Home | History | Annotate | Line # | Download | only in ffs
ffs_inode.c revision 1.62
      1 /*	$NetBSD: ffs_inode.c,v 1.62 2004/01/25 18:06:49 hannken Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 1982, 1986, 1989, 1993
      5  *	The Regents of the University of California.  All rights reserved.
      6  *
      7  * Redistribution and use in source and binary forms, with or without
      8  * modification, are permitted provided that the following conditions
      9  * are met:
     10  * 1. Redistributions of source code must retain the above copyright
     11  *    notice, this list of conditions and the following disclaimer.
     12  * 2. Redistributions in binary form must reproduce the above copyright
     13  *    notice, this list of conditions and the following disclaimer in the
     14  *    documentation and/or other materials provided with the distribution.
     15  * 3. Neither the name of the University nor the names of its contributors
     16  *    may be used to endorse or promote products derived from this software
     17  *    without specific prior written permission.
     18  *
     19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     29  * SUCH DAMAGE.
     30  *
     31  *	@(#)ffs_inode.c	8.13 (Berkeley) 4/21/95
     32  */
     33 
     34 #include <sys/cdefs.h>
     35 __KERNEL_RCSID(0, "$NetBSD: ffs_inode.c,v 1.62 2004/01/25 18:06:49 hannken Exp $");
     36 
     37 #if defined(_KERNEL_OPT)
     38 #include "opt_ffs.h"
     39 #include "opt_quota.h"
     40 #endif
     41 
     42 #include <sys/param.h>
     43 #include <sys/systm.h>
     44 #include <sys/mount.h>
     45 #include <sys/proc.h>
     46 #include <sys/file.h>
     47 #include <sys/buf.h>
     48 #include <sys/vnode.h>
     49 #include <sys/kernel.h>
     50 #include <sys/malloc.h>
     51 #include <sys/trace.h>
     52 #include <sys/resourcevar.h>
     53 
     54 #include <ufs/ufs/quota.h>
     55 #include <ufs/ufs/inode.h>
     56 #include <ufs/ufs/ufsmount.h>
     57 #include <ufs/ufs/ufs_extern.h>
     58 #include <ufs/ufs/ufs_bswap.h>
     59 
     60 #include <ufs/ffs/fs.h>
     61 #include <ufs/ffs/ffs_extern.h>
     62 
     63 static int ffs_indirtrunc __P((struct inode *, daddr_t, daddr_t,
     64 			       daddr_t, int, int64_t *));
     65 
     66 /*
     67  * Update the access, modified, and inode change times as specified
     68  * by the IN_ACCESS, IN_UPDATE, and IN_CHANGE flags respectively.
     69  * The IN_MODIFIED flag is used to specify that the inode needs to be
     70  * updated but that the times have already been set. The access
     71  * and modified times are taken from the second and third parameters;
     72  * the inode change time is always taken from the current time. If
     73  * UPDATE_WAIT flag is set, or UPDATE_DIROP is set and we are not doing
     74  * softupdates, then wait for the disk write of the inode to complete.
     75  */
     76 
     77 int
     78 ffs_update(v)
     79 	void *v;
     80 {
     81 	struct vop_update_args /* {
     82 		struct vnode *a_vp;
     83 		struct timespec *a_access;
     84 		struct timespec *a_modify;
     85 		int a_flags;
     86 	} */ *ap = v;
     87 	struct fs *fs;
     88 	struct buf *bp;
     89 	struct inode *ip;
     90 	int error;
     91 	struct timespec ts;
     92 	caddr_t cp;
     93 	int waitfor, flags;
     94 
     95 	if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
     96 		return (0);
     97 	ip = VTOI(ap->a_vp);
     98 	TIMEVAL_TO_TIMESPEC(&time, &ts);
     99 	FFS_ITIMES(ip,
    100 	    ap->a_access ? ap->a_access : &ts,
    101 	    ap->a_modify ? ap->a_modify : &ts, &ts);
    102 	flags = ip->i_flag & (IN_MODIFIED | IN_ACCESSED);
    103 	if (flags == 0)
    104 		return (0);
    105 	fs = ip->i_fs;
    106 
    107 	if ((flags & IN_MODIFIED) != 0 &&
    108 	    (ap->a_vp->v_mount->mnt_flag & MNT_ASYNC) == 0) {
    109 		waitfor = ap->a_flags & UPDATE_WAIT;
    110 		if ((ap->a_flags & UPDATE_DIROP) && !DOINGSOFTDEP(ap->a_vp))
    111 			waitfor |= UPDATE_WAIT;
    112 	} else
    113 		waitfor = 0;
    114 
    115 	/*
    116 	 * Ensure that uid and gid are correct. This is a temporary
    117 	 * fix until fsck has been changed to do the update.
    118 	 */
    119 	if (fs->fs_magic == FS_UFS1_MAGIC &&			/* XXX */
    120 	    fs->fs_old_inodefmt < FS_44INODEFMT) {		/* XXX */
    121 		ip->i_ffs1_ouid = ip->i_uid;	/* XXX */
    122 		ip->i_ffs1_ogid = ip->i_gid;	/* XXX */
    123 	}							/* XXX */
    124 	error = bread(ip->i_devvp,
    125 		      fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
    126 		      (int)fs->fs_bsize, NOCRED, &bp);
    127 	if (error) {
    128 		brelse(bp);
    129 		return (error);
    130 	}
    131 	ip->i_flag &= ~(IN_MODIFIED | IN_ACCESSED);
    132 	if (DOINGSOFTDEP(ap->a_vp))
    133 		softdep_update_inodeblock(ip, bp, waitfor);
    134 	else if (ip->i_ffs_effnlink != ip->i_nlink)
    135 		panic("ffs_update: bad link cnt");
    136 	if (fs->fs_magic == FS_UFS1_MAGIC) {
    137 		cp = (caddr_t)bp->b_data +
    138 		    (ino_to_fsbo(fs, ip->i_number) * DINODE1_SIZE);
    139 #ifdef FFS_EI
    140 		if (UFS_FSNEEDSWAP(fs))
    141 			ffs_dinode1_swap(ip->i_din.ffs1_din,
    142 			    (struct ufs1_dinode *)cp);
    143 		else
    144 #endif
    145 			memcpy(cp, ip->i_din.ffs1_din, DINODE1_SIZE);
    146 	} else {
    147 		cp = (caddr_t)bp->b_data +
    148 		    (ino_to_fsbo(fs, ip->i_number) * DINODE2_SIZE);
    149 #ifdef FFS_EI
    150 		if (UFS_FSNEEDSWAP(fs))
    151 			ffs_dinode2_swap(ip->i_din.ffs2_din,
    152 			    (struct ufs2_dinode *)cp);
    153 		else
    154 #endif
    155 			memcpy(cp, ip->i_din.ffs2_din, DINODE2_SIZE);
    156 	}
    157 	if (waitfor) {
    158 		return (bwrite(bp));
    159 	} else {
    160 		bdwrite(bp);
    161 		return (0);
    162 	}
    163 }
    164 
    165 #define	SINGLE	0	/* index of single indirect block */
    166 #define	DOUBLE	1	/* index of double indirect block */
    167 #define	TRIPLE	2	/* index of triple indirect block */
    168 /*
    169  * Truncate the inode oip to at most length size, freeing the
    170  * disk blocks.
    171  */
    172 int
    173 ffs_truncate(v)
    174 	void *v;
    175 {
    176 	struct vop_truncate_args /* {
    177 		struct vnode *a_vp;
    178 		off_t a_length;
    179 		int a_flags;
    180 		struct ucred *a_cred;
    181 		struct proc *a_p;
    182 	} */ *ap = v;
    183 	struct vnode *ovp = ap->a_vp;
    184 	struct genfs_node *gp = VTOG(ovp);
    185 	daddr_t lastblock;
    186 	struct inode *oip;
    187 	daddr_t bn, lastiblock[NIADDR], indir_lbn[NIADDR];
    188 	daddr_t oldblks[NDADDR + NIADDR], newblks[NDADDR + NIADDR];
    189 	off_t length = ap->a_length;
    190 	struct fs *fs;
    191 	int offset, size, level;
    192 	int64_t count, blocksreleased = 0;
    193 	int i, ioflag, aflag, nblocks;
    194 	int error, allerror = 0;
    195 	off_t osize;
    196 
    197 	if (length < 0)
    198 		return (EINVAL);
    199 	oip = VTOI(ovp);
    200 	if (ovp->v_type == VLNK &&
    201 	    (oip->i_size < ovp->v_mount->mnt_maxsymlinklen ||
    202 	     (ovp->v_mount->mnt_maxsymlinklen == 0 &&
    203 	      DIP(oip, blocks) == 0))) {
    204 		KDASSERT(length == 0);
    205 		memset(SHORTLINK(oip), 0, (size_t)oip->i_size);
    206 		oip->i_size = 0;
    207 		DIP_ASSIGN(oip, size, 0);
    208 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
    209 		return (VOP_UPDATE(ovp, NULL, NULL, UPDATE_WAIT));
    210 	}
    211 	if (oip->i_size == length) {
    212 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
    213 		return (VOP_UPDATE(ovp, NULL, NULL, 0));
    214 	}
    215 #ifdef QUOTA
    216 	if ((error = getinoquota(oip)) != 0)
    217 		return (error);
    218 #endif
    219 	fs = oip->i_fs;
    220 	if (length > fs->fs_maxfilesize)
    221 		return (EFBIG);
    222 
    223 	osize = oip->i_size;
    224 	ioflag = ap->a_flags;
    225 	aflag = ioflag & IO_SYNC ? B_SYNC : 0;
    226 
    227 	/*
    228 	 * Lengthen the size of the file. We must ensure that the
    229 	 * last byte of the file is allocated. Since the smallest
    230 	 * value of osize is 0, length will be at least 1.
    231 	 */
    232 
    233 	if (osize < length) {
    234 		if (lblkno(fs, osize) < NDADDR &&
    235 		    lblkno(fs, osize) != lblkno(fs, length) &&
    236 		    blkroundup(fs, osize) != osize) {
    237 			error = ufs_balloc_range(ovp, osize,
    238 			    blkroundup(fs, osize) - osize, ap->a_cred, aflag);
    239 			if (error) {
    240 				return error;
    241 			}
    242 			if (ioflag & IO_SYNC) {
    243 				ovp->v_size = blkroundup(fs, osize);
    244 				simple_lock(&ovp->v_interlock);
    245 				VOP_PUTPAGES(ovp,
    246 				    trunc_page(osize & ~(fs->fs_bsize - 1)),
    247 				    round_page(ovp->v_size),
    248 				    PGO_CLEANIT | PGO_SYNCIO);
    249 			}
    250 		}
    251 		error = ufs_balloc_range(ovp, length - 1, 1, ap->a_cred,
    252 		    aflag);
    253 		if (error) {
    254 			(void) VOP_TRUNCATE(ovp, osize, ioflag & IO_SYNC,
    255 			    ap->a_cred, ap->a_p);
    256 			return error;
    257 		}
    258 		uvm_vnp_setsize(ovp, length);
    259 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
    260 		KASSERT(ovp->v_size == oip->i_size);
    261 		return (VOP_UPDATE(ovp, NULL, NULL, 1));
    262 	}
    263 
    264 	/*
    265 	 * When truncating a regular file down to a non-block-aligned size,
    266 	 * we must zero the part of last block which is past the new EOF.
    267 	 * We must synchronously flush the zeroed pages to disk
    268 	 * since the new pages will be invalidated as soon as we
    269 	 * inform the VM system of the new, smaller size.
    270 	 * We must do this before acquiring the GLOCK, since fetching
    271 	 * the pages will acquire the GLOCK internally.
    272 	 * So there is a window where another thread could see a whole
    273 	 * zeroed page past EOF, but that's life.
    274 	 */
    275 
    276 	offset = blkoff(fs, length);
    277 	if (ovp->v_type == VREG && length < osize && offset != 0) {
    278 		voff_t eoz;
    279 
    280 		error = ufs_balloc_range(ovp, length - 1, 1, ap->a_cred,
    281 		    aflag);
    282 		if (error) {
    283 			return error;
    284 		}
    285 		size = blksize(fs, oip, lblkno(fs, length));
    286 		eoz = MIN(lblktosize(fs, lblkno(fs, length)) + size, osize);
    287 		uvm_vnp_zerorange(ovp, length, eoz - length);
    288 		simple_lock(&ovp->v_interlock);
    289 		error = VOP_PUTPAGES(ovp, trunc_page(length), round_page(eoz),
    290 		    PGO_CLEANIT | PGO_DEACTIVATE | PGO_SYNCIO);
    291 		if (error) {
    292 			return error;
    293 		}
    294 	}
    295 
    296 	lockmgr(&gp->g_glock, LK_EXCLUSIVE, NULL);
    297 
    298 	if (DOINGSOFTDEP(ovp)) {
    299 		if (length > 0) {
    300 			/*
    301 			 * If a file is only partially truncated, then
    302 			 * we have to clean up the data structures
    303 			 * describing the allocation past the truncation
    304 			 * point. Finding and deallocating those structures
    305 			 * is a lot of work. Since partial truncation occurs
    306 			 * rarely, we solve the problem by syncing the file
    307 			 * so that it will have no data structures left.
    308 			 */
    309 			if ((error = VOP_FSYNC(ovp, ap->a_cred, FSYNC_WAIT,
    310 			    0, 0, ap->a_p)) != 0) {
    311 				lockmgr(&gp->g_glock, LK_RELEASE, NULL);
    312 				return (error);
    313 			}
    314 			if (oip->i_flag & IN_SPACECOUNTED)
    315 				fs->fs_pendingblocks -= DIP(oip, blocks);
    316 		} else {
    317 			uvm_vnp_setsize(ovp, length);
    318 #ifdef QUOTA
    319  			(void) chkdq(oip, -DIP(oip, blocks), NOCRED, 0);
    320 #endif
    321 			softdep_setup_freeblocks(oip, length, 0);
    322 			(void) vinvalbuf(ovp, 0, ap->a_cred, ap->a_p, 0, 0);
    323 			lockmgr(&gp->g_glock, LK_RELEASE, NULL);
    324 			oip->i_flag |= IN_CHANGE | IN_UPDATE;
    325 			return (VOP_UPDATE(ovp, NULL, NULL, 0));
    326 		}
    327 	}
    328 	oip->i_size = length;
    329 	DIP_ASSIGN(oip, size, length);
    330 	uvm_vnp_setsize(ovp, length);
    331 	/*
    332 	 * Calculate index into inode's block list of
    333 	 * last direct and indirect blocks (if any)
    334 	 * which we want to keep.  Lastblock is -1 when
    335 	 * the file is truncated to 0.
    336 	 */
    337 	lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1;
    338 	lastiblock[SINGLE] = lastblock - NDADDR;
    339 	lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
    340 	lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
    341 	nblocks = btodb(fs->fs_bsize);
    342 	/*
    343 	 * Update file and block pointers on disk before we start freeing
    344 	 * blocks.  If we crash before free'ing blocks below, the blocks
    345 	 * will be returned to the free list.  lastiblock values are also
    346 	 * normalized to -1 for calls to ffs_indirtrunc below.
    347 	 */
    348 	for (level = TRIPLE; level >= SINGLE; level--) {
    349 		oldblks[NDADDR + level] = DIP(oip, ib[level]);
    350 		if (lastiblock[level] < 0) {
    351 			DIP_ASSIGN(oip, ib[level], 0);
    352 			lastiblock[level] = -1;
    353 		}
    354 	}
    355 	for (i = 0; i < NDADDR; i++) {
    356 		oldblks[i] = DIP(oip, db[i]);
    357 		if (i > lastblock)
    358 			DIP_ASSIGN(oip, db[i], 0);
    359 	}
    360 	oip->i_flag |= IN_CHANGE | IN_UPDATE;
    361 	error = VOP_UPDATE(ovp, NULL, NULL, UPDATE_WAIT);
    362 	if (error && !allerror)
    363 		allerror = error;
    364 
    365 	/*
    366 	 * Having written the new inode to disk, save its new configuration
    367 	 * and put back the old block pointers long enough to process them.
    368 	 * Note that we save the new block configuration so we can check it
    369 	 * when we are done.
    370 	 */
    371 	for (i = 0; i < NDADDR; i++) {
    372 		newblks[i] = DIP(oip, db[i]);
    373 		DIP_ASSIGN(oip, db[i], oldblks[i]);
    374 	}
    375 	for (i = 0; i < NIADDR; i++) {
    376 		newblks[NDADDR + i] = DIP(oip, ib[i]);
    377 		DIP_ASSIGN(oip, ib[i], oldblks[NDADDR + i]);
    378 	}
    379 
    380 	oip->i_size = osize;
    381 	DIP_ASSIGN(oip, size, osize);
    382 	error = vtruncbuf(ovp, lastblock + 1, 0, 0);
    383 	if (error && !allerror)
    384 		allerror = error;
    385 
    386 	/*
    387 	 * Indirect blocks first.
    388 	 */
    389 	indir_lbn[SINGLE] = -NDADDR;
    390 	indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
    391 	indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
    392 	for (level = TRIPLE; level >= SINGLE; level--) {
    393 		if (oip->i_ump->um_fstype == UFS1)
    394 			bn = ufs_rw32(oip->i_ffs1_ib[level],UFS_FSNEEDSWAP(fs));
    395 		else
    396 			bn = ufs_rw64(oip->i_ffs2_ib[level],UFS_FSNEEDSWAP(fs));
    397 		if (bn != 0) {
    398 			error = ffs_indirtrunc(oip, indir_lbn[level],
    399 			    fsbtodb(fs, bn), lastiblock[level], level, &count);
    400 			if (error)
    401 				allerror = error;
    402 			blocksreleased += count;
    403 			if (lastiblock[level] < 0) {
    404 				DIP_ASSIGN(oip, ib[level], 0);
    405 				ffs_blkfree(oip, bn, fs->fs_bsize);
    406 				blocksreleased += nblocks;
    407 			}
    408 		}
    409 		if (lastiblock[level] >= 0)
    410 			goto done;
    411 	}
    412 
    413 	/*
    414 	 * All whole direct blocks or frags.
    415 	 */
    416 	for (i = NDADDR - 1; i > lastblock; i--) {
    417 		long bsize;
    418 
    419 		if (oip->i_ump->um_fstype == UFS1)
    420 			bn = ufs_rw32(oip->i_ffs1_db[i], UFS_FSNEEDSWAP(fs));
    421 		else
    422 			bn = ufs_rw64(oip->i_ffs2_db[i], UFS_FSNEEDSWAP(fs));
    423 		if (bn == 0)
    424 			continue;
    425 		DIP_ASSIGN(oip, db[i], 0);
    426 		bsize = blksize(fs, oip, i);
    427 		ffs_blkfree(oip, bn, bsize);
    428 		blocksreleased += btodb(bsize);
    429 	}
    430 	if (lastblock < 0)
    431 		goto done;
    432 
    433 	/*
    434 	 * Finally, look for a change in size of the
    435 	 * last direct block; release any frags.
    436 	 */
    437 	if (oip->i_ump->um_fstype == UFS1)
    438 		bn = ufs_rw32(oip->i_ffs1_db[lastblock], UFS_FSNEEDSWAP(fs));
    439 	else
    440 		bn = ufs_rw64(oip->i_ffs2_db[lastblock], UFS_FSNEEDSWAP(fs));
    441 	if (bn != 0) {
    442 		long oldspace, newspace;
    443 
    444 		/*
    445 		 * Calculate amount of space we're giving
    446 		 * back as old block size minus new block size.
    447 		 */
    448 		oldspace = blksize(fs, oip, lastblock);
    449 		oip->i_size = length;
    450 		DIP_ASSIGN(oip, size, length);
    451 		newspace = blksize(fs, oip, lastblock);
    452 		if (newspace == 0)
    453 			panic("itrunc: newspace");
    454 		if (oldspace - newspace > 0) {
    455 			/*
    456 			 * Block number of space to be free'd is
    457 			 * the old block # plus the number of frags
    458 			 * required for the storage we're keeping.
    459 			 */
    460 			bn += numfrags(fs, newspace);
    461 			ffs_blkfree(oip, bn, oldspace - newspace);
    462 			blocksreleased += btodb(oldspace - newspace);
    463 		}
    464 	}
    465 
    466 done:
    467 #ifdef DIAGNOSTIC
    468 	for (level = SINGLE; level <= TRIPLE; level++)
    469 		if (newblks[NDADDR + level] != DIP(oip, ib[level]))
    470 			panic("itrunc1");
    471 	for (i = 0; i < NDADDR; i++)
    472 		if (newblks[i] != DIP(oip, db[i]))
    473 			panic("itrunc2");
    474 	if (length == 0 &&
    475 	    (!LIST_EMPTY(&ovp->v_cleanblkhd) || !LIST_EMPTY(&ovp->v_dirtyblkhd)))
    476 		panic("itrunc3");
    477 #endif /* DIAGNOSTIC */
    478 	/*
    479 	 * Put back the real size.
    480 	 */
    481 	oip->i_size = length;
    482 	DIP_ASSIGN(oip, size, length);
    483 	DIP_ADD(oip, blocks, -blocksreleased);
    484 	lockmgr(&gp->g_glock, LK_RELEASE, NULL);
    485 	oip->i_flag |= IN_CHANGE;
    486 #ifdef QUOTA
    487 	(void) chkdq(oip, -blocksreleased, NOCRED, 0);
    488 #endif
    489 	KASSERT(ovp->v_type != VREG || ovp->v_size == oip->i_size);
    490 	return (allerror);
    491 }
    492 
    493 /*
    494  * Release blocks associated with the inode ip and stored in the indirect
    495  * block bn.  Blocks are free'd in LIFO order up to (but not including)
    496  * lastbn.  If level is greater than SINGLE, the block is an indirect block
    497  * and recursive calls to indirtrunc must be used to cleanse other indirect
    498  * blocks.
    499  *
    500  * NB: triple indirect blocks are untested.
    501  */
    502 static int
    503 ffs_indirtrunc(ip, lbn, dbn, lastbn, level, countp)
    504 	struct inode *ip;
    505 	daddr_t lbn, lastbn;
    506 	daddr_t dbn;
    507 	int level;
    508 	int64_t *countp;
    509 {
    510 	int i;
    511 	struct buf *bp;
    512 	struct fs *fs = ip->i_fs;
    513 	int32_t *bap1 = NULL;
    514 	int64_t *bap2 = NULL;
    515 	struct vnode *vp;
    516 	daddr_t nb, nlbn, last;
    517 	char *copy = NULL;
    518 	int64_t blkcount, factor, blocksreleased = 0;
    519 	int nblocks;
    520 	int error = 0, allerror = 0;
    521 #ifdef FFS_EI
    522 	const int needswap = UFS_FSNEEDSWAP(fs);
    523 #endif
    524 #define RBAP(ip, i) (((ip)->i_ump->um_fstype == UFS1) ? \
    525 	    ufs_rw32(bap1[i], needswap) : ufs_rw64(bap2[i], needswap))
    526 #define BAP_ASSIGN(ip, i, value)					\
    527 	do {								\
    528 		if ((ip)->i_ump->um_fstype == UFS1)			\
    529 			bap1[i] = (value);				\
    530 		else							\
    531 			bap2[i] = (value);				\
    532 	} while(0)
    533 
    534 	/*
    535 	 * Calculate index in current block of last
    536 	 * block to be kept.  -1 indicates the entire
    537 	 * block so we need not calculate the index.
    538 	 */
    539 	factor = 1;
    540 	for (i = SINGLE; i < level; i++)
    541 		factor *= NINDIR(fs);
    542 	last = lastbn;
    543 	if (lastbn > 0)
    544 		last /= factor;
    545 	nblocks = btodb(fs->fs_bsize);
    546 	/*
    547 	 * Get buffer of block pointers, zero those entries corresponding
    548 	 * to blocks to be free'd, and update on disk copy first.  Since
    549 	 * double(triple) indirect before single(double) indirect, calls
    550 	 * to bmap on these blocks will fail.  However, we already have
    551 	 * the on disk address, so we have to set the b_blkno field
    552 	 * explicitly instead of letting bread do everything for us.
    553 	 */
    554 	vp = ITOV(ip);
    555 	bp = getblk(vp, lbn, (int)fs->fs_bsize, 0, 0);
    556 	if (bp->b_flags & (B_DONE | B_DELWRI)) {
    557 		/* Braces must be here in case trace evaluates to nothing. */
    558 		trace(TR_BREADHIT, pack(vp, fs->fs_bsize), lbn);
    559 	} else {
    560 		trace(TR_BREADMISS, pack(vp, fs->fs_bsize), lbn);
    561 		curproc->p_stats->p_ru.ru_inblock++;	/* pay for read */
    562 		bp->b_flags |= B_READ;
    563 		if (bp->b_bcount > bp->b_bufsize)
    564 			panic("ffs_indirtrunc: bad buffer size");
    565 		bp->b_blkno = dbn;
    566 		BIO_SETPRIO(bp, BPRIO_TIMECRITICAL);
    567 		VOP_STRATEGY(vp, bp);
    568 		error = biowait(bp);
    569 	}
    570 	if (error) {
    571 		brelse(bp);
    572 		*countp = 0;
    573 		return (error);
    574 	}
    575 
    576 	if (ip->i_ump->um_fstype == UFS1)
    577 		bap1 = (int32_t *)bp->b_data;
    578 	else
    579 		bap2 = (int64_t *)bp->b_data;
    580 	if (lastbn >= 0) {
    581 		copy = malloc(fs->fs_bsize, M_TEMP, M_WAITOK);
    582 		memcpy((caddr_t)copy, bp->b_data, (u_int)fs->fs_bsize);
    583 		for (i = last + 1; i < NINDIR(fs); i++)
    584 			BAP_ASSIGN(ip, i, 0);
    585 		error = bwrite(bp);
    586 		if (error)
    587 			allerror = error;
    588 		if (ip->i_ump->um_fstype == UFS1)
    589 			bap1 = (int32_t *)copy;
    590 		else
    591 			bap2 = (int64_t *)copy;
    592 	}
    593 
    594 	/*
    595 	 * Recursively free totally unused blocks.
    596 	 */
    597 	for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
    598 	    i--, nlbn += factor) {
    599 		nb = RBAP(ip, i);
    600 		if (nb == 0)
    601 			continue;
    602 		if (level > SINGLE) {
    603 			error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
    604 					       (daddr_t)-1, level - 1,
    605 					       &blkcount);
    606 			if (error)
    607 				allerror = error;
    608 			blocksreleased += blkcount;
    609 		}
    610 		ffs_blkfree(ip, nb, fs->fs_bsize);
    611 		blocksreleased += nblocks;
    612 	}
    613 
    614 	/*
    615 	 * Recursively free last partial block.
    616 	 */
    617 	if (level > SINGLE && lastbn >= 0) {
    618 		last = lastbn % factor;
    619 		nb = RBAP(ip, i);
    620 		if (nb != 0) {
    621 			error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
    622 					       last, level - 1, &blkcount);
    623 			if (error)
    624 				allerror = error;
    625 			blocksreleased += blkcount;
    626 		}
    627 	}
    628 
    629 	if (copy != NULL) {
    630 		FREE(copy, M_TEMP);
    631 	} else {
    632 		bp->b_flags |= B_INVAL;
    633 		brelse(bp);
    634 	}
    635 
    636 	*countp = blocksreleased;
    637 	return (allerror);
    638 }
    639