1 /* 2 * Copyright (c) 1982, 1986, 1989, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 3. All advertising materials mentioning features or use of this software 14 * must display the following acknowledgement: 15 * This product includes software developed by the University of 16 * California, Berkeley and its contributors. 17 * 4. Neither the name of the University nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 * @(#)ffs_balloc.c 8.8 (Berkeley) 6/16/95 34 * $FreeBSD: src/sys/ufs/ffs/ffs_balloc.c,v 1.26.2.1 2002/10/10 19:48:20 dillon Exp $ 35 * $DragonFly: src/sys/vfs/ufs/ffs_balloc.c,v 1.12 2005/08/28 23:23:10 dillon Exp $ 36 */ 37 38 #include <sys/param.h> 39 #include <sys/systm.h> 40 #include <sys/proc.h> 41 #include <sys/buf.h> 42 #include <sys/lock.h> 43 #include <sys/mount.h> 44 #include <sys/vnode.h> 45 46 #include "quota.h" 47 #include "inode.h" 48 #include "ufs_extern.h" 49 50 #include "fs.h" 51 #include "ffs_extern.h" 52 53 /* 54 * Balloc defines the structure of filesystem storage 55 * by allocating the physical blocks on a device given 56 * the inode and the logical block number in a file. 57 * 58 * ffs_balloc(struct vnode *a_vp, ufs_daddr_t a_lbn, int a_size, 59 * struct ucred *a_cred, int a_flags, struct buf *a_bpp) 60 */ 61 int 62 ffs_balloc(struct vop_balloc_args *ap) 63 { 64 struct inode *ip; 65 ufs_daddr_t lbn; 66 int size; 67 struct ucred *cred; 68 int flags; 69 struct fs *fs; 70 ufs_daddr_t nb; 71 struct buf *bp, *nbp, *dbp; 72 struct vnode *vp; 73 struct indir indirs[NIADDR + 2]; 74 ufs_daddr_t newb, *bap, pref; 75 int deallocated, osize, nsize, num, i, error; 76 ufs_daddr_t *allocib, *blkp, *allocblk, allociblk[NIADDR + 1]; 77 struct thread *td = curthread; /* XXX */ 78 int unwindidx; 79 int seqcount; 80 81 vp = ap->a_vp; 82 ip = VTOI(vp); 83 fs = ip->i_fs; 84 lbn = lblkno(fs, ap->a_startoffset); 85 size = blkoff(fs, ap->a_startoffset) + ap->a_size; 86 if (size > fs->fs_bsize) 87 panic("ffs_balloc: blk too big"); 88 *ap->a_bpp = NULL; 89 if (lbn < 0) 90 return (EFBIG); 91 cred = ap->a_cred; 92 flags = ap->a_flags; 93 94 /* 95 * If the next write will extend the file into a new block, 96 * and the file is currently composed of a fragment 97 * this fragment has to be extended to be a full block. 98 */ 99 nb = lblkno(fs, ip->i_size); 100 if (nb < NDADDR && nb < lbn) { 101 /* 102 * The filesize prior to this write can fit in direct 103 * blocks (ex. fragmentation is possibly done) 104 * we are now extending the file write beyond 105 * the block which has end of the file prior to this write. 106 */ 107 osize = blksize(fs, ip, nb); 108 /* 109 * osize gives disk allocated size in the last block. It is 110 * either in fragments or a file system block size. 111 */ 112 if (osize < fs->fs_bsize && osize > 0) { 113 /* A few fragments are already allocated, since the 114 * current extends beyond this block allocated the 115 * complete block as fragments are on in last block. 116 */ 117 error = ffs_realloccg(ip, nb, 118 ffs_blkpref(ip, nb, (int)nb, &ip->i_db[0]), 119 osize, (int)fs->fs_bsize, cred, &bp); 120 if (error) 121 return (error); 122 if (DOINGSOFTDEP(vp)) 123 softdep_setup_allocdirect(ip, nb, 124 dbtofsb(fs, bp->b_blkno), ip->i_db[nb], 125 fs->fs_bsize, osize, bp); 126 /* adjust the inode size, we just grew */ 127 ip->i_size = smalllblktosize(fs, nb + 1); 128 ip->i_db[nb] = dbtofsb(fs, bp->b_blkno); 129 ip->i_flag |= IN_CHANGE | IN_UPDATE; 130 if (flags & B_SYNC) 131 bwrite(bp); 132 else 133 bawrite(bp); 134 /* bp is already released here */ 135 } 136 } 137 /* 138 * The first NDADDR blocks are direct blocks 139 */ 140 if (lbn < NDADDR) { 141 nb = ip->i_db[lbn]; 142 if (nb != 0 && ip->i_size >= smalllblktosize(fs, lbn + 1)) { 143 error = bread(vp, lbn, fs->fs_bsize, &bp); 144 if (error) { 145 brelse(bp); 146 return (error); 147 } 148 bp->b_blkno = fsbtodb(fs, nb); 149 *ap->a_bpp = bp; 150 return (0); 151 } 152 if (nb != 0) { 153 /* 154 * Consider need to reallocate a fragment. 155 */ 156 osize = fragroundup(fs, blkoff(fs, ip->i_size)); 157 nsize = fragroundup(fs, size); 158 if (nsize <= osize) { 159 error = bread(vp, lbn, osize, &bp); 160 if (error) { 161 brelse(bp); 162 return (error); 163 } 164 bp->b_blkno = fsbtodb(fs, nb); 165 } else { 166 error = ffs_realloccg(ip, lbn, 167 ffs_blkpref(ip, lbn, (int)lbn, 168 &ip->i_db[0]), osize, nsize, cred, &bp); 169 if (error) 170 return (error); 171 if (DOINGSOFTDEP(vp)) 172 softdep_setup_allocdirect(ip, lbn, 173 dbtofsb(fs, bp->b_blkno), nb, 174 nsize, osize, bp); 175 } 176 } else { 177 if (ip->i_size < smalllblktosize(fs, lbn + 1)) 178 nsize = fragroundup(fs, size); 179 else 180 nsize = fs->fs_bsize; 181 error = ffs_alloc(ip, lbn, 182 ffs_blkpref(ip, lbn, (int)lbn, &ip->i_db[0]), 183 nsize, cred, &newb); 184 if (error) 185 return (error); 186 bp = getblk(vp, lbn, nsize, 0, 0); 187 bp->b_blkno = fsbtodb(fs, newb); 188 if (flags & B_CLRBUF) 189 vfs_bio_clrbuf(bp); 190 if (DOINGSOFTDEP(vp)) 191 softdep_setup_allocdirect(ip, lbn, newb, 0, 192 nsize, 0, bp); 193 } 194 ip->i_db[lbn] = dbtofsb(fs, bp->b_blkno); 195 ip->i_flag |= IN_CHANGE | IN_UPDATE; 196 *ap->a_bpp = bp; 197 return (0); 198 } 199 /* 200 * Determine the number of levels of indirection. 201 */ 202 pref = 0; 203 if ((error = ufs_getlbns(vp, lbn, indirs, &num)) != 0) 204 return(error); 205 #ifdef DIAGNOSTIC 206 if (num < 1) 207 panic ("ffs_balloc: ufs_bmaparray returned indirect block"); 208 #endif 209 /* 210 * Get a handle on the data block buffer before working through 211 * indirect blocks to avoid a deadlock between the VM system holding 212 * a locked VM page and issuing a BMAP (which tries to lock the 213 * indirect blocks), and the filesystem holding a locked indirect 214 * block and then trying to read a data block (which tries to lock 215 * the underlying VM pages). 216 */ 217 dbp = getblk(vp, lbn, fs->fs_bsize, 0, 0); 218 219 /* 220 * Setup undo history 221 */ 222 allocib = NULL; 223 allocblk = allociblk; 224 unwindidx = -1; 225 226 /* 227 * Fetch the first indirect block directly from the inode, allocating 228 * one if necessary. 229 */ 230 --num; 231 nb = ip->i_ib[indirs[0].in_off]; 232 if (nb == 0) { 233 pref = ffs_blkpref(ip, lbn, 0, (ufs_daddr_t *)0); 234 /* 235 * If the filesystem has run out of space we can skip the 236 * full fsync/undo of the main [fail] case since no undo 237 * history has been built yet. Hence the goto fail2. 238 */ 239 if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, 240 cred, &newb)) != 0) 241 goto fail2; 242 nb = newb; 243 *allocblk++ = nb; 244 bp = getblk(vp, indirs[1].in_lbn, fs->fs_bsize, 0, 0); 245 bp->b_blkno = fsbtodb(fs, nb); 246 vfs_bio_clrbuf(bp); 247 if (DOINGSOFTDEP(vp)) { 248 softdep_setup_allocdirect(ip, NDADDR + indirs[0].in_off, 249 newb, 0, fs->fs_bsize, 0, bp); 250 bdwrite(bp); 251 } else { 252 /* 253 * Write synchronously so that indirect blocks 254 * never point at garbage. 255 */ 256 if (DOINGASYNC(vp)) 257 bdwrite(bp); 258 else if ((error = bwrite(bp)) != 0) 259 goto fail; 260 } 261 allocib = &ip->i_ib[indirs[0].in_off]; 262 *allocib = nb; 263 ip->i_flag |= IN_CHANGE | IN_UPDATE; 264 } 265 266 /* 267 * Fetch through the indirect blocks, allocating as necessary. 268 */ 269 for (i = 1;;) { 270 error = bread(vp, indirs[i].in_lbn, (int)fs->fs_bsize, &bp); 271 if (error) { 272 brelse(bp); 273 goto fail; 274 } 275 bap = (ufs_daddr_t *)bp->b_data; 276 nb = bap[indirs[i].in_off]; 277 if (i == num) 278 break; 279 i += 1; 280 if (nb != 0) { 281 bqrelse(bp); 282 continue; 283 } 284 if (pref == 0) 285 pref = ffs_blkpref(ip, lbn, 0, (ufs_daddr_t *)0); 286 if ((error = 287 ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred, &newb)) != 0) { 288 brelse(bp); 289 goto fail; 290 } 291 nb = newb; 292 *allocblk++ = nb; 293 nbp = getblk(vp, indirs[i].in_lbn, fs->fs_bsize, 0, 0); 294 nbp->b_blkno = fsbtodb(fs, nb); 295 vfs_bio_clrbuf(nbp); 296 if (DOINGSOFTDEP(vp)) { 297 softdep_setup_allocindir_meta(nbp, ip, bp, 298 indirs[i - 1].in_off, nb); 299 bdwrite(nbp); 300 } else { 301 /* 302 * Write synchronously so that indirect blocks 303 * never point at garbage. 304 */ 305 if ((error = bwrite(nbp)) != 0) { 306 brelse(bp); 307 goto fail; 308 } 309 } 310 bap[indirs[i - 1].in_off] = nb; 311 if (allocib == NULL && unwindidx < 0) 312 unwindidx = i - 1; 313 /* 314 * If required, write synchronously, otherwise use 315 * delayed write. 316 */ 317 if (flags & B_SYNC) { 318 bwrite(bp); 319 } else { 320 if (bp->b_bufsize == fs->fs_bsize) 321 bp->b_flags |= B_CLUSTEROK; 322 bdwrite(bp); 323 } 324 } 325 326 /* 327 * Get the data block, allocating if necessary. We have already 328 * called getblk() on the data block buffer, dbp. If we have to 329 * allocate it and B_CLRBUF has been set the inference is an intention 330 * to zero out the related disk blocks, so we do not have to issue 331 * a read. Instead we simply call vfs_bio_clrbuf(). If B_CLRBUF is 332 * not set the caller intends to overwrite the entire contents of the 333 * buffer and we don't waste time trying to clean up the contents. 334 * 335 * bp references the current indirect block. When allocating, 336 * the block must be updated. 337 */ 338 if (nb == 0) { 339 pref = ffs_blkpref(ip, lbn, indirs[i].in_off, &bap[0]); 340 error = ffs_alloc(ip, 341 lbn, pref, (int)fs->fs_bsize, cred, &newb); 342 if (error) { 343 brelse(bp); 344 goto fail; 345 } 346 nb = newb; 347 *allocblk++ = nb; 348 dbp->b_blkno = fsbtodb(fs, nb); 349 if (flags & B_CLRBUF) 350 vfs_bio_clrbuf(dbp); 351 if (DOINGSOFTDEP(vp)) 352 softdep_setup_allocindir_page(ip, lbn, bp, 353 indirs[i].in_off, nb, 0, dbp); 354 bap[indirs[i].in_off] = nb; 355 /* 356 * If required, write synchronously, otherwise use 357 * delayed write. 358 */ 359 if (flags & B_SYNC) { 360 bwrite(bp); 361 } else { 362 if (bp->b_bufsize == fs->fs_bsize) 363 bp->b_flags |= B_CLUSTEROK; 364 bdwrite(bp); 365 } 366 *ap->a_bpp = dbp; 367 return (0); 368 } 369 brelse(bp); 370 371 /* 372 * At this point all related indirect blocks have been allocated 373 * if necessary and released. bp is no longer valid. dbp holds 374 * our getblk()'d data block. 375 * 376 * XXX we previously performed a cluster_read operation here. 377 */ 378 if (flags & B_CLRBUF) { 379 /* 380 * If B_CLRBUF is set we must validate the invalid portions 381 * of the buffer. This typically requires a read-before- 382 * write. The strategy call will fill in b_blkno in that 383 * case. 384 * 385 * If we hit this case we do a cluster read if possible 386 * since nearby data blocks are likely to be accessed soon 387 * too. 388 */ 389 if ((dbp->b_flags & B_CACHE) == 0) { 390 bqrelse(dbp); 391 seqcount = (flags & B_SEQMASK) >> B_SEQSHIFT; 392 if (seqcount && 393 (vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) { 394 error = cluster_read(vp, ip->i_size, lbn, 395 (int)fs->fs_bsize, 396 MAXBSIZE, seqcount, &dbp); 397 } else { 398 error = bread(vp, lbn, (int)fs->fs_bsize, &dbp); 399 } 400 if (error) 401 goto fail; 402 } else { 403 dbp->b_blkno = fsbtodb(fs, nb); 404 } 405 } else { 406 /* 407 * If B_CLRBUF is not set the caller intends to overwrite 408 * the entire contents of the buffer. We can simply set 409 * b_blkno and we are done. 410 */ 411 dbp->b_blkno = fsbtodb(fs, nb); 412 } 413 *ap->a_bpp = dbp; 414 return (0); 415 fail: 416 /* 417 * If we have failed part way through block allocation, we 418 * have to deallocate any indirect blocks that we have allocated. 419 * We have to fsync the file before we start to get rid of all 420 * of its dependencies so that we do not leave them dangling. 421 * We have to sync it at the end so that the soft updates code 422 * does not find any untracked changes. Although this is really 423 * slow, running out of disk space is not expected to be a common 424 * occurence. The error return from fsync is ignored as we already 425 * have an error to return to the user. 426 */ 427 (void) VOP_FSYNC(vp, MNT_WAIT, td); 428 for (deallocated = 0, blkp = allociblk; blkp < allocblk; blkp++) { 429 ffs_blkfree(ip, *blkp, fs->fs_bsize); 430 deallocated += fs->fs_bsize; 431 } 432 if (allocib != NULL) { 433 *allocib = 0; 434 } else if (unwindidx >= 0) { 435 int r; 436 437 r = bread(vp, indirs[unwindidx].in_lbn, (int)fs->fs_bsize, &bp); 438 if (r) { 439 panic("Could not unwind indirect block, error %d", r); 440 brelse(bp); 441 } else { 442 bap = (ufs_daddr_t *)bp->b_data; 443 bap[indirs[unwindidx].in_off] = 0; 444 if (flags & B_SYNC) { 445 bwrite(bp); 446 } else { 447 if (bp->b_bufsize == fs->fs_bsize) 448 bp->b_flags |= B_CLUSTEROK; 449 bdwrite(bp); 450 } 451 } 452 } 453 if (deallocated) { 454 #ifdef QUOTA 455 /* 456 * Restore user's disk quota because allocation failed. 457 */ 458 (void) chkdq(ip, (long)-btodb(deallocated), cred, FORCE); 459 #endif 460 ip->i_blocks -= btodb(deallocated); 461 ip->i_flag |= IN_CHANGE | IN_UPDATE; 462 } 463 (void) VOP_FSYNC(vp, MNT_WAIT, td); 464 465 /* 466 * Cleanup the data block we getblk()'d before returning. 467 */ 468 fail2: 469 brelse(dbp); 470 return (error); 471 } 472 473