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 1.17 2006/05/06 02:43:14 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 dofftofsb(fs, bp->b_bio2.bio_offset), 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] = dofftofsb(fs, bp->b_bio2.bio_offset); 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, lblktodoff(fs, lbn), fs->fs_bsize, &bp); 152 if (error) { 153 brelse(bp); 154 return (error); 155 } 156 bp->b_bio2.bio_offset = fsbtodoff(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, lblktodoff(fs, lbn), 168 osize, &bp); 169 if (error) { 170 brelse(bp); 171 return (error); 172 } 173 bp->b_bio2.bio_offset = fsbtodoff(fs, nb); 174 } else { 175 error = ffs_realloccg(ip, lbn, 176 ffs_blkpref(ip, lbn, (int)lbn, 177 &ip->i_db[0]), osize, nsize, cred, &bp); 178 if (error) 179 return (error); 180 if (DOINGSOFTDEP(vp)) 181 softdep_setup_allocdirect(ip, lbn, 182 dofftofsb(fs, bp->b_bio2.bio_offset), 183 nb, nsize, osize, bp); 184 } 185 } else { 186 if (ip->i_size < smalllblktosize(fs, lbn + 1)) 187 nsize = fragroundup(fs, size); 188 else 189 nsize = fs->fs_bsize; 190 error = ffs_alloc(ip, lbn, 191 ffs_blkpref(ip, lbn, (int)lbn, &ip->i_db[0]), 192 nsize, cred, &newb); 193 if (error) 194 return (error); 195 bp = getblk(vp, lblktodoff(fs, lbn), nsize, 0, 0); 196 bp->b_bio2.bio_offset = fsbtodoff(fs, newb); 197 if (flags & B_CLRBUF) 198 vfs_bio_clrbuf(bp); 199 if (DOINGSOFTDEP(vp)) 200 softdep_setup_allocdirect(ip, lbn, newb, 0, 201 nsize, 0, bp); 202 } 203 ip->i_db[lbn] = dofftofsb(fs, bp->b_bio2.bio_offset); 204 ip->i_flag |= IN_CHANGE | IN_UPDATE; 205 *ap->a_bpp = bp; 206 return (0); 207 } 208 /* 209 * Determine the number of levels of indirection. 210 */ 211 pref = 0; 212 if ((error = ufs_getlbns(vp, lbn, indirs, &num)) != 0) 213 return(error); 214 #ifdef DIAGNOSTIC 215 if (num < 1) 216 panic ("ffs_balloc: ufs_bmaparray returned indirect block"); 217 #endif 218 /* 219 * Get a handle on the data block buffer before working through 220 * indirect blocks to avoid a deadlock between the VM system holding 221 * a locked VM page and issuing a BMAP (which tries to lock the 222 * indirect blocks), and the filesystem holding a locked indirect 223 * block and then trying to read a data block (which tries to lock 224 * the underlying VM pages). 225 */ 226 dbp = getblk(vp, lblktodoff(fs, lbn), fs->fs_bsize, 0, 0); 227 228 /* 229 * Setup undo history 230 */ 231 allocib = NULL; 232 allocblk = allociblk; 233 unwindidx = -1; 234 235 /* 236 * Fetch the first indirect block directly from the inode, allocating 237 * one if necessary. 238 */ 239 --num; 240 nb = ip->i_ib[indirs[0].in_off]; 241 if (nb == 0) { 242 pref = ffs_blkpref(ip, lbn, 0, (ufs_daddr_t *)0); 243 /* 244 * If the filesystem has run out of space we can skip the 245 * full fsync/undo of the main [fail] case since no undo 246 * history has been built yet. Hence the goto fail2. 247 */ 248 if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, 249 cred, &newb)) != 0) 250 goto fail2; 251 nb = newb; 252 *allocblk++ = nb; 253 bp = getblk(vp, lblktodoff(fs, indirs[1].in_lbn), 254 fs->fs_bsize, 0, 0); 255 bp->b_bio2.bio_offset = fsbtodoff(fs, nb); 256 vfs_bio_clrbuf(bp); 257 if (DOINGSOFTDEP(vp)) { 258 softdep_setup_allocdirect(ip, NDADDR + indirs[0].in_off, 259 newb, 0, fs->fs_bsize, 0, bp); 260 bdwrite(bp); 261 } else { 262 /* 263 * Write synchronously so that indirect blocks 264 * never point at garbage. 265 */ 266 if (DOINGASYNC(vp)) 267 bdwrite(bp); 268 else if ((error = bwrite(bp)) != 0) 269 goto fail; 270 } 271 allocib = &ip->i_ib[indirs[0].in_off]; 272 *allocib = nb; 273 ip->i_flag |= IN_CHANGE | IN_UPDATE; 274 } 275 276 /* 277 * Fetch through the indirect blocks, allocating as necessary. 278 */ 279 for (i = 1;;) { 280 error = bread(vp, lblktodoff(fs, indirs[i].in_lbn), (int)fs->fs_bsize, &bp); 281 if (error) { 282 brelse(bp); 283 goto fail; 284 } 285 bap = (ufs_daddr_t *)bp->b_data; 286 nb = bap[indirs[i].in_off]; 287 if (i == num) 288 break; 289 i += 1; 290 if (nb != 0) { 291 bqrelse(bp); 292 continue; 293 } 294 if (pref == 0) 295 pref = ffs_blkpref(ip, lbn, 0, (ufs_daddr_t *)0); 296 if ((error = 297 ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred, &newb)) != 0) { 298 brelse(bp); 299 goto fail; 300 } 301 nb = newb; 302 *allocblk++ = nb; 303 nbp = getblk(vp, lblktodoff(fs, indirs[i].in_lbn), 304 fs->fs_bsize, 0, 0); 305 nbp->b_bio2.bio_offset = fsbtodoff(fs, nb); 306 vfs_bio_clrbuf(nbp); 307 if (DOINGSOFTDEP(vp)) { 308 softdep_setup_allocindir_meta(nbp, ip, bp, 309 indirs[i - 1].in_off, nb); 310 bdwrite(nbp); 311 } else { 312 /* 313 * Write synchronously so that indirect blocks 314 * never point at garbage. 315 */ 316 if ((error = bwrite(nbp)) != 0) { 317 brelse(bp); 318 goto fail; 319 } 320 } 321 bap[indirs[i - 1].in_off] = nb; 322 if (allocib == NULL && unwindidx < 0) 323 unwindidx = i - 1; 324 /* 325 * If required, write synchronously, otherwise use 326 * delayed write. 327 */ 328 if (flags & B_SYNC) { 329 bwrite(bp); 330 } else { 331 if (bp->b_bufsize == fs->fs_bsize) 332 bp->b_flags |= B_CLUSTEROK; 333 bdwrite(bp); 334 } 335 } 336 337 /* 338 * Get the data block, allocating if necessary. We have already 339 * called getblk() on the data block buffer, dbp. If we have to 340 * allocate it and B_CLRBUF has been set the inference is an intention 341 * to zero out the related disk blocks, so we do not have to issue 342 * a read. Instead we simply call vfs_bio_clrbuf(). If B_CLRBUF is 343 * not set the caller intends to overwrite the entire contents of the 344 * buffer and we don't waste time trying to clean up the contents. 345 * 346 * bp references the current indirect block. When allocating, 347 * the block must be updated. 348 */ 349 if (nb == 0) { 350 pref = ffs_blkpref(ip, lbn, indirs[i].in_off, &bap[0]); 351 error = ffs_alloc(ip, 352 lbn, pref, (int)fs->fs_bsize, cred, &newb); 353 if (error) { 354 brelse(bp); 355 goto fail; 356 } 357 nb = newb; 358 *allocblk++ = nb; 359 dbp->b_bio2.bio_offset = fsbtodoff(fs, nb); 360 if (flags & B_CLRBUF) 361 vfs_bio_clrbuf(dbp); 362 if (DOINGSOFTDEP(vp)) 363 softdep_setup_allocindir_page(ip, lbn, bp, 364 indirs[i].in_off, nb, 0, dbp); 365 bap[indirs[i].in_off] = nb; 366 /* 367 * If required, write synchronously, otherwise use 368 * delayed write. 369 */ 370 if (flags & B_SYNC) { 371 bwrite(bp); 372 } else { 373 if (bp->b_bufsize == fs->fs_bsize) 374 bp->b_flags |= B_CLUSTEROK; 375 bdwrite(bp); 376 } 377 *ap->a_bpp = dbp; 378 return (0); 379 } 380 brelse(bp); 381 382 /* 383 * At this point all related indirect blocks have been allocated 384 * if necessary and released. bp is no longer valid. dbp holds 385 * our getblk()'d data block. 386 * 387 * XXX we previously performed a cluster_read operation here. 388 */ 389 if (flags & B_CLRBUF) { 390 /* 391 * If B_CLRBUF is set we must validate the invalid portions 392 * of the buffer. This typically requires a read-before- 393 * write. The strategy call will fill in bio_offset in that 394 * case. 395 * 396 * If we hit this case we do a cluster read if possible 397 * since nearby data blocks are likely to be accessed soon 398 * too. 399 */ 400 if ((dbp->b_flags & B_CACHE) == 0) { 401 bqrelse(dbp); 402 seqcount = (flags & B_SEQMASK) >> B_SEQSHIFT; 403 if (seqcount && 404 (vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) { 405 error = cluster_read(vp, (off_t)ip->i_size, 406 lblktodoff(fs, lbn), 407 (int)fs->fs_bsize, 408 MAXBSIZE, seqcount, &dbp); 409 } else { 410 error = bread(vp, lblktodoff(fs, lbn), (int)fs->fs_bsize, &dbp); 411 } 412 if (error) 413 goto fail; 414 } else { 415 dbp->b_bio2.bio_offset = fsbtodoff(fs, nb); 416 } 417 } else { 418 /* 419 * If B_CLRBUF is not set the caller intends to overwrite 420 * the entire contents of the buffer. We can simply set 421 * bio_offset and we are done. 422 */ 423 dbp->b_bio2.bio_offset = fsbtodoff(fs, nb); 424 } 425 *ap->a_bpp = dbp; 426 return (0); 427 fail: 428 /* 429 * If we have failed part way through block allocation, we 430 * have to deallocate any indirect blocks that we have allocated. 431 * We have to fsync the file before we start to get rid of all 432 * of its dependencies so that we do not leave them dangling. 433 * We have to sync it at the end so that the soft updates code 434 * does not find any untracked changes. Although this is really 435 * slow, running out of disk space is not expected to be a common 436 * occurence. The error return from fsync is ignored as we already 437 * have an error to return to the user. 438 */ 439 (void) VOP_FSYNC(vp, MNT_WAIT); 440 for (deallocated = 0, blkp = allociblk; blkp < allocblk; blkp++) { 441 ffs_blkfree(ip, *blkp, fs->fs_bsize); 442 deallocated += fs->fs_bsize; 443 } 444 if (allocib != NULL) { 445 *allocib = 0; 446 } else if (unwindidx >= 0) { 447 int r; 448 449 r = bread(vp, lblktodoff(fs, indirs[unwindidx].in_lbn), (int)fs->fs_bsize, &bp); 450 if (r) { 451 panic("Could not unwind indirect block, error %d", r); 452 brelse(bp); 453 } else { 454 bap = (ufs_daddr_t *)bp->b_data; 455 bap[indirs[unwindidx].in_off] = 0; 456 if (flags & B_SYNC) { 457 bwrite(bp); 458 } else { 459 if (bp->b_bufsize == fs->fs_bsize) 460 bp->b_flags |= B_CLUSTEROK; 461 bdwrite(bp); 462 } 463 } 464 } 465 if (deallocated) { 466 #ifdef QUOTA 467 /* 468 * Restore user's disk quota because allocation failed. 469 */ 470 (void) ufs_chkdq(ip, (long)-btodb(deallocated), cred, FORCE); 471 #endif 472 ip->i_blocks -= btodb(deallocated); 473 ip->i_flag |= IN_CHANGE | IN_UPDATE; 474 } 475 (void) VOP_FSYNC(vp, MNT_WAIT); 476 477 /* 478 * Cleanup the data block we getblk()'d before returning. 479 */ 480 fail2: 481 brelse(dbp); 482 return (error); 483 } 484 485