap = v;
    218      1.38  perseant 	struct vnode *ovp = ap->a_vp;
    219      1.38  perseant 	ufs_daddr_t lastblock;
    220      1.38  perseant 	struct inode *oip;
    221      1.38  perseant 	ufs_daddr_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR];
    222      1.38  perseant 	ufs_daddr_t newblks[NDADDR + NIADDR];
    223       1.1   mycroft 	off_t length = ap->a_length;
    224       1.1   mycroft 	struct lfs *fs;
    225      1.38  perseant 	struct buf *bp;
    226      1.38  perseant 	int offset, size, level;
    227      1.40  perseant 	long count, rcount, nblocks, blocksreleased = 0, real_released = 0;
    228      1.38  perseant 	int i;
    229      1.38  perseant 	int aflags, error, allerror = 0;
    230      1.38  perseant 	off_t osize;
    231      1.38  perseant 	long lastseg;
    232      1.38  perseant 	size_t bc;
    233      1.47  perseant 	int obufsize, odb;
    234      1.38  perseant 
    235      1.33  perseant 	if (length < 0)
    236      1.33  perseant 		return (EINVAL);
    237      1.38  perseant 	oip = VTOI(ovp);
    238      1.33  perseant 
    239      1.38  perseant 	/*
    240      1.38  perseant 	 * Just return and not update modification times.
    241      1.38  perseant 	 */
    242      1.38  perseant 	if (oip->i_ffs_size == length)
    243      1.38  perseant 		return (0);
    244       1.1   mycroft 
    245      1.38  perseant 	if (ovp->v_type == VLNK &&
    246      1.38  perseant 	    (oip->i_ffs_size < ovp->v_mount->mnt_maxsymlinklen ||
    247      1.38  perseant 	     (ovp->v_mount->mnt_maxsymlinklen == 0 &&
    248      1.38  perseant 	      oip->i_din.ffs_din.di_blocks == 0))) {
    249       1.1   mycroft #ifdef DIAGNOSTIC
    250       1.1   mycroft 		if (length != 0)
    251       1.1   mycroft 			panic("lfs_truncate: partial truncate of symlink");
    252       1.1   mycroft #endif
    253      1.38  perseant 		memset((char *)&oip->i_ffs_shortlink, 0, (u_int)oip->i_ffs_size);
    254      1.38  perseant 		oip->i_ffs_size = 0;
    255      1.38  perseant 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
    256      1.38  perseant 		return (VOP_UPDATE(ovp, NULL, NULL, 0));
    257      1.38  perseant 	}
    258      1.38  perseant 	if (oip->i_ffs_size == length) {
    259      1.38  perseant 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
    260      1.38  perseant 		return (VOP_UPDATE(ovp, NULL, NULL, 0));
    261       1.1   mycroft 	}
    262      1.38  perseant #ifdef QUOTA
    263      1.38  perseant 	if ((error = getinoquota(oip)) != 0)
    264      1.38  perseant 		return (error);
    265      1.38  perseant #endif
    266      1.38  perseant 	fs = oip->i_lfs;
    267      1.18  perseant 	lfs_imtime(fs);
    268      1.38  perseant 	osize = oip->i_ffs_size;
    269      1.38  perseant 	ovp->v_lasta = ovp->v_clen = ovp->v_cstart = ovp->v_lastw = 0;
    270      1.38  perseant 
    271      1.38  perseant 	/*
    272      1.38  perseant 	 * Lengthen the size of the file. We must ensure that the
    273      1.38  perseant 	 * last byte of the file is allocated. Since the smallest
    274      1.38  perseant 	 * value of osize is 0, length will be at least 1.
    275      1.38  perseant 	 */
    276      1.38  perseant 	if (osize < length) {
    277      1.33  perseant 		if (length > fs->lfs_maxfilesize)
    278      1.33  perseant 			return (EFBIG);
    279      1.38  perseant 		aflags = B_CLRBUF;
    280      1.38  perseant 		if (ap->a_flags & IO_SYNC)
    281      1.38  perseant 			aflags |= B_SYNC;
    282      1.44  perseant 		error = lfs_reserve(fs, ovp, fsbtodb(fs, NIADDR + 2));
    283      1.44  perseant 		if (error)
    284      1.44  perseant 			return (error);
    285      1.38  perseant 		error = VOP_BALLOC(ovp, length - 1, 1, ap->a_cred, aflags, &bp);
    286      1.44  perseant 		lfs_reserve(fs, ovp, -fsbtodb(fs, NIADDR + 2));
    287      1.33  perseant 		if (error)
    288      1.38  perseant 			return (error);
    289      1.38  perseant 		oip->i_ffs_size = length;
    290      1.38  perseant 		uvm_vnp_setsize(ovp, length);
    291      1.44  perseant 		(void) VOP_BWRITE(bp);
    292      1.38  perseant 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
    293      1.38  perseant 		return (VOP_UPDATE(ovp, NULL, NULL, 0));
    294       1.1   mycroft 	}
    295       1.1   mycroft 
    296      1.44  perseant 	if ((error = lfs_reserve(fs, ovp, fsbtodb(fs, 2 * NIADDR + 3))) != 0)
    297      1.44  perseant 		return (error);
    298      1.18  perseant 	/*
    299      1.35  perseant 	 * Make sure no writes to this inode can happen while we're
    300      1.35  perseant 	 * truncating.  Otherwise, blocks which are accounted for on the
    301      1.35  perseant 	 * inode *and* which have been created for cleaning can coexist,
    302      1.35  perseant 	 * and cause an overcounting.
    303      1.20  perseant 	 *
    304      1.35  perseant 	 * (We don't need to *hold* the seglock, though, because we already
    305      1.35  perseant 	 * hold the inode lock; draining the seglock is sufficient.)
    306      1.18  perseant 	 */
    307      1.38  perseant 	if (ovp != fs->lfs_unlockvp) {
    308      1.35  perseant 		while(fs->lfs_seglock) {
    309      1.35  perseant 			tsleep(&fs->lfs_seglock, PRIBIO+1, "lfs_truncate", 0);
    310      1.35  perseant 		}
    311      1.18  perseant 	}
    312      1.18  perseant 
    313       1.1   mycroft 	/*
    314      1.38  perseant 	 * Shorten the size of the file. If the file is not being
    315      1.38  perseant 	 * truncated to a block boundary, the contents of the
    316      1.38  perseant 	 * partial block following the end of the file must be
    317      1.38  perseant 	 * zero'ed in case it ever becomes accessible again because
    318      1.38  perseant 	 * of subsequent file growth. Directories however are not
    319      1.38  perseant 	 * zero'ed as they should grow back initialized to empty.
    320      1.38  perseant 	 */
    321      1.38  perseant 	offset = blkoff(fs, length);
    322      1.38  perseant 	lastseg = -1;
    323      1.38  perseant 	bc = 0;
    324      1.38  perseant 	if (offset == 0) {
    325      1.38  perseant 		oip->i_ffs_size = length;
    326      1.38  perseant 	} else {
    327      1.38  perseant 		lbn = lblkno(fs, length);
    328      1.38  perseant 		aflags = B_CLRBUF;
    329      1.38  perseant 		if (ap->a_flags & IO_SYNC)
    330      1.38  perseant 			aflags |= B_SYNC;
    331      1.38  perseant 		error = VOP_BALLOC(ovp, length - 1, 1, ap->a_cred, aflags, &bp);
    332      1.44  perseant 		if (error) {
    333      1.44  perseant 			lfs_reserve(fs, ovp, -fsbtodb(fs, 2 * NIADDR + 3));
    334      1.38  perseant 			return (error);
    335      1.44  perseant 		}
    336      1.47  perseant 		obufsize = bp->b_bufsize;
    337      1.47  perseant 		odb = btodb(bp->b_bcount);
    338      1.38  perseant 		oip->i_ffs_size = length;
    339      1.38  perseant 		size = blksize(fs, oip, lbn);
    340      1.38  perseant 		if (ovp->v_type != VDIR)
    341      1.38  perseant 			memset((char *)bp->b_data + offset, 0,
    342      1.38  perseant 			       (u_int)(size - offset));
    343      1.38  perseant 		allocbuf(bp, size);
    344      1.47  perseant 		if (bp->b_flags & B_DELWRI) {
    345      1.47  perseant 			if ((bp->b_flags & (B_LOCKED | B_CALL)) == B_LOCKED)
    346      1.47  perseant 				locked_queue_bytes -= obufsize - bp->b_bufsize;
    347      1.47  perseant 			fs->lfs_avail += odb - btodb(size);
    348      1.47  perseant 		}
    349      1.44  perseant 		(void) VOP_BWRITE(bp);
    350      1.38  perseant 	}
    351      1.38  perseant 	uvm_vnp_setsize(ovp, length);
    352      1.38  perseant 	/*
    353      1.38  perseant 	 * Calculate index into inode's block list of
    354      1.38  perseant 	 * last direct and indirect blocks (if any)
    355      1.38  perseant 	 * which we want to keep.  Lastblock is -1 when
    356      1.38  perseant 	 * the file is truncated to 0.
    357      1.38  perseant 	 */
    358      1.38  perseant 	lastblock = lblkno(fs, length + fs->lfs_bsize - 1) - 1;
    359      1.38  perseant 	lastiblock[SINGLE] = lastblock - NDADDR;
    360      1.38  perseant 	lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
    361      1.38  perseant 	lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
    362      1.38  perseant 	nblocks = btodb(fs->lfs_bsize);
    363      1.38  perseant 	/*
    364      1.38  perseant 	 * Record changed file and block pointers before we start
    365      1.38  perseant 	 * freeing blocks.  lastiblock values are also normalized to -1
    366      1.38  perseant 	 * for calls to lfs_indirtrunc below.
    367      1.38  perseant 	 */
    368      1.38  perseant 	memcpy((caddr_t)newblks, (caddr_t)&oip->i_ffs_db[0], sizeof newblks);
    369      1.38  perseant 	for (level = TRIPLE; level >= SINGLE; level--)
    370      1.38  perseant 		if (lastiblock[level] < 0) {
    371      1.38  perseant 			newblks[NDADDR+level] = 0;
    372      1.38  perseant 			lastiblock[level] = -1;
    373      1.38  perseant 		}
    374      1.38  perseant 	for (i = NDADDR - 1; i > lastblock; i--)
    375      1.38  perseant 		newblks[i] = 0;
    376      1.38  perseant 
    377      1.38  perseant 	oip->i_ffs_size = osize;
    378      1.38  perseant 	error = lfs_vtruncbuf(ovp, lastblock + 1, 0, 0);
    379      1.38  perseant 	if (error && !allerror)
    380      1.38  perseant 		allerror = error;
    381       1.1   mycroft 
    382       1.1   mycroft 	/*
    383      1.38  perseant 	 * Indirect blocks first.
    384       1.1   mycroft 	 */
    385      1.38  perseant 	indir_lbn[SINGLE] = -NDADDR;
    386      1.38  perseant 	indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
    387      1.38  perseant 	indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
    388      1.38  perseant 	for (level = TRIPLE; level >= SINGLE; level--) {
    389      1.38  perseant 		bn = oip->i_ffs_ib[level];
    390      1.38  perseant 		if (bn != 0) {
    391      1.38  perseant 			error = lfs_indirtrunc(oip, indir_lbn[level],
    392      1.40  perseant 					       bn, lastiblock[level],
    393      1.40  perseant 					       level, &count, &rcount,
    394      1.48  perseant 					       &lastseg, &bc, ap->a_p);
    395      1.38  perseant 			if (error)
    396      1.38  perseant 				allerror = error;
    397      1.40  perseant 			real_released += rcount;
    398      1.38  perseant 			blocksreleased += count;
    399      1.38  perseant 			if (lastiblock[level] < 0) {
    400      1.40  perseant 				if (oip->i_ffs_ib[level] > 0)
    401      1.40  perseant 					real_released += nblocks;
    402      1.40  perseant 				blocksreleased += nblocks;
    403      1.38  perseant 				oip->i_ffs_ib[level] = 0;
    404      1.38  perseant 				lfs_blkfree(fs, bn, fs->lfs_bsize, &lastseg, &bc);
    405      1.38  perseant 			}
    406      1.38  perseant 		}
    407      1.38  perseant 		if (lastiblock[level] >= 0)
    408      1.38  perseant 			goto done;
    409      1.38  perseant 	}
    410      1.38  perseant 
    411      1.38  perseant 	/*
    412      1.38  perseant 	 * All whole direct blocks or frags.
    413      1.38  perseant 	 */
    414      1.38  perseant 	for (i = NDADDR - 1; i > lastblock; i--) {
    415      1.38  perseant 		long bsize;
    416      1.12      fvdl 
    417      1.38  perseant 		bn = oip->i_ffs_db[i];
    418      1.38  perseant 		if (bn == 0)
    419      1.38  perseant 			continue;
    420      1.40  perseant 		bsize = blksize(fs, oip, i);
    421      1.40  perseant 		if (oip->i_ffs_db[i] > 0)
    422      1.40  perseant 			real_released += btodb(bsize);
    423      1.40  perseant 		blocksreleased += btodb(bsize);
    424      1.38  perseant 		oip->i_ffs_db[i] = 0;
    425      1.38  perseant 		lfs_blkfree(fs, bn, bsize, &lastseg, &bc);
    426       1.1   mycroft 	}
    427      1.38  perseant 	if (lastblock < 0)
    428      1.38  perseant 		goto done;
    429      1.38  perseant 
    430       1.1   mycroft 	/*
    431      1.38  perseant 	 * Finally, look for a change in size of the
    432      1.38  perseant 	 * last direct block; release any frags.
    433       1.1   mycroft 	 */
    434      1.38  perseant 	bn = oip->i_ffs_db[lastblock];
    435      1.38  perseant 	if (bn != 0) {
    436      1.38  perseant 		long oldspace, newspace;
    437       1.1   mycroft 
    438      1.38  perseant 		/*
    439      1.38  perseant 		 * Calculate amount of space we're giving
    440      1.38  perseant 		 * back as old block size minus new block size.
    441      1.38  perseant 		 */
    442      1.38  perseant 		oldspace = blksize(fs, oip, lastblock);
    443      1.38  perseant 		oip->i_ffs_size = length;
    444      1.38  perseant 		newspace = blksize(fs, oip, lastblock);
    445      1.38  perseant 		if (newspace == 0)
    446      1.38  perseant 			panic("itrunc: newspace");
    447      1.38  perseant 		if (oldspace - newspace > 0) {
    448      1.38  perseant 			lfs_blkfree(fs, bn, oldspace - newspace, &lastseg, &bc);
    449      1.45  perseant 			if (bn > 0)
    450      1.45  perseant 				real_released += btodb(oldspace - newspace);
    451      1.38  perseant 			blocksreleased += btodb(oldspace - newspace);
    452       1.1   mycroft 		}
    453       1.1   mycroft 	}
    454      1.38  perseant 
    455      1.38  perseant done:
    456      1.38  perseant 	/* Finish segment accounting corrections */
    457      1.38  perseant 	lfs_update_seguse(fs, lastseg, bc);
    458      1.38  perseant #ifdef DIAGNOSTIC
    459      1.38  perseant 	for (level = SINGLE; level <= TRIPLE; level++)
    460      1.38  perseant 		if (newblks[NDADDR + level] != oip->i_ffs_ib[level])
    461      1.38  perseant 			panic("lfs itrunc1");
    462      1.38  perseant 	for (i = 0; i < NDADDR; i++)
    463      1.38  perseant 		if (newblks[i] != oip->i_ffs_db[i])
    464      1.38  perseant 			panic("lfs itrunc2");
    465      1.38  perseant 	if (length == 0 &&
    466      1.38  perseant 	    (!LIST_EMPTY(&ovp->v_cleanblkhd) || !LIST_EMPTY(&ovp->v_dirtyblkhd)))
    467      1.38  perseant 		panic("lfs itrunc3");
    468      1.38  perseant #endif /* DIAGNOSTIC */
    469      1.38  perseant 	/*
    470      1.38  perseant 	 * Put back the real size.
    471      1.38  perseant 	 */
    472      1.38  perseant 	oip->i_ffs_size = length;
    473      1.40  perseant 	oip->i_lfs_effnblks -= blocksreleased;
    474      1.40  perseant 	oip->i_ffs_blocks -= real_released;
    475      1.38  perseant 	fs->lfs_bfree += blocksreleased;
    476      1.38  perseant #ifdef DIAGNOSTIC
    477      1.45  perseant 	if (oip->i_ffs_size == 0 && oip->i_ffs_blocks != 0) {
    478      1.45  perseant 		printf("lfs_truncate: truncate to 0 but %d blocks on inode\n",
    479      1.38  perseant 		       oip->i_ffs_blocks);
    480      1.38  perseant 		panic("lfs_truncate: persistent blocks\n");
    481       1.1   mycroft 	}
    482      1.38  perseant #endif
    483      1.38  perseant 	oip->i_flag |= IN_CHANGE;
    484      1.38  perseant #ifdef QUOTA
    485      1.38  perseant 	(void) chkdq(oip, -blocksreleased, NOCRED, 0);
    486       1.1   mycroft #endif
    487      1.44  perseant 	lfs_reserve(fs, ovp, -fsbtodb(fs, 2 * NIADDR + 3));
    488      1.38  perseant 	return (allerror);
    489      1.38  perseant }
    490      1.28      fvdl 
    491      1.38  perseant /* Update segment usage information when removing a block. */
    492      1.38  perseant static int
    493      1.38  perseant lfs_blkfree(struct lfs *fs, daddr_t daddr, size_t bsize, long *lastseg,
    494      1.38  perseant 	    size_t *num)
    495      1.38  perseant {
    496      1.38  perseant 	long seg;
    497      1.38  perseant 	int error = 0;
    498      1.24  perseant 
    499      1.38  perseant 	bsize = fragroundup(fs, bsize);
    500      1.38  perseant 	if (daddr > 0) {
    501      1.38  perseant 		if (*lastseg != (seg = datosn(fs, daddr))) {
    502      1.38  perseant 			error = lfs_update_seguse(fs, *lastseg, *num);
    503      1.38  perseant 			*num = bsize;
    504      1.38  perseant 			*lastseg = seg;
    505      1.38  perseant 		} else
    506      1.38  perseant 			*num += bsize;
    507      1.18  perseant 	}
    508      1.38  perseant 	return error;
    509      1.38  perseant }
    510      1.38  perseant 
    511      1.38  perseant /* Finish the accounting updates for a segment. */
    512      1.38  perseant static int
    513      1.38  perseant lfs_update_seguse(struct lfs *fs, long lastseg, size_t num)
    514      1.38  perseant {
    515      1.38  perseant 	SEGUSE *sup;
    516      1.38  perseant 	struct buf *bp;
    517      1.38  perseant 
    518      1.38  perseant 	if (lastseg < 0 || num == 0)
    519      1.38  perseant 		return 0;
    520      1.35  perseant 
    521      1.38  perseant 
    522      1.38  perseant 	LFS_SEGENTRY(sup, fs, lastseg, bp);
    523      1.38  perseant 	if (num > sup->su_nbytes) {
    524      1.38  perseant 		printf("lfs_truncate: segment %ld short by %ld\n",
    525      1.38  perseant 		       lastseg, (long)num - sup->su_nbytes);
    526      1.38  perseant 		panic("lfs_truncate: negative bytes");
    527      1.38  perseant 		sup->su_nbytes = num;
    528      1.12      fvdl 	}
    529      1.38  perseant 	sup->su_nbytes -= num;
    530      1.44  perseant 	return (VOP_BWRITE(bp)); /* Ifile */
    531      1.38  perseant }
    532      1.38  perseant 
    533      1.38  perseant /*
    534      1.38  perseant  * Release blocks associated with the inode ip and stored in the indirect
    535      1.38  perseant  * block bn.  Blocks are free'd in LIFO order up to (but not including)
    536      1.38  perseant  * lastbn.  If level is greater than SINGLE, the block is an indirect block
    537      1.38  perseant  * and recursive calls to indirtrunc must be used to cleanse other indirect
    538      1.38  perseant  * blocks.
    539      1.38  perseant  *
    540      1.38  perseant  * NB: triple indirect blocks are untested.
    541      1.38  perseant  */
    542      1.38  perseant static int
    543      1.38  perseant lfs_indirtrunc(struct inode *ip, ufs_daddr_t lbn, daddr_t dbn,
    544      1.38  perseant 	       ufs_daddr_t lastbn, int level, long *countp,
    545      1.48  perseant 	       long *rcountp, long *lastsegp, size_t *bcp, struct proc *p)
    546      1.38  perseant {
    547      1.38  perseant 	int i;
    548      1.38  perseant 	struct buf *bp;
    549      1.38  perseant 	struct lfs *fs = ip->i_lfs;
    550      1.38  perseant 	ufs_daddr_t *bap;
    551      1.38  perseant 	struct vnode *vp;
    552      1.38  perseant 	ufs_daddr_t *copy = NULL, nb, nlbn, last;
    553      1.40  perseant 	long blkcount, rblkcount, factor;
    554      1.40  perseant 	int nblocks, blocksreleased = 0, real_released = 0;
    555      1.38  perseant 	int error = 0, allerror = 0;
    556      1.38  perseant 
    557      1.38  perseant 	/*
    558      1.38  perseant 	 * Calculate index in current block of last
    559      1.38  perseant 	 * block to be kept.  -1 indicates the entire
    560      1.38  perseant 	 * block so we need not calculate the index.
    561      1.38  perseant 	 */
    562      1.38  perseant 	factor = 1;
    563      1.38  perseant 	for (i = SINGLE; i < level; i++)
    564      1.38  perseant 		factor *= NINDIR(fs);
    565      1.38  perseant 	last = lastbn;
    566      1.38  perseant 	if (lastbn > 0)
    567      1.38  perseant 		last /= factor;
    568      1.38  perseant 	nblocks = btodb(fs->lfs_bsize);
    569      1.38  perseant 	/*
    570      1.38  perseant 	 * Get buffer of block pointers, zero those entries corresponding
    571      1.38  perseant 	 * to blocks to be free'd, and update on disk copy first.  Since
    572      1.38  perseant 	 * double(triple) indirect before single(double) indirect, calls
    573      1.38  perseant 	 * to bmap on these blocks will fail.  However, we already have
    574      1.38  perseant 	 * the on disk address, so we have to set the b_blkno field
    575      1.38  perseant 	 * explicitly instead of letting bread do everything for us.
    576      1.38  perseant 	 */
    577      1.38  perseant 	vp = ITOV(ip);
    578      1.38  perseant 	bp = getblk(vp, lbn, (int)fs->lfs_bsize, 0, 0);
    579      1.38  perseant 	if (bp->b_flags & (B_DONE | B_DELWRI)) {
    580      1.38  perseant 		/* Braces must be here in case trace evaluates to nothing. */
    581      1.38  perseant 		trace(TR_BREADHIT, pack(vp, fs->lfs_bsize), lbn);
    582      1.38  perseant 	} else {
    583      1.38  perseant 		trace(TR_BREADMISS, pack(vp, fs->lfs_bsize), lbn);
    584      1.48  perseant 		p->p_stats->p_ru.ru_inblock++;	/* pay for read */
    585      1.38  perseant 		bp->b_flags |= B_READ;
    586      1.38  perseant 		if (bp->b_bcount > bp->b_bufsize)
    587      1.38  perseant 			panic("lfs_indirtrunc: bad buffer size");
    588      1.38  perseant 		bp->b_blkno = dbn;
    589      1.38  perseant 		VOP_STRATEGY(bp);
    590      1.38  perseant 		error = biowait(bp);
    591      1.38  perseant 	}
    592      1.38  perseant 	if (error) {
    593      1.38  perseant 		brelse(bp);
    594      1.40  perseant 		*countp = *rcountp = 0;
    595      1.38  perseant 		return (error);
    596      1.38  perseant 	}
    597      1.38  perseant 
    598      1.38  perseant 	bap = (ufs_daddr_t *)bp->b_data;
    599      1.38  perseant 	if (lastbn >= 0) {
    600      1.38  perseant 		MALLOC(copy, ufs_daddr_t *, fs->lfs_bsize, M_TEMP, M_WAITOK);
    601      1.38  perseant 		memcpy((caddr_t)copy, (caddr_t)bap, (u_int)fs->lfs_bsize);
    602      1.38  perseant 		memset((caddr_t)&bap[last + 1], 0,
    603      1.38  perseant 		  (u_int)(NINDIR(fs) - (last + 1)) * sizeof (ufs_daddr_t));
    604      1.38  perseant 		error = VOP_BWRITE(bp);
    605      1.38  perseant 		if (error)
    606      1.38  perseant 			allerror = error;
    607      1.38  perseant 		bap = copy;
    608      1.35  perseant 	}
    609      1.35  perseant 
    610      1.38  perseant 	/*
    611      1.38  perseant 	 * Recursively free totally unused blocks.
    612      1.38  perseant 	 */
    613      1.38  perseant 	for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
    614      1.38  perseant 	    i--, nlbn += factor) {
    615      1.38  perseant 		nb = bap[i];
    616      1.38  perseant 		if (nb == 0)
    617      1.38  perseant 			continue;
    618      1.38  perseant 		if (level > SINGLE) {
    619      1.38  perseant 			error = lfs_indirtrunc(ip, nlbn, nb,
    620      1.38  perseant 					       (ufs_daddr_t)-1, level - 1,
    621      1.40  perseant 					       &blkcount, &rblkcount,
    622      1.48  perseant 					       lastsegp, bcp, p);
    623      1.38  perseant 			if (error)
    624      1.38  perseant 				allerror = error;
    625      1.38  perseant 			blocksreleased += blkcount;
    626      1.40  perseant 			real_released += rblkcount;
    627      1.38  perseant 		}
    628      1.38  perseant 		lfs_blkfree(fs, nb, fs->lfs_bsize, lastsegp, bcp);
    629      1.40  perseant 		if (bap[i] > 0)
    630      1.40  perseant 			real_released += nblocks;
    631      1.38  perseant 		blocksreleased += nblocks;
    632      1.35  perseant 	}
    633      1.35  perseant 
    634      1.38  perseant 	/*
    635      1.38  perseant 	 * Recursively free last partial block.
    636      1.38  perseant 	 */
    637      1.38  perseant 	if (level > SINGLE && lastbn >= 0) {
    638      1.38  perseant 		last = lastbn % factor;
    639      1.38  perseant 		nb = bap[i];
    640      1.38  perseant 		if (nb != 0) {
    641      1.38  perseant 			error = lfs_indirtrunc(ip, nlbn, nb,
    642      1.38  perseant 					       last, level - 1, &blkcount,
    643      1.48  perseant 					       &rblkcount, lastsegp, bcp, p);
    644      1.38  perseant 			if (error)
    645      1.38  perseant 				allerror = error;
    646      1.40  perseant 			real_released += rblkcount;
    647      1.38  perseant 			blocksreleased += blkcount;
    648      1.38  perseant 		}
    649      1.38  perseant 	}
    650      1.35  perseant 
    651      1.38  perseant 	if (copy != NULL) {
    652      1.38  perseant 		FREE(copy, M_TEMP);
    653      1.38  perseant 	} else {
    654      1.47  perseant 		if (bp->b_flags & B_DELWRI) {
    655      1.47  perseant 			LFS_UNLOCK_BUF(bp);
    656      1.47  perseant 			fs->lfs_avail += btodb(bp->b_bcount);
    657      1.47  perseant 			wakeup(&fs->lfs_avail);
    658      1.47  perseant 		}
    659      1.38  perseant 		bp->b_flags |= B_INVAL;
    660      1.38  perseant 		brelse(bp);
    661      1.38  perseant 	}
    662      1.18  perseant 
    663      1.38  perseant 	*countp = blocksreleased;
    664      1.40  perseant 	*rcountp = real_released;
    665      1.38  perseant 	return (allerror);
    666      1.21  perseant }
    667      1.21  perseant 
    668      1.21  perseant /*
    669      1.38  perseant  * Destroy any in core blocks past the truncation length.
    670      1.38  perseant  * Inlined from vtruncbuf, so that lfs_avail could be updated.
    671      1.21  perseant  */
    672      1.38  perseant static int
    673      1.38  perseant lfs_vtruncbuf(vp, lbn, slpflag, slptimeo)
    674      1.32  augustss 	struct vnode *vp;
    675      1.38  perseant 	daddr_t lbn;
    676      1.38  perseant 	int slpflag, slptimeo;
    677      1.21  perseant {
    678      1.38  perseant 	struct buf *bp, *nbp;
    679      1.38  perseant 	int s, error;
    680      1.38  perseant 	struct lfs *fs;
    681      1.38  perseant 
    682      1.38  perseant 	fs = VTOI(vp)->i_lfs;
    683      1.38  perseant 	s = splbio();
    684      1.21  perseant 
    685      1.38  perseant restart:
    686      1.38  perseant 	for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
    687      1.38  perseant 		nbp = LIST_NEXT(bp, b_vnbufs);
    688      1.38  perseant 		if (bp->b_lblkno < lbn)
    689      1.38  perseant 			continue;
    690      1.38  perseant 		if (bp->b_flags & B_BUSY) {
    691      1.38  perseant 			bp->b_flags |= B_WANTED;
    692      1.38  perseant 			error = tsleep((caddr_t)bp, slpflag | (PRIBIO + 1),
    693      1.38  perseant 			    "lfs_vtruncbuf", slptimeo);
    694      1.38  perseant 			if (error) {
    695      1.22  perseant 				splx(s);
    696      1.38  perseant 				return (error);
    697      1.22  perseant 			}
    698      1.38  perseant 			goto restart;
    699      1.38  perseant 		}
    700      1.38  perseant 		bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH;
    701      1.43  perseant 		if (bp->b_flags & B_DELWRI) {
    702      1.43  perseant 			bp->b_flags &= ~B_DELWRI;
    703      1.41  perseant 			fs->lfs_avail += btodb(bp->b_bcount);
    704      1.47  perseant 			wakeup(&fs->lfs_avail);
    705      1.43  perseant 		}
    706      1.46  perseant 		LFS_UNLOCK_BUF(bp);
    707      1.38  perseant 		brelse(bp);
    708      1.38  perseant 	}
    709      1.38  perseant 
    710      1.38  perseant 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
    711      1.38  perseant 		nbp = LIST_NEXT(bp, b_vnbufs);
    712      1.38  perseant 		if (bp->b_lblkno < lbn)
    713      1.38  perseant 			continue;
    714      1.38  perseant 		if (bp->b_flags & B_BUSY) {
    715      1.38  perseant 			bp->b_flags |= B_WANTED;
    716      1.38  perseant 			error = tsleep((caddr_t)bp, slpflag | (PRIBIO + 1),
    717      1.38  perseant 			    "lfs_vtruncbuf", slptimeo);
    718      1.38  perseant 			if (error) {
    719      1.35  perseant 				splx(s);
    720      1.38  perseant 				return (error);
    721      1.21  perseant 			}
    722      1.38  perseant 			goto restart;
    723      1.38  perseant 		}
    724      1.38  perseant 		bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH;
    725      1.43  perseant 		if (bp->b_flags & B_DELWRI) {
    726      1.43  perseant 			bp->b_flags &= ~B_DELWRI;
    727      1.41  perseant 			fs->lfs_avail += btodb(bp->b_bcount);
    728      1.47  perseant 			wakeup(&fs->lfs_avail);
    729      1.43  perseant 		}
    730      1.46  perseant 		LFS_UNLOCK_BUF(bp);
    731      1.38  perseant 		brelse(bp);
    732      1.21  perseant 	}
    733      1.38  perseant 
    734      1.38  perseant 	splx(s);
    735      1.38  perseant 
    736      1.21  perseant 	return (0);
    737       1.1   mycroft }
    738      1.38  perseant 
    739