Home | History | Annotate | Line # | Download | only in lfs
lfs_inode.c revision 1.47
      1 /*	$NetBSD: lfs_inode.c,v 1.47 2000/11/21 00:00:31 perseant Exp $	*/
      2 
      3 /*-
      4  * Copyright (c) 1999, 2000 The NetBSD Foundation, Inc.
      5  * All rights reserved.
      6  *
      7  * This code is derived from software contributed to The NetBSD Foundation
      8  * by Konrad E. Schroder <perseant (at) hhhh.org>.
      9  *
     10  * Redistribution and use in source and binary forms, with or without
     11  * modification, are permitted provided that the following conditions
     12  * are met:
     13  * 1. Redistributions of source code must retain the above copyright
     14  *    notice, this list of conditions and the following disclaimer.
     15  * 2. Redistributions in binary form must reproduce the above copyright
     16  *    notice, this list of conditions and the following disclaimer in the
     17  *    documentation and/or other materials provided with the distribution.
     18  * 3. All advertising materials mentioning features or use of this software
     19  *    must display the following acknowledgement:
     20  *      This product includes software developed by the NetBSD
     21  *      Foundation, Inc. and its contributors.
     22  * 4. Neither the name of The NetBSD Foundation nor the names of its
     23  *    contributors may be used to endorse or promote products derived
     24  *    from this software without specific prior written permission.
     25  *
     26  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     27  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     28  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     29  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     30  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     31  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     32  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     33  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     34  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     35  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     36  * POSSIBILITY OF SUCH DAMAGE.
     37  */
     38 /*
     39  * Copyright (c) 1986, 1989, 1991, 1993
     40  *	The Regents of the University of California.  All rights reserved.
     41  *
     42  * Redistribution and use in source and binary forms, with or without
     43  * modification, are permitted provided that the following conditions
     44  * are met:
     45  * 1. Redistributions of source code must retain the above copyright
     46  *    notice, this list of conditions and the following disclaimer.
     47  * 2. Redistributions in binary form must reproduce the above copyright
     48  *    notice, this list of conditions and the following disclaimer in the
     49  *    documentation and/or other materials provided with the distribution.
     50  * 3. All advertising materials mentioning features or use of this software
     51  *    must display the following acknowledgement:
     52  *	This product includes software developed by the University of
     53  *	California, Berkeley and its contributors.
     54  * 4. Neither the name of the University nor the names of its contributors
     55  *    may be used to endorse or promote products derived from this software
     56  *    without specific prior written permission.
     57  *
     58  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     59  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     60  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     61  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     62  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     63  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     64  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     65  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     66  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     67  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     68  * SUCH DAMAGE.
     69  *
     70  *	@(#)lfs_inode.c	8.9 (Berkeley) 5/8/95
     71  */
     72 
     73 #if defined(_KERNEL) && !defined(_LKM)
     74 #include "opt_quota.h"
     75 #endif
     76 
     77 #include <sys/param.h>
     78 #include <sys/systm.h>
     79 #include <sys/mount.h>
     80 #include <sys/proc.h>
     81 #include <sys/file.h>
     82 #include <sys/buf.h>
     83 #include <sys/vnode.h>
     84 #include <sys/kernel.h>
     85 #include <sys/malloc.h>
     86 #include <sys/trace.h>
     87 #include <sys/resourcevar.h>
     88 
     89 #include <ufs/ufs/quota.h>
     90 #include <ufs/ufs/inode.h>
     91 #include <ufs/ufs/ufsmount.h>
     92 #include <ufs/ufs/ufs_extern.h>
     93 
     94 #include <ufs/lfs/lfs.h>
     95 #include <ufs/lfs/lfs_extern.h>
     96 
     97 extern int locked_queue_count;
     98 extern long locked_queue_bytes;
     99 
    100 static int lfs_update_seguse(struct lfs *, long, size_t);
    101 static int lfs_indirtrunc (struct inode *, ufs_daddr_t, ufs_daddr_t,
    102 			   ufs_daddr_t, int, long *, long *, long *, size_t *);
    103 static int lfs_blkfree (struct lfs *, daddr_t, size_t, long *, size_t *);
    104 static int lfs_vtruncbuf(struct vnode *, daddr_t, int, int);
    105 
    106 /* Search a block for a specific dinode. */
    107 struct dinode *
    108 lfs_ifind(fs, ino, bp)
    109 	struct lfs *fs;
    110 	ino_t ino;
    111 	struct buf *bp;
    112 {
    113 	int cnt;
    114 	struct dinode *dip = (struct dinode *)bp->b_data;
    115 	struct dinode *ldip;
    116 
    117 	for (cnt = INOPB(fs), ldip = dip + (cnt - 1); cnt--; --ldip)
    118 		if (ldip->di_inumber == ino)
    119 			return (ldip);
    120 
    121 	printf("offset is 0x%x (seg %d)\n", fs->lfs_offset,
    122 	       datosn(fs,fs->lfs_offset));
    123 	printf("block is 0x%x (seg %d)\n", bp->b_blkno,
    124 	       datosn(fs,bp->b_blkno));
    125 	panic("lfs_ifind: dinode %u not found", ino);
    126 	/* NOTREACHED */
    127 }
    128 
    129 int
    130 lfs_update(v)
    131 	void *v;
    132 {
    133 	struct vop_update_args /* {
    134 				  struct vnode *a_vp;
    135 				  struct timespec *a_access;
    136 				  struct timespec *a_modify;
    137 				  int a_flags;
    138 				  } */ *ap = v;
    139 	struct inode *ip;
    140 	struct vnode *vp = ap->a_vp;
    141 	int oflag;
    142 	struct timespec ts;
    143 	struct lfs *fs = VFSTOUFS(vp->v_mount)->um_lfs;
    144 
    145 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
    146 		return (0);
    147 	ip = VTOI(vp);
    148 
    149 	/*
    150 	 * If we are called from vinvalbuf, and the file's blocks have
    151 	 * already been scheduled for writing, but the writes have not
    152 	 * yet completed, lfs_vflush will not be called, and vinvalbuf
    153 	 * will cause a panic.  So, we must wait until any pending write
    154 	 * for our inode completes, if we are called with UPDATE_WAIT set.
    155 	 */
    156 	while((ap->a_flags & (UPDATE_WAIT|UPDATE_DIROP)) == UPDATE_WAIT &&
    157 	    WRITEINPROG(vp)) {
    158 #ifdef DEBUG_LFS
    159 		printf("lfs_update: sleeping on inode %d (in-progress)\n",
    160 		       ip->i_number);
    161 #endif
    162 		tsleep(vp, (PRIBIO+1), "lfs_update", 0);
    163 	}
    164 	oflag = ip->i_flag;
    165 	TIMEVAL_TO_TIMESPEC(&time, &ts);
    166 	LFS_ITIMES(ip,
    167 		   ap->a_access ? ap->a_access : &ts,
    168 		   ap->a_modify ? ap->a_modify : &ts, &ts);
    169 	if ((ip->i_flag & (IN_MODIFIED | IN_ACCESSED | IN_CLEANING)) == 0) {
    170 		return (0);
    171 	}
    172 
    173 	/* If sync, push back the vnode and any dirty blocks it may have. */
    174 	if((ap->a_flags & (UPDATE_WAIT|UPDATE_DIROP))==UPDATE_WAIT) {
    175 		/* Avoid flushing VDIROP. */
    176 		++fs->lfs_diropwait;
    177 		while(vp->v_flag & VDIROP) {
    178 #ifdef DEBUG_LFS
    179 			printf("lfs_update: sleeping on inode %d (dirops)\n",
    180 			       ip->i_number);
    181 			printf("lfs_update: vflags 0x%lx, iflags 0x%x\n",
    182 				vp->v_flag, ip->i_flag);
    183 #endif
    184 			if(fs->lfs_dirops == 0)
    185 				lfs_flush_fs(fs, SEGM_SYNC);
    186 			else
    187 				tsleep(&fs->lfs_writer, PRIBIO+1, "lfs_fsync",
    188 				       0);
    189 			/* XXX KS - by falling out here, are we writing the vn
    190 			twice? */
    191 		}
    192 		--fs->lfs_diropwait;
    193 		return lfs_vflush(vp);
    194         }
    195 	return 0;
    196 }
    197 
    198 #define	SINGLE	0	/* index of single indirect block */
    199 #define	DOUBLE	1	/* index of double indirect block */
    200 #define	TRIPLE	2	/* index of triple indirect block */
    201 /*
    202  * Truncate the inode oip to at most length size, freeing the
    203  * disk blocks.
    204  */
    205 /* VOP_BWRITE 1 + NIADDR + VOP_BALLOC == 2 + 2*NIADDR times */
    206 int
    207 lfs_truncate(v)
    208 	void *v;
    209 {
    210 	struct vop_truncate_args /* {
    211 		struct vnode *a_vp;
    212 		off_t a_length;
    213 		int a_flags;
    214 		struct ucred *a_cred;
    215 		struct proc *a_p;
    216 	} */ *ap = v;
    217 	struct vnode *ovp = ap->a_vp;
    218 	ufs_daddr_t lastblock;
    219 	struct inode *oip;
    220 	ufs_daddr_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR];
    221 	ufs_daddr_t newblks[NDADDR + NIADDR];
    222 	off_t length = ap->a_length;
    223 	struct lfs *fs;
    224 	struct buf *bp;
    225 	int offset, size, level;
    226 	long count, rcount, nblocks, blocksreleased = 0, real_released = 0;
    227 	int i;
    228 	int aflags, error, allerror = 0;
    229 	off_t osize;
    230 	long lastseg;
    231 	size_t bc;
    232 	int obufsize, odb;
    233 
    234 	if (length < 0)
    235 		return (EINVAL);
    236 	oip = VTOI(ovp);
    237 
    238 	/*
    239 	 * Just return and not update modification times.
    240 	 */
    241 	if (oip->i_ffs_size == length)
    242 		return (0);
    243 
    244 	if (ovp->v_type == VLNK &&
    245 	    (oip->i_ffs_size < ovp->v_mount->mnt_maxsymlinklen ||
    246 	     (ovp->v_mount->mnt_maxsymlinklen == 0 &&
    247 	      oip->i_din.ffs_din.di_blocks == 0))) {
    248 #ifdef DIAGNOSTIC
    249 		if (length != 0)
    250 			panic("lfs_truncate: partial truncate of symlink");
    251 #endif
    252 		memset((char *)&oip->i_ffs_shortlink, 0, (u_int)oip->i_ffs_size);
    253 		oip->i_ffs_size = 0;
    254 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
    255 		return (VOP_UPDATE(ovp, NULL, NULL, 0));
    256 	}
    257 	if (oip->i_ffs_size == length) {
    258 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
    259 		return (VOP_UPDATE(ovp, NULL, NULL, 0));
    260 	}
    261 #ifdef QUOTA
    262 	if ((error = getinoquota(oip)) != 0)
    263 		return (error);
    264 #endif
    265 	fs = oip->i_lfs;
    266 	lfs_imtime(fs);
    267 	osize = oip->i_ffs_size;
    268 	ovp->v_lasta = ovp->v_clen = ovp->v_cstart = ovp->v_lastw = 0;
    269 
    270 	/*
    271 	 * Lengthen the size of the file. We must ensure that the
    272 	 * last byte of the file is allocated. Since the smallest
    273 	 * value of osize is 0, length will be at least 1.
    274 	 */
    275 	if (osize < length) {
    276 		if (length > fs->lfs_maxfilesize)
    277 			return (EFBIG);
    278 		aflags = B_CLRBUF;
    279 		if (ap->a_flags & IO_SYNC)
    280 			aflags |= B_SYNC;
    281 		error = lfs_reserve(fs, ovp, fsbtodb(fs, NIADDR + 2));
    282 		if (error)
    283 			return (error);
    284 		error = VOP_BALLOC(ovp, length - 1, 1, ap->a_cred, aflags, &bp);
    285 		lfs_reserve(fs, ovp, -fsbtodb(fs, NIADDR + 2));
    286 		if (error)
    287 			return (error);
    288 		oip->i_ffs_size = length;
    289 		uvm_vnp_setsize(ovp, length);
    290 		(void) uvm_vnp_uncache(ovp);
    291 		(void) VOP_BWRITE(bp);
    292 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
    293 		return (VOP_UPDATE(ovp, NULL, NULL, 0));
    294 	}
    295 
    296 	if ((error = lfs_reserve(fs, ovp, fsbtodb(fs, 2 * NIADDR + 3))) != 0)
    297 		return (error);
    298 	/*
    299 	 * Make sure no writes to this inode can happen while we're
    300 	 * truncating.  Otherwise, blocks which are accounted for on the
    301 	 * inode *and* which have been created for cleaning can coexist,
    302 	 * and cause an overcounting.
    303 	 *
    304 	 * (We don't need to *hold* the seglock, though, because we already
    305 	 * hold the inode lock; draining the seglock is sufficient.)
    306 	 */
    307 	if (ovp != fs->lfs_unlockvp) {
    308 		while(fs->lfs_seglock) {
    309 			tsleep(&fs->lfs_seglock, PRIBIO+1, "lfs_truncate", 0);
    310 		}
    311 	}
    312 
    313 	/*
    314 	 * Shorten the size of the file. If the file is not being
    315 	 * truncated to a block boundary, the contents of the
    316 	 * partial block following the end of the file must be
    317 	 * zero'ed in case it ever becomes accessible again because
    318 	 * of subsequent file growth. Directories however are not
    319 	 * zero'ed as they should grow back initialized to empty.
    320 	 */
    321 	offset = blkoff(fs, length);
    322 	lastseg = -1;
    323 	bc = 0;
    324 	if (offset == 0) {
    325 		oip->i_ffs_size = length;
    326 	} else {
    327 		lbn = lblkno(fs, length);
    328 		aflags = B_CLRBUF;
    329 		if (ap->a_flags & IO_SYNC)
    330 			aflags |= B_SYNC;
    331 		error = VOP_BALLOC(ovp, length - 1, 1, ap->a_cred, aflags, &bp);
    332 		if (error) {
    333 			lfs_reserve(fs, ovp, -fsbtodb(fs, 2 * NIADDR + 3));
    334 			return (error);
    335 		}
    336 		obufsize = bp->b_bufsize;
    337 		odb = btodb(bp->b_bcount);
    338 		oip->i_ffs_size = length;
    339 		size = blksize(fs, oip, lbn);
    340 		(void) uvm_vnp_uncache(ovp);
    341 		if (ovp->v_type != VDIR)
    342 			memset((char *)bp->b_data + offset, 0,
    343 			       (u_int)(size - offset));
    344 		allocbuf(bp, size);
    345 		if (bp->b_flags & B_DELWRI) {
    346 			if ((bp->b_flags & (B_LOCKED | B_CALL)) == B_LOCKED)
    347 				locked_queue_bytes -= obufsize - bp->b_bufsize;
    348 			fs->lfs_avail += odb - btodb(size);
    349 		}
    350 		(void) VOP_BWRITE(bp);
    351 	}
    352 	uvm_vnp_setsize(ovp, length);
    353 	/*
    354 	 * Calculate index into inode's block list of
    355 	 * last direct and indirect blocks (if any)
    356 	 * which we want to keep.  Lastblock is -1 when
    357 	 * the file is truncated to 0.
    358 	 */
    359 	lastblock = lblkno(fs, length + fs->lfs_bsize - 1) - 1;
    360 	lastiblock[SINGLE] = lastblock - NDADDR;
    361 	lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
    362 	lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
    363 	nblocks = btodb(fs->lfs_bsize);
    364 	/*
    365 	 * Record changed file and block pointers before we start
    366 	 * freeing blocks.  lastiblock values are also normalized to -1
    367 	 * for calls to lfs_indirtrunc below.
    368 	 */
    369 	memcpy((caddr_t)newblks, (caddr_t)&oip->i_ffs_db[0], sizeof newblks);
    370 	for (level = TRIPLE; level >= SINGLE; level--)
    371 		if (lastiblock[level] < 0) {
    372 			newblks[NDADDR+level] = 0;
    373 			lastiblock[level] = -1;
    374 		}
    375 	for (i = NDADDR - 1; i > lastblock; i--)
    376 		newblks[i] = 0;
    377 
    378 	oip->i_ffs_size = osize;
    379 	error = lfs_vtruncbuf(ovp, lastblock + 1, 0, 0);
    380 	if (error && !allerror)
    381 		allerror = error;
    382 
    383 	/*
    384 	 * Indirect blocks first.
    385 	 */
    386 	indir_lbn[SINGLE] = -NDADDR;
    387 	indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
    388 	indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
    389 	for (level = TRIPLE; level >= SINGLE; level--) {
    390 		bn = oip->i_ffs_ib[level];
    391 		if (bn != 0) {
    392 			error = lfs_indirtrunc(oip, indir_lbn[level],
    393 					       bn, lastiblock[level],
    394 					       level, &count, &rcount,
    395 					       &lastseg, &bc);
    396 			if (error)
    397 				allerror = error;
    398 			real_released += rcount;
    399 			blocksreleased += count;
    400 			if (lastiblock[level] < 0) {
    401 				if (oip->i_ffs_ib[level] > 0)
    402 					real_released += nblocks;
    403 				blocksreleased += nblocks;
    404 				oip->i_ffs_ib[level] = 0;
    405 				lfs_blkfree(fs, bn, fs->lfs_bsize, &lastseg, &bc);
    406 			}
    407 		}
    408 		if (lastiblock[level] >= 0)
    409 			goto done;
    410 	}
    411 
    412 	/*
    413 	 * All whole direct blocks or frags.
    414 	 */
    415 	for (i = NDADDR - 1; i > lastblock; i--) {
    416 		long bsize;
    417 
    418 		bn = oip->i_ffs_db[i];
    419 		if (bn == 0)
    420 			continue;
    421 		bsize = blksize(fs, oip, i);
    422 		if (oip->i_ffs_db[i] > 0)
    423 			real_released += btodb(bsize);
    424 		blocksreleased += btodb(bsize);
    425 		oip->i_ffs_db[i] = 0;
    426 		lfs_blkfree(fs, bn, bsize, &lastseg, &bc);
    427 	}
    428 	if (lastblock < 0)
    429 		goto done;
    430 
    431 	/*
    432 	 * Finally, look for a change in size of the
    433 	 * last direct block; release any frags.
    434 	 */
    435 	bn = oip->i_ffs_db[lastblock];
    436 	if (bn != 0) {
    437 		long oldspace, newspace;
    438 
    439 		/*
    440 		 * Calculate amount of space we're giving
    441 		 * back as old block size minus new block size.
    442 		 */
    443 		oldspace = blksize(fs, oip, lastblock);
    444 		oip->i_ffs_size = length;
    445 		newspace = blksize(fs, oip, lastblock);
    446 		if (newspace == 0)
    447 			panic("itrunc: newspace");
    448 		if (oldspace - newspace > 0) {
    449 			lfs_blkfree(fs, bn, oldspace - newspace, &lastseg, &bc);
    450 			if (bn > 0)
    451 				real_released += btodb(oldspace - newspace);
    452 			blocksreleased += btodb(oldspace - newspace);
    453 		}
    454 	}
    455 
    456 done:
    457 	/* Finish segment accounting corrections */
    458 	lfs_update_seguse(fs, lastseg, bc);
    459 #ifdef DIAGNOSTIC
    460 	for (level = SINGLE; level <= TRIPLE; level++)
    461 		if (newblks[NDADDR + level] != oip->i_ffs_ib[level])
    462 			panic("lfs itrunc1");
    463 	for (i = 0; i < NDADDR; i++)
    464 		if (newblks[i] != oip->i_ffs_db[i])
    465 			panic("lfs itrunc2");
    466 	if (length == 0 &&
    467 	    (!LIST_EMPTY(&ovp->v_cleanblkhd) || !LIST_EMPTY(&ovp->v_dirtyblkhd)))
    468 		panic("lfs itrunc3");
    469 #endif /* DIAGNOSTIC */
    470 	/*
    471 	 * Put back the real size.
    472 	 */
    473 	oip->i_ffs_size = length;
    474 	oip->i_lfs_effnblks -= blocksreleased;
    475 	oip->i_ffs_blocks -= real_released;
    476 	fs->lfs_bfree += blocksreleased;
    477 #ifdef DIAGNOSTIC
    478 	if (oip->i_ffs_size == 0 && oip->i_ffs_blocks != 0) {
    479 		printf("lfs_truncate: truncate to 0 but %d blocks on inode\n",
    480 		       oip->i_ffs_blocks);
    481 		panic("lfs_truncate: persistent blocks\n");
    482 	}
    483 #endif
    484 	oip->i_flag |= IN_CHANGE;
    485 #ifdef QUOTA
    486 	(void) chkdq(oip, -blocksreleased, NOCRED, 0);
    487 #endif
    488 	lfs_reserve(fs, ovp, -fsbtodb(fs, 2 * NIADDR + 3));
    489 	return (allerror);
    490 }
    491 
    492 /* Update segment usage information when removing a block. */
    493 static int
    494 lfs_blkfree(struct lfs *fs, daddr_t daddr, size_t bsize, long *lastseg,
    495 	    size_t *num)
    496 {
    497 	long seg;
    498 	int error = 0;
    499 
    500 	bsize = fragroundup(fs, bsize);
    501 	if (daddr > 0) {
    502 		if (*lastseg != (seg = datosn(fs, daddr))) {
    503 			error = lfs_update_seguse(fs, *lastseg, *num);
    504 			*num = bsize;
    505 			*lastseg = seg;
    506 		} else
    507 			*num += bsize;
    508 	}
    509 	return error;
    510 }
    511 
    512 /* Finish the accounting updates for a segment. */
    513 static int
    514 lfs_update_seguse(struct lfs *fs, long lastseg, size_t num)
    515 {
    516 	SEGUSE *sup;
    517 	struct buf *bp;
    518 
    519 	if (lastseg < 0 || num == 0)
    520 		return 0;
    521 
    522 
    523 	LFS_SEGENTRY(sup, fs, lastseg, bp);
    524 	if (num > sup->su_nbytes) {
    525 		printf("lfs_truncate: segment %ld short by %ld\n",
    526 		       lastseg, (long)num - sup->su_nbytes);
    527 		panic("lfs_truncate: negative bytes");
    528 		sup->su_nbytes = num;
    529 	}
    530 	sup->su_nbytes -= num;
    531 	return (VOP_BWRITE(bp)); /* Ifile */
    532 }
    533 
    534 /*
    535  * Release blocks associated with the inode ip and stored in the indirect
    536  * block bn.  Blocks are free'd in LIFO order up to (but not including)
    537  * lastbn.  If level is greater than SINGLE, the block is an indirect block
    538  * and recursive calls to indirtrunc must be used to cleanse other indirect
    539  * blocks.
    540  *
    541  * NB: triple indirect blocks are untested.
    542  */
    543 static int
    544 lfs_indirtrunc(struct inode *ip, ufs_daddr_t lbn, daddr_t dbn,
    545 	       ufs_daddr_t lastbn, int level, long *countp,
    546 	       long *rcountp, long *lastsegp, size_t *bcp)
    547 {
    548 	int i;
    549 	struct buf *bp;
    550 	struct lfs *fs = ip->i_lfs;
    551 	ufs_daddr_t *bap;
    552 	struct vnode *vp;
    553 	ufs_daddr_t *copy = NULL, nb, nlbn, last;
    554 	long blkcount, rblkcount, factor;
    555 	int nblocks, blocksreleased = 0, real_released = 0;
    556 	int error = 0, allerror = 0;
    557 
    558 	/*
    559 	 * Calculate index in current block of last
    560 	 * block to be kept.  -1 indicates the entire
    561 	 * block so we need not calculate the index.
    562 	 */
    563 	factor = 1;
    564 	for (i = SINGLE; i < level; i++)
    565 		factor *= NINDIR(fs);
    566 	last = lastbn;
    567 	if (lastbn > 0)
    568 		last /= factor;
    569 	nblocks = btodb(fs->lfs_bsize);
    570 	/*
    571 	 * Get buffer of block pointers, zero those entries corresponding
    572 	 * to blocks to be free'd, and update on disk copy first.  Since
    573 	 * double(triple) indirect before single(double) indirect, calls
    574 	 * to bmap on these blocks will fail.  However, we already have
    575 	 * the on disk address, so we have to set the b_blkno field
    576 	 * explicitly instead of letting bread do everything for us.
    577 	 */
    578 	vp = ITOV(ip);
    579 	bp = getblk(vp, lbn, (int)fs->lfs_bsize, 0, 0);
    580 	if (bp->b_flags & (B_DONE | B_DELWRI)) {
    581 		/* Braces must be here in case trace evaluates to nothing. */
    582 		trace(TR_BREADHIT, pack(vp, fs->lfs_bsize), lbn);
    583 	} else {
    584 		trace(TR_BREADMISS, pack(vp, fs->lfs_bsize), lbn);
    585 		curproc->p_stats->p_ru.ru_inblock++;	/* pay for read */
    586 		bp->b_flags |= B_READ;
    587 		if (bp->b_bcount > bp->b_bufsize)
    588 			panic("lfs_indirtrunc: bad buffer size");
    589 		bp->b_blkno = dbn;
    590 		VOP_STRATEGY(bp);
    591 		error = biowait(bp);
    592 	}
    593 	if (error) {
    594 		brelse(bp);
    595 		*countp = *rcountp = 0;
    596 		return (error);
    597 	}
    598 
    599 	bap = (ufs_daddr_t *)bp->b_data;
    600 	if (lastbn >= 0) {
    601 		MALLOC(copy, ufs_daddr_t *, fs->lfs_bsize, M_TEMP, M_WAITOK);
    602 		memcpy((caddr_t)copy, (caddr_t)bap, (u_int)fs->lfs_bsize);
    603 		memset((caddr_t)&bap[last + 1], 0,
    604 		  (u_int)(NINDIR(fs) - (last + 1)) * sizeof (ufs_daddr_t));
    605 		error = VOP_BWRITE(bp);
    606 		if (error)
    607 			allerror = error;
    608 		bap = copy;
    609 	}
    610 
    611 	/*
    612 	 * Recursively free totally unused blocks.
    613 	 */
    614 	for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
    615 	    i--, nlbn += factor) {
    616 		nb = bap[i];
    617 		if (nb == 0)
    618 			continue;
    619 		if (level > SINGLE) {
    620 			error = lfs_indirtrunc(ip, nlbn, nb,
    621 					       (ufs_daddr_t)-1, level - 1,
    622 					       &blkcount, &rblkcount,
    623 					       lastsegp, bcp);
    624 			if (error)
    625 				allerror = error;
    626 			blocksreleased += blkcount;
    627 			real_released += rblkcount;
    628 		}
    629 		lfs_blkfree(fs, nb, fs->lfs_bsize, lastsegp, bcp);
    630 		if (bap[i] > 0)
    631 			real_released += nblocks;
    632 		blocksreleased += nblocks;
    633 	}
    634 
    635 	/*
    636 	 * Recursively free last partial block.
    637 	 */
    638 	if (level > SINGLE && lastbn >= 0) {
    639 		last = lastbn % factor;
    640 		nb = bap[i];
    641 		if (nb != 0) {
    642 			error = lfs_indirtrunc(ip, nlbn, nb,
    643 					       last, level - 1, &blkcount,
    644 					       &rblkcount, lastsegp, bcp);
    645 			if (error)
    646 				allerror = error;
    647 			real_released += rblkcount;
    648 			blocksreleased += blkcount;
    649 		}
    650 	}
    651 
    652 	if (copy != NULL) {
    653 		FREE(copy, M_TEMP);
    654 	} else {
    655 		if (bp->b_flags & B_DELWRI) {
    656 			LFS_UNLOCK_BUF(bp);
    657 			fs->lfs_avail += btodb(bp->b_bcount);
    658 			wakeup(&fs->lfs_avail);
    659 		}
    660 		bp->b_flags |= B_INVAL;
    661 		brelse(bp);
    662 	}
    663 
    664 	*countp = blocksreleased;
    665 	*rcountp = real_released;
    666 	return (allerror);
    667 }
    668 
    669 /*
    670  * Destroy any in core blocks past the truncation length.
    671  * Inlined from vtruncbuf, so that lfs_avail could be updated.
    672  */
    673 static int
    674 lfs_vtruncbuf(vp, lbn, slpflag, slptimeo)
    675 	struct vnode *vp;
    676 	daddr_t lbn;
    677 	int slpflag, slptimeo;
    678 {
    679 	struct buf *bp, *nbp;
    680 	int s, error;
    681 	struct lfs *fs;
    682 
    683 	fs = VTOI(vp)->i_lfs;
    684 	s = splbio();
    685 
    686 restart:
    687 	for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
    688 		nbp = LIST_NEXT(bp, b_vnbufs);
    689 		if (bp->b_lblkno < lbn)
    690 			continue;
    691 		if (bp->b_flags & B_BUSY) {
    692 			bp->b_flags |= B_WANTED;
    693 			error = tsleep((caddr_t)bp, slpflag | (PRIBIO + 1),
    694 			    "lfs_vtruncbuf", slptimeo);
    695 			if (error) {
    696 				splx(s);
    697 				return (error);
    698 			}
    699 			goto restart;
    700 		}
    701 		bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH;
    702 		if (bp->b_flags & B_DELWRI) {
    703 			bp->b_flags &= ~B_DELWRI;
    704 			fs->lfs_avail += btodb(bp->b_bcount);
    705 			wakeup(&fs->lfs_avail);
    706 		}
    707 		LFS_UNLOCK_BUF(bp);
    708 		brelse(bp);
    709 	}
    710 
    711 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
    712 		nbp = LIST_NEXT(bp, b_vnbufs);
    713 		if (bp->b_lblkno < lbn)
    714 			continue;
    715 		if (bp->b_flags & B_BUSY) {
    716 			bp->b_flags |= B_WANTED;
    717 			error = tsleep((caddr_t)bp, slpflag | (PRIBIO + 1),
    718 			    "lfs_vtruncbuf", slptimeo);
    719 			if (error) {
    720 				splx(s);
    721 				return (error);
    722 			}
    723 			goto restart;
    724 		}
    725 		bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH;
    726 		if (bp->b_flags & B_DELWRI) {
    727 			bp->b_flags &= ~B_DELWRI;
    728 			fs->lfs_avail += btodb(bp->b_bcount);
    729 			wakeup(&fs->lfs_avail);
    730 		}
    731 		LFS_UNLOCK_BUF(bp);
    732 		brelse(bp);
    733 	}
    734 
    735 	splx(s);
    736 
    737 	return (0);
    738 }
    739 
    740