ffs_balloc.c revision 304668
1/*- 2 * Copyright (c) 2002 Networks Associates Technology, Inc. 3 * All rights reserved. 4 * 5 * This software was developed for the FreeBSD Project by Marshall 6 * Kirk McKusick and Network Associates Laboratories, the Security 7 * Research Division of Network Associates, Inc. under DARPA/SPAWAR 8 * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS 9 * research program 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 * 32 * Copyright (c) 1982, 1986, 1989, 1993 33 * The Regents of the University of California. All rights reserved. 34 * 35 * Redistribution and use in source and binary forms, with or without 36 * modification, are permitted provided that the following conditions 37 * are met: 38 * 1. Redistributions of source code must retain the above copyright 39 * notice, this list of conditions and the following disclaimer. 40 * 2. Redistributions in binary form must reproduce the above copyright 41 * notice, this list of conditions and the following disclaimer in the 42 * documentation and/or other materials provided with the distribution. 43 * 4. Neither the name of the University nor the names of its contributors 44 * may be used to endorse or promote products derived from this software 45 * without specific prior written permission. 46 * 47 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 48 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 49 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 50 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 51 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 52 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 53 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 54 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 55 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 56 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 57 * SUCH DAMAGE. 58 * 59 * @(#)ffs_balloc.c 8.8 (Berkeley) 6/16/95 60 */ 61 62#include <sys/cdefs.h> 63__FBSDID("$FreeBSD: stable/10/sys/ufs/ffs/ffs_balloc.c 304668 2016-08-23 07:52:07Z kib $"); 64 65#include <sys/param.h> 66#include <sys/systm.h> 67#include <sys/bio.h> 68#include <sys/buf.h> 69#include <sys/lock.h> 70#include <sys/mount.h> 71#include <sys/vnode.h> 72 73#include <ufs/ufs/quota.h> 74#include <ufs/ufs/inode.h> 75#include <ufs/ufs/ufs_extern.h> 76#include <ufs/ufs/extattr.h> 77#include <ufs/ufs/ufsmount.h> 78 79#include <ufs/ffs/fs.h> 80#include <ufs/ffs/ffs_extern.h> 81 82/* 83 * Balloc defines the structure of filesystem storage 84 * by allocating the physical blocks on a device given 85 * the inode and the logical block number in a file. 86 * This is the allocation strategy for UFS1. Below is 87 * the allocation strategy for UFS2. 88 */ 89int 90ffs_balloc_ufs1(struct vnode *vp, off_t startoffset, int size, 91 struct ucred *cred, int flags, struct buf **bpp) 92{ 93 struct inode *ip; 94 struct ufs1_dinode *dp; 95 ufs_lbn_t lbn, lastlbn; 96 struct fs *fs; 97 ufs1_daddr_t nb; 98 struct buf *bp, *nbp; 99 struct ufsmount *ump; 100 struct indir indirs[NIADDR + 2]; 101 int deallocated, osize, nsize, num, i, error; 102 ufs2_daddr_t newb; 103 ufs1_daddr_t *bap, pref; 104 ufs1_daddr_t *allocib, *blkp, *allocblk, allociblk[NIADDR + 1]; 105 ufs2_daddr_t *lbns_remfree, lbns[NIADDR + 1]; 106 int unwindidx = -1; 107 int saved_inbdflush; 108 static struct timeval lastfail; 109 static int curfail; 110 int gbflags, reclaimed; 111 112 ip = VTOI(vp); 113 dp = ip->i_din1; 114 fs = ip->i_fs; 115 ump = ip->i_ump; 116 lbn = lblkno(fs, startoffset); 117 size = blkoff(fs, startoffset) + size; 118 reclaimed = 0; 119 if (size > fs->fs_bsize) 120 panic("ffs_balloc_ufs1: blk too big"); 121 *bpp = NULL; 122 if (flags & IO_EXT) 123 return (EOPNOTSUPP); 124 if (lbn < 0) 125 return (EFBIG); 126 gbflags = (flags & BA_UNMAPPED) != 0 ? GB_UNMAPPED : 0; 127 128 if (DOINGSOFTDEP(vp)) 129 softdep_prealloc(vp, MNT_WAIT); 130 /* 131 * If the next write will extend the file into a new block, 132 * and the file is currently composed of a fragment 133 * this fragment has to be extended to be a full block. 134 */ 135 lastlbn = lblkno(fs, ip->i_size); 136 if (lastlbn < NDADDR && lastlbn < lbn) { 137 nb = lastlbn; 138 osize = blksize(fs, ip, nb); 139 if (osize < fs->fs_bsize && osize > 0) { 140 UFS_LOCK(ump); 141 error = ffs_realloccg(ip, nb, dp->di_db[nb], 142 ffs_blkpref_ufs1(ip, lastlbn, (int)nb, 143 &dp->di_db[0]), osize, (int)fs->fs_bsize, flags, 144 cred, &bp); 145 if (error) 146 return (error); 147 if (DOINGSOFTDEP(vp)) 148 softdep_setup_allocdirect(ip, nb, 149 dbtofsb(fs, bp->b_blkno), dp->di_db[nb], 150 fs->fs_bsize, osize, bp); 151 ip->i_size = smalllblktosize(fs, nb + 1); 152 dp->di_size = ip->i_size; 153 dp->di_db[nb] = dbtofsb(fs, bp->b_blkno); 154 ip->i_flag |= IN_CHANGE | IN_UPDATE; 155 if (flags & IO_SYNC) 156 bwrite(bp); 157 else 158 bawrite(bp); 159 } 160 } 161 /* 162 * The first NDADDR blocks are direct blocks 163 */ 164 if (lbn < NDADDR) { 165 if (flags & BA_METAONLY) 166 panic("ffs_balloc_ufs1: BA_METAONLY for direct block"); 167 nb = dp->di_db[lbn]; 168 if (nb != 0 && ip->i_size >= smalllblktosize(fs, lbn + 1)) { 169 error = bread(vp, lbn, fs->fs_bsize, NOCRED, &bp); 170 if (error) { 171 brelse(bp); 172 return (error); 173 } 174 bp->b_blkno = fsbtodb(fs, nb); 175 *bpp = bp; 176 return (0); 177 } 178 if (nb != 0) { 179 /* 180 * Consider need to reallocate a fragment. 181 */ 182 osize = fragroundup(fs, blkoff(fs, ip->i_size)); 183 nsize = fragroundup(fs, size); 184 if (nsize <= osize) { 185 error = bread(vp, lbn, osize, NOCRED, &bp); 186 if (error) { 187 brelse(bp); 188 return (error); 189 } 190 bp->b_blkno = fsbtodb(fs, nb); 191 } else { 192 UFS_LOCK(ump); 193 error = ffs_realloccg(ip, lbn, dp->di_db[lbn], 194 ffs_blkpref_ufs1(ip, lbn, (int)lbn, 195 &dp->di_db[0]), osize, nsize, flags, 196 cred, &bp); 197 if (error) 198 return (error); 199 if (DOINGSOFTDEP(vp)) 200 softdep_setup_allocdirect(ip, lbn, 201 dbtofsb(fs, bp->b_blkno), nb, 202 nsize, osize, bp); 203 } 204 } else { 205 if (ip->i_size < smalllblktosize(fs, lbn + 1)) 206 nsize = fragroundup(fs, size); 207 else 208 nsize = fs->fs_bsize; 209 UFS_LOCK(ump); 210 error = ffs_alloc(ip, lbn, 211 ffs_blkpref_ufs1(ip, lbn, (int)lbn, &dp->di_db[0]), 212 nsize, flags, cred, &newb); 213 if (error) 214 return (error); 215 bp = getblk(vp, lbn, nsize, 0, 0, gbflags); 216 bp->b_blkno = fsbtodb(fs, newb); 217 if (flags & BA_CLRBUF) 218 vfs_bio_clrbuf(bp); 219 if (DOINGSOFTDEP(vp)) 220 softdep_setup_allocdirect(ip, lbn, newb, 0, 221 nsize, 0, bp); 222 } 223 dp->di_db[lbn] = dbtofsb(fs, bp->b_blkno); 224 ip->i_flag |= IN_CHANGE | IN_UPDATE; 225 *bpp = bp; 226 return (0); 227 } 228 /* 229 * Determine the number of levels of indirection. 230 */ 231 pref = 0; 232 if ((error = ufs_getlbns(vp, lbn, indirs, &num)) != 0) 233 return(error); 234#ifdef INVARIANTS 235 if (num < 1) 236 panic ("ffs_balloc_ufs1: ufs_getlbns returned indirect block"); 237#endif 238 saved_inbdflush = curthread_pflags_set(TDP_INBDFLUSH); 239 /* 240 * Fetch the first indirect block allocating if necessary. 241 */ 242 --num; 243 nb = dp->di_ib[indirs[0].in_off]; 244 allocib = NULL; 245 allocblk = allociblk; 246 lbns_remfree = lbns; 247 if (nb == 0) { 248 UFS_LOCK(ump); 249 pref = ffs_blkpref_ufs1(ip, lbn, -indirs[0].in_off - 1, 250 (ufs1_daddr_t *)0); 251 if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, 252 flags, cred, &newb)) != 0) { 253 curthread_pflags_restore(saved_inbdflush); 254 return (error); 255 } 256 pref = newb + fs->fs_frag; 257 nb = newb; 258 MPASS(allocblk < allociblk + nitems(allociblk)); 259 MPASS(lbns_remfree < lbns + nitems(lbns)); 260 *allocblk++ = nb; 261 *lbns_remfree++ = indirs[1].in_lbn; 262 bp = getblk(vp, indirs[1].in_lbn, fs->fs_bsize, 0, 0, gbflags); 263 bp->b_blkno = fsbtodb(fs, nb); 264 vfs_bio_clrbuf(bp); 265 if (DOINGSOFTDEP(vp)) { 266 softdep_setup_allocdirect(ip, NDADDR + indirs[0].in_off, 267 newb, 0, fs->fs_bsize, 0, bp); 268 bdwrite(bp); 269 } else { 270 /* 271 * Write synchronously so that indirect blocks 272 * never point at garbage. 273 */ 274 if (DOINGASYNC(vp)) 275 bdwrite(bp); 276 else if ((error = bwrite(bp)) != 0) 277 goto fail; 278 } 279 allocib = &dp->di_ib[indirs[0].in_off]; 280 *allocib = nb; 281 ip->i_flag |= IN_CHANGE | IN_UPDATE; 282 } 283 /* 284 * Fetch through the indirect blocks, allocating as necessary. 285 */ 286retry: 287 for (i = 1;;) { 288 error = bread(vp, 289 indirs[i].in_lbn, (int)fs->fs_bsize, NOCRED, &bp); 290 if (error) { 291 brelse(bp); 292 goto fail; 293 } 294 bap = (ufs1_daddr_t *)bp->b_data; 295 nb = bap[indirs[i].in_off]; 296 if (i == num) 297 break; 298 i += 1; 299 if (nb != 0) { 300 bqrelse(bp); 301 continue; 302 } 303 UFS_LOCK(ump); 304 /* 305 * If parent indirect has just been allocated, try to cluster 306 * immediately following it. 307 */ 308 if (pref == 0) 309 pref = ffs_blkpref_ufs1(ip, lbn, i - num - 1, 310 (ufs1_daddr_t *)0); 311 if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, 312 flags | IO_BUFLOCKED, cred, &newb)) != 0) { 313 brelse(bp); 314 if (DOINGSOFTDEP(vp) && ++reclaimed == 1) { 315 UFS_LOCK(ump); 316 softdep_request_cleanup(fs, vp, cred, 317 FLUSH_BLOCKS_WAIT); 318 UFS_UNLOCK(ump); 319 goto retry; 320 } 321 if (ppsratecheck(&lastfail, &curfail, 1)) { 322 ffs_fserr(fs, ip->i_number, "filesystem full"); 323 uprintf("\n%s: write failed, filesystem " 324 "is full\n", fs->fs_fsmnt); 325 } 326 goto fail; 327 } 328 pref = newb + fs->fs_frag; 329 nb = newb; 330 MPASS(allocblk < allociblk + nitems(allociblk)); 331 MPASS(lbns_remfree < lbns + nitems(lbns)); 332 *allocblk++ = nb; 333 *lbns_remfree++ = indirs[i].in_lbn; 334 nbp = getblk(vp, indirs[i].in_lbn, fs->fs_bsize, 0, 0, 0); 335 nbp->b_blkno = fsbtodb(fs, nb); 336 vfs_bio_clrbuf(nbp); 337 if (DOINGSOFTDEP(vp)) { 338 softdep_setup_allocindir_meta(nbp, ip, bp, 339 indirs[i - 1].in_off, nb); 340 bdwrite(nbp); 341 } else { 342 /* 343 * Write synchronously so that indirect blocks 344 * never point at garbage. 345 */ 346 if ((error = bwrite(nbp)) != 0) { 347 brelse(bp); 348 goto fail; 349 } 350 } 351 bap[indirs[i - 1].in_off] = nb; 352 if (allocib == NULL && unwindidx < 0) 353 unwindidx = i - 1; 354 /* 355 * If required, write synchronously, otherwise use 356 * delayed write. 357 */ 358 if (flags & IO_SYNC) { 359 bwrite(bp); 360 } else { 361 if (bp->b_bufsize == fs->fs_bsize) 362 bp->b_flags |= B_CLUSTEROK; 363 bdwrite(bp); 364 } 365 } 366 /* 367 * If asked only for the indirect block, then return it. 368 */ 369 if (flags & BA_METAONLY) { 370 curthread_pflags_restore(saved_inbdflush); 371 *bpp = bp; 372 return (0); 373 } 374 /* 375 * Get the data block, allocating if necessary. 376 */ 377 if (nb == 0) { 378 UFS_LOCK(ump); 379 /* 380 * If allocating metadata at the front of the cylinder 381 * group and parent indirect block has just been allocated, 382 * then cluster next to it if it is the first indirect in 383 * the file. Otherwise it has been allocated in the metadata 384 * area, so we want to find our own place out in the data area. 385 */ 386 if (pref == 0 || (lbn > NDADDR && fs->fs_metaspace != 0)) 387 pref = ffs_blkpref_ufs1(ip, lbn, indirs[i].in_off, 388 &bap[0]); 389 error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, 390 flags | IO_BUFLOCKED, cred, &newb); 391 if (error) { 392 brelse(bp); 393 if (DOINGSOFTDEP(vp) && ++reclaimed == 1) { 394 UFS_LOCK(ump); 395 softdep_request_cleanup(fs, vp, cred, 396 FLUSH_BLOCKS_WAIT); 397 UFS_UNLOCK(ump); 398 goto retry; 399 } 400 if (ppsratecheck(&lastfail, &curfail, 1)) { 401 ffs_fserr(fs, ip->i_number, "filesystem full"); 402 uprintf("\n%s: write failed, filesystem " 403 "is full\n", fs->fs_fsmnt); 404 } 405 goto fail; 406 } 407 nb = newb; 408 MPASS(allocblk < allociblk + nitems(allociblk)); 409 MPASS(lbns_remfree < lbns + nitems(lbns)); 410 *allocblk++ = nb; 411 *lbns_remfree++ = lbn; 412 nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0, gbflags); 413 nbp->b_blkno = fsbtodb(fs, nb); 414 if (flags & BA_CLRBUF) 415 vfs_bio_clrbuf(nbp); 416 if (DOINGSOFTDEP(vp)) 417 softdep_setup_allocindir_page(ip, lbn, bp, 418 indirs[i].in_off, nb, 0, nbp); 419 bap[indirs[i].in_off] = nb; 420 /* 421 * If required, write synchronously, otherwise use 422 * delayed write. 423 */ 424 if (flags & IO_SYNC) { 425 bwrite(bp); 426 } else { 427 if (bp->b_bufsize == fs->fs_bsize) 428 bp->b_flags |= B_CLUSTEROK; 429 bdwrite(bp); 430 } 431 curthread_pflags_restore(saved_inbdflush); 432 *bpp = nbp; 433 return (0); 434 } 435 brelse(bp); 436 if (flags & BA_CLRBUF) { 437 int seqcount = (flags & BA_SEQMASK) >> BA_SEQSHIFT; 438 if (seqcount != 0 && 439 (vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0 && 440 !(vm_page_count_severe() || buf_dirty_count_severe())) { 441 error = cluster_read(vp, ip->i_size, lbn, 442 (int)fs->fs_bsize, NOCRED, 443 MAXBSIZE, seqcount, gbflags, &nbp); 444 } else { 445 error = bread_gb(vp, lbn, (int)fs->fs_bsize, NOCRED, 446 gbflags, &nbp); 447 } 448 if (error) { 449 brelse(nbp); 450 goto fail; 451 } 452 } else { 453 nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0, gbflags); 454 nbp->b_blkno = fsbtodb(fs, nb); 455 } 456 curthread_pflags_restore(saved_inbdflush); 457 *bpp = nbp; 458 return (0); 459fail: 460 curthread_pflags_restore(saved_inbdflush); 461 /* 462 * If we have failed to allocate any blocks, simply return the error. 463 * This is the usual case and avoids the need to fsync the file. 464 */ 465 if (allocblk == allociblk && allocib == NULL && unwindidx == -1) 466 return (error); 467 /* 468 * If we have failed part way through block allocation, we 469 * have to deallocate any indirect blocks that we have allocated. 470 * We have to fsync the file before we start to get rid of all 471 * of its dependencies so that we do not leave them dangling. 472 * We have to sync it at the end so that the soft updates code 473 * does not find any untracked changes. Although this is really 474 * slow, running out of disk space is not expected to be a common 475 * occurrence. The error return from fsync is ignored as we already 476 * have an error to return to the user. 477 * 478 * XXX Still have to journal the free below 479 */ 480 (void) ffs_syncvnode(vp, MNT_WAIT, 0); 481 for (deallocated = 0, blkp = allociblk, lbns_remfree = lbns; 482 blkp < allocblk; blkp++, lbns_remfree++) { 483 /* 484 * We shall not leave the freed blocks on the vnode 485 * buffer object lists. 486 */ 487 bp = getblk(vp, *lbns_remfree, fs->fs_bsize, 0, 0, GB_NOCREAT); 488 if (bp != NULL) { 489 bp->b_flags |= (B_INVAL | B_RELBUF); 490 bp->b_flags &= ~B_ASYNC; 491 brelse(bp); 492 } 493 deallocated += fs->fs_bsize; 494 } 495 if (allocib != NULL) { 496 *allocib = 0; 497 } else if (unwindidx >= 0) { 498 int r; 499 500 r = bread(vp, indirs[unwindidx].in_lbn, 501 (int)fs->fs_bsize, NOCRED, &bp); 502 if (r) { 503 panic("Could not unwind indirect block, error %d", r); 504 brelse(bp); 505 } else { 506 bap = (ufs1_daddr_t *)bp->b_data; 507 bap[indirs[unwindidx].in_off] = 0; 508 if (flags & IO_SYNC) { 509 bwrite(bp); 510 } else { 511 if (bp->b_bufsize == fs->fs_bsize) 512 bp->b_flags |= B_CLUSTEROK; 513 bdwrite(bp); 514 } 515 } 516 } 517 if (deallocated) { 518#ifdef QUOTA 519 /* 520 * Restore user's disk quota because allocation failed. 521 */ 522 (void) chkdq(ip, -btodb(deallocated), cred, FORCE); 523#endif 524 dp->di_blocks -= btodb(deallocated); 525 ip->i_flag |= IN_CHANGE | IN_UPDATE; 526 } 527 (void) ffs_syncvnode(vp, MNT_WAIT, 0); 528 /* 529 * After the buffers are invalidated and on-disk pointers are 530 * cleared, free the blocks. 531 */ 532 for (blkp = allociblk; blkp < allocblk; blkp++) { 533 ffs_blkfree(ump, fs, ip->i_devvp, *blkp, fs->fs_bsize, 534 ip->i_number, vp->v_type, NULL); 535 } 536 return (error); 537} 538 539/* 540 * Balloc defines the structure of file system storage 541 * by allocating the physical blocks on a device given 542 * the inode and the logical block number in a file. 543 * This is the allocation strategy for UFS2. Above is 544 * the allocation strategy for UFS1. 545 */ 546int 547ffs_balloc_ufs2(struct vnode *vp, off_t startoffset, int size, 548 struct ucred *cred, int flags, struct buf **bpp) 549{ 550 struct inode *ip; 551 struct ufs2_dinode *dp; 552 ufs_lbn_t lbn, lastlbn; 553 struct fs *fs; 554 struct buf *bp, *nbp; 555 struct ufsmount *ump; 556 struct indir indirs[NIADDR + 2]; 557 ufs2_daddr_t nb, newb, *bap, pref; 558 ufs2_daddr_t *allocib, *blkp, *allocblk, allociblk[NIADDR + 1]; 559 ufs2_daddr_t *lbns_remfree, lbns[NIADDR + 1]; 560 int deallocated, osize, nsize, num, i, error; 561 int unwindidx = -1; 562 int saved_inbdflush; 563 static struct timeval lastfail; 564 static int curfail; 565 int gbflags, reclaimed; 566 567 ip = VTOI(vp); 568 dp = ip->i_din2; 569 fs = ip->i_fs; 570 ump = ip->i_ump; 571 lbn = lblkno(fs, startoffset); 572 size = blkoff(fs, startoffset) + size; 573 reclaimed = 0; 574 if (size > fs->fs_bsize) 575 panic("ffs_balloc_ufs2: blk too big"); 576 *bpp = NULL; 577 if (lbn < 0) 578 return (EFBIG); 579 gbflags = (flags & BA_UNMAPPED) != 0 ? GB_UNMAPPED : 0; 580 581 if (DOINGSOFTDEP(vp)) 582 softdep_prealloc(vp, MNT_WAIT); 583 584 /* 585 * Check for allocating external data. 586 */ 587 if (flags & IO_EXT) { 588 if (lbn >= NXADDR) 589 return (EFBIG); 590 /* 591 * If the next write will extend the data into a new block, 592 * and the data is currently composed of a fragment 593 * this fragment has to be extended to be a full block. 594 */ 595 lastlbn = lblkno(fs, dp->di_extsize); 596 if (lastlbn < lbn) { 597 nb = lastlbn; 598 osize = sblksize(fs, dp->di_extsize, nb); 599 if (osize < fs->fs_bsize && osize > 0) { 600 UFS_LOCK(ump); 601 error = ffs_realloccg(ip, -1 - nb, 602 dp->di_extb[nb], 603 ffs_blkpref_ufs2(ip, lastlbn, (int)nb, 604 &dp->di_extb[0]), osize, 605 (int)fs->fs_bsize, flags, cred, &bp); 606 if (error) 607 return (error); 608 if (DOINGSOFTDEP(vp)) 609 softdep_setup_allocext(ip, nb, 610 dbtofsb(fs, bp->b_blkno), 611 dp->di_extb[nb], 612 fs->fs_bsize, osize, bp); 613 dp->di_extsize = smalllblktosize(fs, nb + 1); 614 dp->di_extb[nb] = dbtofsb(fs, bp->b_blkno); 615 bp->b_xflags |= BX_ALTDATA; 616 ip->i_flag |= IN_CHANGE; 617 if (flags & IO_SYNC) 618 bwrite(bp); 619 else 620 bawrite(bp); 621 } 622 } 623 /* 624 * All blocks are direct blocks 625 */ 626 if (flags & BA_METAONLY) 627 panic("ffs_balloc_ufs2: BA_METAONLY for ext block"); 628 nb = dp->di_extb[lbn]; 629 if (nb != 0 && dp->di_extsize >= smalllblktosize(fs, lbn + 1)) { 630 error = bread_gb(vp, -1 - lbn, fs->fs_bsize, NOCRED, 631 gbflags, &bp); 632 if (error) { 633 brelse(bp); 634 return (error); 635 } 636 bp->b_blkno = fsbtodb(fs, nb); 637 bp->b_xflags |= BX_ALTDATA; 638 *bpp = bp; 639 return (0); 640 } 641 if (nb != 0) { 642 /* 643 * Consider need to reallocate a fragment. 644 */ 645 osize = fragroundup(fs, blkoff(fs, dp->di_extsize)); 646 nsize = fragroundup(fs, size); 647 if (nsize <= osize) { 648 error = bread_gb(vp, -1 - lbn, osize, NOCRED, 649 gbflags, &bp); 650 if (error) { 651 brelse(bp); 652 return (error); 653 } 654 bp->b_blkno = fsbtodb(fs, nb); 655 bp->b_xflags |= BX_ALTDATA; 656 } else { 657 UFS_LOCK(ump); 658 error = ffs_realloccg(ip, -1 - lbn, 659 dp->di_extb[lbn], 660 ffs_blkpref_ufs2(ip, lbn, (int)lbn, 661 &dp->di_extb[0]), osize, nsize, flags, 662 cred, &bp); 663 if (error) 664 return (error); 665 bp->b_xflags |= BX_ALTDATA; 666 if (DOINGSOFTDEP(vp)) 667 softdep_setup_allocext(ip, lbn, 668 dbtofsb(fs, bp->b_blkno), nb, 669 nsize, osize, bp); 670 } 671 } else { 672 if (dp->di_extsize < smalllblktosize(fs, lbn + 1)) 673 nsize = fragroundup(fs, size); 674 else 675 nsize = fs->fs_bsize; 676 UFS_LOCK(ump); 677 error = ffs_alloc(ip, lbn, 678 ffs_blkpref_ufs2(ip, lbn, (int)lbn, &dp->di_extb[0]), 679 nsize, flags, cred, &newb); 680 if (error) 681 return (error); 682 bp = getblk(vp, -1 - lbn, nsize, 0, 0, gbflags); 683 bp->b_blkno = fsbtodb(fs, newb); 684 bp->b_xflags |= BX_ALTDATA; 685 if (flags & BA_CLRBUF) 686 vfs_bio_clrbuf(bp); 687 if (DOINGSOFTDEP(vp)) 688 softdep_setup_allocext(ip, lbn, newb, 0, 689 nsize, 0, bp); 690 } 691 dp->di_extb[lbn] = dbtofsb(fs, bp->b_blkno); 692 ip->i_flag |= IN_CHANGE; 693 *bpp = bp; 694 return (0); 695 } 696 /* 697 * If the next write will extend the file into a new block, 698 * and the file is currently composed of a fragment 699 * this fragment has to be extended to be a full block. 700 */ 701 lastlbn = lblkno(fs, ip->i_size); 702 if (lastlbn < NDADDR && lastlbn < lbn) { 703 nb = lastlbn; 704 osize = blksize(fs, ip, nb); 705 if (osize < fs->fs_bsize && osize > 0) { 706 UFS_LOCK(ump); 707 error = ffs_realloccg(ip, nb, dp->di_db[nb], 708 ffs_blkpref_ufs2(ip, lastlbn, (int)nb, 709 &dp->di_db[0]), osize, (int)fs->fs_bsize, 710 flags, cred, &bp); 711 if (error) 712 return (error); 713 if (DOINGSOFTDEP(vp)) 714 softdep_setup_allocdirect(ip, nb, 715 dbtofsb(fs, bp->b_blkno), 716 dp->di_db[nb], 717 fs->fs_bsize, osize, bp); 718 ip->i_size = smalllblktosize(fs, nb + 1); 719 dp->di_size = ip->i_size; 720 dp->di_db[nb] = dbtofsb(fs, bp->b_blkno); 721 ip->i_flag |= IN_CHANGE | IN_UPDATE; 722 if (flags & IO_SYNC) 723 bwrite(bp); 724 else 725 bawrite(bp); 726 } 727 } 728 /* 729 * The first NDADDR blocks are direct blocks 730 */ 731 if (lbn < NDADDR) { 732 if (flags & BA_METAONLY) 733 panic("ffs_balloc_ufs2: BA_METAONLY for direct block"); 734 nb = dp->di_db[lbn]; 735 if (nb != 0 && ip->i_size >= smalllblktosize(fs, lbn + 1)) { 736 error = bread_gb(vp, lbn, fs->fs_bsize, NOCRED, 737 gbflags, &bp); 738 if (error) { 739 brelse(bp); 740 return (error); 741 } 742 bp->b_blkno = fsbtodb(fs, nb); 743 *bpp = bp; 744 return (0); 745 } 746 if (nb != 0) { 747 /* 748 * Consider need to reallocate a fragment. 749 */ 750 osize = fragroundup(fs, blkoff(fs, ip->i_size)); 751 nsize = fragroundup(fs, size); 752 if (nsize <= osize) { 753 error = bread_gb(vp, lbn, osize, NOCRED, 754 gbflags, &bp); 755 if (error) { 756 brelse(bp); 757 return (error); 758 } 759 bp->b_blkno = fsbtodb(fs, nb); 760 } else { 761 UFS_LOCK(ump); 762 error = ffs_realloccg(ip, lbn, dp->di_db[lbn], 763 ffs_blkpref_ufs2(ip, lbn, (int)lbn, 764 &dp->di_db[0]), osize, nsize, flags, 765 cred, &bp); 766 if (error) 767 return (error); 768 if (DOINGSOFTDEP(vp)) 769 softdep_setup_allocdirect(ip, lbn, 770 dbtofsb(fs, bp->b_blkno), nb, 771 nsize, osize, bp); 772 } 773 } else { 774 if (ip->i_size < smalllblktosize(fs, lbn + 1)) 775 nsize = fragroundup(fs, size); 776 else 777 nsize = fs->fs_bsize; 778 UFS_LOCK(ump); 779 error = ffs_alloc(ip, lbn, 780 ffs_blkpref_ufs2(ip, lbn, (int)lbn, 781 &dp->di_db[0]), nsize, flags, cred, &newb); 782 if (error) 783 return (error); 784 bp = getblk(vp, lbn, nsize, 0, 0, gbflags); 785 bp->b_blkno = fsbtodb(fs, newb); 786 if (flags & BA_CLRBUF) 787 vfs_bio_clrbuf(bp); 788 if (DOINGSOFTDEP(vp)) 789 softdep_setup_allocdirect(ip, lbn, newb, 0, 790 nsize, 0, bp); 791 } 792 dp->di_db[lbn] = dbtofsb(fs, bp->b_blkno); 793 ip->i_flag |= IN_CHANGE | IN_UPDATE; 794 *bpp = bp; 795 return (0); 796 } 797 /* 798 * Determine the number of levels of indirection. 799 */ 800 pref = 0; 801 if ((error = ufs_getlbns(vp, lbn, indirs, &num)) != 0) 802 return(error); 803#ifdef INVARIANTS 804 if (num < 1) 805 panic ("ffs_balloc_ufs2: ufs_getlbns returned indirect block"); 806#endif 807 saved_inbdflush = curthread_pflags_set(TDP_INBDFLUSH); 808 /* 809 * Fetch the first indirect block allocating if necessary. 810 */ 811 --num; 812 nb = dp->di_ib[indirs[0].in_off]; 813 allocib = NULL; 814 allocblk = allociblk; 815 lbns_remfree = lbns; 816 if (nb == 0) { 817 UFS_LOCK(ump); 818 pref = ffs_blkpref_ufs2(ip, lbn, -indirs[0].in_off - 1, 819 (ufs2_daddr_t *)0); 820 if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, 821 flags, cred, &newb)) != 0) { 822 curthread_pflags_restore(saved_inbdflush); 823 return (error); 824 } 825 pref = newb + fs->fs_frag; 826 nb = newb; 827 MPASS(allocblk < allociblk + nitems(allociblk)); 828 MPASS(lbns_remfree < lbns + nitems(lbns)); 829 *allocblk++ = nb; 830 *lbns_remfree++ = indirs[1].in_lbn; 831 bp = getblk(vp, indirs[1].in_lbn, fs->fs_bsize, 0, 0, 832 GB_UNMAPPED); 833 bp->b_blkno = fsbtodb(fs, nb); 834 vfs_bio_clrbuf(bp); 835 if (DOINGSOFTDEP(vp)) { 836 softdep_setup_allocdirect(ip, NDADDR + indirs[0].in_off, 837 newb, 0, fs->fs_bsize, 0, bp); 838 bdwrite(bp); 839 } else { 840 /* 841 * Write synchronously so that indirect blocks 842 * never point at garbage. 843 */ 844 if (DOINGASYNC(vp)) 845 bdwrite(bp); 846 else if ((error = bwrite(bp)) != 0) 847 goto fail; 848 } 849 allocib = &dp->di_ib[indirs[0].in_off]; 850 *allocib = nb; 851 ip->i_flag |= IN_CHANGE | IN_UPDATE; 852 } 853 /* 854 * Fetch through the indirect blocks, allocating as necessary. 855 */ 856retry: 857 for (i = 1;;) { 858 error = bread(vp, 859 indirs[i].in_lbn, (int)fs->fs_bsize, NOCRED, &bp); 860 if (error) { 861 brelse(bp); 862 goto fail; 863 } 864 bap = (ufs2_daddr_t *)bp->b_data; 865 nb = bap[indirs[i].in_off]; 866 if (i == num) 867 break; 868 i += 1; 869 if (nb != 0) { 870 bqrelse(bp); 871 continue; 872 } 873 UFS_LOCK(ump); 874 /* 875 * If parent indirect has just been allocated, try to cluster 876 * immediately following it. 877 */ 878 if (pref == 0) 879 pref = ffs_blkpref_ufs2(ip, lbn, i - num - 1, 880 (ufs2_daddr_t *)0); 881 if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, 882 flags | IO_BUFLOCKED, cred, &newb)) != 0) { 883 brelse(bp); 884 if (DOINGSOFTDEP(vp) && ++reclaimed == 1) { 885 UFS_LOCK(ump); 886 softdep_request_cleanup(fs, vp, cred, 887 FLUSH_BLOCKS_WAIT); 888 UFS_UNLOCK(ump); 889 goto retry; 890 } 891 if (ppsratecheck(&lastfail, &curfail, 1)) { 892 ffs_fserr(fs, ip->i_number, "filesystem full"); 893 uprintf("\n%s: write failed, filesystem " 894 "is full\n", fs->fs_fsmnt); 895 } 896 goto fail; 897 } 898 pref = newb + fs->fs_frag; 899 nb = newb; 900 MPASS(allocblk < allociblk + nitems(allociblk)); 901 MPASS(lbns_remfree < lbns + nitems(lbns)); 902 *allocblk++ = nb; 903 *lbns_remfree++ = indirs[i].in_lbn; 904 nbp = getblk(vp, indirs[i].in_lbn, fs->fs_bsize, 0, 0, 905 GB_UNMAPPED); 906 nbp->b_blkno = fsbtodb(fs, nb); 907 vfs_bio_clrbuf(nbp); 908 if (DOINGSOFTDEP(vp)) { 909 softdep_setup_allocindir_meta(nbp, ip, bp, 910 indirs[i - 1].in_off, nb); 911 bdwrite(nbp); 912 } else { 913 /* 914 * Write synchronously so that indirect blocks 915 * never point at garbage. 916 */ 917 if ((error = bwrite(nbp)) != 0) { 918 brelse(bp); 919 goto fail; 920 } 921 } 922 bap[indirs[i - 1].in_off] = nb; 923 if (allocib == NULL && unwindidx < 0) 924 unwindidx = i - 1; 925 /* 926 * If required, write synchronously, otherwise use 927 * delayed write. 928 */ 929 if (flags & IO_SYNC) { 930 bwrite(bp); 931 } else { 932 if (bp->b_bufsize == fs->fs_bsize) 933 bp->b_flags |= B_CLUSTEROK; 934 bdwrite(bp); 935 } 936 } 937 /* 938 * If asked only for the indirect block, then return it. 939 */ 940 if (flags & BA_METAONLY) { 941 curthread_pflags_restore(saved_inbdflush); 942 *bpp = bp; 943 return (0); 944 } 945 /* 946 * Get the data block, allocating if necessary. 947 */ 948 if (nb == 0) { 949 UFS_LOCK(ump); 950 /* 951 * If allocating metadata at the front of the cylinder 952 * group and parent indirect block has just been allocated, 953 * then cluster next to it if it is the first indirect in 954 * the file. Otherwise it has been allocated in the metadata 955 * area, so we want to find our own place out in the data area. 956 */ 957 if (pref == 0 || (lbn > NDADDR && fs->fs_metaspace != 0)) 958 pref = ffs_blkpref_ufs2(ip, lbn, indirs[i].in_off, 959 &bap[0]); 960 error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, 961 flags | IO_BUFLOCKED, cred, &newb); 962 if (error) { 963 brelse(bp); 964 if (DOINGSOFTDEP(vp) && ++reclaimed == 1) { 965 UFS_LOCK(ump); 966 softdep_request_cleanup(fs, vp, cred, 967 FLUSH_BLOCKS_WAIT); 968 UFS_UNLOCK(ump); 969 goto retry; 970 } 971 if (ppsratecheck(&lastfail, &curfail, 1)) { 972 ffs_fserr(fs, ip->i_number, "filesystem full"); 973 uprintf("\n%s: write failed, filesystem " 974 "is full\n", fs->fs_fsmnt); 975 } 976 goto fail; 977 } 978 nb = newb; 979 MPASS(allocblk < allociblk + nitems(allociblk)); 980 MPASS(lbns_remfree < lbns + nitems(lbns)); 981 *allocblk++ = nb; 982 *lbns_remfree++ = lbn; 983 nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0, gbflags); 984 nbp->b_blkno = fsbtodb(fs, nb); 985 if (flags & BA_CLRBUF) 986 vfs_bio_clrbuf(nbp); 987 if (DOINGSOFTDEP(vp)) 988 softdep_setup_allocindir_page(ip, lbn, bp, 989 indirs[i].in_off, nb, 0, nbp); 990 bap[indirs[i].in_off] = nb; 991 /* 992 * If required, write synchronously, otherwise use 993 * delayed write. 994 */ 995 if (flags & IO_SYNC) { 996 bwrite(bp); 997 } else { 998 if (bp->b_bufsize == fs->fs_bsize) 999 bp->b_flags |= B_CLUSTEROK; 1000 bdwrite(bp); 1001 } 1002 curthread_pflags_restore(saved_inbdflush); 1003 *bpp = nbp; 1004 return (0); 1005 } 1006 brelse(bp); 1007 /* 1008 * If requested clear invalid portions of the buffer. If we 1009 * have to do a read-before-write (typical if BA_CLRBUF is set), 1010 * try to do some read-ahead in the sequential case to reduce 1011 * the number of I/O transactions. 1012 */ 1013 if (flags & BA_CLRBUF) { 1014 int seqcount = (flags & BA_SEQMASK) >> BA_SEQSHIFT; 1015 if (seqcount != 0 && 1016 (vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0 && 1017 !(vm_page_count_severe() || buf_dirty_count_severe())) { 1018 error = cluster_read(vp, ip->i_size, lbn, 1019 (int)fs->fs_bsize, NOCRED, 1020 MAXBSIZE, seqcount, gbflags, &nbp); 1021 } else { 1022 error = bread_gb(vp, lbn, (int)fs->fs_bsize, 1023 NOCRED, gbflags, &nbp); 1024 } 1025 if (error) { 1026 brelse(nbp); 1027 goto fail; 1028 } 1029 } else { 1030 nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0, gbflags); 1031 nbp->b_blkno = fsbtodb(fs, nb); 1032 } 1033 curthread_pflags_restore(saved_inbdflush); 1034 *bpp = nbp; 1035 return (0); 1036fail: 1037 curthread_pflags_restore(saved_inbdflush); 1038 /* 1039 * If we have failed to allocate any blocks, simply return the error. 1040 * This is the usual case and avoids the need to fsync the file. 1041 */ 1042 if (allocblk == allociblk && allocib == NULL && unwindidx == -1) 1043 return (error); 1044 /* 1045 * If we have failed part way through block allocation, we 1046 * have to deallocate any indirect blocks that we have allocated. 1047 * We have to fsync the file before we start to get rid of all 1048 * of its dependencies so that we do not leave them dangling. 1049 * We have to sync it at the end so that the soft updates code 1050 * does not find any untracked changes. Although this is really 1051 * slow, running out of disk space is not expected to be a common 1052 * occurrence. The error return from fsync is ignored as we already 1053 * have an error to return to the user. 1054 * 1055 * XXX Still have to journal the free below 1056 */ 1057 (void) ffs_syncvnode(vp, MNT_WAIT, 0); 1058 for (deallocated = 0, blkp = allociblk, lbns_remfree = lbns; 1059 blkp < allocblk; blkp++, lbns_remfree++) { 1060 /* 1061 * We shall not leave the freed blocks on the vnode 1062 * buffer object lists. 1063 */ 1064 bp = getblk(vp, *lbns_remfree, fs->fs_bsize, 0, 0, GB_NOCREAT); 1065 if (bp != NULL) { 1066 bp->b_flags |= (B_INVAL | B_RELBUF); 1067 bp->b_flags &= ~B_ASYNC; 1068 brelse(bp); 1069 } 1070 deallocated += fs->fs_bsize; 1071 } 1072 if (allocib != NULL) { 1073 *allocib = 0; 1074 } else if (unwindidx >= 0) { 1075 int r; 1076 1077 r = bread(vp, indirs[unwindidx].in_lbn, 1078 (int)fs->fs_bsize, NOCRED, &bp); 1079 if (r) { 1080 panic("Could not unwind indirect block, error %d", r); 1081 brelse(bp); 1082 } else { 1083 bap = (ufs2_daddr_t *)bp->b_data; 1084 bap[indirs[unwindidx].in_off] = 0; 1085 if (flags & IO_SYNC) { 1086 bwrite(bp); 1087 } else { 1088 if (bp->b_bufsize == fs->fs_bsize) 1089 bp->b_flags |= B_CLUSTEROK; 1090 bdwrite(bp); 1091 } 1092 } 1093 } 1094 if (deallocated) { 1095#ifdef QUOTA 1096 /* 1097 * Restore user's disk quota because allocation failed. 1098 */ 1099 (void) chkdq(ip, -btodb(deallocated), cred, FORCE); 1100#endif 1101 dp->di_blocks -= btodb(deallocated); 1102 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1103 } 1104 (void) ffs_syncvnode(vp, MNT_WAIT, 0); 1105 /* 1106 * After the buffers are invalidated and on-disk pointers are 1107 * cleared, free the blocks. 1108 */ 1109 for (blkp = allociblk; blkp < allocblk; blkp++) { 1110 ffs_blkfree(ump, fs, ip->i_devvp, *blkp, fs->fs_bsize, 1111 ip->i_number, vp->v_type, NULL); 1112 } 1113 return (error); 1114} 1115