Home | History | Annotate | Line # | Download | only in ffs
ffs_inode.c revision 1.108
      1 /*	$NetBSD: ffs_inode.c,v 1.108 2011/11/23 19:42:10 bouyer Exp $	*/
      2 
      3 /*-
      4  * Copyright (c) 2008 The NetBSD Foundation, Inc.
      5  * All rights reserved.
      6  *
      7  * This code is derived from software contributed to The NetBSD Foundation
      8  * by Wasabi Systems, Inc.
      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  *
     19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     29  * POSSIBILITY OF SUCH DAMAGE.
     30  */
     31 
     32 /*
     33  * Copyright (c) 1982, 1986, 1989, 1993
     34  *	The Regents of the University of California.  All rights reserved.
     35  *
     36  * Redistribution and use in source and binary forms, with or without
     37  * modification, are permitted provided that the following conditions
     38  * are met:
     39  * 1. Redistributions of source code must retain the above copyright
     40  *    notice, this list of conditions and the following disclaimer.
     41  * 2. Redistributions in binary form must reproduce the above copyright
     42  *    notice, this list of conditions and the following disclaimer in the
     43  *    documentation and/or other materials provided with the distribution.
     44  * 3. Neither the name of the University nor the names of its contributors
     45  *    may be used to endorse or promote products derived from this software
     46  *    without specific prior written permission.
     47  *
     48  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     49  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     50  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     51  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     52  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     53  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     54  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     55  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     56  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     57  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     58  * SUCH DAMAGE.
     59  *
     60  *	@(#)ffs_inode.c	8.13 (Berkeley) 4/21/95
     61  */
     62 
     63 #include <sys/cdefs.h>
     64 __KERNEL_RCSID(0, "$NetBSD: ffs_inode.c,v 1.108 2011/11/23 19:42:10 bouyer Exp $");
     65 
     66 #if defined(_KERNEL_OPT)
     67 #include "opt_ffs.h"
     68 #include "opt_quota.h"
     69 #endif
     70 
     71 #include <sys/param.h>
     72 #include <sys/systm.h>
     73 #include <sys/buf.h>
     74 #include <sys/file.h>
     75 #include <sys/fstrans.h>
     76 #include <sys/kauth.h>
     77 #include <sys/kernel.h>
     78 #include <sys/malloc.h>
     79 #include <sys/mount.h>
     80 #include <sys/proc.h>
     81 #include <sys/resourcevar.h>
     82 #include <sys/trace.h>
     83 #include <sys/vnode.h>
     84 #include <sys/wapbl.h>
     85 
     86 #include <ufs/ufs/quota.h>
     87 #include <ufs/ufs/inode.h>
     88 #include <ufs/ufs/ufsmount.h>
     89 #include <ufs/ufs/ufs_extern.h>
     90 #include <ufs/ufs/ufs_bswap.h>
     91 #include <ufs/ufs/ufs_wapbl.h>
     92 
     93 #include <ufs/ffs/fs.h>
     94 #include <ufs/ffs/ffs_extern.h>
     95 
     96 static int ffs_indirtrunc(struct inode *, daddr_t, daddr_t, daddr_t, int,
     97 			  int64_t *);
     98 
     99 /*
    100  * Update the access, modified, and inode change times as specified
    101  * by the IN_ACCESS, IN_UPDATE, and IN_CHANGE flags respectively.
    102  * The IN_MODIFIED flag is used to specify that the inode needs to be
    103  * updated but that the times have already been set. The access
    104  * and modified times are taken from the second and third parameters;
    105  * the inode change time is always taken from the current time. If
    106  * UPDATE_WAIT flag is set, or UPDATE_DIROP is set then wait for the
    107  * disk write of the inode to complete.
    108  */
    109 
    110 int
    111 ffs_update(struct vnode *vp, const struct timespec *acc,
    112     const struct timespec *mod, int updflags)
    113 {
    114 	struct fs *fs;
    115 	struct buf *bp;
    116 	struct inode *ip;
    117 	int error;
    118 	void *cp;
    119 	int waitfor, flags;
    120 
    121 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
    122 		return (0);
    123 	ip = VTOI(vp);
    124 	FFS_ITIMES(ip, acc, mod, NULL);
    125 	if (updflags & UPDATE_CLOSE)
    126 		flags = ip->i_flag & (IN_MODIFIED | IN_ACCESSED);
    127 	else
    128 		flags = ip->i_flag & IN_MODIFIED;
    129 	if (flags == 0)
    130 		return (0);
    131 	fs = ip->i_fs;
    132 
    133 	if ((flags & IN_MODIFIED) != 0 &&
    134 	    (vp->v_mount->mnt_flag & MNT_ASYNC) == 0) {
    135 		waitfor = updflags & UPDATE_WAIT;
    136 		if ((updflags & UPDATE_DIROP) != 0)
    137 			waitfor |= UPDATE_WAIT;
    138 	} else
    139 		waitfor = 0;
    140 
    141 	/*
    142 	 * Ensure that uid and gid are correct. This is a temporary
    143 	 * fix until fsck has been changed to do the update.
    144 	 */
    145 	if (fs->fs_magic == FS_UFS1_MAGIC &&			/* XXX */
    146 	    fs->fs_old_inodefmt < FS_44INODEFMT) {		/* XXX */
    147 		ip->i_ffs1_ouid = ip->i_uid;	/* XXX */
    148 		ip->i_ffs1_ogid = ip->i_gid;	/* XXX */
    149 	}							/* XXX */
    150 	error = bread(ip->i_devvp,
    151 		      fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
    152 		      (int)fs->fs_bsize, NOCRED, B_MODIFY, &bp);
    153 	if (error) {
    154 		brelse(bp, 0);
    155 		return (error);
    156 	}
    157 	ip->i_flag &= ~(IN_MODIFIED | IN_ACCESSED);
    158 	/* Keep unlinked inode list up to date */
    159 	KDASSERT(DIP(ip, nlink) == ip->i_nlink);
    160 	if (ip->i_mode) {
    161 		if (ip->i_nlink > 0) {
    162 			UFS_WAPBL_UNREGISTER_INODE(ip->i_ump->um_mountp,
    163 			    ip->i_number, ip->i_mode);
    164 		} else {
    165 			UFS_WAPBL_REGISTER_INODE(ip->i_ump->um_mountp,
    166 			    ip->i_number, ip->i_mode);
    167 		}
    168 	}
    169 	if (fs->fs_magic == FS_UFS1_MAGIC) {
    170 		cp = (char *)bp->b_data +
    171 		    (ino_to_fsbo(fs, ip->i_number) * DINODE1_SIZE);
    172 #ifdef FFS_EI
    173 		if (UFS_FSNEEDSWAP(fs))
    174 			ffs_dinode1_swap(ip->i_din.ffs1_din,
    175 			    (struct ufs1_dinode *)cp);
    176 		else
    177 #endif
    178 			memcpy(cp, ip->i_din.ffs1_din, DINODE1_SIZE);
    179 	} else {
    180 		cp = (char *)bp->b_data +
    181 		    (ino_to_fsbo(fs, ip->i_number) * DINODE2_SIZE);
    182 #ifdef FFS_EI
    183 		if (UFS_FSNEEDSWAP(fs))
    184 			ffs_dinode2_swap(ip->i_din.ffs2_din,
    185 			    (struct ufs2_dinode *)cp);
    186 		else
    187 #endif
    188 			memcpy(cp, ip->i_din.ffs2_din, DINODE2_SIZE);
    189 	}
    190 	if (waitfor) {
    191 		return (bwrite(bp));
    192 	} else {
    193 		bdwrite(bp);
    194 		return (0);
    195 	}
    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 int
    206 ffs_truncate(struct vnode *ovp, off_t length, int ioflag, kauth_cred_t cred)
    207 {
    208 	daddr_t lastblock;
    209 	struct inode *oip = VTOI(ovp);
    210 	daddr_t bn, lastiblock[NIADDR], indir_lbn[NIADDR];
    211 	daddr_t blks[NDADDR + NIADDR];
    212 	struct fs *fs;
    213 	int offset, pgoffset, level;
    214 	int64_t count, blocksreleased = 0;
    215 	int i, aflag, nblocks;
    216 	int error, allerror = 0;
    217 	off_t osize;
    218 	int sync;
    219 	struct ufsmount *ump = oip->i_ump;
    220 
    221 	if (ovp->v_type == VCHR || ovp->v_type == VBLK ||
    222 	    ovp->v_type == VFIFO || ovp->v_type == VSOCK) {
    223 		KASSERT(oip->i_size == 0);
    224 		return 0;
    225 	}
    226 
    227 	if (length < 0)
    228 		return (EINVAL);
    229 
    230 	if (ovp->v_type == VLNK &&
    231 	    (oip->i_size < ump->um_maxsymlinklen ||
    232 	     (ump->um_maxsymlinklen == 0 && DIP(oip, blocks) == 0))) {
    233 		KDASSERT(length == 0);
    234 		memset(SHORTLINK(oip), 0, (size_t)oip->i_size);
    235 		oip->i_size = 0;
    236 		DIP_ASSIGN(oip, size, 0);
    237 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
    238 		return (ffs_update(ovp, NULL, NULL, 0));
    239 	}
    240 	if (oip->i_size == length) {
    241 		/* still do a uvm_vnp_setsize() as writesize may be larger */
    242 		uvm_vnp_setsize(ovp, length);
    243 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
    244 		return (ffs_update(ovp, NULL, NULL, 0));
    245 	}
    246 	fs = oip->i_fs;
    247 	if (length > ump->um_maxfilesize)
    248 		return (EFBIG);
    249 
    250 	if ((oip->i_flags & SF_SNAPSHOT) != 0)
    251 		ffs_snapremove(ovp);
    252 
    253 	osize = oip->i_size;
    254 	aflag = ioflag & IO_SYNC ? B_SYNC : 0;
    255 
    256 	/*
    257 	 * Lengthen the size of the file. We must ensure that the
    258 	 * last byte of the file is allocated. Since the smallest
    259 	 * value of osize is 0, length will be at least 1.
    260 	 */
    261 
    262 	if (osize < length) {
    263 		if (lblkno(fs, osize) < NDADDR &&
    264 		    lblkno(fs, osize) != lblkno(fs, length) &&
    265 		    blkroundup(fs, osize) != osize) {
    266 			off_t eob;
    267 
    268 			eob = blkroundup(fs, osize);
    269 			uvm_vnp_setwritesize(ovp, eob);
    270 			error = ufs_balloc_range(ovp, osize, eob - osize,
    271 			    cred, aflag);
    272 			if (error) {
    273 				(void) ffs_truncate(ovp, osize,
    274 				    ioflag & IO_SYNC, cred);
    275 				return error;
    276 			}
    277 			if (ioflag & IO_SYNC) {
    278 				mutex_enter(ovp->v_interlock);
    279 				VOP_PUTPAGES(ovp,
    280 				    trunc_page(osize & fs->fs_bmask),
    281 				    round_page(eob), PGO_CLEANIT | PGO_SYNCIO |
    282 				    PGO_JOURNALLOCKED);
    283 			}
    284 		}
    285 		uvm_vnp_setwritesize(ovp, length);
    286 		error = ufs_balloc_range(ovp, length - 1, 1, cred, aflag);
    287 		if (error) {
    288 			(void) ffs_truncate(ovp, osize, ioflag & IO_SYNC, cred);
    289 			return (error);
    290 		}
    291 		uvm_vnp_setsize(ovp, length);
    292 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
    293 		KASSERT(ovp->v_size == oip->i_size);
    294 		return (ffs_update(ovp, NULL, NULL, 0));
    295 	}
    296 
    297 	/*
    298 	 * When truncating a regular file down to a non-block-aligned size,
    299 	 * we must zero the part of last block which is past the new EOF.
    300 	 * We must synchronously flush the zeroed pages to disk
    301 	 * since the new pages will be invalidated as soon as we
    302 	 * inform the VM system of the new, smaller size.
    303 	 * We must do this before acquiring the GLOCK, since fetching
    304 	 * the pages will acquire the GLOCK internally.
    305 	 * So there is a window where another thread could see a whole
    306 	 * zeroed page past EOF, but that's life.
    307 	 */
    308 
    309 	offset = blkoff(fs, length);
    310 	pgoffset = length & PAGE_MASK;
    311 	if (ovp->v_type == VREG && (pgoffset != 0 || offset != 0) &&
    312 	    osize > length) {
    313 		daddr_t lbn;
    314 		voff_t eoz;
    315 		int size;
    316 
    317 		if (offset != 0) {
    318 			error = ufs_balloc_range(ovp, length - 1, 1, cred,
    319 			    aflag);
    320 			if (error)
    321 				return error;
    322 		}
    323 		lbn = lblkno(fs, length);
    324 		size = blksize(fs, oip, lbn);
    325 		eoz = MIN(MAX(lblktosize(fs, lbn) + size, round_page(pgoffset)),
    326 		    osize);
    327 		ubc_zerorange(&ovp->v_uobj, length, eoz - length,
    328 		    UBC_UNMAP_FLAG(ovp));
    329 		if (round_page(eoz) > round_page(length)) {
    330 			mutex_enter(ovp->v_interlock);
    331 			error = VOP_PUTPAGES(ovp, round_page(length),
    332 			    round_page(eoz),
    333 			    PGO_CLEANIT | PGO_DEACTIVATE | PGO_JOURNALLOCKED |
    334 			    ((ioflag & IO_SYNC) ? PGO_SYNCIO : 0));
    335 			if (error)
    336 				return error;
    337 		}
    338 	}
    339 
    340 	genfs_node_wrlock(ovp);
    341 	oip->i_size = length;
    342 	DIP_ASSIGN(oip, size, length);
    343 	uvm_vnp_setsize(ovp, length);
    344 	/*
    345 	 * Calculate index into inode's block list of
    346 	 * last direct and indirect blocks (if any)
    347 	 * which we want to keep.  Lastblock is -1 when
    348 	 * the file is truncated to 0.
    349 	 */
    350 	lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1;
    351 	lastiblock[SINGLE] = lastblock - NDADDR;
    352 	lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
    353 	lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
    354 	nblocks = btodb(fs->fs_bsize);
    355 	/*
    356 	 * Update file and block pointers on disk before we start freeing
    357 	 * blocks.  If we crash before free'ing blocks below, the blocks
    358 	 * will be returned to the free list.  lastiblock values are also
    359 	 * normalized to -1 for calls to ffs_indirtrunc below.
    360 	 */
    361 	sync = 0;
    362 	for (level = TRIPLE; level >= SINGLE; level--) {
    363 		blks[NDADDR + level] = DIP(oip, ib[level]);
    364 		if (lastiblock[level] < 0 && blks[NDADDR + level] != 0) {
    365 			sync = 1;
    366 			DIP_ASSIGN(oip, ib[level], 0);
    367 			lastiblock[level] = -1;
    368 		}
    369 	}
    370 	for (i = 0; i < NDADDR; i++) {
    371 		blks[i] = DIP(oip, db[i]);
    372 		if (i > lastblock && blks[i] != 0) {
    373 			sync = 1;
    374 			DIP_ASSIGN(oip, db[i], 0);
    375 		}
    376 	}
    377 	oip->i_flag |= IN_CHANGE | IN_UPDATE;
    378 	if (sync) {
    379 		error = ffs_update(ovp, NULL, NULL, UPDATE_WAIT);
    380 		if (error && !allerror)
    381 			allerror = error;
    382 	}
    383 
    384 	/*
    385 	 * Having written the new inode to disk, save its new configuration
    386 	 * and put back the old block pointers long enough to process them.
    387 	 * Note that we save the new block configuration so we can check it
    388 	 * when we are done.
    389 	 */
    390 	for (i = 0; i < NDADDR; i++) {
    391 		bn = DIP(oip, db[i]);
    392 		DIP_ASSIGN(oip, db[i], blks[i]);
    393 		blks[i] = bn;
    394 	}
    395 	for (i = 0; i < NIADDR; i++) {
    396 		bn = DIP(oip, ib[i]);
    397 		DIP_ASSIGN(oip, ib[i], blks[NDADDR + i]);
    398 		blks[NDADDR + i] = bn;
    399 	}
    400 
    401 	oip->i_size = osize;
    402 	DIP_ASSIGN(oip, size, osize);
    403 	error = vtruncbuf(ovp, lastblock + 1, 0, 0);
    404 	if (error && !allerror)
    405 		allerror = error;
    406 
    407 	/*
    408 	 * Indirect blocks first.
    409 	 */
    410 	indir_lbn[SINGLE] = -NDADDR;
    411 	indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
    412 	indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
    413 	for (level = TRIPLE; level >= SINGLE; level--) {
    414 		if (oip->i_ump->um_fstype == UFS1)
    415 			bn = ufs_rw32(oip->i_ffs1_ib[level],UFS_FSNEEDSWAP(fs));
    416 		else
    417 			bn = ufs_rw64(oip->i_ffs2_ib[level],UFS_FSNEEDSWAP(fs));
    418 		if (bn != 0) {
    419 			error = ffs_indirtrunc(oip, indir_lbn[level],
    420 			    fsbtodb(fs, bn), lastiblock[level], level, &count);
    421 			if (error)
    422 				allerror = error;
    423 			blocksreleased += count;
    424 			if (lastiblock[level] < 0) {
    425 				DIP_ASSIGN(oip, ib[level], 0);
    426 				if (oip->i_ump->um_mountp->mnt_wapbl) {
    427 					UFS_WAPBL_REGISTER_DEALLOCATION(
    428 					    oip->i_ump->um_mountp,
    429 					    fsbtodb(fs, bn), fs->fs_bsize);
    430 				} else
    431 					ffs_blkfree(fs, oip->i_devvp, bn,
    432 					    fs->fs_bsize, oip->i_number);
    433 				blocksreleased += nblocks;
    434 			}
    435 		}
    436 		if (lastiblock[level] >= 0)
    437 			goto done;
    438 	}
    439 
    440 	/*
    441 	 * All whole direct blocks or frags.
    442 	 */
    443 	for (i = NDADDR - 1; i > lastblock; i--) {
    444 		long bsize;
    445 
    446 		if (oip->i_ump->um_fstype == UFS1)
    447 			bn = ufs_rw32(oip->i_ffs1_db[i], UFS_FSNEEDSWAP(fs));
    448 		else
    449 			bn = ufs_rw64(oip->i_ffs2_db[i], UFS_FSNEEDSWAP(fs));
    450 		if (bn == 0)
    451 			continue;
    452 		DIP_ASSIGN(oip, db[i], 0);
    453 		bsize = blksize(fs, oip, i);
    454 		if ((oip->i_ump->um_mountp->mnt_wapbl) &&
    455 		    (ovp->v_type != VREG)) {
    456 			UFS_WAPBL_REGISTER_DEALLOCATION(oip->i_ump->um_mountp,
    457 			    fsbtodb(fs, bn), bsize);
    458 		} else
    459 			ffs_blkfree(fs, oip->i_devvp, bn, bsize, oip->i_number);
    460 		blocksreleased += btodb(bsize);
    461 	}
    462 	if (lastblock < 0)
    463 		goto done;
    464 
    465 	/*
    466 	 * Finally, look for a change in size of the
    467 	 * last direct block; release any frags.
    468 	 */
    469 	if (oip->i_ump->um_fstype == UFS1)
    470 		bn = ufs_rw32(oip->i_ffs1_db[lastblock], UFS_FSNEEDSWAP(fs));
    471 	else
    472 		bn = ufs_rw64(oip->i_ffs2_db[lastblock], UFS_FSNEEDSWAP(fs));
    473 	if (bn != 0) {
    474 		long oldspace, newspace;
    475 
    476 		/*
    477 		 * Calculate amount of space we're giving
    478 		 * back as old block size minus new block size.
    479 		 */
    480 		oldspace = blksize(fs, oip, lastblock);
    481 		oip->i_size = length;
    482 		DIP_ASSIGN(oip, size, length);
    483 		newspace = blksize(fs, oip, lastblock);
    484 		if (newspace == 0)
    485 			panic("itrunc: newspace");
    486 		if (oldspace - newspace > 0) {
    487 			/*
    488 			 * Block number of space to be free'd is
    489 			 * the old block # plus the number of frags
    490 			 * required for the storage we're keeping.
    491 			 */
    492 			bn += numfrags(fs, newspace);
    493 			if ((oip->i_ump->um_mountp->mnt_wapbl) &&
    494 			    (ovp->v_type != VREG)) {
    495 				UFS_WAPBL_REGISTER_DEALLOCATION(
    496 				    oip->i_ump->um_mountp, fsbtodb(fs, bn),
    497 				    oldspace - newspace);
    498 			} else
    499 				ffs_blkfree(fs, oip->i_devvp, bn,
    500 				    oldspace - newspace, oip->i_number);
    501 			blocksreleased += btodb(oldspace - newspace);
    502 		}
    503 	}
    504 
    505 done:
    506 #ifdef DIAGNOSTIC
    507 	for (level = SINGLE; level <= TRIPLE; level++)
    508 		if (blks[NDADDR + level] != DIP(oip, ib[level]))
    509 			panic("itrunc1");
    510 	for (i = 0; i < NDADDR; i++)
    511 		if (blks[i] != DIP(oip, db[i]))
    512 			panic("itrunc2");
    513 	if (length == 0 &&
    514 	    (!LIST_EMPTY(&ovp->v_cleanblkhd) || !LIST_EMPTY(&ovp->v_dirtyblkhd)))
    515 		panic("itrunc3");
    516 #endif /* DIAGNOSTIC */
    517 	/*
    518 	 * Put back the real size.
    519 	 */
    520 	oip->i_size = length;
    521 	DIP_ASSIGN(oip, size, length);
    522 	DIP_ADD(oip, blocks, -blocksreleased);
    523 	genfs_node_unlock(ovp);
    524 	oip->i_flag |= IN_CHANGE;
    525 	UFS_WAPBL_UPDATE(ovp, NULL, NULL, 0);
    526 #if defined(QUOTA) || defined(QUOTA2)
    527 	(void) chkdq(oip, -blocksreleased, NOCRED, 0);
    528 #endif
    529 	KASSERT(ovp->v_type != VREG || ovp->v_size == oip->i_size);
    530 	return (allerror);
    531 }
    532 
    533 /*
    534  * Release blocks associated with the inode ip and stored in the indirect
    535  * block bn.  Blocks are free'd in LIFO order up to (but not including)
    536  * lastbn.  If level is greater than SINGLE, the block is an indirect block
    537  * and recursive calls to indirtrunc must be used to cleanse other indirect
    538  * blocks.
    539  *
    540  * NB: triple indirect blocks are untested.
    541  */
    542 static int
    543 ffs_indirtrunc(struct inode *ip, daddr_t lbn, daddr_t dbn, daddr_t lastbn,
    544     int level, int64_t *countp)
    545 {
    546 	int i;
    547 	struct buf *bp;
    548 	struct fs *fs = ip->i_fs;
    549 	int32_t *bap1 = NULL;
    550 	int64_t *bap2 = NULL;
    551 	struct vnode *vp;
    552 	daddr_t nb, nlbn, last;
    553 	char *copy = NULL;
    554 	int64_t blkcount, factor, blocksreleased = 0;
    555 	int nblocks;
    556 	int error = 0, allerror = 0;
    557 #ifdef FFS_EI
    558 	const int needswap = UFS_FSNEEDSWAP(fs);
    559 #endif
    560 #define RBAP(ip, i) (((ip)->i_ump->um_fstype == UFS1) ? \
    561 	    ufs_rw32(bap1[i], needswap) : ufs_rw64(bap2[i], needswap))
    562 #define BAP_ASSIGN(ip, i, value)					\
    563 	do {								\
    564 		if ((ip)->i_ump->um_fstype == UFS1)			\
    565 			bap1[i] = (value);				\
    566 		else							\
    567 			bap2[i] = (value);				\
    568 	} while(0)
    569 
    570 	/*
    571 	 * Calculate index in current block of last
    572 	 * block to be kept.  -1 indicates the entire
    573 	 * block so we need not calculate the index.
    574 	 */
    575 	factor = 1;
    576 	for (i = SINGLE; i < level; i++)
    577 		factor *= NINDIR(fs);
    578 	last = lastbn;
    579 	if (lastbn > 0)
    580 		last /= factor;
    581 	nblocks = btodb(fs->fs_bsize);
    582 	/*
    583 	 * Get buffer of block pointers, zero those entries corresponding
    584 	 * to blocks to be free'd, and update on disk copy first.  Since
    585 	 * double(triple) indirect before single(double) indirect, calls
    586 	 * to bmap on these blocks will fail.  However, we already have
    587 	 * the on disk address, so we have to set the b_blkno field
    588 	 * explicitly instead of letting bread do everything for us.
    589 	 */
    590 	vp = ITOV(ip);
    591 	error = ffs_getblk(vp, lbn, FFS_NOBLK, fs->fs_bsize, false, &bp);
    592 	if (error) {
    593 		*countp = 0;
    594 		return error;
    595 	}
    596 	if (bp->b_oflags & (BO_DONE | BO_DELWRI)) {
    597 		/* Braces must be here in case trace evaluates to nothing. */
    598 		trace(TR_BREADHIT, pack(vp, fs->fs_bsize), lbn);
    599 	} else {
    600 		trace(TR_BREADMISS, pack(vp, fs->fs_bsize), lbn);
    601 		curlwp->l_ru.ru_inblock++;	/* pay for read */
    602 		bp->b_flags |= B_READ;
    603 		bp->b_flags &= ~B_COWDONE;	/* we change blkno below */
    604 		if (bp->b_bcount > bp->b_bufsize)
    605 			panic("ffs_indirtrunc: bad buffer size");
    606 		bp->b_blkno = dbn;
    607 		BIO_SETPRIO(bp, BPRIO_TIMECRITICAL);
    608 		VOP_STRATEGY(vp, bp);
    609 		error = biowait(bp);
    610 		if (error == 0)
    611 			error = fscow_run(bp, true);
    612 	}
    613 	if (error) {
    614 		brelse(bp, 0);
    615 		*countp = 0;
    616 		return (error);
    617 	}
    618 
    619 	if (ip->i_ump->um_fstype == UFS1)
    620 		bap1 = (int32_t *)bp->b_data;
    621 	else
    622 		bap2 = (int64_t *)bp->b_data;
    623 	if (lastbn >= 0) {
    624 		copy = malloc(fs->fs_bsize, M_TEMP, M_WAITOK);
    625 		memcpy((void *)copy, bp->b_data, (u_int)fs->fs_bsize);
    626 		for (i = last + 1; i < NINDIR(fs); i++)
    627 			BAP_ASSIGN(ip, i, 0);
    628 		error = bwrite(bp);
    629 		if (error)
    630 			allerror = error;
    631 		if (ip->i_ump->um_fstype == UFS1)
    632 			bap1 = (int32_t *)copy;
    633 		else
    634 			bap2 = (int64_t *)copy;
    635 	}
    636 
    637 	/*
    638 	 * Recursively free totally unused blocks.
    639 	 */
    640 	for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
    641 	    i--, nlbn += factor) {
    642 		nb = RBAP(ip, i);
    643 		if (nb == 0)
    644 			continue;
    645 		if (level > SINGLE) {
    646 			error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
    647 					       (daddr_t)-1, level - 1,
    648 					       &blkcount);
    649 			if (error)
    650 				allerror = error;
    651 			blocksreleased += blkcount;
    652 		}
    653 		if ((ip->i_ump->um_mountp->mnt_wapbl) &&
    654 		    ((level > SINGLE) || (ITOV(ip)->v_type != VREG))) {
    655 			UFS_WAPBL_REGISTER_DEALLOCATION(ip->i_ump->um_mountp,
    656 			    fsbtodb(fs, nb), fs->fs_bsize);
    657 		} else
    658 			ffs_blkfree(fs, ip->i_devvp, nb, fs->fs_bsize,
    659 			    ip->i_number);
    660 		blocksreleased += nblocks;
    661 	}
    662 
    663 	/*
    664 	 * Recursively free last partial block.
    665 	 */
    666 	if (level > SINGLE && lastbn >= 0) {
    667 		last = lastbn % factor;
    668 		nb = RBAP(ip, i);
    669 		if (nb != 0) {
    670 			error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
    671 					       last, level - 1, &blkcount);
    672 			if (error)
    673 				allerror = error;
    674 			blocksreleased += blkcount;
    675 		}
    676 	}
    677 
    678 	if (copy != NULL) {
    679 		free(copy, M_TEMP);
    680 	} else {
    681 		brelse(bp, BC_INVAL);
    682 	}
    683 
    684 	*countp = blocksreleased;
    685 	return (allerror);
    686 }
    687 
    688 void
    689 ffs_itimes(struct inode *ip, const struct timespec *acc,
    690     const struct timespec *mod, const struct timespec *cre)
    691 {
    692 	struct timespec now;
    693 
    694 	if (!(ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE | IN_MODIFY))) {
    695 		return;
    696 	}
    697 
    698 	vfs_timestamp(&now);
    699 	if (ip->i_flag & IN_ACCESS) {
    700 		if (acc == NULL)
    701 			acc = &now;
    702 		DIP_ASSIGN(ip, atime, acc->tv_sec);
    703 		DIP_ASSIGN(ip, atimensec, acc->tv_nsec);
    704 	}
    705 	if (ip->i_flag & (IN_UPDATE | IN_MODIFY)) {
    706 		if ((ip->i_flags & SF_SNAPSHOT) == 0) {
    707 			if (mod == NULL)
    708 				mod = &now;
    709 			DIP_ASSIGN(ip, mtime, mod->tv_sec);
    710 			DIP_ASSIGN(ip, mtimensec, mod->tv_nsec);
    711 		}
    712 		ip->i_modrev++;
    713 	}
    714 	if (ip->i_flag & (IN_CHANGE | IN_MODIFY)) {
    715 		if (cre == NULL)
    716 			cre = &now;
    717 		DIP_ASSIGN(ip, ctime, cre->tv_sec);
    718 		DIP_ASSIGN(ip, ctimensec, cre->tv_nsec);
    719 	}
    720 	if (ip->i_flag & (IN_ACCESS | IN_MODIFY))
    721 		ip->i_flag |= IN_ACCESSED;
    722 	if (ip->i_flag & (IN_UPDATE | IN_CHANGE))
    723 		ip->i_flag |= IN_MODIFIED;
    724 	ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE | IN_MODIFY);
    725 }
    726