ffs_balloc.c revision 1.40.8.1
1 /* $NetBSD: ffs_balloc.c,v 1.40.8.1 2006/04/01 12:07:51 yamt 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 1.40.8.1 2006/04/01 12:07:51 yamt 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(struct vnode *, off_t, int, struct ucred *, int, 70 struct buf **); 71 static int ffs_balloc_ufs2(struct vnode *, off_t, int, struct ucred *, int, 72 struct buf **); 73 74 /* 75 * Balloc defines the structure of file system storage 76 * by allocating the physical blocks on a device given 77 * the inode and the logical block number in a file. 78 */ 79 80 int 81 ffs_balloc(struct vnode *vp, off_t off, int size, struct ucred *cred, int flags, 82 struct buf **bpp) 83 { 84 85 if (VTOI(vp)->i_fs->fs_magic == FS_UFS2_MAGIC) 86 return ffs_balloc_ufs2(vp, off, size, cred, flags, bpp); 87 else 88 return ffs_balloc_ufs1(vp, off, size, cred, flags, bpp); 89 } 90 91 static int 92 ffs_balloc_ufs1(struct vnode *vp, off_t off, int size, struct ucred *cred, 93 int flags, struct buf **bpp) 94 { 95 daddr_t lbn, lastlbn; 96 struct buf *bp, *nbp; 97 struct inode *ip = VTOI(vp); 98 struct fs *fs = ip->i_fs; 99 struct indir indirs[NIADDR + 2]; 100 daddr_t newb, pref, nb; 101 int32_t *bap; /* XXX ondisk32 */ 102 int deallocated, osize, nsize, num, i, error; 103 int32_t *blkp, *allocblk, allociblk[NIADDR + 1]; 104 int32_t *allocib; 105 int unwindidx = -1; 106 #ifdef FFS_EI 107 const int needswap = UFS_FSNEEDSWAP(fs); 108 #endif 109 UVMHIST_FUNC("ffs_balloc"); UVMHIST_CALLED(ubchist); 110 111 lbn = lblkno(fs, off); 112 size = blkoff(fs, off) + size; 113 if (size > fs->fs_bsize) 114 panic("ffs_balloc: blk too big"); 115 if (bpp != NULL) { 116 *bpp = NULL; 117 } 118 UVMHIST_LOG(ubchist, "vp %p lbn 0x%x size 0x%x", vp, lbn, size,0); 119 120 KASSERT(size <= fs->fs_bsize); 121 if (lbn < 0) 122 return (EFBIG); 123 124 /* 125 * If the next write will extend the file into a new block, 126 * and the file is currently composed of a fragment 127 * this fragment has to be extended to be a full block. 128 */ 129 130 lastlbn = lblkno(fs, ip->i_size); 131 if (lastlbn < NDADDR && lastlbn < lbn) { 132 nb = lastlbn; 133 osize = blksize(fs, ip, nb); 134 if (osize < fs->fs_bsize && osize > 0) { 135 error = ffs_realloccg(ip, nb, 136 ffs_blkpref_ufs1(ip, lastlbn, nb, 137 &ip->i_ffs1_db[0]), 138 osize, (int)fs->fs_bsize, cred, bpp, &newb); 139 if (error) 140 return (error); 141 if (DOINGSOFTDEP(vp)) 142 softdep_setup_allocdirect(ip, nb, newb, 143 ufs_rw32(ip->i_ffs1_db[nb], needswap), 144 fs->fs_bsize, osize, bpp ? *bpp : NULL); 145 ip->i_size = lblktosize(fs, nb + 1); 146 ip->i_ffs1_size = ip->i_size; 147 uvm_vnp_setsize(vp, ip->i_ffs1_size); 148 ip->i_ffs1_db[nb] = ufs_rw32((u_int32_t)newb, needswap); 149 ip->i_flag |= IN_CHANGE | IN_UPDATE; 150 if (bpp) { 151 if (flags & B_SYNC) 152 bwrite(*bpp); 153 else 154 bawrite(*bpp); 155 } 156 } 157 } 158 159 /* 160 * The first NDADDR blocks are direct blocks 161 */ 162 163 if (lbn < NDADDR) { 164 nb = ufs_rw32(ip->i_ffs1_db[lbn], needswap); 165 if (nb != 0 && ip->i_size >= lblktosize(fs, lbn + 1)) { 166 167 /* 168 * The block is an already-allocated direct block 169 * and the file already extends past this block, 170 * thus this must be a whole block. 171 * Just read the block (if requested). 172 */ 173 174 if (bpp != NULL) { 175 error = bread(vp, lbn, fs->fs_bsize, NOCRED, 176 bpp); 177 if (error) { 178 brelse(*bpp); 179 return (error); 180 } 181 } 182 return (0); 183 } 184 if (nb != 0) { 185 186 /* 187 * Consider need to reallocate a fragment. 188 */ 189 190 osize = fragroundup(fs, blkoff(fs, ip->i_size)); 191 nsize = fragroundup(fs, size); 192 if (nsize <= osize) { 193 194 /* 195 * The existing block is already 196 * at least as big as we want. 197 * Just read the block (if requested). 198 */ 199 200 if (bpp != NULL) { 201 error = bread(vp, lbn, osize, NOCRED, 202 bpp); 203 if (error) { 204 brelse(*bpp); 205 return (error); 206 } 207 } 208 return 0; 209 } else { 210 211 /* 212 * The existing block is smaller than we want, 213 * grow it. 214 */ 215 216 error = ffs_realloccg(ip, lbn, 217 ffs_blkpref_ufs1(ip, lbn, (int)lbn, 218 &ip->i_ffs1_db[0]), osize, nsize, cred, 219 bpp, &newb); 220 if (error) 221 return (error); 222 if (DOINGSOFTDEP(vp)) 223 softdep_setup_allocdirect(ip, lbn, 224 newb, nb, nsize, osize, 225 bpp ? *bpp : NULL); 226 } 227 } else { 228 229 /* 230 * the block was not previously allocated, 231 * allocate a new block or fragment. 232 */ 233 234 if (ip->i_size < lblktosize(fs, lbn + 1)) 235 nsize = fragroundup(fs, size); 236 else 237 nsize = fs->fs_bsize; 238 error = ffs_alloc(ip, lbn, 239 ffs_blkpref_ufs1(ip, lbn, (int)lbn, 240 &ip->i_ffs1_db[0]), 241 nsize, cred, &newb); 242 if (error) 243 return (error); 244 if (bpp != NULL) { 245 bp = getblk(vp, lbn, nsize, 0, 0); 246 bp->b_blkno = fsbtodb(fs, newb); 247 if (flags & B_CLRBUF) 248 clrbuf(bp); 249 *bpp = bp; 250 } 251 if (DOINGSOFTDEP(vp)) { 252 softdep_setup_allocdirect(ip, lbn, newb, 0, 253 nsize, 0, bpp ? *bpp : NULL); 254 } 255 } 256 ip->i_ffs1_db[lbn] = ufs_rw32((u_int32_t)newb, needswap); 257 ip->i_flag |= IN_CHANGE | IN_UPDATE; 258 return (0); 259 } 260 261 /* 262 * Determine the number of levels of indirection. 263 */ 264 265 pref = 0; 266 if ((error = ufs_getlbns(vp, lbn, indirs, &num)) != 0) 267 return (error); 268 269 /* 270 * Fetch the first indirect block allocating if necessary. 271 */ 272 273 --num; 274 nb = ufs_rw32(ip->i_ffs1_ib[indirs[0].in_off], needswap); 275 allocib = NULL; 276 allocblk = allociblk; 277 if (nb == 0) { 278 pref = ffs_blkpref_ufs1(ip, lbn, 0, (int32_t *)0); 279 error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred, 280 &newb); 281 if (error) 282 goto fail; 283 nb = newb; 284 *allocblk++ = nb; 285 bp = getblk(vp, indirs[1].in_lbn, fs->fs_bsize, 0, 0); 286 bp->b_blkno = fsbtodb(fs, nb); 287 clrbuf(bp); 288 if (DOINGSOFTDEP(vp)) { 289 softdep_setup_allocdirect(ip, NDADDR + indirs[0].in_off, 290 newb, 0, fs->fs_bsize, 0, bp); 291 bdwrite(bp); 292 } else { 293 294 /* 295 * Write synchronously so that indirect blocks 296 * never point at garbage. 297 */ 298 299 if ((error = bwrite(bp)) != 0) 300 goto fail; 301 } 302 unwindidx = 0; 303 allocib = &ip->i_ffs1_ib[indirs[0].in_off]; 304 *allocib = ufs_rw32(nb, needswap); 305 ip->i_flag |= IN_CHANGE | IN_UPDATE; 306 } 307 308 /* 309 * Fetch through the indirect blocks, allocating as necessary. 310 */ 311 312 for (i = 1;;) { 313 error = bread(vp, 314 indirs[i].in_lbn, (int)fs->fs_bsize, NOCRED, &bp); 315 if (error) { 316 brelse(bp); 317 goto fail; 318 } 319 bap = (int32_t *)bp->b_data; /* XXX ondisk32 */ 320 nb = ufs_rw32(bap[indirs[i].in_off], needswap); 321 if (i == num) 322 break; 323 i++; 324 if (nb != 0) { 325 brelse(bp); 326 continue; 327 } 328 if (pref == 0) 329 pref = ffs_blkpref_ufs1(ip, lbn, 0, (int32_t *)0); 330 error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred, 331 &newb); 332 if (error) { 333 brelse(bp); 334 goto fail; 335 } 336 nb = newb; 337 *allocblk++ = nb; 338 nbp = getblk(vp, indirs[i].in_lbn, fs->fs_bsize, 0, 0); 339 nbp->b_blkno = fsbtodb(fs, nb); 340 clrbuf(nbp); 341 if (DOINGSOFTDEP(vp)) { 342 softdep_setup_allocindir_meta(nbp, ip, bp, 343 indirs[i - 1].in_off, nb); 344 bdwrite(nbp); 345 } else { 346 347 /* 348 * Write synchronously so that indirect blocks 349 * never point at garbage. 350 */ 351 352 if ((error = bwrite(nbp)) != 0) { 353 brelse(bp); 354 goto fail; 355 } 356 } 357 if (unwindidx < 0) 358 unwindidx = i - 1; 359 bap[indirs[i - 1].in_off] = ufs_rw32(nb, needswap); 360 361 /* 362 * If required, write synchronously, otherwise use 363 * delayed write. 364 */ 365 366 if (flags & B_SYNC) { 367 bwrite(bp); 368 } else { 369 bdwrite(bp); 370 } 371 } 372 373 if (flags & B_METAONLY) { 374 KASSERT(bpp != NULL); 375 *bpp = bp; 376 return (0); 377 } 378 379 /* 380 * Get the data block, allocating if necessary. 381 */ 382 383 if (nb == 0) { 384 pref = ffs_blkpref_ufs1(ip, lbn, indirs[num].in_off, &bap[0]); 385 error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred, 386 &newb); 387 if (error) { 388 brelse(bp); 389 goto fail; 390 } 391 nb = newb; 392 *allocblk++ = nb; 393 if (bpp != NULL) { 394 nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0); 395 nbp->b_blkno = fsbtodb(fs, nb); 396 if (flags & B_CLRBUF) 397 clrbuf(nbp); 398 *bpp = nbp; 399 } 400 if (DOINGSOFTDEP(vp)) 401 softdep_setup_allocindir_page(ip, lbn, bp, 402 indirs[num].in_off, nb, 0, bpp ? *bpp : NULL); 403 bap[indirs[num].in_off] = ufs_rw32(nb, needswap); 404 if (allocib == NULL && unwindidx < 0) { 405 unwindidx = i - 1; 406 } 407 408 /* 409 * If required, write synchronously, otherwise use 410 * delayed write. 411 */ 412 413 if (flags & B_SYNC) { 414 bwrite(bp); 415 } else { 416 bdwrite(bp); 417 } 418 return (0); 419 } 420 brelse(bp); 421 if (bpp != NULL) { 422 if (flags & B_CLRBUF) { 423 error = bread(vp, lbn, (int)fs->fs_bsize, NOCRED, &nbp); 424 if (error) { 425 brelse(nbp); 426 goto fail; 427 } 428 } else { 429 nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0); 430 nbp->b_blkno = fsbtodb(fs, nb); 431 clrbuf(nbp); 432 } 433 *bpp = nbp; 434 } 435 return (0); 436 437 fail: 438 /* 439 * If we have failed part way through block allocation, we 440 * have to deallocate any indirect blocks that we have allocated. 441 */ 442 443 if (unwindidx >= 0) { 444 445 /* 446 * First write out any buffers we've created to resolve their 447 * softdeps. This must be done in reverse order of creation 448 * so that we resolve the dependencies in one pass. 449 * Write the cylinder group buffers for these buffers too. 450 */ 451 452 for (i = num; i >= unwindidx; i--) { 453 if (i == 0) { 454 break; 455 } 456 bp = getblk(vp, indirs[i].in_lbn, (int)fs->fs_bsize, 0, 457 0); 458 if (bp->b_flags & B_DELWRI) { 459 nb = fsbtodb(fs, cgtod(fs, dtog(fs, 460 dbtofsb(fs, bp->b_blkno)))); 461 bwrite(bp); 462 bp = getblk(ip->i_devvp, nb, (int)fs->fs_cgsize, 463 0, 0); 464 if (bp->b_flags & B_DELWRI) { 465 bwrite(bp); 466 } else { 467 bp->b_flags |= B_INVAL; 468 brelse(bp); 469 } 470 } else { 471 bp->b_flags |= B_INVAL; 472 brelse(bp); 473 } 474 } 475 if (DOINGSOFTDEP(vp) && unwindidx == 0) { 476 ip->i_flag |= IN_CHANGE | IN_UPDATE; 477 ffs_update(vp, NULL, NULL, UPDATE_WAIT); 478 } 479 480 /* 481 * Now that any dependencies that we created have been 482 * resolved, we can undo the partial allocation. 483 */ 484 485 if (unwindidx == 0) { 486 *allocib = 0; 487 ip->i_flag |= IN_CHANGE | IN_UPDATE; 488 if (DOINGSOFTDEP(vp)) 489 ffs_update(vp, NULL, NULL, UPDATE_WAIT); 490 } else { 491 int r; 492 493 r = bread(vp, indirs[unwindidx].in_lbn, 494 (int)fs->fs_bsize, NOCRED, &bp); 495 if (r) { 496 panic("Could not unwind indirect block, error %d", r); 497 brelse(bp); 498 } else { 499 bap = (int32_t *)bp->b_data; /* XXX ondisk32 */ 500 bap[indirs[unwindidx].in_off] = 0; 501 bwrite(bp); 502 } 503 } 504 for (i = unwindidx + 1; i <= num; i++) { 505 bp = getblk(vp, indirs[i].in_lbn, (int)fs->fs_bsize, 0, 506 0); 507 bp->b_flags |= B_INVAL; 508 brelse(bp); 509 } 510 } 511 for (deallocated = 0, blkp = allociblk; blkp < allocblk; blkp++) { 512 ffs_blkfree(fs, ip->i_devvp, *blkp, fs->fs_bsize, ip->i_number); 513 deallocated += fs->fs_bsize; 514 } 515 if (deallocated) { 516 #ifdef QUOTA 517 /* 518 * Restore user's disk quota because allocation failed. 519 */ 520 (void)chkdq(ip, -btodb(deallocated), cred, FORCE); 521 #endif 522 ip->i_ffs1_blocks -= btodb(deallocated); 523 ip->i_flag |= IN_CHANGE | IN_UPDATE; 524 } 525 return (error); 526 } 527 528 static int 529 ffs_balloc_ufs2(struct vnode *vp, off_t off, int size, struct ucred *cred, 530 int flags, struct buf **bpp) 531 { 532 daddr_t lbn, lastlbn; 533 struct buf *bp, *nbp; 534 struct inode *ip = VTOI(vp); 535 struct fs *fs = ip->i_fs; 536 struct indir indirs[NIADDR + 2]; 537 daddr_t newb, pref, nb; 538 int64_t *bap; 539 int deallocated, osize, nsize, num, i, error; 540 daddr_t *blkp, *allocblk, allociblk[NIADDR + 1]; 541 int64_t *allocib; 542 int unwindidx = -1; 543 #ifdef FFS_EI 544 const int needswap = UFS_FSNEEDSWAP(fs); 545 #endif 546 UVMHIST_FUNC("ffs_balloc"); UVMHIST_CALLED(ubchist); 547 548 lbn = lblkno(fs, off); 549 size = blkoff(fs, off) + size; 550 if (size > fs->fs_bsize) 551 panic("ffs_balloc: blk too big"); 552 if (bpp != NULL) { 553 *bpp = NULL; 554 } 555 UVMHIST_LOG(ubchist, "vp %p lbn 0x%x size 0x%x", vp, lbn, size,0); 556 557 KASSERT(size <= fs->fs_bsize); 558 if (lbn < 0) 559 return (EFBIG); 560 561 #ifdef notyet 562 /* 563 * Check for allocating external data. 564 */ 565 if (flags & IO_EXT) { 566 if (lbn >= NXADDR) 567 return (EFBIG); 568 /* 569 * If the next write will extend the data into a new block, 570 * and the data is currently composed of a fragment 571 * this fragment has to be extended to be a full block. 572 */ 573 lastlbn = lblkno(fs, dp->di_extsize); 574 if (lastlbn < lbn) { 575 nb = lastlbn; 576 osize = sblksize(fs, dp->di_extsize, nb); 577 if (osize < fs->fs_bsize && osize > 0) { 578 error = ffs_realloccg(ip, -1 - nb, 579 dp->di_extb[nb], 580 ffs_blkpref_ufs2(ip, lastlbn, (int)nb, 581 &dp->di_extb[0]), osize, 582 (int)fs->fs_bsize, cred, &bp); 583 if (error) 584 return (error); 585 if (DOINGSOFTDEP(vp)) 586 softdep_setup_allocext(ip, nb, 587 dbtofsb(fs, bp->b_blkno), 588 dp->di_extb[nb], 589 fs->fs_bsize, osize, bp); 590 dp->di_extsize = smalllblktosize(fs, nb + 1); 591 dp->di_extb[nb] = dbtofsb(fs, bp->b_blkno); 592 bp->b_xflags |= BX_ALTDATA; 593 ip->i_flag |= IN_CHANGE | IN_UPDATE; 594 if (flags & IO_SYNC) 595 bwrite(bp); 596 else 597 bawrite(bp); 598 } 599 } 600 /* 601 * All blocks are direct blocks 602 */ 603 if (flags & BA_METAONLY) 604 panic("ffs_balloc_ufs2: BA_METAONLY for ext block"); 605 nb = dp->di_extb[lbn]; 606 if (nb != 0 && dp->di_extsize >= smalllblktosize(fs, lbn + 1)) { 607 error = bread(vp, -1 - lbn, fs->fs_bsize, NOCRED, &bp); 608 if (error) { 609 brelse(bp); 610 return (error); 611 } 612 bp->b_blkno = fsbtodb(fs, nb); 613 bp->b_xflags |= BX_ALTDATA; 614 *bpp = bp; 615 return (0); 616 } 617 if (nb != 0) { 618 /* 619 * Consider need to reallocate a fragment. 620 */ 621 osize = fragroundup(fs, blkoff(fs, dp->di_extsize)); 622 nsize = fragroundup(fs, size); 623 if (nsize <= osize) { 624 error = bread(vp, -1 - lbn, osize, NOCRED, &bp); 625 if (error) { 626 brelse(bp); 627 return (error); 628 } 629 bp->b_blkno = fsbtodb(fs, nb); 630 bp->b_xflags |= BX_ALTDATA; 631 } else { 632 error = ffs_realloccg(ip, -1 - lbn, 633 dp->di_extb[lbn], 634 ffs_blkpref_ufs2(ip, lbn, (int)lbn, 635 &dp->di_extb[0]), osize, nsize, cred, &bp); 636 if (error) 637 return (error); 638 bp->b_xflags |= BX_ALTDATA; 639 if (DOINGSOFTDEP(vp)) 640 softdep_setup_allocext(ip, lbn, 641 dbtofsb(fs, bp->b_blkno), nb, 642 nsize, osize, bp); 643 } 644 } else { 645 if (dp->di_extsize < smalllblktosize(fs, lbn + 1)) 646 nsize = fragroundup(fs, size); 647 else 648 nsize = fs->fs_bsize; 649 error = ffs_alloc(ip, lbn, 650 ffs_blkpref_ufs2(ip, lbn, (int)lbn, &dp->di_extb[0]), 651 nsize, cred, &newb); 652 if (error) 653 return (error); 654 bp = getblk(vp, -1 - lbn, nsize, 0, 0); 655 bp->b_blkno = fsbtodb(fs, newb); 656 bp->b_xflags |= BX_ALTDATA; 657 if (flags & BA_CLRBUF) 658 vfs_bio_clrbuf(bp); 659 if (DOINGSOFTDEP(vp)) 660 softdep_setup_allocext(ip, lbn, newb, 0, 661 nsize, 0, bp); 662 } 663 dp->di_extb[lbn] = dbtofsb(fs, bp->b_blkno); 664 ip->i_flag |= IN_CHANGE | IN_UPDATE; 665 *bpp = bp; 666 return (0); 667 } 668 #endif 669 /* 670 * If the next write will extend the file into a new block, 671 * and the file is currently composed of a fragment 672 * this fragment has to be extended to be a full block. 673 */ 674 675 lastlbn = lblkno(fs, ip->i_size); 676 if (lastlbn < NDADDR && lastlbn < lbn) { 677 nb = lastlbn; 678 osize = blksize(fs, ip, nb); 679 if (osize < fs->fs_bsize && osize > 0) { 680 error = ffs_realloccg(ip, nb, 681 ffs_blkpref_ufs2(ip, lastlbn, nb, 682 &ip->i_ffs2_db[0]), 683 osize, (int)fs->fs_bsize, cred, bpp, &newb); 684 if (error) 685 return (error); 686 if (DOINGSOFTDEP(vp)) 687 softdep_setup_allocdirect(ip, nb, newb, 688 ufs_rw64(ip->i_ffs2_db[nb], needswap), 689 fs->fs_bsize, osize, bpp ? *bpp : NULL); 690 ip->i_size = lblktosize(fs, nb + 1); 691 ip->i_ffs2_size = ip->i_size; 692 uvm_vnp_setsize(vp, ip->i_size); 693 ip->i_ffs2_db[nb] = ufs_rw64(newb, needswap); 694 ip->i_flag |= IN_CHANGE | IN_UPDATE; 695 if (bpp) { 696 if (flags & B_SYNC) 697 bwrite(*bpp); 698 else 699 bawrite(*bpp); 700 } 701 } 702 } 703 704 /* 705 * The first NDADDR blocks are direct blocks 706 */ 707 708 if (lbn < NDADDR) { 709 nb = ufs_rw64(ip->i_ffs2_db[lbn], needswap); 710 if (nb != 0 && ip->i_size >= lblktosize(fs, lbn + 1)) { 711 712 /* 713 * The block is an already-allocated direct block 714 * and the file already extends past this block, 715 * thus this must be a whole block. 716 * Just read the block (if requested). 717 */ 718 719 if (bpp != NULL) { 720 error = bread(vp, lbn, fs->fs_bsize, NOCRED, 721 bpp); 722 if (error) { 723 brelse(*bpp); 724 return (error); 725 } 726 } 727 return (0); 728 } 729 if (nb != 0) { 730 731 /* 732 * Consider need to reallocate a fragment. 733 */ 734 735 osize = fragroundup(fs, blkoff(fs, ip->i_size)); 736 nsize = fragroundup(fs, size); 737 if (nsize <= osize) { 738 739 /* 740 * The existing block is already 741 * at least as big as we want. 742 * Just read the block (if requested). 743 */ 744 745 if (bpp != NULL) { 746 error = bread(vp, lbn, osize, NOCRED, 747 bpp); 748 if (error) { 749 brelse(*bpp); 750 return (error); 751 } 752 } 753 return 0; 754 } else { 755 756 /* 757 * The existing block is smaller than we want, 758 * grow it. 759 */ 760 761 error = ffs_realloccg(ip, lbn, 762 ffs_blkpref_ufs2(ip, lbn, (int)lbn, 763 &ip->i_ffs2_db[0]), osize, nsize, cred, 764 bpp, &newb); 765 if (error) 766 return (error); 767 if (DOINGSOFTDEP(vp)) 768 softdep_setup_allocdirect(ip, lbn, 769 newb, nb, nsize, osize, 770 bpp ? *bpp : NULL); 771 } 772 } else { 773 774 /* 775 * the block was not previously allocated, 776 * allocate a new block or fragment. 777 */ 778 779 if (ip->i_size < lblktosize(fs, lbn + 1)) 780 nsize = fragroundup(fs, size); 781 else 782 nsize = fs->fs_bsize; 783 error = ffs_alloc(ip, lbn, 784 ffs_blkpref_ufs2(ip, lbn, (int)lbn, 785 &ip->i_ffs2_db[0]), nsize, cred, &newb); 786 if (error) 787 return (error); 788 if (bpp != NULL) { 789 bp = getblk(vp, lbn, nsize, 0, 0); 790 bp->b_blkno = fsbtodb(fs, newb); 791 if (flags & B_CLRBUF) 792 clrbuf(bp); 793 *bpp = bp; 794 } 795 if (DOINGSOFTDEP(vp)) { 796 softdep_setup_allocdirect(ip, lbn, newb, 0, 797 nsize, 0, bpp ? *bpp : NULL); 798 } 799 } 800 ip->i_ffs2_db[lbn] = ufs_rw64(newb, needswap); 801 ip->i_flag |= IN_CHANGE | IN_UPDATE; 802 return (0); 803 } 804 805 /* 806 * Determine the number of levels of indirection. 807 */ 808 809 pref = 0; 810 if ((error = ufs_getlbns(vp, lbn, indirs, &num)) != 0) 811 return (error); 812 813 /* 814 * Fetch the first indirect block allocating if necessary. 815 */ 816 817 --num; 818 nb = ufs_rw64(ip->i_ffs2_ib[indirs[0].in_off], needswap); 819 allocib = NULL; 820 allocblk = allociblk; 821 if (nb == 0) { 822 pref = ffs_blkpref_ufs2(ip, lbn, 0, (int64_t *)0); 823 error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred, 824 &newb); 825 if (error) 826 goto fail; 827 nb = newb; 828 *allocblk++ = nb; 829 bp = getblk(vp, indirs[1].in_lbn, fs->fs_bsize, 0, 0); 830 bp->b_blkno = fsbtodb(fs, nb); 831 clrbuf(bp); 832 if (DOINGSOFTDEP(vp)) { 833 softdep_setup_allocdirect(ip, NDADDR + indirs[0].in_off, 834 newb, 0, fs->fs_bsize, 0, bp); 835 bdwrite(bp); 836 } else { 837 838 /* 839 * Write synchronously so that indirect blocks 840 * never point at garbage. 841 */ 842 843 if ((error = bwrite(bp)) != 0) 844 goto fail; 845 } 846 unwindidx = 0; 847 allocib = &ip->i_ffs2_ib[indirs[0].in_off]; 848 *allocib = ufs_rw64(nb, needswap); 849 ip->i_flag |= IN_CHANGE | IN_UPDATE; 850 } 851 852 /* 853 * Fetch through the indirect blocks, allocating as necessary. 854 */ 855 856 for (i = 1;;) { 857 error = bread(vp, 858 indirs[i].in_lbn, (int)fs->fs_bsize, NOCRED, &bp); 859 if (error) { 860 brelse(bp); 861 goto fail; 862 } 863 bap = (int64_t *)bp->b_data; 864 nb = ufs_rw64(bap[indirs[i].in_off], needswap); 865 if (i == num) 866 break; 867 i++; 868 if (nb != 0) { 869 brelse(bp); 870 continue; 871 } 872 if (pref == 0) 873 pref = ffs_blkpref_ufs2(ip, lbn, 0, (int64_t *)0); 874 error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred, 875 &newb); 876 if (error) { 877 brelse(bp); 878 goto fail; 879 } 880 nb = newb; 881 *allocblk++ = nb; 882 nbp = getblk(vp, indirs[i].in_lbn, fs->fs_bsize, 0, 0); 883 nbp->b_blkno = fsbtodb(fs, nb); 884 clrbuf(nbp); 885 if (DOINGSOFTDEP(vp)) { 886 softdep_setup_allocindir_meta(nbp, ip, bp, 887 indirs[i - 1].in_off, nb); 888 bdwrite(nbp); 889 } else { 890 891 /* 892 * Write synchronously so that indirect blocks 893 * never point at garbage. 894 */ 895 896 if ((error = bwrite(nbp)) != 0) { 897 brelse(bp); 898 goto fail; 899 } 900 } 901 if (unwindidx < 0) 902 unwindidx = i - 1; 903 bap[indirs[i - 1].in_off] = ufs_rw64(nb, needswap); 904 905 /* 906 * If required, write synchronously, otherwise use 907 * delayed write. 908 */ 909 910 if (flags & B_SYNC) { 911 bwrite(bp); 912 } else { 913 bdwrite(bp); 914 } 915 } 916 917 if (flags & B_METAONLY) { 918 KASSERT(bpp != NULL); 919 *bpp = bp; 920 return (0); 921 } 922 923 /* 924 * Get the data block, allocating if necessary. 925 */ 926 927 if (nb == 0) { 928 pref = ffs_blkpref_ufs2(ip, lbn, indirs[num].in_off, &bap[0]); 929 error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred, 930 &newb); 931 if (error) { 932 brelse(bp); 933 goto fail; 934 } 935 nb = newb; 936 *allocblk++ = nb; 937 if (bpp != NULL) { 938 nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0); 939 nbp->b_blkno = fsbtodb(fs, nb); 940 if (flags & B_CLRBUF) 941 clrbuf(nbp); 942 *bpp = nbp; 943 } 944 if (DOINGSOFTDEP(vp)) 945 softdep_setup_allocindir_page(ip, lbn, bp, 946 indirs[num].in_off, nb, 0, bpp ? *bpp : NULL); 947 bap[indirs[num].in_off] = ufs_rw64(nb, needswap); 948 if (allocib == NULL && unwindidx < 0) { 949 unwindidx = i - 1; 950 } 951 952 /* 953 * If required, write synchronously, otherwise use 954 * delayed write. 955 */ 956 957 if (flags & B_SYNC) { 958 bwrite(bp); 959 } else { 960 bdwrite(bp); 961 } 962 return (0); 963 } 964 brelse(bp); 965 if (bpp != NULL) { 966 if (flags & B_CLRBUF) { 967 error = bread(vp, lbn, (int)fs->fs_bsize, NOCRED, &nbp); 968 if (error) { 969 brelse(nbp); 970 goto fail; 971 } 972 } else { 973 nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0); 974 nbp->b_blkno = fsbtodb(fs, nb); 975 clrbuf(nbp); 976 } 977 *bpp = nbp; 978 } 979 return (0); 980 981 fail: 982 /* 983 * If we have failed part way through block allocation, we 984 * have to deallocate any indirect blocks that we have allocated. 985 */ 986 987 if (unwindidx >= 0) { 988 989 /* 990 * First write out any buffers we've created to resolve their 991 * softdeps. This must be done in reverse order of creation 992 * so that we resolve the dependencies in one pass. 993 * Write the cylinder group buffers for these buffers too. 994 */ 995 996 for (i = num; i >= unwindidx; i--) { 997 if (i == 0) { 998 break; 999 } 1000 bp = getblk(vp, indirs[i].in_lbn, (int)fs->fs_bsize, 0, 1001 0); 1002 if (bp->b_flags & B_DELWRI) { 1003 nb = fsbtodb(fs, cgtod(fs, dtog(fs, 1004 dbtofsb(fs, bp->b_blkno)))); 1005 bwrite(bp); 1006 bp = getblk(ip->i_devvp, nb, (int)fs->fs_cgsize, 1007 0, 0); 1008 if (bp->b_flags & B_DELWRI) { 1009 bwrite(bp); 1010 } else { 1011 bp->b_flags |= B_INVAL; 1012 brelse(bp); 1013 } 1014 } else { 1015 bp->b_flags |= B_INVAL; 1016 brelse(bp); 1017 } 1018 } 1019 if (DOINGSOFTDEP(vp) && unwindidx == 0) { 1020 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1021 ffs_update(vp, NULL, NULL, UPDATE_WAIT); 1022 } 1023 1024 /* 1025 * Now that any dependencies that we created have been 1026 * resolved, we can undo the partial allocation. 1027 */ 1028 1029 if (unwindidx == 0) { 1030 *allocib = 0; 1031 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1032 if (DOINGSOFTDEP(vp)) 1033 ffs_update(vp, NULL, NULL, UPDATE_WAIT); 1034 } else { 1035 int r; 1036 1037 r = bread(vp, indirs[unwindidx].in_lbn, 1038 (int)fs->fs_bsize, NOCRED, &bp); 1039 if (r) { 1040 panic("Could not unwind indirect block, error %d", r); 1041 brelse(bp); 1042 } else { 1043 bap = (int64_t *)bp->b_data; 1044 bap[indirs[unwindidx].in_off] = 0; 1045 bwrite(bp); 1046 } 1047 } 1048 for (i = unwindidx + 1; i <= num; i++) { 1049 bp = getblk(vp, indirs[i].in_lbn, (int)fs->fs_bsize, 0, 1050 0); 1051 bp->b_flags |= B_INVAL; 1052 brelse(bp); 1053 } 1054 } 1055 for (deallocated = 0, blkp = allociblk; blkp < allocblk; blkp++) { 1056 ffs_blkfree(fs, ip->i_devvp, *blkp, fs->fs_bsize, ip->i_number); 1057 deallocated += fs->fs_bsize; 1058 } 1059 if (deallocated) { 1060 #ifdef QUOTA 1061 /* 1062 * Restore user's disk quota because allocation failed. 1063 */ 1064 (void)chkdq(ip, -btodb(deallocated), cred, FORCE); 1065 #endif 1066 ip->i_ffs2_blocks -= btodb(deallocated); 1067 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1068 } 1069 return (error); 1070 } 1071
Indexes created Mon Oct 13 08:09:50 GMT 2025