ffs_balloc.c revision 1.48.12.2
1 /* $NetBSD: ffs_balloc.c,v 1.48.12.2 2008/09/18 04:37:05 wrstuden 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 1.48.12.2 2008/09/18 04:37:05 wrstuden 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/kauth.h> 57 #include <sys/fstrans.h> 58 59 #include <ufs/ufs/quota.h> 60 #include <ufs/ufs/ufsmount.h> 61 #include <ufs/ufs/inode.h> 62 #include <ufs/ufs/ufs_extern.h> 63 #include <ufs/ufs/ufs_bswap.h> 64 65 #include <ufs/ffs/fs.h> 66 #include <ufs/ffs/ffs_extern.h> 67 68 #include <uvm/uvm.h> 69 70 static int ffs_balloc_ufs1(struct vnode *, off_t, int, kauth_cred_t, int, 71 struct buf **); 72 static int ffs_balloc_ufs2(struct vnode *, off_t, int, kauth_cred_t, int, 73 struct buf **); 74 75 /* 76 * Balloc defines the structure of file system storage 77 * by allocating the physical blocks on a device given 78 * the inode and the logical block number in a file. 79 */ 80 81 int 82 ffs_balloc(struct vnode *vp, off_t off, int size, kauth_cred_t cred, int flags, 83 struct buf **bpp) 84 { 85 int error; 86 87 if (VTOI(vp)->i_fs->fs_magic == FS_UFS2_MAGIC) 88 error = ffs_balloc_ufs2(vp, off, size, cred, flags, bpp); 89 else 90 error = ffs_balloc_ufs1(vp, off, size, cred, flags, bpp); 91 92 if (error == 0 && bpp != NULL && (error = fscow_run(*bpp, false)) != 0) 93 brelse(*bpp, 0); 94 95 return error; 96 } 97 98 static int 99 ffs_balloc_ufs1(struct vnode *vp, off_t off, int size, kauth_cred_t cred, 100 int flags, struct buf **bpp) 101 { 102 daddr_t lbn, lastlbn; 103 struct buf *bp, *nbp; 104 struct inode *ip = VTOI(vp); 105 struct fs *fs = ip->i_fs; 106 struct ufsmount *ump = ip->i_ump; 107 struct indir indirs[NIADDR + 2]; 108 daddr_t newb, pref, nb; 109 int32_t *bap; /* XXX ondisk32 */ 110 int deallocated, osize, nsize, num, i, error; 111 int32_t *blkp, *allocblk, allociblk[NIADDR + 1]; 112 int32_t *allocib; 113 int unwindidx = -1; 114 #ifdef FFS_EI 115 const int needswap = UFS_FSNEEDSWAP(fs); 116 #endif 117 UVMHIST_FUNC("ffs_balloc"); UVMHIST_CALLED(ubchist); 118 119 lbn = lblkno(fs, off); 120 size = blkoff(fs, off) + size; 121 if (size > fs->fs_bsize) 122 panic("ffs_balloc: blk too big"); 123 if (bpp != NULL) { 124 *bpp = NULL; 125 } 126 UVMHIST_LOG(ubchist, "vp %p lbn 0x%x size 0x%x", vp, lbn, size,0); 127 128 if (lbn < 0) 129 return (EFBIG); 130 131 /* 132 * If the next write will extend the file into a new block, 133 * and the file is currently composed of a fragment 134 * this fragment has to be extended to be a full block. 135 */ 136 137 lastlbn = lblkno(fs, ip->i_size); 138 if (lastlbn < NDADDR && lastlbn < lbn) { 139 nb = lastlbn; 140 osize = blksize(fs, ip, nb); 141 if (osize < fs->fs_bsize && osize > 0) { 142 mutex_enter(&ump->um_lock); 143 error = ffs_realloccg(ip, nb, 144 ffs_blkpref_ufs1(ip, lastlbn, nb, flags, 145 &ip->i_ffs1_db[0]), 146 osize, (int)fs->fs_bsize, cred, bpp, &newb); 147 if (error) 148 return (error); 149 if (DOINGSOFTDEP(vp)) 150 softdep_setup_allocdirect(ip, nb, newb, 151 ufs_rw32(ip->i_ffs1_db[nb], needswap), 152 fs->fs_bsize, osize, bpp ? *bpp : NULL); 153 ip->i_size = lblktosize(fs, nb + 1); 154 ip->i_ffs1_size = ip->i_size; 155 uvm_vnp_setsize(vp, ip->i_ffs1_size); 156 ip->i_ffs1_db[nb] = ufs_rw32((u_int32_t)newb, needswap); 157 ip->i_flag |= IN_CHANGE | IN_UPDATE; 158 if (bpp && *bpp) { 159 if (flags & B_SYNC) 160 bwrite(*bpp); 161 else 162 bawrite(*bpp); 163 } 164 } 165 } 166 167 /* 168 * The first NDADDR blocks are direct blocks 169 */ 170 171 if (lbn < NDADDR) { 172 nb = ufs_rw32(ip->i_ffs1_db[lbn], needswap); 173 if (nb != 0 && ip->i_size >= lblktosize(fs, lbn + 1)) { 174 175 /* 176 * The block is an already-allocated direct block 177 * and the file already extends past this block, 178 * thus this must be a whole block. 179 * Just read the block (if requested). 180 */ 181 182 if (bpp != NULL) { 183 error = bread(vp, lbn, fs->fs_bsize, NOCRED, 184 B_MODIFY, bpp); 185 if (error) { 186 brelse(*bpp, 0); 187 return (error); 188 } 189 } 190 return (0); 191 } 192 if (nb != 0) { 193 194 /* 195 * Consider need to reallocate a fragment. 196 */ 197 198 osize = fragroundup(fs, blkoff(fs, ip->i_size)); 199 nsize = fragroundup(fs, size); 200 if (nsize <= osize) { 201 202 /* 203 * The existing block is already 204 * at least as big as we want. 205 * Just read the block (if requested). 206 */ 207 208 if (bpp != NULL) { 209 error = bread(vp, lbn, osize, NOCRED, 210 B_MODIFY, bpp); 211 if (error) { 212 brelse(*bpp, 0); 213 return (error); 214 } 215 } 216 return 0; 217 } else { 218 219 /* 220 * The existing block is smaller than we want, 221 * grow it. 222 */ 223 mutex_enter(&ump->um_lock); 224 error = ffs_realloccg(ip, lbn, 225 ffs_blkpref_ufs1(ip, lbn, (int)lbn, flags, 226 &ip->i_ffs1_db[0]), 227 osize, nsize, cred, bpp, &newb); 228 if (error) 229 return (error); 230 if (DOINGSOFTDEP(vp)) 231 softdep_setup_allocdirect(ip, lbn, 232 newb, nb, nsize, osize, 233 bpp ? *bpp : NULL); 234 } 235 } else { 236 237 /* 238 * the block was not previously allocated, 239 * allocate a new block or fragment. 240 */ 241 242 if (ip->i_size < lblktosize(fs, lbn + 1)) 243 nsize = fragroundup(fs, size); 244 else 245 nsize = fs->fs_bsize; 246 mutex_enter(&ump->um_lock); 247 error = ffs_alloc(ip, lbn, 248 ffs_blkpref_ufs1(ip, lbn, (int)lbn, flags, 249 &ip->i_ffs1_db[0]), 250 nsize, flags, cred, &newb); 251 if (error) 252 return (error); 253 if (bpp != NULL) { 254 error = ffs_getblk(vp, lbn, fsbtodb(fs, newb), 255 nsize, (flags & B_CLRBUF) != 0, bpp); 256 if (error) 257 return error; 258 } 259 if (DOINGSOFTDEP(vp)) { 260 softdep_setup_allocdirect(ip, lbn, newb, 0, 261 nsize, 0, bpp ? *bpp : NULL); 262 } 263 } 264 ip->i_ffs1_db[lbn] = ufs_rw32((u_int32_t)newb, needswap); 265 ip->i_flag |= IN_CHANGE | IN_UPDATE; 266 return (0); 267 } 268 269 /* 270 * Determine the number of levels of indirection. 271 */ 272 273 pref = 0; 274 if ((error = ufs_getlbns(vp, lbn, indirs, &num)) != 0) 275 return (error); 276 277 /* 278 * Fetch the first indirect block allocating if necessary. 279 */ 280 281 --num; 282 nb = ufs_rw32(ip->i_ffs1_ib[indirs[0].in_off], needswap); 283 allocib = NULL; 284 allocblk = allociblk; 285 if (nb == 0) { 286 mutex_enter(&ump->um_lock); 287 pref = ffs_blkpref_ufs1(ip, lbn, 0, flags | B_METAONLY, NULL); 288 error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, 289 flags | B_METAONLY, cred, &newb); 290 if (error) 291 goto fail; 292 nb = newb; 293 *allocblk++ = nb; 294 error = ffs_getblk(vp, indirs[1].in_lbn, fsbtodb(fs, nb), 295 fs->fs_bsize, true, &bp); 296 if (error) 297 goto fail; 298 if (DOINGSOFTDEP(vp)) { 299 softdep_setup_allocdirect(ip, NDADDR + indirs[0].in_off, 300 newb, 0, fs->fs_bsize, 0, bp); 301 bdwrite(bp); 302 } else { 303 304 /* 305 * Write synchronously so that indirect blocks 306 * never point at garbage. 307 */ 308 309 if ((error = bwrite(bp)) != 0) 310 goto fail; 311 } 312 unwindidx = 0; 313 allocib = &ip->i_ffs1_ib[indirs[0].in_off]; 314 *allocib = ufs_rw32(nb, needswap); 315 ip->i_flag |= IN_CHANGE | IN_UPDATE; 316 } 317 318 /* 319 * Fetch through the indirect blocks, allocating as necessary. 320 */ 321 322 for (i = 1;;) { 323 error = bread(vp, 324 indirs[i].in_lbn, (int)fs->fs_bsize, NOCRED, 0, &bp); 325 if (error) { 326 brelse(bp, 0); 327 goto fail; 328 } 329 bap = (int32_t *)bp->b_data; /* XXX ondisk32 */ 330 nb = ufs_rw32(bap[indirs[i].in_off], needswap); 331 if (i == num) 332 break; 333 i++; 334 if (nb != 0) { 335 brelse(bp, 0); 336 continue; 337 } 338 if (fscow_run(bp, true) != 0) { 339 brelse(bp, 0); 340 goto fail; 341 } 342 mutex_enter(&ump->um_lock); 343 if (pref == 0) 344 pref = ffs_blkpref_ufs1(ip, lbn, 0, flags | B_METAONLY, 345 NULL); 346 error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, 347 flags | B_METAONLY, cred, &newb); 348 if (error) { 349 brelse(bp, 0); 350 goto fail; 351 } 352 nb = newb; 353 *allocblk++ = nb; 354 error = ffs_getblk(vp, indirs[i].in_lbn, fsbtodb(fs, nb), 355 fs->fs_bsize, true, &nbp); 356 if (error) { 357 brelse(bp, 0); 358 goto fail; 359 } 360 if (DOINGSOFTDEP(vp)) { 361 softdep_setup_allocindir_meta(nbp, ip, bp, 362 indirs[i - 1].in_off, nb); 363 bdwrite(nbp); 364 } else { 365 366 /* 367 * Write synchronously so that indirect blocks 368 * never point at garbage. 369 */ 370 371 if ((error = bwrite(nbp)) != 0) { 372 brelse(bp, 0); 373 goto fail; 374 } 375 } 376 if (unwindidx < 0) 377 unwindidx = i - 1; 378 bap[indirs[i - 1].in_off] = ufs_rw32(nb, needswap); 379 380 /* 381 * If required, write synchronously, otherwise use 382 * delayed write. 383 */ 384 385 if (flags & B_SYNC) { 386 bwrite(bp); 387 } else { 388 bdwrite(bp); 389 } 390 } 391 392 if (flags & B_METAONLY) { 393 KASSERT(bpp != NULL); 394 *bpp = bp; 395 return (0); 396 } 397 398 /* 399 * Get the data block, allocating if necessary. 400 */ 401 402 if (nb == 0) { 403 if (fscow_run(bp, true) != 0) { 404 brelse(bp, 0); 405 goto fail; 406 } 407 mutex_enter(&ump->um_lock); 408 pref = ffs_blkpref_ufs1(ip, lbn, indirs[num].in_off, flags, 409 &bap[0]); 410 error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, flags, cred, 411 &newb); 412 if (error) { 413 brelse(bp, 0); 414 goto fail; 415 } 416 nb = newb; 417 *allocblk++ = nb; 418 if (bpp != NULL) { 419 error = ffs_getblk(vp, lbn, fsbtodb(fs, nb), 420 fs->fs_bsize, (flags & B_CLRBUF) != 0, bpp); 421 if (error) { 422 brelse(bp, 0); 423 goto fail; 424 } 425 } 426 if (DOINGSOFTDEP(vp)) 427 softdep_setup_allocindir_page(ip, lbn, bp, 428 indirs[num].in_off, nb, 0, bpp ? *bpp : NULL); 429 bap[indirs[num].in_off] = ufs_rw32(nb, needswap); 430 if (allocib == NULL && unwindidx < 0) { 431 unwindidx = i - 1; 432 } 433 434 /* 435 * If required, write synchronously, otherwise use 436 * delayed write. 437 */ 438 439 if (flags & B_SYNC) { 440 bwrite(bp); 441 } else { 442 bdwrite(bp); 443 } 444 return (0); 445 } 446 brelse(bp, 0); 447 if (bpp != NULL) { 448 if (flags & B_CLRBUF) { 449 error = bread(vp, lbn, (int)fs->fs_bsize, 450 NOCRED, B_MODIFY, &nbp); 451 if (error) { 452 brelse(nbp, 0); 453 goto fail; 454 } 455 } else { 456 error = ffs_getblk(vp, lbn, fsbtodb(fs, nb), 457 fs->fs_bsize, true, &nbp); 458 if (error) 459 goto fail; 460 } 461 *bpp = nbp; 462 } 463 return (0); 464 465 fail: 466 /* 467 * If we have failed part way through block allocation, we 468 * have to deallocate any indirect blocks that we have allocated. 469 */ 470 471 if (unwindidx >= 0) { 472 473 /* 474 * First write out any buffers we've created to resolve their 475 * softdeps. This must be done in reverse order of creation 476 * so that we resolve the dependencies in one pass. 477 * Write the cylinder group buffers for these buffers too. 478 */ 479 480 for (i = num; i >= unwindidx; i--) { 481 if (i == 0) { 482 break; 483 } 484 if (ffs_getblk(vp, indirs[i].in_lbn, FFS_NOBLK, 485 fs->fs_bsize, false, &bp) != 0) 486 continue; 487 if (bp->b_oflags & BO_DELWRI) { 488 nb = fsbtodb(fs, cgtod(fs, dtog(fs, 489 dbtofsb(fs, bp->b_blkno)))); 490 bwrite(bp); 491 if (ffs_getblk(ip->i_devvp, nb, FFS_NOBLK, 492 fs->fs_cgsize, false, &bp) != 0) 493 continue; 494 if (bp->b_oflags & BO_DELWRI) { 495 bwrite(bp); 496 } else { 497 brelse(bp, BC_INVAL); 498 } 499 } else { 500 brelse(bp, BC_INVAL); 501 } 502 } 503 504 /* Now flush all dependencies to disk. */ 505 #ifdef notyet 506 /* XXX pages locked */ 507 (void)softdep_sync_metadata(vp); 508 #endif 509 510 if (DOINGSOFTDEP(vp) && unwindidx == 0) { 511 ip->i_flag |= IN_CHANGE | IN_UPDATE; 512 ffs_update(vp, NULL, NULL, UPDATE_WAIT); 513 } 514 515 /* 516 * Now that any dependencies that we created have been 517 * resolved, we can undo the partial allocation. 518 */ 519 520 if (unwindidx == 0) { 521 *allocib = 0; 522 ip->i_flag |= IN_CHANGE | IN_UPDATE; 523 if (DOINGSOFTDEP(vp)) 524 ffs_update(vp, NULL, NULL, UPDATE_WAIT); 525 } else { 526 int r; 527 528 r = bread(vp, indirs[unwindidx].in_lbn, 529 (int)fs->fs_bsize, NOCRED, 0, &bp); 530 if (r) { 531 panic("Could not unwind indirect block, error %d", r); 532 brelse(bp, 0); 533 } else { 534 bap = (int32_t *)bp->b_data; /* XXX ondisk32 */ 535 bap[indirs[unwindidx].in_off] = 0; 536 bwrite(bp); 537 } 538 } 539 for (i = unwindidx + 1; i <= num; i++) { 540 if (ffs_getblk(vp, indirs[i].in_lbn, FFS_NOBLK, 541 fs->fs_bsize, false, &bp) == 0) 542 brelse(bp, BC_INVAL); 543 } 544 } 545 for (deallocated = 0, blkp = allociblk; blkp < allocblk; blkp++) { 546 ffs_blkfree(fs, ip->i_devvp, *blkp, fs->fs_bsize, ip->i_number); 547 deallocated += fs->fs_bsize; 548 } 549 if (deallocated) { 550 #ifdef QUOTA 551 /* 552 * Restore user's disk quota because allocation failed. 553 */ 554 (void)chkdq(ip, -btodb(deallocated), cred, FORCE); 555 #endif 556 ip->i_ffs1_blocks -= btodb(deallocated); 557 ip->i_flag |= IN_CHANGE | IN_UPDATE; 558 } 559 /* 560 * Flush all dependencies again so that the soft updates code 561 * doesn't find any untracked changes. 562 */ 563 #ifdef notyet 564 /* XXX pages locked */ 565 (void)softdep_sync_metadata(vp); 566 #endif 567 return (error); 568 } 569 570 static int 571 ffs_balloc_ufs2(struct vnode *vp, off_t off, int size, kauth_cred_t cred, 572 int flags, struct buf **bpp) 573 { 574 daddr_t lbn, lastlbn; 575 struct buf *bp, *nbp; 576 struct inode *ip = VTOI(vp); 577 struct fs *fs = ip->i_fs; 578 struct ufsmount *ump = ip->i_ump; 579 struct indir indirs[NIADDR + 2]; 580 daddr_t newb, pref, nb; 581 int64_t *bap; 582 int deallocated, osize, nsize, num, i, error; 583 daddr_t *blkp, *allocblk, allociblk[NIADDR + 1]; 584 int64_t *allocib; 585 int unwindidx = -1; 586 #ifdef FFS_EI 587 const int needswap = UFS_FSNEEDSWAP(fs); 588 #endif 589 UVMHIST_FUNC("ffs_balloc"); UVMHIST_CALLED(ubchist); 590 591 lbn = lblkno(fs, off); 592 size = blkoff(fs, off) + size; 593 if (size > fs->fs_bsize) 594 panic("ffs_balloc: blk too big"); 595 if (bpp != NULL) { 596 *bpp = NULL; 597 } 598 UVMHIST_LOG(ubchist, "vp %p lbn 0x%x size 0x%x", vp, lbn, size,0); 599 600 if (lbn < 0) 601 return (EFBIG); 602 603 #ifdef notyet 604 /* 605 * Check for allocating external data. 606 */ 607 if (flags & IO_EXT) { 608 if (lbn >= NXADDR) 609 return (EFBIG); 610 /* 611 * If the next write will extend the data into a new block, 612 * and the data is currently composed of a fragment 613 * this fragment has to be extended to be a full block. 614 */ 615 lastlbn = lblkno(fs, dp->di_extsize); 616 if (lastlbn < lbn) { 617 nb = lastlbn; 618 osize = sblksize(fs, dp->di_extsize, nb); 619 if (osize < fs->fs_bsize && osize > 0) { 620 mutex_enter(&ump->um_lock); 621 error = ffs_realloccg(ip, -1 - nb, 622 dp->di_extb[nb], 623 ffs_blkpref_ufs2(ip, lastlbn, (int)nb, 624 flags, &dp->di_extb[0]), 625 osize, 626 (int)fs->fs_bsize, cred, &bp); 627 if (error) 628 return (error); 629 if (DOINGSOFTDEP(vp)) 630 softdep_setup_allocext(ip, nb, 631 dbtofsb(fs, bp->b_blkno), 632 dp->di_extb[nb], 633 fs->fs_bsize, osize, bp); 634 dp->di_extsize = smalllblktosize(fs, nb + 1); 635 dp->di_extb[nb] = dbtofsb(fs, bp->b_blkno); 636 bp->b_xflags |= BX_ALTDATA; 637 ip->i_flag |= IN_CHANGE | IN_UPDATE; 638 if (flags & IO_SYNC) 639 bwrite(bp); 640 else 641 bawrite(bp); 642 } 643 } 644 /* 645 * All blocks are direct blocks 646 */ 647 if (flags & BA_METAONLY) 648 panic("ffs_balloc_ufs2: BA_METAONLY for ext block"); 649 nb = dp->di_extb[lbn]; 650 if (nb != 0 && dp->di_extsize >= smalllblktosize(fs, lbn + 1)) { 651 error = bread(vp, -1 - lbn, fs->fs_bsize, 652 NOCRED, 0, &bp); 653 if (error) { 654 brelse(bp, 0); 655 return (error); 656 } 657 mutex_enter(&bp->b_interlock); 658 bp->b_blkno = fsbtodb(fs, nb); 659 bp->b_xflags |= BX_ALTDATA; 660 mutex_exit(&bp->b_interlock); 661 *bpp = bp; 662 return (0); 663 } 664 if (nb != 0) { 665 /* 666 * Consider need to reallocate a fragment. 667 */ 668 osize = fragroundup(fs, blkoff(fs, dp->di_extsize)); 669 nsize = fragroundup(fs, size); 670 if (nsize <= osize) { 671 error = bread(vp, -1 - lbn, osize, 672 NOCRED, 0, &bp); 673 if (error) { 674 brelse(bp, 0); 675 return (error); 676 } 677 mutex_enter(&bp->b_interlock); 678 bp->b_blkno = fsbtodb(fs, nb); 679 bp->b_xflags |= BX_ALTDATA; 680 mutex_exit(&bp->b_interlock); 681 } else { 682 mutex_enter(&ump->um_lock); 683 error = ffs_realloccg(ip, -1 - lbn, 684 dp->di_extb[lbn], 685 ffs_blkpref_ufs2(ip, lbn, (int)lbn, flags, 686 &dp->di_extb[0]), 687 osize, nsize, cred, &bp); 688 if (error) 689 return (error); 690 bp->b_xflags |= BX_ALTDATA; 691 if (DOINGSOFTDEP(vp)) 692 softdep_setup_allocext(ip, lbn, 693 dbtofsb(fs, bp->b_blkno), nb, 694 nsize, osize, bp); 695 } 696 } else { 697 if (dp->di_extsize < smalllblktosize(fs, lbn + 1)) 698 nsize = fragroundup(fs, size); 699 else 700 nsize = fs->fs_bsize; 701 mutex_enter(&ump->um_lock); 702 error = ffs_alloc(ip, lbn, 703 ffs_blkpref_ufs2(ip, lbn, (int)lbn, flags, 704 &dp->di_extb[0]), 705 nsize, flags, cred, &newb); 706 if (error) 707 return (error); 708 error = ffs_getblk(vp, -1 - lbn, fsbtodb(fs, newb), 709 nsize, (flags & BA_CLRBUF) != 0, &bp); 710 if (error) 711 return error; 712 bp->b_xflags |= BX_ALTDATA; 713 if (DOINGSOFTDEP(vp)) 714 softdep_setup_allocext(ip, lbn, newb, 0, 715 nsize, 0, bp); 716 } 717 dp->di_extb[lbn] = dbtofsb(fs, bp->b_blkno); 718 ip->i_flag |= IN_CHANGE | IN_UPDATE; 719 *bpp = bp; 720 return (0); 721 } 722 #endif 723 /* 724 * If the next write will extend the file into a new block, 725 * and the file is currently composed of a fragment 726 * this fragment has to be extended to be a full block. 727 */ 728 729 lastlbn = lblkno(fs, ip->i_size); 730 if (lastlbn < NDADDR && lastlbn < lbn) { 731 nb = lastlbn; 732 osize = blksize(fs, ip, nb); 733 if (osize < fs->fs_bsize && osize > 0) { 734 mutex_enter(&ump->um_lock); 735 error = ffs_realloccg(ip, nb, 736 ffs_blkpref_ufs2(ip, lastlbn, nb, flags, 737 &ip->i_ffs2_db[0]), 738 osize, (int)fs->fs_bsize, cred, bpp, &newb); 739 if (error) 740 return (error); 741 if (DOINGSOFTDEP(vp)) 742 softdep_setup_allocdirect(ip, nb, newb, 743 ufs_rw64(ip->i_ffs2_db[nb], needswap), 744 fs->fs_bsize, osize, bpp ? *bpp : NULL); 745 ip->i_size = lblktosize(fs, nb + 1); 746 ip->i_ffs2_size = ip->i_size; 747 uvm_vnp_setsize(vp, ip->i_size); 748 ip->i_ffs2_db[nb] = ufs_rw64(newb, needswap); 749 ip->i_flag |= IN_CHANGE | IN_UPDATE; 750 if (bpp) { 751 if (flags & B_SYNC) 752 bwrite(*bpp); 753 else 754 bawrite(*bpp); 755 } 756 } 757 } 758 759 /* 760 * The first NDADDR blocks are direct blocks 761 */ 762 763 if (lbn < NDADDR) { 764 nb = ufs_rw64(ip->i_ffs2_db[lbn], needswap); 765 if (nb != 0 && ip->i_size >= lblktosize(fs, lbn + 1)) { 766 767 /* 768 * The block is an already-allocated direct block 769 * and the file already extends past this block, 770 * thus this must be a whole block. 771 * Just read the block (if requested). 772 */ 773 774 if (bpp != NULL) { 775 error = bread(vp, lbn, fs->fs_bsize, NOCRED, 776 B_MODIFY, bpp); 777 if (error) { 778 brelse(*bpp, 0); 779 return (error); 780 } 781 } 782 return (0); 783 } 784 if (nb != 0) { 785 786 /* 787 * Consider need to reallocate a fragment. 788 */ 789 790 osize = fragroundup(fs, blkoff(fs, ip->i_size)); 791 nsize = fragroundup(fs, size); 792 if (nsize <= osize) { 793 794 /* 795 * The existing block is already 796 * at least as big as we want. 797 * Just read the block (if requested). 798 */ 799 800 if (bpp != NULL) { 801 error = bread(vp, lbn, osize, NOCRED, 802 B_MODIFY, bpp); 803 if (error) { 804 brelse(*bpp, 0); 805 return (error); 806 } 807 } 808 return 0; 809 } else { 810 811 /* 812 * The existing block is smaller than we want, 813 * grow it. 814 */ 815 mutex_enter(&ump->um_lock); 816 error = ffs_realloccg(ip, lbn, 817 ffs_blkpref_ufs2(ip, lbn, (int)lbn, flags, 818 &ip->i_ffs2_db[0]), 819 osize, nsize, cred, bpp, &newb); 820 if (error) 821 return (error); 822 if (DOINGSOFTDEP(vp)) 823 softdep_setup_allocdirect(ip, lbn, 824 newb, nb, nsize, osize, 825 bpp ? *bpp : NULL); 826 } 827 } else { 828 829 /* 830 * the block was not previously allocated, 831 * allocate a new block or fragment. 832 */ 833 834 if (ip->i_size < lblktosize(fs, lbn + 1)) 835 nsize = fragroundup(fs, size); 836 else 837 nsize = fs->fs_bsize; 838 mutex_enter(&ump->um_lock); 839 error = ffs_alloc(ip, lbn, 840 ffs_blkpref_ufs2(ip, lbn, (int)lbn, flags, 841 &ip->i_ffs2_db[0]), 842 nsize, flags, cred, &newb); 843 if (error) 844 return (error); 845 if (bpp != NULL) { 846 error = ffs_getblk(vp, lbn, fsbtodb(fs, newb), 847 nsize, (flags & B_CLRBUF) != 0, bpp); 848 if (error) 849 return error; 850 } 851 if (DOINGSOFTDEP(vp)) { 852 softdep_setup_allocdirect(ip, lbn, newb, 0, 853 nsize, 0, bpp ? *bpp : NULL); 854 } 855 } 856 ip->i_ffs2_db[lbn] = ufs_rw64(newb, needswap); 857 ip->i_flag |= IN_CHANGE | IN_UPDATE; 858 return (0); 859 } 860 861 /* 862 * Determine the number of levels of indirection. 863 */ 864 865 pref = 0; 866 if ((error = ufs_getlbns(vp, lbn, indirs, &num)) != 0) 867 return (error); 868 869 /* 870 * Fetch the first indirect block allocating if necessary. 871 */ 872 873 --num; 874 nb = ufs_rw64(ip->i_ffs2_ib[indirs[0].in_off], needswap); 875 allocib = NULL; 876 allocblk = allociblk; 877 if (nb == 0) { 878 mutex_enter(&ump->um_lock); 879 pref = ffs_blkpref_ufs2(ip, lbn, 0, flags | B_METAONLY, NULL); 880 error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, 881 flags | B_METAONLY, cred, &newb); 882 if (error) 883 goto fail; 884 nb = newb; 885 *allocblk++ = nb; 886 error = ffs_getblk(vp, indirs[1].in_lbn, fsbtodb(fs, nb), 887 fs->fs_bsize, true, &bp); 888 if (error) 889 goto fail; 890 if (DOINGSOFTDEP(vp)) { 891 softdep_setup_allocdirect(ip, NDADDR + indirs[0].in_off, 892 newb, 0, fs->fs_bsize, 0, bp); 893 bdwrite(bp); 894 } else { 895 896 /* 897 * Write synchronously so that indirect blocks 898 * never point at garbage. 899 */ 900 901 if ((error = bwrite(bp)) != 0) 902 goto fail; 903 } 904 unwindidx = 0; 905 allocib = &ip->i_ffs2_ib[indirs[0].in_off]; 906 *allocib = ufs_rw64(nb, needswap); 907 ip->i_flag |= IN_CHANGE | IN_UPDATE; 908 } 909 910 /* 911 * Fetch through the indirect blocks, allocating as necessary. 912 */ 913 914 for (i = 1;;) { 915 error = bread(vp, 916 indirs[i].in_lbn, (int)fs->fs_bsize, NOCRED, 0, &bp); 917 if (error) { 918 brelse(bp, 0); 919 goto fail; 920 } 921 bap = (int64_t *)bp->b_data; 922 nb = ufs_rw64(bap[indirs[i].in_off], needswap); 923 if (i == num) 924 break; 925 i++; 926 if (nb != 0) { 927 brelse(bp, 0); 928 continue; 929 } 930 if (fscow_run(bp, true) != 0) { 931 brelse(bp, 0); 932 goto fail; 933 } 934 mutex_enter(&ump->um_lock); 935 if (pref == 0) 936 pref = ffs_blkpref_ufs2(ip, lbn, 0, flags | B_METAONLY, 937 NULL); 938 error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, 939 flags | B_METAONLY, cred, &newb); 940 if (error) { 941 brelse(bp, 0); 942 goto fail; 943 } 944 nb = newb; 945 *allocblk++ = nb; 946 error = ffs_getblk(vp, indirs[i].in_lbn, fsbtodb(fs, nb), 947 fs->fs_bsize, true, &nbp); 948 if (error) { 949 brelse(bp, 0); 950 goto fail; 951 } 952 if (DOINGSOFTDEP(vp)) { 953 softdep_setup_allocindir_meta(nbp, ip, bp, 954 indirs[i - 1].in_off, nb); 955 bdwrite(nbp); 956 } else { 957 958 /* 959 * Write synchronously so that indirect blocks 960 * never point at garbage. 961 */ 962 963 if ((error = bwrite(nbp)) != 0) { 964 brelse(bp, 0); 965 goto fail; 966 } 967 } 968 if (unwindidx < 0) 969 unwindidx = i - 1; 970 bap[indirs[i - 1].in_off] = ufs_rw64(nb, needswap); 971 972 /* 973 * If required, write synchronously, otherwise use 974 * delayed write. 975 */ 976 977 if (flags & B_SYNC) { 978 bwrite(bp); 979 } else { 980 bdwrite(bp); 981 } 982 } 983 984 if (flags & B_METAONLY) { 985 KASSERT(bpp != NULL); 986 *bpp = bp; 987 return (0); 988 } 989 990 /* 991 * Get the data block, allocating if necessary. 992 */ 993 994 if (nb == 0) { 995 if (fscow_run(bp, true) != 0) { 996 brelse(bp, 0); 997 goto fail; 998 } 999 mutex_enter(&ump->um_lock); 1000 pref = ffs_blkpref_ufs2(ip, lbn, indirs[num].in_off, flags, 1001 &bap[0]); 1002 error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, flags, cred, 1003 &newb); 1004 if (error) { 1005 brelse(bp, 0); 1006 goto fail; 1007 } 1008 nb = newb; 1009 *allocblk++ = nb; 1010 if (bpp != NULL) { 1011 error = ffs_getblk(vp, lbn, fsbtodb(fs, nb), 1012 fs->fs_bsize, (flags & B_CLRBUF) != 0, bpp); 1013 if (error) { 1014 brelse(bp, 0); 1015 goto fail; 1016 } 1017 } 1018 if (DOINGSOFTDEP(vp)) 1019 softdep_setup_allocindir_page(ip, lbn, bp, 1020 indirs[num].in_off, nb, 0, bpp ? *bpp : NULL); 1021 bap[indirs[num].in_off] = ufs_rw64(nb, needswap); 1022 if (allocib == NULL && unwindidx < 0) { 1023 unwindidx = i - 1; 1024 } 1025 1026 /* 1027 * If required, write synchronously, otherwise use 1028 * delayed write. 1029 */ 1030 1031 if (flags & B_SYNC) { 1032 bwrite(bp); 1033 } else { 1034 bdwrite(bp); 1035 } 1036 return (0); 1037 } 1038 brelse(bp, 0); 1039 if (bpp != NULL) { 1040 if (flags & B_CLRBUF) { 1041 error = bread(vp, lbn, (int)fs->fs_bsize, 1042 NOCRED, B_MODIFY, &nbp); 1043 if (error) { 1044 brelse(nbp, 0); 1045 goto fail; 1046 } 1047 } else { 1048 error = ffs_getblk(vp, lbn, fsbtodb(fs, nb), 1049 fs->fs_bsize, true, &nbp); 1050 if (error) 1051 goto fail; 1052 } 1053 *bpp = nbp; 1054 } 1055 return (0); 1056 1057 fail: 1058 /* 1059 * If we have failed part way through block allocation, we 1060 * have to deallocate any indirect blocks that we have allocated. 1061 */ 1062 1063 if (unwindidx >= 0) { 1064 1065 /* 1066 * First write out any buffers we've created to resolve their 1067 * softdeps. This must be done in reverse order of creation 1068 * so that we resolve the dependencies in one pass. 1069 * Write the cylinder group buffers for these buffers too. 1070 */ 1071 1072 for (i = num; i >= unwindidx; i--) { 1073 if (i == 0) { 1074 break; 1075 } 1076 if (ffs_getblk(vp, indirs[i].in_lbn, FFS_NOBLK, 1077 fs->fs_bsize, false, &bp) != 0) 1078 continue; 1079 if (bp->b_oflags & BO_DELWRI) { 1080 nb = fsbtodb(fs, cgtod(fs, dtog(fs, 1081 dbtofsb(fs, bp->b_blkno)))); 1082 bwrite(bp); 1083 if (ffs_getblk(ip->i_devvp, nb, FFS_NOBLK, 1084 fs->fs_cgsize, false, &bp) != 0) 1085 continue; 1086 if (bp->b_oflags & BO_DELWRI) { 1087 bwrite(bp); 1088 } else { 1089 brelse(bp, BC_INVAL); 1090 } 1091 } else { 1092 brelse(bp, BC_INVAL); 1093 } 1094 } 1095 1096 /* Now flush the dependencies to disk. */ 1097 #ifdef notyet 1098 /* XXX pages locked */ 1099 (void)softdep_sync_metadata(vp); 1100 #endif 1101 1102 if (DOINGSOFTDEP(vp) && unwindidx == 0) { 1103 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1104 ffs_update(vp, NULL, NULL, UPDATE_WAIT); 1105 } 1106 1107 /* 1108 * Now that any dependencies that we created have been 1109 * resolved, we can undo the partial allocation. 1110 */ 1111 1112 if (unwindidx == 0) { 1113 *allocib = 0; 1114 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1115 if (DOINGSOFTDEP(vp)) 1116 ffs_update(vp, NULL, NULL, UPDATE_WAIT); 1117 } else { 1118 int r; 1119 1120 r = bread(vp, indirs[unwindidx].in_lbn, 1121 (int)fs->fs_bsize, NOCRED, 0, &bp); 1122 if (r) { 1123 panic("Could not unwind indirect block, error %d", r); 1124 brelse(bp, 0); 1125 } else { 1126 bap = (int64_t *)bp->b_data; 1127 bap[indirs[unwindidx].in_off] = 0; 1128 bwrite(bp); 1129 } 1130 } 1131 for (i = unwindidx + 1; i <= num; i++) { 1132 if (ffs_getblk(vp, indirs[i].in_lbn, FFS_NOBLK, 1133 fs->fs_bsize, false, &bp) == 0) 1134 brelse(bp, BC_INVAL); 1135 } 1136 } 1137 for (deallocated = 0, blkp = allociblk; blkp < allocblk; blkp++) { 1138 ffs_blkfree(fs, ip->i_devvp, *blkp, fs->fs_bsize, ip->i_number); 1139 deallocated += fs->fs_bsize; 1140 } 1141 if (deallocated) { 1142 #ifdef QUOTA 1143 /* 1144 * Restore user's disk quota because allocation failed. 1145 */ 1146 (void)chkdq(ip, -btodb(deallocated), cred, FORCE); 1147 #endif 1148 ip->i_ffs2_blocks -= btodb(deallocated); 1149 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1150 } 1151 1152 /* 1153 * Flush all dependencies again so that the soft updates code 1154 * doesn't find any untracked changes. 1155 */ 1156 #ifdef notyet 1157 /* XXX pages locked */ 1158 (void)softdep_sync_metadata(vp); 1159 #endif 1160 return (error); 1161 } 1162
Indexes created Mon Oct 13 14:09:54 GMT 2025