ffs_balloc.c revision 1.38.2.1
1 /* $NetBSD: ffs_balloc.c,v 1.38.2.1 2005/10/20 03:00:30 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 1.38.2.1 2005/10/20 03:00:30 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 *bpp = bp; 375 return (0); 376 } 377 378 /* 379 * Get the data block, allocating if necessary. 380 */ 381 382 if (nb == 0) { 383 pref = ffs_blkpref_ufs1(ip, lbn, indirs[num].in_off, &bap[0]); 384 error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred, 385 &newb); 386 if (error) { 387 brelse(bp); 388 goto fail; 389 } 390 nb = newb; 391 *allocblk++ = nb; 392 if (bpp != NULL) { 393 nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0); 394 nbp->b_blkno = fsbtodb(fs, nb); 395 if (flags & B_CLRBUF) 396 clrbuf(nbp); 397 *bpp = nbp; 398 } 399 if (DOINGSOFTDEP(vp)) 400 softdep_setup_allocindir_page(ip, lbn, bp, 401 indirs[num].in_off, nb, 0, bpp ? *bpp : NULL); 402 bap[indirs[num].in_off] = ufs_rw32(nb, needswap); 403 if (allocib == NULL && unwindidx < 0) { 404 unwindidx = i - 1; 405 } 406 407 /* 408 * If required, write synchronously, otherwise use 409 * delayed write. 410 */ 411 412 if (flags & B_SYNC) { 413 bwrite(bp); 414 } else { 415 bdwrite(bp); 416 } 417 return (0); 418 } 419 brelse(bp); 420 if (bpp != NULL) { 421 if (flags & B_CLRBUF) { 422 error = bread(vp, lbn, (int)fs->fs_bsize, NOCRED, &nbp); 423 if (error) { 424 brelse(nbp); 425 goto fail; 426 } 427 } else { 428 nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0); 429 nbp->b_blkno = fsbtodb(fs, nb); 430 clrbuf(nbp); 431 } 432 *bpp = nbp; 433 } 434 return (0); 435 436 fail: 437 /* 438 * If we have failed part way through block allocation, we 439 * have to deallocate any indirect blocks that we have allocated. 440 */ 441 442 if (unwindidx >= 0) { 443 444 /* 445 * First write out any buffers we've created to resolve their 446 * softdeps. This must be done in reverse order of creation 447 * so that we resolve the dependencies in one pass. 448 * Write the cylinder group buffers for these buffers too. 449 */ 450 451 for (i = num; i >= unwindidx; i--) { 452 if (i == 0) { 453 break; 454 } 455 bp = getblk(vp, indirs[i].in_lbn, (int)fs->fs_bsize, 0, 456 0); 457 if (bp->b_flags & B_DELWRI) { 458 nb = fsbtodb(fs, cgtod(fs, dtog(fs, 459 dbtofsb(fs, bp->b_blkno)))); 460 bwrite(bp); 461 bp = getblk(ip->i_devvp, nb, (int)fs->fs_cgsize, 462 0, 0); 463 if (bp->b_flags & B_DELWRI) { 464 bwrite(bp); 465 } else { 466 bp->b_flags |= B_INVAL; 467 brelse(bp); 468 } 469 } else { 470 bp->b_flags |= B_INVAL; 471 brelse(bp); 472 } 473 } 474 if (DOINGSOFTDEP(vp) && unwindidx == 0) { 475 ip->i_flag |= IN_CHANGE | IN_UPDATE; 476 UFS_UPDATE(vp, NULL, NULL, UPDATE_WAIT); 477 } 478 479 /* 480 * Now that any dependencies that we created have been 481 * resolved, we can undo the partial allocation. 482 */ 483 484 if (unwindidx == 0) { 485 *allocib = 0; 486 ip->i_flag |= IN_CHANGE | IN_UPDATE; 487 if (DOINGSOFTDEP(vp)) 488 UFS_UPDATE(vp, NULL, NULL, UPDATE_WAIT); 489 } else { 490 int r; 491 492 r = bread(vp, indirs[unwindidx].in_lbn, 493 (int)fs->fs_bsize, NOCRED, &bp); 494 if (r) { 495 panic("Could not unwind indirect block, error %d", r); 496 brelse(bp); 497 } else { 498 bap = (int32_t *)bp->b_data; /* XXX ondisk32 */ 499 bap[indirs[unwindidx].in_off] = 0; 500 bwrite(bp); 501 } 502 } 503 for (i = unwindidx + 1; i <= num; i++) { 504 bp = getblk(vp, indirs[i].in_lbn, (int)fs->fs_bsize, 0, 505 0); 506 bp->b_flags |= B_INVAL; 507 brelse(bp); 508 } 509 } 510 for (deallocated = 0, blkp = allociblk; blkp < allocblk; blkp++) { 511 ffs_blkfree(fs, ip->i_devvp, *blkp, fs->fs_bsize, ip->i_number); 512 deallocated += fs->fs_bsize; 513 } 514 if (deallocated) { 515 #ifdef QUOTA 516 /* 517 * Restore user's disk quota because allocation failed. 518 */ 519 (void)chkdq(ip, -btodb(deallocated), cred, FORCE); 520 #endif 521 ip->i_ffs1_blocks -= btodb(deallocated); 522 ip->i_flag |= IN_CHANGE | IN_UPDATE; 523 } 524 return (error); 525 } 526 527 static int 528 ffs_balloc_ufs2(struct vnode *vp, off_t off, int size, struct ucred *cred, 529 int flags, struct buf **bpp) 530 { 531 daddr_t lbn, lastlbn; 532 struct buf *bp, *nbp; 533 struct inode *ip = VTOI(vp); 534 struct fs *fs = ip->i_fs; 535 struct indir indirs[NIADDR + 2]; 536 daddr_t newb, pref, nb; 537 int64_t *bap; 538 int deallocated, osize, nsize, num, i, error; 539 daddr_t *blkp, *allocblk, allociblk[NIADDR + 1]; 540 int64_t *allocib; 541 int unwindidx = -1; 542 #ifdef FFS_EI 543 const int needswap = UFS_FSNEEDSWAP(fs); 544 #endif 545 UVMHIST_FUNC("ffs_balloc"); UVMHIST_CALLED(ubchist); 546 547 lbn = lblkno(fs, off); 548 size = blkoff(fs, off) + size; 549 if (size > fs->fs_bsize) 550 panic("ffs_balloc: blk too big"); 551 if (bpp != NULL) { 552 *bpp = NULL; 553 } 554 UVMHIST_LOG(ubchist, "vp %p lbn 0x%x size 0x%x", vp, lbn, size,0); 555 556 KASSERT(size <= fs->fs_bsize); 557 if (lbn < 0) 558 return (EFBIG); 559 560 #ifdef notyet 561 /* 562 * Check for allocating external data. 563 */ 564 if (flags & IO_EXT) { 565 if (lbn >= NXADDR) 566 return (EFBIG); 567 /* 568 * If the next write will extend the data into a new block, 569 * and the data is currently composed of a fragment 570 * this fragment has to be extended to be a full block. 571 */ 572 lastlbn = lblkno(fs, dp->di_extsize); 573 if (lastlbn < lbn) { 574 nb = lastlbn; 575 osize = sblksize(fs, dp->di_extsize, nb); 576 if (osize < fs->fs_bsize && osize > 0) { 577 error = ffs_realloccg(ip, -1 - nb, 578 dp->di_extb[nb], 579 ffs_blkpref_ufs2(ip, lastlbn, (int)nb, 580 &dp->di_extb[0]), osize, 581 (int)fs->fs_bsize, cred, &bp); 582 if (error) 583 return (error); 584 if (DOINGSOFTDEP(vp)) 585 softdep_setup_allocext(ip, nb, 586 dbtofsb(fs, bp->b_blkno), 587 dp->di_extb[nb], 588 fs->fs_bsize, osize, bp); 589 dp->di_extsize = smalllblktosize(fs, nb + 1); 590 dp->di_extb[nb] = dbtofsb(fs, bp->b_blkno); 591 bp->b_xflags |= BX_ALTDATA; 592 ip->i_flag |= IN_CHANGE | IN_UPDATE; 593 if (flags & IO_SYNC) 594 bwrite(bp); 595 else 596 bawrite(bp); 597 } 598 } 599 /* 600 * All blocks are direct blocks 601 */ 602 if (flags & BA_METAONLY) 603 panic("ffs_balloc_ufs2: BA_METAONLY for ext block"); 604 nb = dp->di_extb[lbn]; 605 if (nb != 0 && dp->di_extsize >= smalllblktosize(fs, lbn + 1)) { 606 error = bread(vp, -1 - lbn, fs->fs_bsize, NOCRED, &bp); 607 if (error) { 608 brelse(bp); 609 return (error); 610 } 611 bp->b_blkno = fsbtodb(fs, nb); 612 bp->b_xflags |= BX_ALTDATA; 613 *bpp = bp; 614 return (0); 615 } 616 if (nb != 0) { 617 /* 618 * Consider need to reallocate a fragment. 619 */ 620 osize = fragroundup(fs, blkoff(fs, dp->di_extsize)); 621 nsize = fragroundup(fs, size); 622 if (nsize <= osize) { 623 error = bread(vp, -1 - lbn, osize, NOCRED, &bp); 624 if (error) { 625 brelse(bp); 626 return (error); 627 } 628 bp->b_blkno = fsbtodb(fs, nb); 629 bp->b_xflags |= BX_ALTDATA; 630 } else { 631 error = ffs_realloccg(ip, -1 - lbn, 632 dp->di_extb[lbn], 633 ffs_blkpref_ufs2(ip, lbn, (int)lbn, 634 &dp->di_extb[0]), osize, nsize, cred, &bp); 635 if (error) 636 return (error); 637 bp->b_xflags |= BX_ALTDATA; 638 if (DOINGSOFTDEP(vp)) 639 softdep_setup_allocext(ip, lbn, 640 dbtofsb(fs, bp->b_blkno), nb, 641 nsize, osize, bp); 642 } 643 } else { 644 if (dp->di_extsize < smalllblktosize(fs, lbn + 1)) 645 nsize = fragroundup(fs, size); 646 else 647 nsize = fs->fs_bsize; 648 error = ffs_alloc(ip, lbn, 649 ffs_blkpref_ufs2(ip, lbn, (int)lbn, &dp->di_extb[0]), 650 nsize, cred, &newb); 651 if (error) 652 return (error); 653 bp = getblk(vp, -1 - lbn, nsize, 0, 0); 654 bp->b_blkno = fsbtodb(fs, newb); 655 bp->b_xflags |= BX_ALTDATA; 656 if (flags & BA_CLRBUF) 657 vfs_bio_clrbuf(bp); 658 if (DOINGSOFTDEP(vp)) 659 softdep_setup_allocext(ip, lbn, newb, 0, 660 nsize, 0, bp); 661 } 662 dp->di_extb[lbn] = dbtofsb(fs, bp->b_blkno); 663 ip->i_flag |= IN_CHANGE | IN_UPDATE; 664 *bpp = bp; 665 return (0); 666 } 667 #endif 668 /* 669 * If the next write will extend the file into a new block, 670 * and the file is currently composed of a fragment 671 * this fragment has to be extended to be a full block. 672 */ 673 674 lastlbn = lblkno(fs, ip->i_size); 675 if (lastlbn < NDADDR && lastlbn < lbn) { 676 nb = lastlbn; 677 osize = blksize(fs, ip, nb); 678 if (osize < fs->fs_bsize && osize > 0) { 679 error = ffs_realloccg(ip, nb, 680 ffs_blkpref_ufs2(ip, lastlbn, nb, 681 &ip->i_ffs2_db[0]), 682 osize, (int)fs->fs_bsize, cred, bpp, &newb); 683 if (error) 684 return (error); 685 if (DOINGSOFTDEP(vp)) 686 softdep_setup_allocdirect(ip, nb, newb, 687 ufs_rw64(ip->i_ffs2_db[nb], needswap), 688 fs->fs_bsize, osize, bpp ? *bpp : NULL); 689 ip->i_size = lblktosize(fs, nb + 1); 690 ip->i_ffs2_size = ip->i_size; 691 uvm_vnp_setsize(vp, ip->i_size); 692 ip->i_ffs2_db[nb] = ufs_rw64(newb, needswap); 693 ip->i_flag |= IN_CHANGE | IN_UPDATE; 694 if (bpp) { 695 if (flags & B_SYNC) 696 bwrite(*bpp); 697 else 698 bawrite(*bpp); 699 } 700 } 701 } 702 703 /* 704 * The first NDADDR blocks are direct blocks 705 */ 706 707 if (lbn < NDADDR) { 708 nb = ufs_rw64(ip->i_ffs2_db[lbn], needswap); 709 if (nb != 0 && ip->i_size >= lblktosize(fs, lbn + 1)) { 710 711 /* 712 * The block is an already-allocated direct block 713 * and the file already extends past this block, 714 * thus this must be a whole block. 715 * Just read the block (if requested). 716 */ 717 718 if (bpp != NULL) { 719 error = bread(vp, lbn, fs->fs_bsize, NOCRED, 720 bpp); 721 if (error) { 722 brelse(*bpp); 723 return (error); 724 } 725 } 726 return (0); 727 } 728 if (nb != 0) { 729 730 /* 731 * Consider need to reallocate a fragment. 732 */ 733 734 osize = fragroundup(fs, blkoff(fs, ip->i_size)); 735 nsize = fragroundup(fs, size); 736 if (nsize <= osize) { 737 738 /* 739 * The existing block is already 740 * at least as big as we want. 741 * Just read the block (if requested). 742 */ 743 744 if (bpp != NULL) { 745 error = bread(vp, lbn, osize, NOCRED, 746 bpp); 747 if (error) { 748 brelse(*bpp); 749 return (error); 750 } 751 } 752 return 0; 753 } else { 754 755 /* 756 * The existing block is smaller than we want, 757 * grow it. 758 */ 759 760 error = ffs_realloccg(ip, lbn, 761 ffs_blkpref_ufs2(ip, lbn, (int)lbn, 762 &ip->i_ffs2_db[0]), osize, nsize, cred, 763 bpp, &newb); 764 if (error) 765 return (error); 766 if (DOINGSOFTDEP(vp)) 767 softdep_setup_allocdirect(ip, lbn, 768 newb, nb, nsize, osize, 769 bpp ? *bpp : NULL); 770 } 771 } else { 772 773 /* 774 * the block was not previously allocated, 775 * allocate a new block or fragment. 776 */ 777 778 if (ip->i_size < lblktosize(fs, lbn + 1)) 779 nsize = fragroundup(fs, size); 780 else 781 nsize = fs->fs_bsize; 782 error = ffs_alloc(ip, lbn, 783 ffs_blkpref_ufs2(ip, lbn, (int)lbn, 784 &ip->i_ffs2_db[0]), nsize, cred, &newb); 785 if (error) 786 return (error); 787 if (bpp != NULL) { 788 bp = getblk(vp, lbn, nsize, 0, 0); 789 bp->b_blkno = fsbtodb(fs, newb); 790 if (flags & B_CLRBUF) 791 clrbuf(bp); 792 *bpp = bp; 793 } 794 if (DOINGSOFTDEP(vp)) { 795 softdep_setup_allocdirect(ip, lbn, newb, 0, 796 nsize, 0, bpp ? *bpp : NULL); 797 } 798 } 799 ip->i_ffs2_db[lbn] = ufs_rw64(newb, needswap); 800 ip->i_flag |= IN_CHANGE | IN_UPDATE; 801 return (0); 802 } 803 804 /* 805 * Determine the number of levels of indirection. 806 */ 807 808 pref = 0; 809 if ((error = ufs_getlbns(vp, lbn, indirs, &num)) != 0) 810 return (error); 811 812 /* 813 * Fetch the first indirect block allocating if necessary. 814 */ 815 816 --num; 817 nb = ufs_rw64(ip->i_ffs2_ib[indirs[0].in_off], needswap); 818 allocib = NULL; 819 allocblk = allociblk; 820 if (nb == 0) { 821 pref = ffs_blkpref_ufs2(ip, lbn, 0, (int64_t *)0); 822 error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred, 823 &newb); 824 if (error) 825 goto fail; 826 nb = newb; 827 *allocblk++ = nb; 828 bp = getblk(vp, indirs[1].in_lbn, fs->fs_bsize, 0, 0); 829 bp->b_blkno = fsbtodb(fs, nb); 830 clrbuf(bp); 831 if (DOINGSOFTDEP(vp)) { 832 softdep_setup_allocdirect(ip, NDADDR + indirs[0].in_off, 833 newb, 0, fs->fs_bsize, 0, bp); 834 bdwrite(bp); 835 } else { 836 837 /* 838 * Write synchronously so that indirect blocks 839 * never point at garbage. 840 */ 841 842 if ((error = bwrite(bp)) != 0) 843 goto fail; 844 } 845 unwindidx = 0; 846 allocib = &ip->i_ffs2_ib[indirs[0].in_off]; 847 *allocib = ufs_rw64(nb, needswap); 848 ip->i_flag |= IN_CHANGE | IN_UPDATE; 849 } 850 851 /* 852 * Fetch through the indirect blocks, allocating as necessary. 853 */ 854 855 for (i = 1;;) { 856 error = bread(vp, 857 indirs[i].in_lbn, (int)fs->fs_bsize, NOCRED, &bp); 858 if (error) { 859 brelse(bp); 860 goto fail; 861 } 862 bap = (int64_t *)bp->b_data; 863 nb = ufs_rw64(bap[indirs[i].in_off], needswap); 864 if (i == num) 865 break; 866 i++; 867 if (nb != 0) { 868 brelse(bp); 869 continue; 870 } 871 if (pref == 0) 872 pref = ffs_blkpref_ufs2(ip, lbn, 0, (int64_t *)0); 873 error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred, 874 &newb); 875 if (error) { 876 brelse(bp); 877 goto fail; 878 } 879 nb = newb; 880 *allocblk++ = nb; 881 nbp = getblk(vp, indirs[i].in_lbn, fs->fs_bsize, 0, 0); 882 nbp->b_blkno = fsbtodb(fs, nb); 883 clrbuf(nbp); 884 if (DOINGSOFTDEP(vp)) { 885 softdep_setup_allocindir_meta(nbp, ip, bp, 886 indirs[i - 1].in_off, nb); 887 bdwrite(nbp); 888 } else { 889 890 /* 891 * Write synchronously so that indirect blocks 892 * never point at garbage. 893 */ 894 895 if ((error = bwrite(nbp)) != 0) { 896 brelse(bp); 897 goto fail; 898 } 899 } 900 if (unwindidx < 0) 901 unwindidx = i - 1; 902 bap[indirs[i - 1].in_off] = ufs_rw64(nb, needswap); 903 904 /* 905 * If required, write synchronously, otherwise use 906 * delayed write. 907 */ 908 909 if (flags & B_SYNC) { 910 bwrite(bp); 911 } else { 912 bdwrite(bp); 913 } 914 } 915 916 if (flags & B_METAONLY) { 917 *bpp = bp; 918 return (0); 919 } 920 921 /* 922 * Get the data block, allocating if necessary. 923 */ 924 925 if (nb == 0) { 926 pref = ffs_blkpref_ufs2(ip, lbn, indirs[num].in_off, &bap[0]); 927 error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred, 928 &newb); 929 if (error) { 930 brelse(bp); 931 goto fail; 932 } 933 nb = newb; 934 *allocblk++ = nb; 935 if (bpp != NULL) { 936 nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0); 937 nbp->b_blkno = fsbtodb(fs, nb); 938 if (flags & B_CLRBUF) 939 clrbuf(nbp); 940 *bpp = nbp; 941 } 942 if (DOINGSOFTDEP(vp)) 943 softdep_setup_allocindir_page(ip, lbn, bp, 944 indirs[num].in_off, nb, 0, bpp ? *bpp : NULL); 945 bap[indirs[num].in_off] = ufs_rw64(nb, needswap); 946 if (allocib == NULL && unwindidx < 0) { 947 unwindidx = i - 1; 948 } 949 950 /* 951 * If required, write synchronously, otherwise use 952 * delayed write. 953 */ 954 955 if (flags & B_SYNC) { 956 bwrite(bp); 957 } else { 958 bdwrite(bp); 959 } 960 return (0); 961 } 962 brelse(bp); 963 if (bpp != NULL) { 964 if (flags & B_CLRBUF) { 965 error = bread(vp, lbn, (int)fs->fs_bsize, NOCRED, &nbp); 966 if (error) { 967 brelse(nbp); 968 goto fail; 969 } 970 } else { 971 nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0); 972 nbp->b_blkno = fsbtodb(fs, nb); 973 clrbuf(nbp); 974 } 975 *bpp = nbp; 976 } 977 return (0); 978 979 fail: 980 /* 981 * If we have failed part way through block allocation, we 982 * have to deallocate any indirect blocks that we have allocated. 983 */ 984 985 if (unwindidx >= 0) { 986 987 /* 988 * First write out any buffers we've created to resolve their 989 * softdeps. This must be done in reverse order of creation 990 * so that we resolve the dependencies in one pass. 991 * Write the cylinder group buffers for these buffers too. 992 */ 993 994 for (i = num; i >= unwindidx; i--) { 995 if (i == 0) { 996 break; 997 } 998 bp = getblk(vp, indirs[i].in_lbn, (int)fs->fs_bsize, 0, 999 0); 1000 if (bp->b_flags & B_DELWRI) { 1001 nb = fsbtodb(fs, cgtod(fs, dtog(fs, 1002 dbtofsb(fs, bp->b_blkno)))); 1003 bwrite(bp); 1004 bp = getblk(ip->i_devvp, nb, (int)fs->fs_cgsize, 1005 0, 0); 1006 if (bp->b_flags & B_DELWRI) { 1007 bwrite(bp); 1008 } else { 1009 bp->b_flags |= B_INVAL; 1010 brelse(bp); 1011 } 1012 } else { 1013 bp->b_flags |= B_INVAL; 1014 brelse(bp); 1015 } 1016 } 1017 if (DOINGSOFTDEP(vp) && unwindidx == 0) { 1018 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1019 UFS_UPDATE(vp, NULL, NULL, UPDATE_WAIT); 1020 } 1021 1022 /* 1023 * Now that any dependencies that we created have been 1024 * resolved, we can undo the partial allocation. 1025 */ 1026 1027 if (unwindidx == 0) { 1028 *allocib = 0; 1029 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1030 if (DOINGSOFTDEP(vp)) 1031 UFS_UPDATE(vp, NULL, NULL, UPDATE_WAIT); 1032 } else { 1033 int r; 1034 1035 r = bread(vp, indirs[unwindidx].in_lbn, 1036 (int)fs->fs_bsize, NOCRED, &bp); 1037 if (r) { 1038 panic("Could not unwind indirect block, error %d", r); 1039 brelse(bp); 1040 } else { 1041 bap = (int64_t *)bp->b_data; 1042 bap[indirs[unwindidx].in_off] = 0; 1043 bwrite(bp); 1044 } 1045 } 1046 for (i = unwindidx + 1; i <= num; i++) { 1047 bp = getblk(vp, indirs[i].in_lbn, (int)fs->fs_bsize, 0, 1048 0); 1049 bp->b_flags |= B_INVAL; 1050 brelse(bp); 1051 } 1052 } 1053 for (deallocated = 0, blkp = allociblk; blkp < allocblk; blkp++) { 1054 ffs_blkfree(fs, ip->i_devvp, *blkp, fs->fs_bsize, ip->i_number); 1055 deallocated += fs->fs_bsize; 1056 } 1057 if (deallocated) { 1058 #ifdef QUOTA 1059 /* 1060 * Restore user's disk quota because allocation failed. 1061 */ 1062 (void)chkdq(ip, -btodb(deallocated), cred, FORCE); 1063 #endif 1064 ip->i_ffs2_blocks -= btodb(deallocated); 1065 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1066 } 1067 return (error); 1068 } 1069
Indexes created Sun Oct 12 22:09:51 GMT 2025