Home | History | Annotate | Line # | Download | only in ffs
ffs_balloc.c revision 1.21
      1 /*	$NetBSD: ffs_balloc.c,v 1.21 2000/06/28 14:16:40 mrg 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. All advertising materials mentioning features or use of this software
     16  *    must display the following acknowledgement:
     17  *	This product includes software developed by the University of
     18  *	California, Berkeley and its contributors.
     19  * 4. Neither the name of the University nor the names of its contributors
     20  *    may be used to endorse or promote products derived from this software
     21  *    without specific prior written permission.
     22  *
     23  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     26  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     27  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     32  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     33  * SUCH DAMAGE.
     34  *
     35  *	@(#)ffs_balloc.c	8.8 (Berkeley) 6/16/95
     36  */
     37 
     38 #if defined(_KERNEL) && !defined(_LKM)
     39 #include "opt_quota.h"
     40 #endif
     41 
     42 #include <sys/param.h>
     43 #include <sys/systm.h>
     44 #include <sys/buf.h>
     45 #include <sys/proc.h>
     46 #include <sys/file.h>
     47 #include <sys/mount.h>
     48 #include <sys/vnode.h>
     49 #include <sys/mount.h>
     50 
     51 #include <ufs/ufs/quota.h>
     52 #include <ufs/ufs/ufsmount.h>
     53 #include <ufs/ufs/inode.h>
     54 #include <ufs/ufs/ufs_extern.h>
     55 #include <ufs/ufs/ufs_bswap.h>
     56 
     57 #include <ufs/ffs/fs.h>
     58 #include <ufs/ffs/ffs_extern.h>
     59 
     60 /*
     61  * Balloc defines the structure of file system storage
     62  * by allocating the physical blocks on a device given
     63  * the inode and the logical block number in a file.
     64  */
     65 int
     66 ffs_balloc(v)
     67 	void *v;
     68 {
     69 	struct vop_balloc_args /* {
     70 		struct vnode *a_vp;
     71 		off_t a_startpoint;
     72 		int a_size;
     73 		struct ucred *a_cred;
     74 		int a_flags;
     75 		struct buf *a_bpp;
     76 	} */ *ap = v;
     77 	ufs_daddr_t lbn;
     78 	int size;
     79 	struct ucred *cred;
     80 	int flags;
     81 	ufs_daddr_t nb;
     82 	struct buf *bp, *nbp;
     83 	struct vnode *vp = ap->a_vp;
     84 	struct inode *ip = VTOI(vp);
     85 	struct fs *fs = ip->i_fs;
     86 	struct indir indirs[NIADDR + 2];
     87 	ufs_daddr_t newb, *bap, pref;
     88 	int deallocated, osize, nsize, num, i, error;
     89 	ufs_daddr_t *allocib, *blkp, *allocblk, allociblk[NIADDR + 1];
     90 	int unwindidx = -1;
     91 #ifdef FFS_EI
     92 	const int needswap = UFS_FSNEEDSWAP(fs);
     93 #endif
     94 
     95 	lbn = lblkno(fs, ap->a_startoffset);
     96 	size = blkoff(fs, ap->a_startoffset) + ap->a_size;
     97 	if (size > fs->fs_bsize)
     98 		panic("ffs_balloc: blk too big");
     99 	*ap->a_bpp = NULL;
    100 	if (lbn < 0)
    101 		return (EFBIG);
    102 	cred = ap->a_cred;
    103 	flags = ap->a_flags;
    104 
    105 	/*
    106 	 * If the next write will extend the file into a new block,
    107 	 * and the file is currently composed of a fragment
    108 	 * this fragment has to be extended to be a full block.
    109 	 */
    110 	nb = lblkno(fs, ip->i_ffs_size);
    111 	if (nb < NDADDR && nb < lbn) {
    112 		osize = blksize(fs, ip, nb);
    113 		if (osize < fs->fs_bsize && osize > 0) {
    114 			error = ffs_realloccg(ip, nb,
    115 				ffs_blkpref(ip, nb, (int)nb, &ip->i_ffs_db[0]),
    116 				osize, (int)fs->fs_bsize, cred, &bp);
    117 			if (error)
    118 				return (error);
    119 			if (DOINGSOFTDEP(vp))
    120 				softdep_setup_allocdirect(ip, nb,
    121 				    dbtofsb(fs, bp->b_blkno),
    122 				    ufs_rw32(ip->i_ffs_db[nb], needswap),
    123 				    fs->fs_bsize, osize, bp);
    124 			ip->i_ffs_size = (nb + 1) * fs->fs_bsize;
    125 			uvm_vnp_setsize(vp, ip->i_ffs_size);
    126 			ip->i_ffs_db[nb] = ufs_rw32(dbtofsb(fs, bp->b_blkno),
    127 			    needswap);
    128 			ip->i_flag |= IN_CHANGE | IN_UPDATE;
    129 			if (flags & B_SYNC)
    130 				bwrite(bp);
    131 			else
    132 				bawrite(bp);
    133 		}
    134 	}
    135 	/*
    136 	 * The first NDADDR blocks are direct blocks
    137 	 */
    138 	if (lbn < NDADDR) {
    139 		nb = ufs_rw32(ip->i_ffs_db[lbn], needswap);
    140 		if (nb != 0 && ip->i_ffs_size >= (lbn + 1) * fs->fs_bsize) {
    141 			error = bread(vp, lbn, fs->fs_bsize, NOCRED, &bp);
    142 			if (error) {
    143 				brelse(bp);
    144 				return (error);
    145 			}
    146 			*ap->a_bpp = bp;
    147 			return (0);
    148 		}
    149 		if (nb != 0) {
    150 			/*
    151 			 * Consider need to reallocate a fragment.
    152 			 */
    153 			osize = fragroundup(fs, blkoff(fs, ip->i_ffs_size));
    154 			nsize = fragroundup(fs, size);
    155 			if (nsize <= osize) {
    156 				error = bread(vp, lbn, osize, NOCRED, &bp);
    157 				if (error) {
    158 					brelse(bp);
    159 					return (error);
    160 				}
    161 			} else {
    162 				error = ffs_realloccg(ip, lbn,
    163 				    ffs_blkpref(ip, lbn, (int)lbn,
    164 					&ip->i_ffs_db[0]), osize, nsize, cred,
    165 					&bp);
    166 				if (error)
    167 					return (error);
    168 				if (DOINGSOFTDEP(vp))
    169 					softdep_setup_allocdirect(ip, lbn,
    170 					    dbtofsb(fs, bp->b_blkno), nb,
    171 					    nsize, osize, bp);
    172 			}
    173 		} else {
    174 			if (ip->i_ffs_size < (lbn + 1) * fs->fs_bsize)
    175 				nsize = fragroundup(fs, size);
    176 			else
    177 				nsize = fs->fs_bsize;
    178 			error = ffs_alloc(ip, lbn,
    179 			    ffs_blkpref(ip, lbn, (int)lbn, &ip->i_ffs_db[0]),
    180 				nsize, cred, &newb);
    181 			if (error)
    182 				return (error);
    183 			bp = getblk(vp, lbn, nsize, 0, 0);
    184 			bp->b_blkno = fsbtodb(fs, newb);
    185 			if (flags & B_CLRBUF)
    186 				clrbuf(bp);
    187 			if (DOINGSOFTDEP(vp))
    188 				softdep_setup_allocdirect(ip, lbn, newb, 0,
    189 				    nsize, 0, bp);
    190 		}
    191 		ip->i_ffs_db[lbn] = ufs_rw32(dbtofsb(fs, bp->b_blkno),
    192 			needswap);
    193 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
    194 		*ap->a_bpp = bp;
    195 		return (0);
    196 	}
    197 	/*
    198 	 * Determine the number of levels of indirection.
    199 	 */
    200 	pref = 0;
    201 	if ((error = ufs_getlbns(vp, lbn, indirs, &num)) != 0)
    202 		return(error);
    203 #ifdef DIAGNOSTIC
    204 	if (num < 1)
    205 		panic ("ffs_balloc: ufs_bmaparray returned indirect block\n");
    206 #endif
    207 	/*
    208 	 * Fetch the first indirect block allocating if necessary.
    209 	 */
    210 	--num;
    211 	nb = ufs_rw32(ip->i_ffs_ib[indirs[0].in_off], needswap);
    212 	allocib = NULL;
    213 	allocblk = allociblk;
    214 	if (nb == 0) {
    215 		pref = ffs_blkpref(ip, lbn, 0, (ufs_daddr_t *)0);
    216 		error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred,
    217 		    &newb);
    218 		if (error)
    219 			return (error);
    220 		nb = newb;
    221 		*allocblk++ = nb;
    222 		bp = getblk(vp, indirs[1].in_lbn, fs->fs_bsize, 0, 0);
    223 		bp->b_blkno = fsbtodb(fs, nb);
    224 		clrbuf(bp);
    225 		if (DOINGSOFTDEP(vp)) {
    226 			softdep_setup_allocdirect(ip, NDADDR + indirs[0].in_off,
    227 			    newb, 0, fs->fs_bsize, 0, bp);
    228 			bdwrite(bp);
    229 		} else {
    230 			/*
    231 			 * Write synchronously so that indirect blocks
    232 			 * never point at garbage.
    233 			 */
    234 			if ((error = bwrite(bp)) != 0)
    235 				goto fail;
    236 		}
    237 		unwindidx = 0;
    238 		allocib = &ip->i_ffs_ib[indirs[0].in_off];
    239 		*allocib = ufs_rw32(nb, needswap);
    240 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
    241 	}
    242 	/*
    243 	 * Fetch through the indirect blocks, allocating as necessary.
    244 	 */
    245 	for (i = 1;;) {
    246 		error = bread(vp,
    247 		    indirs[i].in_lbn, (int)fs->fs_bsize, NOCRED, &bp);
    248 		if (error) {
    249 			brelse(bp);
    250 			goto fail;
    251 		}
    252 		bap = (ufs_daddr_t *)bp->b_data;
    253 		nb = ufs_rw32(bap[indirs[i].in_off], needswap);
    254 		if (i == num)
    255 			break;
    256 		i++;
    257 		if (nb != 0) {
    258 			brelse(bp);
    259 			continue;
    260 		}
    261 		if (pref == 0)
    262 			pref = ffs_blkpref(ip, lbn, 0, (ufs_daddr_t *)0);
    263 		error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred,
    264 		    &newb);
    265 		if (error) {
    266 			brelse(bp);
    267 			goto fail;
    268 		}
    269 		nb = newb;
    270 		*allocblk++ = nb;
    271 		nbp = getblk(vp, indirs[i].in_lbn, fs->fs_bsize, 0, 0);
    272 		nbp->b_blkno = fsbtodb(fs, nb);
    273 		clrbuf(nbp);
    274 		if (DOINGSOFTDEP(vp)) {
    275 			softdep_setup_allocindir_meta(nbp, ip, bp,
    276 			    indirs[i - 1].in_off, nb);
    277 			bdwrite(nbp);
    278 		} else {
    279 			/*
    280 			 * Write synchronously so that indirect blocks
    281 			 * never point at garbage.
    282 			 */
    283 			if ((error = bwrite(nbp)) != 0) {
    284 				brelse(bp);
    285 				goto fail;
    286 			}
    287 		}
    288 		if (unwindidx < 0)
    289 			unwindidx = i - 1;
    290 		bap[indirs[i - 1].in_off] = ufs_rw32(nb, needswap);
    291 		/*
    292 		 * If required, write synchronously, otherwise use
    293 		 * delayed write.
    294 		 */
    295 		if (flags & B_SYNC) {
    296 			bwrite(bp);
    297 		} else {
    298 			bdwrite(bp);
    299 		}
    300 	}
    301 	/*
    302 	 * Get the data block, allocating if necessary.
    303 	 */
    304 	if (nb == 0) {
    305 		pref = ffs_blkpref(ip, lbn, indirs[num].in_off, &bap[0]);
    306 		error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred,
    307 		    &newb);
    308 		if (error) {
    309 			brelse(bp);
    310 			goto fail;
    311 		}
    312 		nb = newb;
    313 		*allocblk++ = nb;
    314 		nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0);
    315 		nbp->b_blkno = fsbtodb(fs, nb);
    316 		if (flags & B_CLRBUF)
    317 			clrbuf(nbp);
    318 		if (DOINGSOFTDEP(vp))
    319 			softdep_setup_allocindir_page(ip, lbn, bp,
    320 			    indirs[num].in_off, nb, 0, nbp);
    321 		bap[indirs[num].in_off] = ufs_rw32(nb, needswap);
    322 		/*
    323 		 * If required, write synchronously, otherwise use
    324 		 * delayed write.
    325 		 */
    326 		if (flags & B_SYNC) {
    327 			bwrite(bp);
    328 		} else {
    329 			bdwrite(bp);
    330 		}
    331 		*ap->a_bpp = nbp;
    332 		return (0);
    333 	}
    334 	brelse(bp);
    335 	if (flags & B_CLRBUF) {
    336 		error = bread(vp, lbn, (int)fs->fs_bsize, NOCRED, &nbp);
    337 		if (error) {
    338 			brelse(nbp);
    339 			goto fail;
    340 		}
    341 	} else {
    342 		nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0);
    343 		nbp->b_blkno = fsbtodb(fs, nb);
    344 	}
    345 	*ap->a_bpp = nbp;
    346 	return (0);
    347 fail:
    348 	/*
    349 	 * If we have failed part way through block allocation, we
    350 	 * have to deallocate any indirect blocks that we have allocated.
    351 	 * We have to fsync the file before we start to get rid of all
    352 	 * of its dependencies so that we do not leave them dangling.
    353 	 * We have to sync it at the end so that the soft updates code
    354 	 * does not find any untracked changes. Although this is really
    355 	 * slow, running out of disk space is not expected to be a common
    356 	 * occurence. The error return from fsync is ignored as we already
    357 	 * have an error to return to the user.
    358 	 */
    359 	(void) VOP_FSYNC(vp, cred, FSYNC_WAIT, curproc);
    360 	for (deallocated = 0, blkp = allociblk; blkp < allocblk; blkp++) {
    361 		ffs_blkfree(ip, *blkp, fs->fs_bsize);
    362 		deallocated += fs->fs_bsize;
    363 	}
    364 	if (unwindidx >= 0) {
    365 		if (unwindidx == 0) {
    366 			*allocib = 0;
    367 		} else {
    368 			int r;
    369 
    370 			r = bread(vp, indirs[unwindidx].in_lbn,
    371 			    (int)fs->fs_bsize, NOCRED, &bp);
    372 			if (r) {
    373 				panic("Could not unwind indirect block, error %d", r);
    374 				brelse(bp);
    375 			} else {
    376 				bap = (ufs_daddr_t *)bp->b_data;
    377 				bap[indirs[unwindidx].in_off] = 0;
    378 				if (flags & B_SYNC)
    379 					bwrite(bp);
    380 				else
    381 					bdwrite(bp);
    382 			}
    383 		}
    384 		for (i = unwindidx + 1; i <= num; i++) {
    385 			bp = getblk(vp, indirs[i].in_lbn, (int)fs->fs_bsize, 0,
    386 			    0);
    387 			bp->b_flags |= B_INVAL;
    388 			brelse(bp);
    389 		}
    390 	}
    391 	if (deallocated) {
    392 #ifdef QUOTA
    393 		/*
    394 		 * Restore user's disk quota because allocation failed.
    395 		 */
    396 		(void)chkdq(ip, (long)-btodb(deallocated), cred, FORCE);
    397 #endif
    398 		ip->i_ffs_blocks -= btodb(deallocated);
    399 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
    400 	}
    401 	(void) VOP_FSYNC(vp, cred, FSYNC_WAIT, curproc);
    402 	return (error);
    403 }
    404