Home | History | Annotate | Line # | Download | only in ffs
ffs_inode.c revision 1.86.2.4
      1 /*	$NetBSD: ffs_inode.c,v 1.86.2.4 2007/06/09 23:58:19 ad 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 1.86.2.4 2007/06/09 23:58:19 ad 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 #include <sys/kauth.h>
     54 
     55 #include <ufs/ufs/quota.h>
     56 #include <ufs/ufs/inode.h>
     57 #include <ufs/ufs/ufsmount.h>
     58 #include <ufs/ufs/ufs_extern.h>
     59 #include <ufs/ufs/ufs_bswap.h>
     60 
     61 #include <ufs/ffs/fs.h>
     62 #include <ufs/ffs/ffs_extern.h>
     63 
     64 static int ffs_indirtrunc(struct inode *, daddr_t, daddr_t, daddr_t, int,
     65 			  int64_t *);
     66 
     67 /*
     68  * Update the access, modified, and inode change times as specified
     69  * by the IN_ACCESS, IN_UPDATE, and IN_CHANGE flags respectively.
     70  * The IN_MODIFIED flag is used to specify that the inode needs to be
     71  * updated but that the times have already been set. The access
     72  * and modified times are taken from the second and third parameters;
     73  * the inode change time is always taken from the current time. If
     74  * UPDATE_WAIT flag is set, or UPDATE_DIROP is set and we are not doing
     75  * softupdates, then wait for the disk write of the inode to complete.
     76  */
     77 
     78 int
     79 ffs_update(struct vnode *vp, const struct timespec *acc,
     80     const struct timespec *mod, int updflags)
     81 {
     82 	struct fs *fs;
     83 	struct buf *bp;
     84 	struct inode *ip;
     85 	int error;
     86 	void *cp;
     87 	int waitfor, flags;
     88 
     89 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
     90 		return (0);
     91 	ip = VTOI(vp);
     92 	FFS_ITIMES(ip, acc, mod, NULL);
     93 	if (updflags & UPDATE_CLOSE)
     94 		flags = ip->i_flag & (IN_MODIFIED | IN_ACCESSED);
     95 	else
     96 		flags = ip->i_flag & IN_MODIFIED;
     97 	if (flags == 0)
     98 		return (0);
     99 	fs = ip->i_fs;
    100 
    101 	if ((flags & IN_MODIFIED) != 0 &&
    102 	    (vp->v_mount->mnt_flag & MNT_ASYNC) == 0) {
    103 		waitfor = updflags & UPDATE_WAIT;
    104 		if ((updflags & UPDATE_DIROP) && !DOINGSOFTDEP(vp))
    105 			waitfor |= UPDATE_WAIT;
    106 	} else
    107 		waitfor = 0;
    108 
    109 	/*
    110 	 * Ensure that uid and gid are correct. This is a temporary
    111 	 * fix until fsck has been changed to do the update.
    112 	 */
    113 	if (fs->fs_magic == FS_UFS1_MAGIC &&			/* XXX */
    114 	    fs->fs_old_inodefmt < FS_44INODEFMT) {		/* XXX */
    115 		ip->i_ffs1_ouid = ip->i_uid;	/* XXX */
    116 		ip->i_ffs1_ogid = ip->i_gid;	/* XXX */
    117 	}							/* XXX */
    118 	error = bread(ip->i_devvp,
    119 		      fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
    120 		      (int)fs->fs_bsize, NOCRED, &bp);
    121 	if (error) {
    122 		brelse(bp, 0);
    123 		return (error);
    124 	}
    125 	ip->i_flag &= ~(IN_MODIFIED | IN_ACCESSED);
    126 	if (DOINGSOFTDEP(vp))
    127 		softdep_update_inodeblock(ip, bp, waitfor);
    128 	else if (ip->i_ffs_effnlink != ip->i_nlink)
    129 		panic("ffs_update: bad link cnt");
    130 	if (fs->fs_magic == FS_UFS1_MAGIC) {
    131 		cp = (char *)bp->b_data +
    132 		    (ino_to_fsbo(fs, ip->i_number) * DINODE1_SIZE);
    133 #ifdef FFS_EI
    134 		if (UFS_FSNEEDSWAP(fs))
    135 			ffs_dinode1_swap(ip->i_din.ffs1_din,
    136 			    (struct ufs1_dinode *)cp);
    137 		else
    138 #endif
    139 			memcpy(cp, ip->i_din.ffs1_din, DINODE1_SIZE);
    140 	} else {
    141 		cp = (char *)bp->b_data +
    142 		    (ino_to_fsbo(fs, ip->i_number) * DINODE2_SIZE);
    143 #ifdef FFS_EI
    144 		if (UFS_FSNEEDSWAP(fs))
    145 			ffs_dinode2_swap(ip->i_din.ffs2_din,
    146 			    (struct ufs2_dinode *)cp);
    147 		else
    148 #endif
    149 			memcpy(cp, ip->i_din.ffs2_din, DINODE2_SIZE);
    150 	}
    151 	if (waitfor) {
    152 		return (bwrite(bp));
    153 	} else {
    154 		bdwrite(bp);
    155 		return (0);
    156 	}
    157 }
    158 
    159 #define	SINGLE	0	/* index of single indirect block */
    160 #define	DOUBLE	1	/* index of double indirect block */
    161 #define	TRIPLE	2	/* index of triple indirect block */
    162 /*
    163  * Truncate the inode oip to at most length size, freeing the
    164  * disk blocks.
    165  */
    166 int
    167 ffs_truncate(struct vnode *ovp, off_t length, int ioflag, kauth_cred_t cred,
    168     struct lwp *l)
    169 {
    170 	daddr_t lastblock;
    171 	struct inode *oip = VTOI(ovp);
    172 	daddr_t bn, lastiblock[NIADDR], indir_lbn[NIADDR];
    173 	daddr_t blks[NDADDR + NIADDR];
    174 	struct fs *fs;
    175 	int offset, pgoffset, level;
    176 	int64_t count, blocksreleased = 0;
    177 	int i, aflag, nblocks;
    178 	int error, allerror = 0;
    179 	off_t osize;
    180 	int sync;
    181 	struct ufsmount *ump = oip->i_ump;
    182 
    183 	if (ovp->v_type == VCHR || ovp->v_type == VBLK ||
    184 	    ovp->v_type == VFIFO || ovp->v_type == VSOCK) {
    185 		KASSERT(oip->i_size == 0);
    186 		return 0;
    187 	}
    188 
    189 	if (length < 0)
    190 		return (EINVAL);
    191 
    192 	if (ovp->v_type == VLNK &&
    193 	    (oip->i_size < ump->um_maxsymlinklen ||
    194 	     (ump->um_maxsymlinklen == 0 && DIP(oip, blocks) == 0))) {
    195 		KDASSERT(length == 0);
    196 		memset(SHORTLINK(oip), 0, (size_t)oip->i_size);
    197 		oip->i_size = 0;
    198 		DIP_ASSIGN(oip, size, 0);
    199 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
    200 		return (ffs_update(ovp, NULL, NULL, 0));
    201 	}
    202 	if (oip->i_size == length) {
    203 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
    204 		return (ffs_update(ovp, NULL, NULL, 0));
    205 	}
    206 #ifdef QUOTA
    207 	if ((error = getinoquota(oip)) != 0)
    208 		return (error);
    209 #endif
    210 	fs = oip->i_fs;
    211 	if (length > ump->um_maxfilesize)
    212 		return (EFBIG);
    213 
    214 	if ((oip->i_flags & SF_SNAPSHOT) != 0)
    215 		ffs_snapremove(ovp);
    216 
    217 	osize = oip->i_size;
    218 	aflag = ioflag & IO_SYNC ? B_SYNC : 0;
    219 
    220 	/*
    221 	 * Lengthen the size of the file. We must ensure that the
    222 	 * last byte of the file is allocated. Since the smallest
    223 	 * value of osize is 0, length will be at least 1.
    224 	 */
    225 
    226 	if (osize < length) {
    227 		if (lblkno(fs, osize) < NDADDR &&
    228 		    lblkno(fs, osize) != lblkno(fs, length) &&
    229 		    blkroundup(fs, osize) != osize) {
    230 			off_t eob;
    231 
    232 			eob = blkroundup(fs, osize);
    233 			uvm_vnp_setwritesize(ovp, eob);
    234 			error = ufs_balloc_range(ovp, osize, eob - osize,
    235 			    cred, aflag);
    236 			if (error)
    237 				return error;
    238 			if (ioflag & IO_SYNC) {
    239 				mutex_enter(&ovp->v_interlock);
    240 				VOP_PUTPAGES(ovp,
    241 				    trunc_page(osize & fs->fs_bmask),
    242 				    round_page(eob), PGO_CLEANIT | PGO_SYNCIO);
    243 			}
    244 		}
    245 		uvm_vnp_setwritesize(ovp, length);
    246 		error = ufs_balloc_range(ovp, length - 1, 1, cred, aflag);
    247 		if (error) {
    248 			(void) ffs_truncate(ovp, osize, ioflag & IO_SYNC,
    249 			    cred, l);
    250 			return (error);
    251 		}
    252 		uvm_vnp_setsize(ovp, length);
    253 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
    254 		KASSERT(ovp->v_size == oip->i_size);
    255 		return (ffs_update(ovp, NULL, NULL, 0));
    256 	}
    257 
    258 	/*
    259 	 * When truncating a regular file down to a non-block-aligned size,
    260 	 * we must zero the part of last block which is past the new EOF.
    261 	 * We must synchronously flush the zeroed pages to disk
    262 	 * since the new pages will be invalidated as soon as we
    263 	 * inform the VM system of the new, smaller size.
    264 	 * We must do this before acquiring the GLOCK, since fetching
    265 	 * the pages will acquire the GLOCK internally.
    266 	 * So there is a window where another thread could see a whole
    267 	 * zeroed page past EOF, but that's life.
    268 	 */
    269 
    270 	offset = blkoff(fs, length);
    271 	pgoffset = length & PAGE_MASK;
    272 	if (ovp->v_type == VREG && (pgoffset != 0 || offset != 0) &&
    273 	    osize > length) {
    274 		daddr_t lbn;
    275 		voff_t eoz;
    276 		int size;
    277 
    278 		if (offset != 0) {
    279 			error = ufs_balloc_range(ovp, length - 1, 1, cred,
    280 			    aflag);
    281 			if (error)
    282 				return error;
    283 		}
    284 		lbn = lblkno(fs, length);
    285 		size = blksize(fs, oip, lbn);
    286 		eoz = MIN(MAX(lblktosize(fs, lbn) + size, round_page(pgoffset)),
    287 		    osize);
    288 		uvm_vnp_zerorange(ovp, length, eoz - length);
    289 		if (round_page(eoz) > round_page(length)) {
    290 			mutex_enter(&ovp->v_interlock);
    291 			error = VOP_PUTPAGES(ovp, round_page(length),
    292 			    round_page(eoz),
    293 			    PGO_CLEANIT | PGO_DEACTIVATE |
    294 			    ((ioflag & IO_SYNC) ? PGO_SYNCIO : 0));
    295 			if (error)
    296 				return error;
    297 		}
    298 	}
    299 
    300 	genfs_node_wrlock(ovp);
    301 
    302 	if (DOINGSOFTDEP(ovp)) {
    303 		if (length > 0) {
    304 			/*
    305 			 * If a file is only partially truncated, then
    306 			 * we have to clean up the data structures
    307 			 * describing the allocation past the truncation
    308 			 * point. Finding and deallocating those structures
    309 			 * is a lot of work. Since partial truncation occurs
    310 			 * rarely, we solve the problem by syncing the file
    311 			 * so that it will have no data structures left.
    312 			 */
    313 			if ((error = VOP_FSYNC(ovp, cred, FSYNC_WAIT,
    314 			    0, 0, l)) != 0) {
    315 				genfs_node_unlock(ovp);
    316 				return (error);
    317 			}
    318 			mutex_enter(&ump->um_lock);
    319 			if (oip->i_flag & IN_SPACECOUNTED)
    320 				fs->fs_pendingblocks -= DIP(oip, blocks);
    321 			mutex_exit(&ump->um_lock);
    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, cred, l, 0, 0);
    329 			genfs_node_unlock(ovp);
    330 			oip->i_flag |= IN_CHANGE | IN_UPDATE;
    331 			return (ffs_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 	oip->i_flag |= IN_CHANGE | IN_UPDATE;
    371 	if (sync) {
    372 		error = ffs_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 	genfs_node_unlock(ovp);
    501 	oip->i_flag |= IN_CHANGE;
    502 #ifdef QUOTA
    503 	(void) chkdq(oip, -blocksreleased, NOCRED, 0);
    504 #endif
    505 	KASSERT(ovp->v_type != VREG || ovp->v_size == oip->i_size);
    506 	return (allerror);
    507 }
    508 
    509 /*
    510  * Release blocks associated with the inode ip and stored in the indirect
    511  * block bn.  Blocks are free'd in LIFO order up to (but not including)
    512  * lastbn.  If level is greater than SINGLE, the block is an indirect block
    513  * and recursive calls to indirtrunc must be used to cleanse other indirect
    514  * blocks.
    515  *
    516  * NB: triple indirect blocks are untested.
    517  */
    518 static int
    519 ffs_indirtrunc(struct inode *ip, daddr_t lbn, daddr_t dbn, daddr_t lastbn,
    520     int level, int64_t *countp)
    521 {
    522 	int i;
    523 	struct buf *bp;
    524 	struct fs *fs = ip->i_fs;
    525 	int32_t *bap1 = NULL;
    526 	int64_t *bap2 = NULL;
    527 	struct vnode *vp;
    528 	daddr_t nb, nlbn, last;
    529 	char *copy = NULL;
    530 	int64_t blkcount, factor, blocksreleased = 0;
    531 	int nblocks;
    532 	int error = 0, allerror = 0;
    533 #ifdef FFS_EI
    534 	const int needswap = UFS_FSNEEDSWAP(fs);
    535 #endif
    536 #define RBAP(ip, i) (((ip)->i_ump->um_fstype == UFS1) ? \
    537 	    ufs_rw32(bap1[i], needswap) : ufs_rw64(bap2[i], needswap))
    538 #define BAP_ASSIGN(ip, i, value)					\
    539 	do {								\
    540 		if ((ip)->i_ump->um_fstype == UFS1)			\
    541 			bap1[i] = (value);				\
    542 		else							\
    543 			bap2[i] = (value);				\
    544 	} while(0)
    545 
    546 	/*
    547 	 * Calculate index in current block of last
    548 	 * block to be kept.  -1 indicates the entire
    549 	 * block so we need not calculate the index.
    550 	 */
    551 	factor = 1;
    552 	for (i = SINGLE; i < level; i++)
    553 		factor *= NINDIR(fs);
    554 	last = lastbn;
    555 	if (lastbn > 0)
    556 		last /= factor;
    557 	nblocks = btodb(fs->fs_bsize);
    558 	/*
    559 	 * Get buffer of block pointers, zero those entries corresponding
    560 	 * to blocks to be free'd, and update on disk copy first.  Since
    561 	 * double(triple) indirect before single(double) indirect, calls
    562 	 * to bmap on these blocks will fail.  However, we already have
    563 	 * the on disk address, so we have to set the b_blkno field
    564 	 * explicitly instead of letting bread do everything for us.
    565 	 */
    566 	vp = ITOV(ip);
    567 	bp = getblk(vp, lbn, (int)fs->fs_bsize, 0, 0);
    568 	mutex_enter(&bp->b_interlock);
    569 	if (bp->b_flags & (B_DONE | B_DELWRI)) {
    570 		/* Braces must be here in case trace evaluates to nothing. */
    571 		trace(TR_BREADHIT, pack(vp, fs->fs_bsize), lbn);
    572 		mutex_exit(&bp->b_interlock);
    573 	} else {
    574 		trace(TR_BREADMISS, pack(vp, fs->fs_bsize), lbn);
    575 		curproc->p_stats->p_ru.ru_inblock++;	/* pay for read */
    576 		bp->b_flags |= B_READ;
    577 		if (bp->b_bcount > bp->b_bufsize)
    578 			panic("ffs_indirtrunc: bad buffer size");
    579 		bp->b_blkno = dbn;
    580 		BIO_SETPRIO(bp, BPRIO_TIMECRITICAL);
    581 		mutex_exit(&bp->b_interlock);
    582 		VOP_STRATEGY(vp, bp);
    583 		error = biowait(bp);
    584 	}
    585 	if (error) {
    586 		brelse(bp, 0);
    587 		*countp = 0;
    588 		return (error);
    589 	}
    590 
    591 	if (ip->i_ump->um_fstype == UFS1)
    592 		bap1 = (int32_t *)bp->b_data;
    593 	else
    594 		bap2 = (int64_t *)bp->b_data;
    595 	if (lastbn >= 0) {
    596 		copy = malloc(fs->fs_bsize, M_TEMP, M_WAITOK);
    597 		memcpy((void *)copy, bp->b_data, (u_int)fs->fs_bsize);
    598 		for (i = last + 1; i < NINDIR(fs); i++)
    599 			BAP_ASSIGN(ip, i, 0);
    600 		error = bwrite(bp);
    601 		if (error)
    602 			allerror = error;
    603 		if (ip->i_ump->um_fstype == UFS1)
    604 			bap1 = (int32_t *)copy;
    605 		else
    606 			bap2 = (int64_t *)copy;
    607 	}
    608 
    609 	/*
    610 	 * Recursively free totally unused blocks.
    611 	 */
    612 	for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
    613 	    i--, nlbn += factor) {
    614 		nb = RBAP(ip, i);
    615 		if (nb == 0)
    616 			continue;
    617 		if (level > SINGLE) {
    618 			error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
    619 					       (daddr_t)-1, level - 1,
    620 					       &blkcount);
    621 			if (error)
    622 				allerror = error;
    623 			blocksreleased += blkcount;
    624 		}
    625 		ffs_blkfree(fs, ip->i_devvp, nb, fs->fs_bsize, ip->i_number);
    626 		blocksreleased += nblocks;
    627 	}
    628 
    629 	/*
    630 	 * Recursively free last partial block.
    631 	 */
    632 	if (level > SINGLE && lastbn >= 0) {
    633 		last = lastbn % factor;
    634 		nb = RBAP(ip, i);
    635 		if (nb != 0) {
    636 			error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
    637 					       last, level - 1, &blkcount);
    638 			if (error)
    639 				allerror = error;
    640 			blocksreleased += blkcount;
    641 		}
    642 	}
    643 
    644 	if (copy != NULL) {
    645 		FREE(copy, M_TEMP);
    646 	} else {
    647 		brelse(bp, B_INVAL);
    648 	}
    649 
    650 	*countp = blocksreleased;
    651 	return (allerror);
    652 }
    653 
    654 void
    655 ffs_itimes(struct inode *ip, const struct timespec *acc,
    656     const struct timespec *mod, const struct timespec *cre)
    657 {
    658 	struct timespec now;
    659 
    660 	if (!(ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE | IN_MODIFY))) {
    661 		return;
    662 	}
    663 
    664 	vfs_timestamp(&now);
    665 	if (ip->i_flag & IN_ACCESS) {
    666 		if (acc == NULL)
    667 			acc = &now;
    668 		DIP_ASSIGN(ip, atime, acc->tv_sec);
    669 		DIP_ASSIGN(ip, atimensec, acc->tv_nsec);
    670 	}
    671 	if (ip->i_flag & (IN_UPDATE | IN_MODIFY)) {
    672 		if ((ip->i_flags & SF_SNAPSHOT) == 0) {
    673 			if (mod == NULL)
    674 				mod = &now;
    675 			DIP_ASSIGN(ip, mtime, mod->tv_sec);
    676 			DIP_ASSIGN(ip, mtimensec, mod->tv_nsec);
    677 		}
    678 		ip->i_modrev++;
    679 	}
    680 	if (ip->i_flag & (IN_CHANGE | IN_MODIFY)) {
    681 		if (cre == NULL)
    682 			cre = &now;
    683 		DIP_ASSIGN(ip, ctime, cre->tv_sec);
    684 		DIP_ASSIGN(ip, ctimensec, cre->tv_nsec);
    685 	}
    686 	if (ip->i_flag & (IN_ACCESS | IN_MODIFY))
    687 		ip->i_flag |= IN_ACCESSED;
    688 	if (ip->i_flag & (IN_UPDATE | IN_CHANGE))
    689 		ip->i_flag |= IN_MODIFIED;
    690 	ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE | IN_MODIFY);
    691 }
    692