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 1.14 2006/02/17 19:18:08 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 * The vnode must be locked for us to be able to safely mess 96 * around with the inode. 97 */ 98 if (VOP_ISLOCKED(vp, td) != LK_EXCLUSIVE) { 99 panic("ffs_balloc: vnode %p not exclusively locked!", vp); 100 } 101 102 /* 103 * If the next write will extend the file into a new block, 104 * and the file is currently composed of a fragment 105 * this fragment has to be extended to be a full block. 106 */ 107 nb = lblkno(fs, ip->i_size); 108 if (nb < NDADDR && nb < lbn) { 109 /* 110 * The filesize prior to this write can fit in direct 111 * blocks (ex. fragmentation is possibly done) 112 * we are now extending the file write beyond 113 * the block which has end of the file prior to this write. 114 */ 115 osize = blksize(fs, ip, nb); 116 /* 117 * osize gives disk allocated size in the last block. It is 118 * either in fragments or a file system block size. 119 */ 120 if (osize < fs->fs_bsize && osize > 0) { 121 /* A few fragments are already allocated, since the 122 * current extends beyond this block allocated the 123 * complete block as fragments are on in last block. 124 */ 125 error = ffs_realloccg(ip, nb, 126 ffs_blkpref(ip, nb, (int)nb, &ip->i_db[0]), 127 osize, (int)fs->fs_bsize, cred, &bp); 128 if (error) 129 return (error); 130 if (DOINGSOFTDEP(vp)) 131 softdep_setup_allocdirect(ip, nb, 132 dbtofsb(fs, bp->b_bio2.bio_blkno), 133 ip->i_db[nb], fs->fs_bsize, osize, bp); 134 /* adjust the inode size, we just grew */ 135 ip->i_size = smalllblktosize(fs, nb + 1); 136 ip->i_db[nb] = dbtofsb(fs, bp->b_bio2.bio_blkno); 137 ip->i_flag |= IN_CHANGE | IN_UPDATE; 138 if (flags & B_SYNC) 139 bwrite(bp); 140 else 141 bawrite(bp); 142 /* bp is already released here */ 143 } 144 } 145 /* 146 * The first NDADDR blocks are direct blocks 147 */ 148 if (lbn < NDADDR) { 149 nb = ip->i_db[lbn]; 150 if (nb != 0 && ip->i_size >= smalllblktosize(fs, lbn + 1)) { 151 error = bread(vp, lbn, fs->fs_bsize, &bp); 152 if (error) { 153 brelse(bp); 154 return (error); 155 } 156 bp->b_bio2.bio_blkno = fsbtodb(fs, nb); 157 *ap->a_bpp = bp; 158 return (0); 159 } 160 if (nb != 0) { 161 /* 162 * Consider need to reallocate a fragment. 163 */ 164 osize = fragroundup(fs, blkoff(fs, ip->i_size)); 165 nsize = fragroundup(fs, size); 166 if (nsize <= osize) { 167 error = bread(vp, lbn, osize, &bp); 168 if (error) { 169 brelse(bp); 170 return (error); 171 } 172 bp->b_bio2.bio_blkno = fsbtodb(fs, nb); 173 } else { 174 error = ffs_realloccg(ip, lbn, 175 ffs_blkpref(ip, lbn, (int)lbn, 176 &ip->i_db[0]), osize, nsize, cred, &bp); 177 if (error) 178 return (error); 179 if (DOINGSOFTDEP(vp)) 180 softdep_setup_allocdirect(ip, lbn, 181 dbtofsb(fs, bp->b_bio2.bio_blkno), 182 nb, nsize, osize, bp); 183 } 184 } else { 185 if (ip->i_size < smalllblktosize(fs, lbn + 1)) 186 nsize = fragroundup(fs, size); 187 else 188 nsize = fs->fs_bsize; 189 error = ffs_alloc(ip, lbn, 190 ffs_blkpref(ip, lbn, (int)lbn, &ip->i_db[0]), 191 nsize, cred, &newb); 192 if (error) 193 return (error); 194 bp = getblk(vp, lbn, nsize, 0, 0); 195 bp->b_bio2.bio_blkno = fsbtodb(fs, newb); 196 if (flags & B_CLRBUF) 197 vfs_bio_clrbuf(bp); 198 if (DOINGSOFTDEP(vp)) 199 softdep_setup_allocdirect(ip, lbn, newb, 0, 200 nsize, 0, bp); 201 } 202 ip->i_db[lbn] = dbtofsb(fs, bp->b_bio2.bio_blkno); 203 ip->i_flag |= IN_CHANGE | IN_UPDATE; 204 *ap->a_bpp = bp; 205 return (0); 206 } 207 /* 208 * Determine the number of levels of indirection. 209 */ 210 pref = 0; 211 if ((error = ufs_getlbns(vp, lbn, indirs, &num)) != 0) 212 return(error); 213 #ifdef DIAGNOSTIC 214 if (num < 1) 215 panic ("ffs_balloc: ufs_bmaparray returned indirect block"); 216 #endif 217 /* 218 * Get a handle on the data block buffer before working through 219 * indirect blocks to avoid a deadlock between the VM system holding 220 * a locked VM page and issuing a BMAP (which tries to lock the 221 * indirect blocks), and the filesystem holding a locked indirect 222 * block and then trying to read a data block (which tries to lock 223 * the underlying VM pages). 224 */ 225 dbp = getblk(vp, lbn, fs->fs_bsize, 0, 0); 226 227 /* 228 * Setup undo history 229 */ 230 allocib = NULL; 231 allocblk = allociblk; 232 unwindidx = -1; 233 234 /* 235 * Fetch the first indirect block directly from the inode, allocating 236 * one if necessary. 237 */ 238 --num; 239 nb = ip->i_ib[indirs[0].in_off]; 240 if (nb == 0) { 241 pref = ffs_blkpref(ip, lbn, 0, (ufs_daddr_t *)0); 242 /* 243 * If the filesystem has run out of space we can skip the 244 * full fsync/undo of the main [fail] case since no undo 245 * history has been built yet. Hence the goto fail2. 246 */ 247 if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, 248 cred, &newb)) != 0) 249 goto fail2; 250 nb = newb; 251 *allocblk++ = nb; 252 bp = getblk(vp, indirs[1].in_lbn, fs->fs_bsize, 0, 0); 253 bp->b_bio2.bio_blkno = fsbtodb(fs, nb); 254 vfs_bio_clrbuf(bp); 255 if (DOINGSOFTDEP(vp)) { 256 softdep_setup_allocdirect(ip, NDADDR + indirs[0].in_off, 257 newb, 0, fs->fs_bsize, 0, bp); 258 bdwrite(bp); 259 } else { 260 /* 261 * Write synchronously so that indirect blocks 262 * never point at garbage. 263 */ 264 if (DOINGASYNC(vp)) 265 bdwrite(bp); 266 else if ((error = bwrite(bp)) != 0) 267 goto fail; 268 } 269 allocib = &ip->i_ib[indirs[0].in_off]; 270 *allocib = nb; 271 ip->i_flag |= IN_CHANGE | IN_UPDATE; 272 } 273 274 /* 275 * Fetch through the indirect blocks, allocating as necessary. 276 */ 277 for (i = 1;;) { 278 error = bread(vp, indirs[i].in_lbn, (int)fs->fs_bsize, &bp); 279 if (error) { 280 brelse(bp); 281 goto fail; 282 } 283 bap = (ufs_daddr_t *)bp->b_data; 284 nb = bap[indirs[i].in_off]; 285 if (i == num) 286 break; 287 i += 1; 288 if (nb != 0) { 289 bqrelse(bp); 290 continue; 291 } 292 if (pref == 0) 293 pref = ffs_blkpref(ip, lbn, 0, (ufs_daddr_t *)0); 294 if ((error = 295 ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred, &newb)) != 0) { 296 brelse(bp); 297 goto fail; 298 } 299 nb = newb; 300 *allocblk++ = nb; 301 nbp = getblk(vp, indirs[i].in_lbn, fs->fs_bsize, 0, 0); 302 nbp->b_bio2.bio_blkno = fsbtodb(fs, nb); 303 vfs_bio_clrbuf(nbp); 304 if (DOINGSOFTDEP(vp)) { 305 softdep_setup_allocindir_meta(nbp, ip, bp, 306 indirs[i - 1].in_off, nb); 307 bdwrite(nbp); 308 } else { 309 /* 310 * Write synchronously so that indirect blocks 311 * never point at garbage. 312 */ 313 if ((error = bwrite(nbp)) != 0) { 314 brelse(bp); 315 goto fail; 316 } 317 } 318 bap[indirs[i - 1].in_off] = nb; 319 if (allocib == NULL && unwindidx < 0) 320 unwindidx = i - 1; 321 /* 322 * If required, write synchronously, otherwise use 323 * delayed write. 324 */ 325 if (flags & B_SYNC) { 326 bwrite(bp); 327 } else { 328 if (bp->b_bufsize == fs->fs_bsize) 329 bp->b_flags |= B_CLUSTEROK; 330 bdwrite(bp); 331 } 332 } 333 334 /* 335 * Get the data block, allocating if necessary. We have already 336 * called getblk() on the data block buffer, dbp. If we have to 337 * allocate it and B_CLRBUF has been set the inference is an intention 338 * to zero out the related disk blocks, so we do not have to issue 339 * a read. Instead we simply call vfs_bio_clrbuf(). If B_CLRBUF is 340 * not set the caller intends to overwrite the entire contents of the 341 * buffer and we don't waste time trying to clean up the contents. 342 * 343 * bp references the current indirect block. When allocating, 344 * the block must be updated. 345 */ 346 if (nb == 0) { 347 pref = ffs_blkpref(ip, lbn, indirs[i].in_off, &bap[0]); 348 error = ffs_alloc(ip, 349 lbn, pref, (int)fs->fs_bsize, cred, &newb); 350 if (error) { 351 brelse(bp); 352 goto fail; 353 } 354 nb = newb; 355 *allocblk++ = nb; 356 dbp->b_bio2.bio_blkno = fsbtodb(fs, nb); 357 if (flags & B_CLRBUF) 358 vfs_bio_clrbuf(dbp); 359 if (DOINGSOFTDEP(vp)) 360 softdep_setup_allocindir_page(ip, lbn, bp, 361 indirs[i].in_off, nb, 0, dbp); 362 bap[indirs[i].in_off] = nb; 363 /* 364 * If required, write synchronously, otherwise use 365 * delayed write. 366 */ 367 if (flags & B_SYNC) { 368 bwrite(bp); 369 } else { 370 if (bp->b_bufsize == fs->fs_bsize) 371 bp->b_flags |= B_CLUSTEROK; 372 bdwrite(bp); 373 } 374 *ap->a_bpp = dbp; 375 return (0); 376 } 377 brelse(bp); 378 379 /* 380 * At this point all related indirect blocks have been allocated 381 * if necessary and released. bp is no longer valid. dbp holds 382 * our getblk()'d data block. 383 * 384 * XXX we previously performed a cluster_read operation here. 385 */ 386 if (flags & B_CLRBUF) { 387 /* 388 * If B_CLRBUF is set we must validate the invalid portions 389 * of the buffer. This typically requires a read-before- 390 * write. The strategy call will fill in bio_blkno in that 391 * case. 392 * 393 * If we hit this case we do a cluster read if possible 394 * since nearby data blocks are likely to be accessed soon 395 * too. 396 */ 397 if ((dbp->b_flags & B_CACHE) == 0) { 398 bqrelse(dbp); 399 seqcount = (flags & B_SEQMASK) >> B_SEQSHIFT; 400 if (seqcount && 401 (vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) { 402 error = cluster_read(vp, ip->i_size, lbn, 403 (int)fs->fs_bsize, 404 MAXBSIZE, seqcount, &dbp); 405 } else { 406 error = bread(vp, lbn, (int)fs->fs_bsize, &dbp); 407 } 408 if (error) 409 goto fail; 410 } else { 411 dbp->b_bio2.bio_blkno = fsbtodb(fs, nb); 412 } 413 } else { 414 /* 415 * If B_CLRBUF is not set the caller intends to overwrite 416 * the entire contents of the buffer. We can simply set 417 * bio_blkno and we are done. 418 */ 419 dbp->b_bio2.bio_blkno = fsbtodb(fs, nb); 420 } 421 *ap->a_bpp = dbp; 422 return (0); 423 fail: 424 /* 425 * If we have failed part way through block allocation, we 426 * have to deallocate any indirect blocks that we have allocated. 427 * We have to fsync the file before we start to get rid of all 428 * of its dependencies so that we do not leave them dangling. 429 * We have to sync it at the end so that the soft updates code 430 * does not find any untracked changes. Although this is really 431 * slow, running out of disk space is not expected to be a common 432 * occurence. The error return from fsync is ignored as we already 433 * have an error to return to the user. 434 */ 435 (void) VOP_FSYNC(vp, MNT_WAIT, td); 436 for (deallocated = 0, blkp = allociblk; blkp < allocblk; blkp++) { 437 ffs_blkfree(ip, *blkp, fs->fs_bsize); 438 deallocated += fs->fs_bsize; 439 } 440 if (allocib != NULL) { 441 *allocib = 0; 442 } else if (unwindidx >= 0) { 443 int r; 444 445 r = bread(vp, indirs[unwindidx].in_lbn, (int)fs->fs_bsize, &bp); 446 if (r) { 447 panic("Could not unwind indirect block, error %d", r); 448 brelse(bp); 449 } else { 450 bap = (ufs_daddr_t *)bp->b_data; 451 bap[indirs[unwindidx].in_off] = 0; 452 if (flags & B_SYNC) { 453 bwrite(bp); 454 } else { 455 if (bp->b_bufsize == fs->fs_bsize) 456 bp->b_flags |= B_CLUSTEROK; 457 bdwrite(bp); 458 } 459 } 460 } 461 if (deallocated) { 462 #ifdef QUOTA 463 /* 464 * Restore user's disk quota because allocation failed. 465 */ 466 (void) chkdq(ip, (long)-btodb(deallocated), cred, FORCE); 467 #endif 468 ip->i_blocks -= btodb(deallocated); 469 ip->i_flag |= IN_CHANGE | IN_UPDATE; 470 } 471 (void) VOP_FSYNC(vp, MNT_WAIT, td); 472 473 /* 474 * Cleanup the data block we getblk()'d before returning. 475 */ 476 fail2: 477 brelse(dbp); 478 return (error); 479 } 480 481