Home | History | Annotate | Line # | Download | only in ffs
ffs_balloc.c revision 1.33.2.3
      1 /*	$NetBSD: ffs_balloc.c,v 1.33.2.3 2004/09/18 14:56:52 skrll Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 2002 Networks Associates Technology, Inc.
      5  * All rights reserved.
      6  *
      7  * This software was developed for the FreeBSD Project by Marshall
      8  * Kirk McKusick and Network Associates Laboratories, the Security
      9  * Research Division of Network Associates, Inc. under DARPA/SPAWAR
     10  * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS
     11  * research program
     12  *
     13  * Copyright (c) 1982, 1986, 1989, 1993
     14  *	The Regents of the University of California.  All rights reserved.
     15  *
     16  * Redistribution and use in source and binary forms, with or without
     17  * modification, are permitted provided that the following conditions
     18  * are met:
     19  * 1. Redistributions of source code must retain the above copyright
     20  *    notice, this list of conditions and the following disclaimer.
     21  * 2. Redistributions in binary form must reproduce the above copyright
     22  *    notice, this list of conditions and the following disclaimer in the
     23  *    documentation and/or other materials provided with the distribution.
     24  * 3. Neither the name of the University nor the names of its contributors
     25  *    may be used to endorse or promote products derived from this software
     26  *    without specific prior written permission.
     27  *
     28  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     29  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     30  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     31  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     32  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     33  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     34  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     35  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     36  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     37  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     38  * SUCH DAMAGE.
     39  *
     40  *	@(#)ffs_balloc.c	8.8 (Berkeley) 6/16/95
     41  */
     42 
     43 #include <sys/cdefs.h>
     44 __KERNEL_RCSID(0, "$NetBSD: ffs_balloc.c,v 1.33.2.3 2004/09/18 14:56:52 skrll Exp $");
     45 
     46 #if defined(_KERNEL_OPT)
     47 #include "opt_quota.h"
     48 #endif
     49 
     50 #include <sys/param.h>
     51 #include <sys/systm.h>
     52 #include <sys/buf.h>
     53 #include <sys/file.h>
     54 #include <sys/mount.h>
     55 #include <sys/vnode.h>
     56 #include <sys/mount.h>
     57 
     58 #include <ufs/ufs/quota.h>
     59 #include <ufs/ufs/ufsmount.h>
     60 #include <ufs/ufs/inode.h>
     61 #include <ufs/ufs/ufs_extern.h>
     62 #include <ufs/ufs/ufs_bswap.h>
     63 
     64 #include <ufs/ffs/fs.h>
     65 #include <ufs/ffs/ffs_extern.h>
     66 
     67 #include <uvm/uvm.h>
     68 
     69 static int ffs_balloc_ufs1(void *);
     70 static int ffs_balloc_ufs2(void *);
     71 
     72 /*
     73  * Balloc defines the structure of file system storage
     74  * by allocating the physical blocks on a device given
     75  * the inode and the logical block number in a file.
     76  */
     77 
     78 int
     79 ffs_balloc(v)
     80 	void *v;
     81 {
     82 	struct vop_balloc_args *ap = v;
     83 
     84 	if (VTOI(ap->a_vp)->i_fs->fs_magic == FS_UFS2_MAGIC)
     85 		return ffs_balloc_ufs2(v);
     86 	else
     87 		return ffs_balloc_ufs1(v);
     88 }
     89 
     90 static int
     91 ffs_balloc_ufs1(v)
     92 	void *v;
     93 {
     94 	struct vop_balloc_args /* {
     95 		struct vnode *a_vp;
     96 		off_t a_startoffset;
     97 		int a_size;
     98 		struct ucred *a_cred;
     99 		int a_flags;
    100 		struct buf **a_bpp;
    101 	} */ *ap = v;
    102 	daddr_t lbn, lastlbn;
    103 	int size;
    104 	struct ucred *cred;
    105 	int flags;
    106 	int32_t nb;
    107 	struct buf *bp, *nbp;
    108 	struct vnode *vp = ap->a_vp;
    109 	struct inode *ip = VTOI(vp);
    110 	struct fs *fs = ip->i_fs;
    111 	struct indir indirs[NIADDR + 2];
    112 	daddr_t newb, pref;
    113 	int32_t *bap;	/* XXX ondisk32 */
    114 	int deallocated, osize, nsize, num, i, error;
    115 	int32_t *blkp, *allocblk, allociblk[NIADDR + 1];
    116 	int32_t *allocib;
    117 	int unwindidx = -1;
    118 	struct buf **bpp = ap->a_bpp;
    119 #ifdef FFS_EI
    120 	const int needswap = UFS_FSNEEDSWAP(fs);
    121 #endif
    122 	UVMHIST_FUNC("ffs_balloc"); UVMHIST_CALLED(ubchist);
    123 
    124 	lbn = lblkno(fs, ap->a_startoffset);
    125 	size = blkoff(fs, ap->a_startoffset) + ap->a_size;
    126 	if (size > fs->fs_bsize)
    127 		panic("ffs_balloc: blk too big");
    128 	if (bpp != NULL) {
    129 		*bpp = NULL;
    130 	}
    131 	UVMHIST_LOG(ubchist, "vp %p lbn 0x%x size 0x%x", vp, lbn, size,0);
    132 
    133 	KASSERT(size <= fs->fs_bsize);
    134 	if (lbn < 0)
    135 		return (EFBIG);
    136 	cred = ap->a_cred;
    137 	flags = ap->a_flags;
    138 
    139 	/*
    140 	 * If the next write will extend the file into a new block,
    141 	 * and the file is currently composed of a fragment
    142 	 * this fragment has to be extended to be a full block.
    143 	 */
    144 
    145 	lastlbn = lblkno(fs, ip->i_size);
    146 	if (lastlbn < NDADDR && lastlbn < lbn) {
    147 		nb = lastlbn;
    148 		osize = blksize(fs, ip, nb);
    149 		if (osize < fs->fs_bsize && osize > 0) {
    150 			error = ffs_realloccg(ip, nb,
    151 				    ffs_blkpref_ufs1(ip, lastlbn, nb,
    152 					&ip->i_ffs1_db[0]),
    153 				    osize, (int)fs->fs_bsize, cred, bpp, &newb);
    154 			if (error)
    155 				return (error);
    156 			if (DOINGSOFTDEP(vp))
    157 				softdep_setup_allocdirect(ip, nb, newb,
    158 				    ufs_rw32(ip->i_ffs1_db[nb], needswap),
    159 				    fs->fs_bsize, osize, bpp ? *bpp : NULL);
    160 			ip->i_size = lblktosize(fs, nb + 1);
    161 			ip->i_ffs1_size = ip->i_size;
    162 			uvm_vnp_setsize(vp, ip->i_ffs1_size);
    163 			ip->i_ffs1_db[nb] = ufs_rw32((int32_t)newb, needswap);
    164 			ip->i_flag |= IN_CHANGE | IN_UPDATE;
    165 			if (bpp) {
    166 				if (flags & B_SYNC)
    167 					bwrite(*bpp);
    168 				else
    169 					bawrite(*bpp);
    170 			}
    171 		}
    172 	}
    173 
    174 	/*
    175 	 * The first NDADDR blocks are direct blocks
    176 	 */
    177 
    178 	if (lbn < NDADDR) {
    179 		nb = ufs_rw32(ip->i_ffs1_db[lbn], needswap);
    180 		if (nb != 0 && ip->i_size >= lblktosize(fs, lbn + 1)) {
    181 
    182 			/*
    183 			 * The block is an already-allocated direct block
    184 			 * and the file already extends past this block,
    185 			 * thus this must be a whole block.
    186 			 * Just read the block (if requested).
    187 			 */
    188 
    189 			if (bpp != NULL) {
    190 				error = bread(vp, lbn, fs->fs_bsize, NOCRED,
    191 					      bpp);
    192 				if (error) {
    193 					brelse(*bpp);
    194 					return (error);
    195 				}
    196 			}
    197 			return (0);
    198 		}
    199 		if (nb != 0) {
    200 
    201 			/*
    202 			 * Consider need to reallocate a fragment.
    203 			 */
    204 
    205 			osize = fragroundup(fs, blkoff(fs, ip->i_size));
    206 			nsize = fragroundup(fs, size);
    207 			if (nsize <= osize) {
    208 
    209 				/*
    210 				 * The existing block is already
    211 				 * at least as big as we want.
    212 				 * Just read the block (if requested).
    213 				 */
    214 
    215 				if (bpp != NULL) {
    216 					error = bread(vp, lbn, osize, NOCRED,
    217 						      bpp);
    218 					if (error) {
    219 						brelse(*bpp);
    220 						return (error);
    221 					}
    222 				}
    223 				return 0;
    224 			} else {
    225 
    226 				/*
    227 				 * The existing block is smaller than we want,
    228 				 * grow it.
    229 				 */
    230 
    231 				error = ffs_realloccg(ip, lbn,
    232 				    ffs_blkpref_ufs1(ip, lbn, (int)lbn,
    233 					&ip->i_ffs1_db[0]), osize, nsize, cred,
    234 					bpp, &newb);
    235 				if (error)
    236 					return (error);
    237 				if (DOINGSOFTDEP(vp))
    238 					softdep_setup_allocdirect(ip, lbn,
    239 					    newb, nb, nsize, osize,
    240 					    bpp ? *bpp : NULL);
    241 			}
    242 		} else {
    243 
    244 			/*
    245 			 * the block was not previously allocated,
    246 			 * allocate a new block or fragment.
    247 			 */
    248 
    249 			if (ip->i_size < lblktosize(fs, lbn + 1))
    250 				nsize = fragroundup(fs, size);
    251 			else
    252 				nsize = fs->fs_bsize;
    253 			error = ffs_alloc(ip, lbn,
    254 			    ffs_blkpref_ufs1(ip, lbn, (int)lbn,
    255 				&ip->i_ffs1_db[0]),
    256 				nsize, cred, &newb);
    257 			if (error)
    258 				return (error);
    259 			if (bpp != NULL) {
    260 				bp = getblk(vp, lbn, nsize, 0, 0);
    261 				bp->b_blkno = fsbtodb(fs, newb);
    262 				if (flags & B_CLRBUF)
    263 					clrbuf(bp);
    264 				*bpp = bp;
    265 			}
    266 			if (DOINGSOFTDEP(vp)) {
    267 				softdep_setup_allocdirect(ip, lbn, newb, 0,
    268 				    nsize, 0, bpp ? *bpp : NULL);
    269 			}
    270 		}
    271 		ip->i_ffs1_db[lbn] = ufs_rw32((int32_t)newb, needswap);
    272 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
    273 		return (0);
    274 	}
    275 
    276 	/*
    277 	 * Determine the number of levels of indirection.
    278 	 */
    279 
    280 	pref = 0;
    281 	if ((error = ufs_getlbns(vp, lbn, indirs, &num)) != 0)
    282 		return (error);
    283 
    284 	/*
    285 	 * Fetch the first indirect block allocating if necessary.
    286 	 */
    287 
    288 	--num;
    289 	nb = ufs_rw32(ip->i_ffs1_ib[indirs[0].in_off], needswap);
    290 	allocib = NULL;
    291 	allocblk = allociblk;
    292 	if (nb == 0) {
    293 		pref = ffs_blkpref_ufs1(ip, lbn, 0, (int32_t *)0);
    294 		error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred,
    295 		    &newb);
    296 		if (error)
    297 			goto fail;
    298 		nb = newb;
    299 		*allocblk++ = nb;
    300 		bp = getblk(vp, indirs[1].in_lbn, fs->fs_bsize, 0, 0);
    301 		bp->b_blkno = fsbtodb(fs, nb);
    302 		clrbuf(bp);
    303 		if (DOINGSOFTDEP(vp)) {
    304 			softdep_setup_allocdirect(ip, NDADDR + indirs[0].in_off,
    305 			    newb, 0, fs->fs_bsize, 0, bp);
    306 			bdwrite(bp);
    307 		} else {
    308 
    309 			/*
    310 			 * Write synchronously so that indirect blocks
    311 			 * never point at garbage.
    312 			 */
    313 
    314 			if ((error = bwrite(bp)) != 0)
    315 				goto fail;
    316 		}
    317 		unwindidx = 0;
    318 		allocib = &ip->i_ffs1_ib[indirs[0].in_off];
    319 		*allocib = ufs_rw32(nb, needswap);
    320 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
    321 	}
    322 
    323 	/*
    324 	 * Fetch through the indirect blocks, allocating as necessary.
    325 	 */
    326 
    327 	for (i = 1;;) {
    328 		error = bread(vp,
    329 		    indirs[i].in_lbn, (int)fs->fs_bsize, NOCRED, &bp);
    330 		if (error) {
    331 			brelse(bp);
    332 			goto fail;
    333 		}
    334 		bap = (int32_t *)bp->b_data;	/* XXX ondisk32 */
    335 		nb = ufs_rw32(bap[indirs[i].in_off], needswap);
    336 		if (i == num)
    337 			break;
    338 		i++;
    339 		if (nb != 0) {
    340 			brelse(bp);
    341 			continue;
    342 		}
    343 		if (pref == 0)
    344 			pref = ffs_blkpref_ufs1(ip, lbn, 0, (int32_t *)0);
    345 		error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred,
    346 		    &newb);
    347 		if (error) {
    348 			brelse(bp);
    349 			goto fail;
    350 		}
    351 		nb = newb;
    352 		*allocblk++ = nb;
    353 		nbp = getblk(vp, indirs[i].in_lbn, fs->fs_bsize, 0, 0);
    354 		nbp->b_blkno = fsbtodb(fs, nb);
    355 		clrbuf(nbp);
    356 		if (DOINGSOFTDEP(vp)) {
    357 			softdep_setup_allocindir_meta(nbp, ip, bp,
    358 			    indirs[i - 1].in_off, nb);
    359 			bdwrite(nbp);
    360 		} else {
    361 
    362 			/*
    363 			 * Write synchronously so that indirect blocks
    364 			 * never point at garbage.
    365 			 */
    366 
    367 			if ((error = bwrite(nbp)) != 0) {
    368 				brelse(bp);
    369 				goto fail;
    370 			}
    371 		}
    372 		if (unwindidx < 0)
    373 			unwindidx = i - 1;
    374 		bap[indirs[i - 1].in_off] = ufs_rw32(nb, needswap);
    375 
    376 		/*
    377 		 * If required, write synchronously, otherwise use
    378 		 * delayed write.
    379 		 */
    380 
    381 		if (flags & B_SYNC) {
    382 			bwrite(bp);
    383 		} else {
    384 			bdwrite(bp);
    385 		}
    386 	}
    387 
    388 	if (flags & B_METAONLY) {
    389 		*bpp = bp;
    390 		return (0);
    391 	}
    392 
    393 	/*
    394 	 * Get the data block, allocating if necessary.
    395 	 */
    396 
    397 	if (nb == 0) {
    398 		pref = ffs_blkpref_ufs1(ip, lbn, indirs[num].in_off, &bap[0]);
    399 		error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred,
    400 		    &newb);
    401 		if (error) {
    402 			brelse(bp);
    403 			goto fail;
    404 		}
    405 		nb = newb;
    406 		*allocblk++ = nb;
    407 		if (bpp != NULL) {
    408 			nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0);
    409 			nbp->b_blkno = fsbtodb(fs, nb);
    410 			if (flags & B_CLRBUF)
    411 				clrbuf(nbp);
    412 			*bpp = nbp;
    413 		}
    414 		if (DOINGSOFTDEP(vp))
    415 			softdep_setup_allocindir_page(ip, lbn, bp,
    416 			    indirs[num].in_off, nb, 0, bpp ? *bpp : NULL);
    417 		bap[indirs[num].in_off] = ufs_rw32(nb, needswap);
    418 		if (allocib == NULL && unwindidx < 0) {
    419 			unwindidx = i - 1;
    420 		}
    421 
    422 		/*
    423 		 * If required, write synchronously, otherwise use
    424 		 * delayed write.
    425 		 */
    426 
    427 		if (flags & B_SYNC) {
    428 			bwrite(bp);
    429 		} else {
    430 			bdwrite(bp);
    431 		}
    432 		return (0);
    433 	}
    434 	brelse(bp);
    435 	if (bpp != NULL) {
    436 		if (flags & B_CLRBUF) {
    437 			error = bread(vp, lbn, (int)fs->fs_bsize, NOCRED, &nbp);
    438 			if (error) {
    439 				brelse(nbp);
    440 				goto fail;
    441 			}
    442 		} else {
    443 			nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0);
    444 			nbp->b_blkno = fsbtodb(fs, nb);
    445 			clrbuf(nbp);
    446 		}
    447 		*bpp = nbp;
    448 	}
    449 	return (0);
    450 
    451 fail:
    452 	/*
    453 	 * If we have failed part way through block allocation, we
    454 	 * have to deallocate any indirect blocks that we have allocated.
    455 	 */
    456 
    457 	if (unwindidx >= 0) {
    458 
    459 		/*
    460 		 * First write out any buffers we've created to resolve their
    461 		 * softdeps.  This must be done in reverse order of creation
    462 		 * so that we resolve the dependencies in one pass.
    463 		 * Write the cylinder group buffers for these buffers too.
    464 		 */
    465 
    466 		for (i = num; i >= unwindidx; i--) {
    467 			if (i == 0) {
    468 				break;
    469 			}
    470 			bp = getblk(vp, indirs[i].in_lbn, (int)fs->fs_bsize, 0,
    471 			    0);
    472 			if (bp->b_flags & B_DELWRI) {
    473 				nb = fsbtodb(fs, cgtod(fs, dtog(fs,
    474 				    dbtofsb(fs, bp->b_blkno))));
    475 				bwrite(bp);
    476 				bp = getblk(ip->i_devvp, nb, (int)fs->fs_cgsize,
    477 				    0, 0);
    478 				if (bp->b_flags & B_DELWRI) {
    479 					bwrite(bp);
    480 				} else {
    481 					bp->b_flags |= B_INVAL;
    482 					brelse(bp);
    483 				}
    484 			} else {
    485 				bp->b_flags |= B_INVAL;
    486 				brelse(bp);
    487 			}
    488 		}
    489 		if (DOINGSOFTDEP(vp) && unwindidx == 0) {
    490 			ip->i_flag |= IN_CHANGE | IN_UPDATE;
    491 			VOP_UPDATE(vp, NULL, NULL, UPDATE_WAIT);
    492 		}
    493 
    494 		/*
    495 		 * Now that any dependencies that we created have been
    496 		 * resolved, we can undo the partial allocation.
    497 		 */
    498 
    499 		if (unwindidx == 0) {
    500 			*allocib = 0;
    501 			ip->i_flag |= IN_CHANGE | IN_UPDATE;
    502 			if (DOINGSOFTDEP(vp))
    503 				VOP_UPDATE(vp, NULL, NULL, UPDATE_WAIT);
    504 		} else {
    505 			int r;
    506 
    507 			r = bread(vp, indirs[unwindidx].in_lbn,
    508 			    (int)fs->fs_bsize, NOCRED, &bp);
    509 			if (r) {
    510 				panic("Could not unwind indirect block, error %d", r);
    511 				brelse(bp);
    512 			} else {
    513 				bap = (int32_t *)bp->b_data; /* XXX ondisk32 */
    514 				bap[indirs[unwindidx].in_off] = 0;
    515 				bwrite(bp);
    516 			}
    517 		}
    518 		for (i = unwindidx + 1; i <= num; i++) {
    519 			bp = getblk(vp, indirs[i].in_lbn, (int)fs->fs_bsize, 0,
    520 			    0);
    521 			bp->b_flags |= B_INVAL;
    522 			brelse(bp);
    523 		}
    524 	}
    525 	for (deallocated = 0, blkp = allociblk; blkp < allocblk; blkp++) {
    526 		ffs_blkfree(fs, ip->i_devvp, *blkp, fs->fs_bsize, ip->i_number);
    527 		deallocated += fs->fs_bsize;
    528 	}
    529 	if (deallocated) {
    530 #ifdef QUOTA
    531 		/*
    532 		 * Restore user's disk quota because allocation failed.
    533 		 */
    534 		(void)chkdq(ip, -btodb(deallocated), cred, FORCE);
    535 #endif
    536 		ip->i_ffs1_blocks -= btodb(deallocated);
    537 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
    538 	}
    539 	return (error);
    540 }
    541 
    542 static int
    543 ffs_balloc_ufs2(v)
    544 	void *v;
    545 {
    546 	struct vop_balloc_args /* {
    547 		struct vnode *a_vp;
    548 		off_t a_startoffset;
    549 		int a_size;
    550 		struct ucred *a_cred;
    551 		int a_flags;
    552 		struct buf **a_bpp;
    553 	} */ *ap = v;
    554 	daddr_t lbn, lastlbn;
    555 	int size;
    556 	struct ucred *cred;
    557 	int flags;
    558 	struct buf *bp, *nbp;
    559 	struct vnode *vp = ap->a_vp;
    560 	struct inode *ip = VTOI(vp);
    561 	struct fs *fs = ip->i_fs;
    562 	struct indir indirs[NIADDR + 2];
    563 	daddr_t newb, pref, nb;
    564 	int64_t *bap;
    565 	int deallocated, osize, nsize, num, i, error;
    566 	daddr_t *blkp, *allocblk, allociblk[NIADDR + 1];
    567 	int64_t *allocib;
    568 	int unwindidx = -1;
    569 	struct buf **bpp = ap->a_bpp;
    570 #ifdef FFS_EI
    571 	const int needswap = UFS_FSNEEDSWAP(fs);
    572 #endif
    573 	UVMHIST_FUNC("ffs_balloc"); UVMHIST_CALLED(ubchist);
    574 
    575 	lbn = lblkno(fs, ap->a_startoffset);
    576 	size = blkoff(fs, ap->a_startoffset) + ap->a_size;
    577 	if (size > fs->fs_bsize)
    578 		panic("ffs_balloc: blk too big");
    579 	if (bpp != NULL) {
    580 		*bpp = NULL;
    581 	}
    582 	UVMHIST_LOG(ubchist, "vp %p lbn 0x%x size 0x%x", vp, lbn, size,0);
    583 
    584 	KASSERT(size <= fs->fs_bsize);
    585 	if (lbn < 0)
    586 		return (EFBIG);
    587 	cred = ap->a_cred;
    588 	flags = ap->a_flags;
    589 
    590 #ifdef notyet
    591 	/*
    592 	 * Check for allocating external data.
    593 	 */
    594 	if (flags & IO_EXT) {
    595 		if (lbn >= NXADDR)
    596 			return (EFBIG);
    597 		/*
    598 		 * If the next write will extend the data into a new block,
    599 		 * and the data is currently composed of a fragment
    600 		 * this fragment has to be extended to be a full block.
    601 		 */
    602 		lastlbn = lblkno(fs, dp->di_extsize);
    603 		if (lastlbn < lbn) {
    604 			nb = lastlbn;
    605 			osize = sblksize(fs, dp->di_extsize, nb);
    606 			if (osize < fs->fs_bsize && osize > 0) {
    607 				error = ffs_realloccg(ip, -1 - nb,
    608 				    dp->di_extb[nb],
    609 				    ffs_blkpref_ufs2(ip, lastlbn, (int)nb,
    610 				    &dp->di_extb[0]), osize,
    611 				    (int)fs->fs_bsize, cred, &bp);
    612 				if (error)
    613 					return (error);
    614 				if (DOINGSOFTDEP(vp))
    615 					softdep_setup_allocext(ip, nb,
    616 					    dbtofsb(fs, bp->b_blkno),
    617 					    dp->di_extb[nb],
    618 					    fs->fs_bsize, osize, bp);
    619 				dp->di_extsize = smalllblktosize(fs, nb + 1);
    620 				dp->di_extb[nb] = dbtofsb(fs, bp->b_blkno);
    621 				bp->b_xflags |= BX_ALTDATA;
    622 				ip->i_flag |= IN_CHANGE | IN_UPDATE;
    623 				if (flags & IO_SYNC)
    624 					bwrite(bp);
    625 				else
    626 					bawrite(bp);
    627 			}
    628 		}
    629 		/*
    630 		 * All blocks are direct blocks
    631 		 */
    632 		if (flags & BA_METAONLY)
    633 			panic("ffs_balloc_ufs2: BA_METAONLY for ext block");
    634 		nb = dp->di_extb[lbn];
    635 		if (nb != 0 && dp->di_extsize >= smalllblktosize(fs, lbn + 1)) {
    636 			error = bread(vp, -1 - lbn, fs->fs_bsize, NOCRED, &bp);
    637 			if (error) {
    638 				brelse(bp);
    639 				return (error);
    640 			}
    641 			bp->b_blkno = fsbtodb(fs, nb);
    642 			bp->b_xflags |= BX_ALTDATA;
    643 			*bpp = bp;
    644 			return (0);
    645 		}
    646 		if (nb != 0) {
    647 			/*
    648 			 * Consider need to reallocate a fragment.
    649 			 */
    650 			osize = fragroundup(fs, blkoff(fs, dp->di_extsize));
    651 			nsize = fragroundup(fs, size);
    652 			if (nsize <= osize) {
    653 				error = bread(vp, -1 - lbn, osize, NOCRED, &bp);
    654 				if (error) {
    655 					brelse(bp);
    656 					return (error);
    657 				}
    658 				bp->b_blkno = fsbtodb(fs, nb);
    659 				bp->b_xflags |= BX_ALTDATA;
    660 			} else {
    661 				error = ffs_realloccg(ip, -1 - lbn,
    662 				    dp->di_extb[lbn],
    663 				    ffs_blkpref_ufs2(ip, lbn, (int)lbn,
    664 				    &dp->di_extb[0]), osize, nsize, cred, &bp);
    665 				if (error)
    666 					return (error);
    667 				bp->b_xflags |= BX_ALTDATA;
    668 				if (DOINGSOFTDEP(vp))
    669 					softdep_setup_allocext(ip, lbn,
    670 					    dbtofsb(fs, bp->b_blkno), nb,
    671 					    nsize, osize, bp);
    672 			}
    673 		} else {
    674 			if (dp->di_extsize < smalllblktosize(fs, lbn + 1))
    675 				nsize = fragroundup(fs, size);
    676 			else
    677 				nsize = fs->fs_bsize;
    678 			error = ffs_alloc(ip, lbn,
    679 			   ffs_blkpref_ufs2(ip, lbn, (int)lbn, &dp->di_extb[0]),
    680 			   nsize, cred, &newb);
    681 			if (error)
    682 				return (error);
    683 			bp = getblk(vp, -1 - lbn, nsize, 0, 0);
    684 			bp->b_blkno = fsbtodb(fs, newb);
    685 			bp->b_xflags |= BX_ALTDATA;
    686 			if (flags & BA_CLRBUF)
    687 				vfs_bio_clrbuf(bp);
    688 			if (DOINGSOFTDEP(vp))
    689 				softdep_setup_allocext(ip, lbn, newb, 0,
    690 				    nsize, 0, bp);
    691 		}
    692 		dp->di_extb[lbn] = dbtofsb(fs, bp->b_blkno);
    693 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
    694 		*bpp = bp;
    695 		return (0);
    696 	}
    697 #endif
    698 	/*
    699 	 * If the next write will extend the file into a new block,
    700 	 * and the file is currently composed of a fragment
    701 	 * this fragment has to be extended to be a full block.
    702 	 */
    703 
    704 	lastlbn = lblkno(fs, ip->i_size);
    705 	if (lastlbn < NDADDR && lastlbn < lbn) {
    706 		nb = lastlbn;
    707 		osize = blksize(fs, ip, nb);
    708 		if (osize < fs->fs_bsize && osize > 0) {
    709 			error = ffs_realloccg(ip, nb,
    710 				    ffs_blkpref_ufs2(ip, lastlbn, nb,
    711 					&ip->i_ffs2_db[0]),
    712 				    osize, (int)fs->fs_bsize, cred, bpp, &newb);
    713 			if (error)
    714 				return (error);
    715 			if (DOINGSOFTDEP(vp))
    716 				softdep_setup_allocdirect(ip, nb, newb,
    717 				    ufs_rw64(ip->i_ffs2_db[nb], needswap),
    718 				    fs->fs_bsize, osize, bpp ? *bpp : NULL);
    719 			ip->i_size = lblktosize(fs, nb + 1);
    720 			ip->i_ffs2_size = ip->i_size;
    721 			uvm_vnp_setsize(vp, ip->i_size);
    722 			ip->i_ffs2_db[nb] = ufs_rw64(newb, needswap);
    723 			ip->i_flag |= IN_CHANGE | IN_UPDATE;
    724 			if (bpp) {
    725 				if (flags & B_SYNC)
    726 					bwrite(*bpp);
    727 				else
    728 					bawrite(*bpp);
    729 			}
    730 		}
    731 	}
    732 
    733 	/*
    734 	 * The first NDADDR blocks are direct blocks
    735 	 */
    736 
    737 	if (lbn < NDADDR) {
    738 		nb = ufs_rw64(ip->i_ffs2_db[lbn], needswap);
    739 		if (nb != 0 && ip->i_size >= lblktosize(fs, lbn + 1)) {
    740 
    741 			/*
    742 			 * The block is an already-allocated direct block
    743 			 * and the file already extends past this block,
    744 			 * thus this must be a whole block.
    745 			 * Just read the block (if requested).
    746 			 */
    747 
    748 			if (bpp != NULL) {
    749 				error = bread(vp, lbn, fs->fs_bsize, NOCRED,
    750 					      bpp);
    751 				if (error) {
    752 					brelse(*bpp);
    753 					return (error);
    754 				}
    755 			}
    756 			return (0);
    757 		}
    758 		if (nb != 0) {
    759 
    760 			/*
    761 			 * Consider need to reallocate a fragment.
    762 			 */
    763 
    764 			osize = fragroundup(fs, blkoff(fs, ip->i_size));
    765 			nsize = fragroundup(fs, size);
    766 			if (nsize <= osize) {
    767 
    768 				/*
    769 				 * The existing block is already
    770 				 * at least as big as we want.
    771 				 * Just read the block (if requested).
    772 				 */
    773 
    774 				if (bpp != NULL) {
    775 					error = bread(vp, lbn, osize, NOCRED,
    776 						      bpp);
    777 					if (error) {
    778 						brelse(*bpp);
    779 						return (error);
    780 					}
    781 				}
    782 				return 0;
    783 			} else {
    784 
    785 				/*
    786 				 * The existing block is smaller than we want,
    787 				 * grow it.
    788 				 */
    789 
    790 				error = ffs_realloccg(ip, lbn,
    791 				    ffs_blkpref_ufs2(ip, lbn, (int)lbn,
    792 					&ip->i_ffs2_db[0]), osize, nsize, cred,
    793 					bpp, &newb);
    794 				if (error)
    795 					return (error);
    796 				if (DOINGSOFTDEP(vp))
    797 					softdep_setup_allocdirect(ip, lbn,
    798 					    newb, nb, nsize, osize,
    799 					    bpp ? *bpp : NULL);
    800 			}
    801 		} else {
    802 
    803 			/*
    804 			 * the block was not previously allocated,
    805 			 * allocate a new block or fragment.
    806 			 */
    807 
    808 			if (ip->i_size < lblktosize(fs, lbn + 1))
    809 				nsize = fragroundup(fs, size);
    810 			else
    811 				nsize = fs->fs_bsize;
    812 			error = ffs_alloc(ip, lbn,
    813 			    ffs_blkpref_ufs2(ip, lbn, (int)lbn,
    814 				&ip->i_ffs2_db[0]), nsize, cred, &newb);
    815 			if (error)
    816 				return (error);
    817 			if (bpp != NULL) {
    818 				bp = getblk(vp, lbn, nsize, 0, 0);
    819 				bp->b_blkno = fsbtodb(fs, newb);
    820 				if (flags & B_CLRBUF)
    821 					clrbuf(bp);
    822 				*bpp = bp;
    823 			}
    824 			if (DOINGSOFTDEP(vp)) {
    825 				softdep_setup_allocdirect(ip, lbn, newb, 0,
    826 				    nsize, 0, bpp ? *bpp : NULL);
    827 			}
    828 		}
    829 		ip->i_ffs2_db[lbn] = ufs_rw64(newb, needswap);
    830 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
    831 		return (0);
    832 	}
    833 
    834 	/*
    835 	 * Determine the number of levels of indirection.
    836 	 */
    837 
    838 	pref = 0;
    839 	if ((error = ufs_getlbns(vp, lbn, indirs, &num)) != 0)
    840 		return (error);
    841 
    842 	/*
    843 	 * Fetch the first indirect block allocating if necessary.
    844 	 */
    845 
    846 	--num;
    847 	nb = ufs_rw64(ip->i_ffs2_ib[indirs[0].in_off], needswap);
    848 	allocib = NULL;
    849 	allocblk = allociblk;
    850 	if (nb == 0) {
    851 		pref = ffs_blkpref_ufs2(ip, lbn, 0, (int64_t *)0);
    852 		error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred,
    853 		    &newb);
    854 		if (error)
    855 			goto fail;
    856 		nb = newb;
    857 		*allocblk++ = nb;
    858 		bp = getblk(vp, indirs[1].in_lbn, fs->fs_bsize, 0, 0);
    859 		bp->b_blkno = fsbtodb(fs, nb);
    860 		clrbuf(bp);
    861 		if (DOINGSOFTDEP(vp)) {
    862 			softdep_setup_allocdirect(ip, NDADDR + indirs[0].in_off,
    863 			    newb, 0, fs->fs_bsize, 0, bp);
    864 			bdwrite(bp);
    865 		} else {
    866 
    867 			/*
    868 			 * Write synchronously so that indirect blocks
    869 			 * never point at garbage.
    870 			 */
    871 
    872 			if ((error = bwrite(bp)) != 0)
    873 				goto fail;
    874 		}
    875 		unwindidx = 0;
    876 		allocib = &ip->i_ffs2_ib[indirs[0].in_off];
    877 		*allocib = ufs_rw64(nb, needswap);
    878 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
    879 	}
    880 
    881 	/*
    882 	 * Fetch through the indirect blocks, allocating as necessary.
    883 	 */
    884 
    885 	for (i = 1;;) {
    886 		error = bread(vp,
    887 		    indirs[i].in_lbn, (int)fs->fs_bsize, NOCRED, &bp);
    888 		if (error) {
    889 			brelse(bp);
    890 			goto fail;
    891 		}
    892 		bap = (int64_t *)bp->b_data;
    893 		nb = ufs_rw64(bap[indirs[i].in_off], needswap);
    894 		if (i == num)
    895 			break;
    896 		i++;
    897 		if (nb != 0) {
    898 			brelse(bp);
    899 			continue;
    900 		}
    901 		if (pref == 0)
    902 			pref = ffs_blkpref_ufs2(ip, lbn, 0, (int64_t *)0);
    903 		error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred,
    904 		    &newb);
    905 		if (error) {
    906 			brelse(bp);
    907 			goto fail;
    908 		}
    909 		nb = newb;
    910 		*allocblk++ = nb;
    911 		nbp = getblk(vp, indirs[i].in_lbn, fs->fs_bsize, 0, 0);
    912 		nbp->b_blkno = fsbtodb(fs, nb);
    913 		clrbuf(nbp);
    914 		if (DOINGSOFTDEP(vp)) {
    915 			softdep_setup_allocindir_meta(nbp, ip, bp,
    916 			    indirs[i - 1].in_off, nb);
    917 			bdwrite(nbp);
    918 		} else {
    919 
    920 			/*
    921 			 * Write synchronously so that indirect blocks
    922 			 * never point at garbage.
    923 			 */
    924 
    925 			if ((error = bwrite(nbp)) != 0) {
    926 				brelse(bp);
    927 				goto fail;
    928 			}
    929 		}
    930 		if (unwindidx < 0)
    931 			unwindidx = i - 1;
    932 		bap[indirs[i - 1].in_off] = ufs_rw64(nb, needswap);
    933 
    934 		/*
    935 		 * If required, write synchronously, otherwise use
    936 		 * delayed write.
    937 		 */
    938 
    939 		if (flags & B_SYNC) {
    940 			bwrite(bp);
    941 		} else {
    942 			bdwrite(bp);
    943 		}
    944 	}
    945 
    946 	if (flags & B_METAONLY) {
    947 		*bpp = bp;
    948 		return (0);
    949 	}
    950 
    951 	/*
    952 	 * Get the data block, allocating if necessary.
    953 	 */
    954 
    955 	if (nb == 0) {
    956 		pref = ffs_blkpref_ufs2(ip, lbn, indirs[num].in_off, &bap[0]);
    957 		error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred,
    958 		    &newb);
    959 		if (error) {
    960 			brelse(bp);
    961 			goto fail;
    962 		}
    963 		nb = newb;
    964 		*allocblk++ = nb;
    965 		if (bpp != NULL) {
    966 			nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0);
    967 			nbp->b_blkno = fsbtodb(fs, nb);
    968 			if (flags & B_CLRBUF)
    969 				clrbuf(nbp);
    970 			*bpp = nbp;
    971 		}
    972 		if (DOINGSOFTDEP(vp))
    973 			softdep_setup_allocindir_page(ip, lbn, bp,
    974 			    indirs[num].in_off, nb, 0, bpp ? *bpp : NULL);
    975 		bap[indirs[num].in_off] = ufs_rw64(nb, needswap);
    976 		if (allocib == NULL && unwindidx < 0) {
    977 			unwindidx = i - 1;
    978 		}
    979 
    980 		/*
    981 		 * If required, write synchronously, otherwise use
    982 		 * delayed write.
    983 		 */
    984 
    985 		if (flags & B_SYNC) {
    986 			bwrite(bp);
    987 		} else {
    988 			bdwrite(bp);
    989 		}
    990 		return (0);
    991 	}
    992 	brelse(bp);
    993 	if (bpp != NULL) {
    994 		if (flags & B_CLRBUF) {
    995 			error = bread(vp, lbn, (int)fs->fs_bsize, NOCRED, &nbp);
    996 			if (error) {
    997 				brelse(nbp);
    998 				goto fail;
    999 			}
   1000 		} else {
   1001 			nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0);
   1002 			nbp->b_blkno = fsbtodb(fs, nb);
   1003 			clrbuf(nbp);
   1004 		}
   1005 		*bpp = nbp;
   1006 	}
   1007 	return (0);
   1008 
   1009 fail:
   1010 	/*
   1011 	 * If we have failed part way through block allocation, we
   1012 	 * have to deallocate any indirect blocks that we have allocated.
   1013 	 */
   1014 
   1015 	if (unwindidx >= 0) {
   1016 
   1017 		/*
   1018 		 * First write out any buffers we've created to resolve their
   1019 		 * softdeps.  This must be done in reverse order of creation
   1020 		 * so that we resolve the dependencies in one pass.
   1021 		 * Write the cylinder group buffers for these buffers too.
   1022 		 */
   1023 
   1024 		for (i = num; i >= unwindidx; i--) {
   1025 			if (i == 0) {
   1026 				break;
   1027 			}
   1028 			bp = getblk(vp, indirs[i].in_lbn, (int)fs->fs_bsize, 0,
   1029 			    0);
   1030 			if (bp->b_flags & B_DELWRI) {
   1031 				nb = fsbtodb(fs, cgtod(fs, dtog(fs,
   1032 				    dbtofsb(fs, bp->b_blkno))));
   1033 				bwrite(bp);
   1034 				bp = getblk(ip->i_devvp, nb, (int)fs->fs_cgsize,
   1035 				    0, 0);
   1036 				if (bp->b_flags & B_DELWRI) {
   1037 					bwrite(bp);
   1038 				} else {
   1039 					bp->b_flags |= B_INVAL;
   1040 					brelse(bp);
   1041 				}
   1042 			} else {
   1043 				bp->b_flags |= B_INVAL;
   1044 				brelse(bp);
   1045 			}
   1046 		}
   1047 		if (DOINGSOFTDEP(vp) && unwindidx == 0) {
   1048 			ip->i_flag |= IN_CHANGE | IN_UPDATE;
   1049 			VOP_UPDATE(vp, NULL, NULL, UPDATE_WAIT);
   1050 		}
   1051 
   1052 		/*
   1053 		 * Now that any dependencies that we created have been
   1054 		 * resolved, we can undo the partial allocation.
   1055 		 */
   1056 
   1057 		if (unwindidx == 0) {
   1058 			*allocib = 0;
   1059 			ip->i_flag |= IN_CHANGE | IN_UPDATE;
   1060 			if (DOINGSOFTDEP(vp))
   1061 				VOP_UPDATE(vp, NULL, NULL, UPDATE_WAIT);
   1062 		} else {
   1063 			int r;
   1064 
   1065 			r = bread(vp, indirs[unwindidx].in_lbn,
   1066 			    (int)fs->fs_bsize, NOCRED, &bp);
   1067 			if (r) {
   1068 				panic("Could not unwind indirect block, error %d", r);
   1069 				brelse(bp);
   1070 			} else {
   1071 				bap = (int64_t *)bp->b_data;
   1072 				bap[indirs[unwindidx].in_off] = 0;
   1073 				bwrite(bp);
   1074 			}
   1075 		}
   1076 		for (i = unwindidx + 1; i <= num; i++) {
   1077 			bp = getblk(vp, indirs[i].in_lbn, (int)fs->fs_bsize, 0,
   1078 			    0);
   1079 			bp->b_flags |= B_INVAL;
   1080 			brelse(bp);
   1081 		}
   1082 	}
   1083 	for (deallocated = 0, blkp = allociblk; blkp < allocblk; blkp++) {
   1084 		ffs_blkfree(fs, ip->i_devvp, *blkp, fs->fs_bsize, ip->i_number);
   1085 		deallocated += fs->fs_bsize;
   1086 	}
   1087 	if (deallocated) {
   1088 #ifdef QUOTA
   1089 		/*
   1090 		 * Restore user's disk quota because allocation failed.
   1091 		 */
   1092 		(void)chkdq(ip, -btodb(deallocated), cred, FORCE);
   1093 #endif
   1094 		ip->i_ffs2_blocks -= btodb(deallocated);
   1095 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
   1096 	}
   1097 	return (error);
   1098 }
   1099