Home | History | Annotate | Line # | Download | only in ffs
ffs_inode.c revision 1.67
      1 /*	$NetBSD: ffs_inode.c,v 1.67 2004/08/14 02:26:57 mycroft 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 1.67 2004/08/14 02:26:57 mycroft 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 	if (ap->a_flags & UPDATE_CLOSE)
    103 		flags = ip->i_flag & (IN_MODIFIED | IN_ACCESSED);
    104 	else
    105 		flags = ip->i_flag & IN_MODIFIED;
    106 	if (flags == 0)
    107 		return (0);
    108 	fs = ip->i_fs;
    109 
    110 	if ((flags & IN_MODIFIED) != 0 &&
    111 	    (ap->a_vp->v_mount->mnt_flag & MNT_ASYNC) == 0) {
    112 		waitfor = ap->a_flags & UPDATE_WAIT;
    113 		if ((ap->a_flags & UPDATE_DIROP) && !DOINGSOFTDEP(ap->a_vp))
    114 			waitfor |= UPDATE_WAIT;
    115 	} else
    116 		waitfor = 0;
    117 
    118 	/*
    119 	 * Ensure that uid and gid are correct. This is a temporary
    120 	 * fix until fsck has been changed to do the update.
    121 	 */
    122 	if (fs->fs_magic == FS_UFS1_MAGIC &&			/* XXX */
    123 	    fs->fs_old_inodefmt < FS_44INODEFMT) {		/* XXX */
    124 		ip->i_ffs1_ouid = ip->i_uid;	/* XXX */
    125 		ip->i_ffs1_ogid = ip->i_gid;	/* XXX */
    126 	}							/* XXX */
    127 	error = bread(ip->i_devvp,
    128 		      fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
    129 		      (int)fs->fs_bsize, NOCRED, &bp);
    130 	if (error) {
    131 		brelse(bp);
    132 		return (error);
    133 	}
    134 	ip->i_flag &= ~(IN_MODIFIED | IN_ACCESSED);
    135 	if (DOINGSOFTDEP(ap->a_vp))
    136 		softdep_update_inodeblock(ip, bp, waitfor);
    137 	else if (ip->i_ffs_effnlink != ip->i_nlink)
    138 		panic("ffs_update: bad link cnt");
    139 	if (fs->fs_magic == FS_UFS1_MAGIC) {
    140 		cp = (caddr_t)bp->b_data +
    141 		    (ino_to_fsbo(fs, ip->i_number) * DINODE1_SIZE);
    142 #ifdef FFS_EI
    143 		if (UFS_FSNEEDSWAP(fs))
    144 			ffs_dinode1_swap(ip->i_din.ffs1_din,
    145 			    (struct ufs1_dinode *)cp);
    146 		else
    147 #endif
    148 			memcpy(cp, ip->i_din.ffs1_din, DINODE1_SIZE);
    149 	} else {
    150 		cp = (caddr_t)bp->b_data +
    151 		    (ino_to_fsbo(fs, ip->i_number) * DINODE2_SIZE);
    152 #ifdef FFS_EI
    153 		if (UFS_FSNEEDSWAP(fs))
    154 			ffs_dinode2_swap(ip->i_din.ffs2_din,
    155 			    (struct ufs2_dinode *)cp);
    156 		else
    157 #endif
    158 			memcpy(cp, ip->i_din.ffs2_din, DINODE2_SIZE);
    159 	}
    160 	if (waitfor) {
    161 		return (bwrite(bp));
    162 	} else {
    163 		bdwrite(bp);
    164 		return (0);
    165 	}
    166 }
    167 
    168 #define	SINGLE	0	/* index of single indirect block */
    169 #define	DOUBLE	1	/* index of double indirect block */
    170 #define	TRIPLE	2	/* index of triple indirect block */
    171 /*
    172  * Truncate the inode oip to at most length size, freeing the
    173  * disk blocks.
    174  */
    175 int
    176 ffs_truncate(v)
    177 	void *v;
    178 {
    179 	struct vop_truncate_args /* {
    180 		struct vnode *a_vp;
    181 		off_t a_length;
    182 		int a_flags;
    183 		struct ucred *a_cred;
    184 		struct proc *a_p;
    185 	} */ *ap = v;
    186 	struct vnode *ovp = ap->a_vp;
    187 	struct genfs_node *gp = VTOG(ovp);
    188 	daddr_t lastblock;
    189 	struct inode *oip;
    190 	daddr_t bn, lastiblock[NIADDR], indir_lbn[NIADDR];
    191 	daddr_t blks[NDADDR + NIADDR];
    192 	off_t length = ap->a_length;
    193 	struct fs *fs;
    194 	int offset, size, level;
    195 	int64_t count, blocksreleased = 0;
    196 	int i, ioflag, aflag, nblocks;
    197 	int error, allerror = 0;
    198 	off_t osize;
    199 	int sync;
    200 
    201 	if (length < 0)
    202 		return (EINVAL);
    203 	oip = VTOI(ovp);
    204 	if (ovp->v_type == VLNK &&
    205 	    (oip->i_size < ovp->v_mount->mnt_maxsymlinklen ||
    206 	     (ovp->v_mount->mnt_maxsymlinklen == 0 &&
    207 	      DIP(oip, blocks) == 0))) {
    208 		KDASSERT(length == 0);
    209 		memset(SHORTLINK(oip), 0, (size_t)oip->i_size);
    210 		oip->i_size = 0;
    211 		DIP_ASSIGN(oip, size, 0);
    212 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
    213 		return (VOP_UPDATE(ovp, NULL, NULL, 0));
    214 	}
    215 	if (oip->i_size == length) {
    216 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
    217 		return (VOP_UPDATE(ovp, NULL, NULL, 0));
    218 	}
    219 #ifdef QUOTA
    220 	if ((error = getinoquota(oip)) != 0)
    221 		return (error);
    222 #endif
    223 	fs = oip->i_fs;
    224 	if (length > fs->fs_maxfilesize)
    225 		return (EFBIG);
    226 
    227 	if ((oip->i_flags & SF_SNAPSHOT) != 0)
    228 		ffs_snapremove(ovp);
    229 
    230 	osize = oip->i_size;
    231 	ioflag = ap->a_flags;
    232 	aflag = ioflag & IO_SYNC ? B_SYNC : 0;
    233 
    234 	/*
    235 	 * Lengthen the size of the file. We must ensure that the
    236 	 * last byte of the file is allocated. Since the smallest
    237 	 * value of osize is 0, length will be at least 1.
    238 	 */
    239 
    240 	if (osize < length) {
    241 		if (lblkno(fs, osize) < NDADDR &&
    242 		    lblkno(fs, osize) != lblkno(fs, length) &&
    243 		    blkroundup(fs, osize) != osize) {
    244 			error = ufs_balloc_range(ovp, osize,
    245 			    blkroundup(fs, osize) - osize, ap->a_cred, aflag);
    246 			if (error) {
    247 				return error;
    248 			}
    249 			if (ioflag & IO_SYNC) {
    250 				ovp->v_size = blkroundup(fs, osize);
    251 				simple_lock(&ovp->v_interlock);
    252 				VOP_PUTPAGES(ovp,
    253 				    trunc_page(osize & ~(fs->fs_bsize - 1)),
    254 				    round_page(ovp->v_size),
    255 				    PGO_CLEANIT | PGO_SYNCIO);
    256 			}
    257 		}
    258 		error = ufs_balloc_range(ovp, length - 1, 1, ap->a_cred,
    259 		    aflag);
    260 		if (error) {
    261 			(void) VOP_TRUNCATE(ovp, osize, ioflag & IO_SYNC,
    262 			    ap->a_cred, ap->a_p);
    263 			return error;
    264 		}
    265 		uvm_vnp_setsize(ovp, length);
    266 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
    267 		KASSERT(ovp->v_size == oip->i_size);
    268 		return (VOP_UPDATE(ovp, NULL, NULL, 1));
    269 	}
    270 
    271 	/*
    272 	 * When truncating a regular file down to a non-block-aligned size,
    273 	 * we must zero the part of last block which is past the new EOF.
    274 	 * We must synchronously flush the zeroed pages to disk
    275 	 * since the new pages will be invalidated as soon as we
    276 	 * inform the VM system of the new, smaller size.
    277 	 * We must do this before acquiring the GLOCK, since fetching
    278 	 * the pages will acquire the GLOCK internally.
    279 	 * So there is a window where another thread could see a whole
    280 	 * zeroed page past EOF, but that's life.
    281 	 */
    282 
    283 	offset = blkoff(fs, length);
    284 	if (ovp->v_type == VREG && offset != 0 && osize > length &&
    285 	    round_page(osize) > round_page(length)) {
    286 		voff_t eoz;
    287 
    288 		error = ufs_balloc_range(ovp, length - 1, 1, ap->a_cred,
    289 		    aflag);
    290 		if (error)
    291 			return error;
    292 		size = blksize(fs, oip, lblkno(fs, length));
    293 		eoz = MIN(lblktosize(fs, lblkno(fs, length)) + size, osize);
    294 		uvm_vnp_zerorange(ovp, length, eoz - length);
    295 		simple_lock(&ovp->v_interlock);
    296 		error = VOP_PUTPAGES(ovp, trunc_page(length), round_page(eoz),
    297 		    PGO_CLEANIT | PGO_DEACTIVATE | PGO_SYNCIO);
    298 		if (error)
    299 			return error;
    300 	}
    301 
    302 	lockmgr(&gp->g_glock, LK_EXCLUSIVE, NULL);
    303 
    304 	if (DOINGSOFTDEP(ovp)) {
    305 		if (length > 0) {
    306 			/*
    307 			 * If a file is only partially truncated, then
    308 			 * we have to clean up the data structures
    309 			 * describing the allocation past the truncation
    310 			 * point. Finding and deallocating those structures
    311 			 * is a lot of work. Since partial truncation occurs
    312 			 * rarely, we solve the problem by syncing the file
    313 			 * so that it will have no data structures left.
    314 			 */
    315 			if ((error = VOP_FSYNC(ovp, ap->a_cred, FSYNC_WAIT,
    316 			    0, 0, ap->a_p)) != 0) {
    317 				lockmgr(&gp->g_glock, LK_RELEASE, NULL);
    318 				return (error);
    319 			}
    320 			if (oip->i_flag & IN_SPACECOUNTED)
    321 				fs->fs_pendingblocks -= DIP(oip, blocks);
    322 		} else {
    323 			uvm_vnp_setsize(ovp, length);
    324 #ifdef QUOTA
    325  			(void) chkdq(oip, -DIP(oip, blocks), NOCRED, 0);
    326 #endif
    327 			softdep_setup_freeblocks(oip, length, 0);
    328 			(void) vinvalbuf(ovp, 0, ap->a_cred, ap->a_p, 0, 0);
    329 			lockmgr(&gp->g_glock, LK_RELEASE, NULL);
    330 			oip->i_flag |= IN_CHANGE | IN_UPDATE;
    331 			return (VOP_UPDATE(ovp, NULL, NULL, 0));
    332 		}
    333 	}
    334 	oip->i_size = length;
    335 	DIP_ASSIGN(oip, size, length);
    336 	uvm_vnp_setsize(ovp, length);
    337 	/*
    338 	 * Calculate index into inode's block list of
    339 	 * last direct and indirect blocks (if any)
    340 	 * which we want to keep.  Lastblock is -1 when
    341 	 * the file is truncated to 0.
    342 	 */
    343 	lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1;
    344 	lastiblock[SINGLE] = lastblock - NDADDR;
    345 	lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
    346 	lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
    347 	nblocks = btodb(fs->fs_bsize);
    348 	/*
    349 	 * Update file and block pointers on disk before we start freeing
    350 	 * blocks.  If we crash before free'ing blocks below, the blocks
    351 	 * will be returned to the free list.  lastiblock values are also
    352 	 * normalized to -1 for calls to ffs_indirtrunc below.
    353 	 */
    354 	sync = 0;
    355 	for (level = TRIPLE; level >= SINGLE; level--) {
    356 		blks[NDADDR + level] = DIP(oip, ib[level]);
    357 		if (lastiblock[level] < 0 && blks[NDADDR + level] != 0) {
    358 			sync = 1;
    359 			DIP_ASSIGN(oip, ib[level], 0);
    360 			lastiblock[level] = -1;
    361 		}
    362 	}
    363 	for (i = 0; i < NDADDR; i++) {
    364 		blks[i] = DIP(oip, db[i]);
    365 		if (i > lastblock && blks[i] != 0) {
    366 			sync = 1;
    367 			DIP_ASSIGN(oip, db[i], 0);
    368 		}
    369 	}
    370 	if (sync) {
    371 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
    372 		error = VOP_UPDATE(ovp, NULL, NULL, UPDATE_WAIT);
    373 		if (error && !allerror)
    374 			allerror = error;
    375 	}
    376 
    377 	/*
    378 	 * Having written the new inode to disk, save its new configuration
    379 	 * and put back the old block pointers long enough to process them.
    380 	 * Note that we save the new block configuration so we can check it
    381 	 * when we are done.
    382 	 */
    383 	for (i = 0; i < NDADDR; i++) {
    384 		bn = DIP(oip, db[i]);
    385 		DIP_ASSIGN(oip, db[i], blks[i]);
    386 		blks[i] = bn;
    387 	}
    388 	for (i = 0; i < NIADDR; i++) {
    389 		bn = DIP(oip, ib[i]);
    390 		DIP_ASSIGN(oip, ib[i], blks[NDADDR + i]);
    391 		blks[NDADDR + i] = bn;
    392 	}
    393 
    394 	oip->i_size = osize;
    395 	DIP_ASSIGN(oip, size, osize);
    396 	error = vtruncbuf(ovp, lastblock + 1, 0, 0);
    397 	if (error && !allerror)
    398 		allerror = error;
    399 
    400 	/*
    401 	 * Indirect blocks first.
    402 	 */
    403 	indir_lbn[SINGLE] = -NDADDR;
    404 	indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
    405 	indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
    406 	for (level = TRIPLE; level >= SINGLE; level--) {
    407 		if (oip->i_ump->um_fstype == UFS1)
    408 			bn = ufs_rw32(oip->i_ffs1_ib[level],UFS_FSNEEDSWAP(fs));
    409 		else
    410 			bn = ufs_rw64(oip->i_ffs2_ib[level],UFS_FSNEEDSWAP(fs));
    411 		if (bn != 0) {
    412 			error = ffs_indirtrunc(oip, indir_lbn[level],
    413 			    fsbtodb(fs, bn), lastiblock[level], level, &count);
    414 			if (error)
    415 				allerror = error;
    416 			blocksreleased += count;
    417 			if (lastiblock[level] < 0) {
    418 				DIP_ASSIGN(oip, ib[level], 0);
    419 				ffs_blkfree(fs, oip->i_devvp, bn, fs->fs_bsize,
    420 				    oip->i_number);
    421 				blocksreleased += nblocks;
    422 			}
    423 		}
    424 		if (lastiblock[level] >= 0)
    425 			goto done;
    426 	}
    427 
    428 	/*
    429 	 * All whole direct blocks or frags.
    430 	 */
    431 	for (i = NDADDR - 1; i > lastblock; i--) {
    432 		long bsize;
    433 
    434 		if (oip->i_ump->um_fstype == UFS1)
    435 			bn = ufs_rw32(oip->i_ffs1_db[i], UFS_FSNEEDSWAP(fs));
    436 		else
    437 			bn = ufs_rw64(oip->i_ffs2_db[i], UFS_FSNEEDSWAP(fs));
    438 		if (bn == 0)
    439 			continue;
    440 		DIP_ASSIGN(oip, db[i], 0);
    441 		bsize = blksize(fs, oip, i);
    442 		ffs_blkfree(fs, oip->i_devvp, bn, bsize, oip->i_number);
    443 		blocksreleased += btodb(bsize);
    444 	}
    445 	if (lastblock < 0)
    446 		goto done;
    447 
    448 	/*
    449 	 * Finally, look for a change in size of the
    450 	 * last direct block; release any frags.
    451 	 */
    452 	if (oip->i_ump->um_fstype == UFS1)
    453 		bn = ufs_rw32(oip->i_ffs1_db[lastblock], UFS_FSNEEDSWAP(fs));
    454 	else
    455 		bn = ufs_rw64(oip->i_ffs2_db[lastblock], UFS_FSNEEDSWAP(fs));
    456 	if (bn != 0) {
    457 		long oldspace, newspace;
    458 
    459 		/*
    460 		 * Calculate amount of space we're giving
    461 		 * back as old block size minus new block size.
    462 		 */
    463 		oldspace = blksize(fs, oip, lastblock);
    464 		oip->i_size = length;
    465 		DIP_ASSIGN(oip, size, length);
    466 		newspace = blksize(fs, oip, lastblock);
    467 		if (newspace == 0)
    468 			panic("itrunc: newspace");
    469 		if (oldspace - newspace > 0) {
    470 			/*
    471 			 * Block number of space to be free'd is
    472 			 * the old block # plus the number of frags
    473 			 * required for the storage we're keeping.
    474 			 */
    475 			bn += numfrags(fs, newspace);
    476 			ffs_blkfree(fs, oip->i_devvp, bn, oldspace - newspace,
    477 			    oip->i_number);
    478 			blocksreleased += btodb(oldspace - newspace);
    479 		}
    480 	}
    481 
    482 done:
    483 #ifdef DIAGNOSTIC
    484 	for (level = SINGLE; level <= TRIPLE; level++)
    485 		if (blks[NDADDR + level] != DIP(oip, ib[level]))
    486 			panic("itrunc1");
    487 	for (i = 0; i < NDADDR; i++)
    488 		if (blks[i] != DIP(oip, db[i]))
    489 			panic("itrunc2");
    490 	if (length == 0 &&
    491 	    (!LIST_EMPTY(&ovp->v_cleanblkhd) || !LIST_EMPTY(&ovp->v_dirtyblkhd)))
    492 		panic("itrunc3");
    493 #endif /* DIAGNOSTIC */
    494 	/*
    495 	 * Put back the real size.
    496 	 */
    497 	oip->i_size = length;
    498 	DIP_ASSIGN(oip, size, length);
    499 	DIP_ADD(oip, blocks, -blocksreleased);
    500 	lockmgr(&gp->g_glock, LK_RELEASE, NULL);
    501 	oip->i_flag |= IN_CHANGE;
    502 	if (!sync)
    503 		oip->i_flag |= IN_UPDATE;
    504 #ifdef QUOTA
    505 	(void) chkdq(oip, -blocksreleased, NOCRED, 0);
    506 #endif
    507 	KASSERT(ovp->v_type != VREG || ovp->v_size == oip->i_size);
    508 	return (allerror);
    509 }
    510 
    511 /*
    512  * Release blocks associated with the inode ip and stored in the indirect
    513  * block bn.  Blocks are free'd in LIFO order up to (but not including)
    514  * lastbn.  If level is greater than SINGLE, the block is an indirect block
    515  * and recursive calls to indirtrunc must be used to cleanse other indirect
    516  * blocks.
    517  *
    518  * NB: triple indirect blocks are untested.
    519  */
    520 static int
    521 ffs_indirtrunc(ip, lbn, dbn, lastbn, level, countp)
    522 	struct inode *ip;
    523 	daddr_t lbn, lastbn;
    524 	daddr_t dbn;
    525 	int level;
    526 	int64_t *countp;
    527 {
    528 	int i;
    529 	struct buf *bp;
    530 	struct fs *fs = ip->i_fs;
    531 	int32_t *bap1 = NULL;
    532 	int64_t *bap2 = NULL;
    533 	struct vnode *vp;
    534 	daddr_t nb, nlbn, last;
    535 	char *copy = NULL;
    536 	int64_t blkcount, factor, blocksreleased = 0;
    537 	int nblocks;
    538 	int error = 0, allerror = 0;
    539 #ifdef FFS_EI
    540 	const int needswap = UFS_FSNEEDSWAP(fs);
    541 #endif
    542 #define RBAP(ip, i) (((ip)->i_ump->um_fstype == UFS1) ? \
    543 	    ufs_rw32(bap1[i], needswap) : ufs_rw64(bap2[i], needswap))
    544 #define BAP_ASSIGN(ip, i, value)					\
    545 	do {								\
    546 		if ((ip)->i_ump->um_fstype == UFS1)			\
    547 			bap1[i] = (value);				\
    548 		else							\
    549 			bap2[i] = (value);				\
    550 	} while(0)
    551 
    552 	/*
    553 	 * Calculate index in current block of last
    554 	 * block to be kept.  -1 indicates the entire
    555 	 * block so we need not calculate the index.
    556 	 */
    557 	factor = 1;
    558 	for (i = SINGLE; i < level; i++)
    559 		factor *= NINDIR(fs);
    560 	last = lastbn;
    561 	if (lastbn > 0)
    562 		last /= factor;
    563 	nblocks = btodb(fs->fs_bsize);
    564 	/*
    565 	 * Get buffer of block pointers, zero those entries corresponding
    566 	 * to blocks to be free'd, and update on disk copy first.  Since
    567 	 * double(triple) indirect before single(double) indirect, calls
    568 	 * to bmap on these blocks will fail.  However, we already have
    569 	 * the on disk address, so we have to set the b_blkno field
    570 	 * explicitly instead of letting bread do everything for us.
    571 	 */
    572 	vp = ITOV(ip);
    573 	bp = getblk(vp, lbn, (int)fs->fs_bsize, 0, 0);
    574 	if (bp->b_flags & (B_DONE | B_DELWRI)) {
    575 		/* Braces must be here in case trace evaluates to nothing. */
    576 		trace(TR_BREADHIT, pack(vp, fs->fs_bsize), lbn);
    577 	} else {
    578 		trace(TR_BREADMISS, pack(vp, fs->fs_bsize), lbn);
    579 		curproc->p_stats->p_ru.ru_inblock++;	/* pay for read */
    580 		bp->b_flags |= B_READ;
    581 		if (bp->b_bcount > bp->b_bufsize)
    582 			panic("ffs_indirtrunc: bad buffer size");
    583 		bp->b_blkno = dbn;
    584 		BIO_SETPRIO(bp, BPRIO_TIMECRITICAL);
    585 		VOP_STRATEGY(vp, bp);
    586 		error = biowait(bp);
    587 	}
    588 	if (error) {
    589 		brelse(bp);
    590 		*countp = 0;
    591 		return (error);
    592 	}
    593 
    594 	if (ip->i_ump->um_fstype == UFS1)
    595 		bap1 = (int32_t *)bp->b_data;
    596 	else
    597 		bap2 = (int64_t *)bp->b_data;
    598 	if (lastbn >= 0) {
    599 		copy = malloc(fs->fs_bsize, M_TEMP, M_WAITOK);
    600 		memcpy((caddr_t)copy, bp->b_data, (u_int)fs->fs_bsize);
    601 		for (i = last + 1; i < NINDIR(fs); i++)
    602 			BAP_ASSIGN(ip, i, 0);
    603 		error = bwrite(bp);
    604 		if (error)
    605 			allerror = error;
    606 		if (ip->i_ump->um_fstype == UFS1)
    607 			bap1 = (int32_t *)copy;
    608 		else
    609 			bap2 = (int64_t *)copy;
    610 	}
    611 
    612 	/*
    613 	 * Recursively free totally unused blocks.
    614 	 */
    615 	for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
    616 	    i--, nlbn += factor) {
    617 		nb = RBAP(ip, i);
    618 		if (nb == 0)
    619 			continue;
    620 		if (level > SINGLE) {
    621 			error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
    622 					       (daddr_t)-1, level - 1,
    623 					       &blkcount);
    624 			if (error)
    625 				allerror = error;
    626 			blocksreleased += blkcount;
    627 		}
    628 		ffs_blkfree(fs, ip->i_devvp, nb, fs->fs_bsize, ip->i_number);
    629 		blocksreleased += nblocks;
    630 	}
    631 
    632 	/*
    633 	 * Recursively free last partial block.
    634 	 */
    635 	if (level > SINGLE && lastbn >= 0) {
    636 		last = lastbn % factor;
    637 		nb = RBAP(ip, i);
    638 		if (nb != 0) {
    639 			error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
    640 					       last, level - 1, &blkcount);
    641 			if (error)
    642 				allerror = error;
    643 			blocksreleased += blkcount;
    644 		}
    645 	}
    646 
    647 	if (copy != NULL) {
    648 		FREE(copy, M_TEMP);
    649 	} else {
    650 		bp->b_flags |= B_INVAL;
    651 		brelse(bp);
    652 	}
    653 
    654 	*countp = blocksreleased;
    655 	return (allerror);
    656 }
    657