1 /* $NetBSD: lfs.c,v 1.77 2025/11/04 00:50:36 perseant Exp $ */ 2 /*- 3 * Copyright (c) 2003 The NetBSD Foundation, Inc. 4 * All rights reserved. 5 * 6 * This code is derived from software contributed to The NetBSD Foundation 7 * by Konrad E. Schroder <perseant (at) hhhh.org>. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 18 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 19 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 20 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 21 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 22 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 23 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 24 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 25 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 26 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 27 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 28 * POSSIBILITY OF SUCH DAMAGE. 29 */ 30 /* 31 * Copyright (c) 1989, 1991, 1993 32 * The Regents of the University of California. All rights reserved. 33 * (c) UNIX System Laboratories, Inc. 34 * All or some portions of this file are derived from material licensed 35 * to the University of California by American Telephone and Telegraph 36 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 37 * the permission of UNIX System Laboratories, Inc. 38 * 39 * Redistribution and use in source and binary forms, with or without 40 * modification, are permitted provided that the following conditions 41 * are met: 42 * 1. Redistributions of source code must retain the above copyright 43 * notice, this list of conditions and the following disclaimer. 44 * 2. Redistributions in binary form must reproduce the above copyright 45 * notice, this list of conditions and the following disclaimer in the 46 * documentation and/or other materials provided with the distribution. 47 * 3. Neither the name of the University nor the names of its contributors 48 * may be used to endorse or promote products derived from this software 49 * without specific prior written permission. 50 * 51 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 52 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 53 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 54 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 55 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 56 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 57 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 58 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 59 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 60 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 61 * SUCH DAMAGE. 62 * 63 * @(#)ufs_bmap.c 8.8 (Berkeley) 8/11/95 64 */ 65 66 67 #include <sys/types.h> 68 #include <sys/param.h> 69 #include <sys/time.h> 70 #include <sys/buf.h> 71 #include <sys/mount.h> 72 73 #define vnode uvnode 74 #include <ufs/lfs/lfs.h> 75 #include <ufs/lfs/lfs_inode.h> 76 #include <ufs/lfs/lfs_accessors.h> 77 #undef vnode 78 79 #include <assert.h> 80 #include <err.h> 81 #include <errno.h> 82 #include <stdarg.h> 83 #include <stdbool.h> 84 #include <stdio.h> 85 #include <stdlib.h> 86 #include <string.h> 87 #include <unistd.h> 88 #include <util.h> 89 90 #include "bufcache.h" 91 #include "extern.h" 92 #include "lfs_user.h" 93 #include "segwrite.h" 94 #include "kernelops.h" 95 96 #define panic call_panic 97 98 long dev_bsize = DEV_BSIZE; 99 100 static int 101 lfs_fragextend(struct uvnode *, int, int, daddr_t, struct ubuf **); 102 103 int fsdirty = 0; 104 void (*panic_func)(int, const char *, va_list) = my_vpanic; 105 106 /* 107 * LFS buffer and uvnode operations 108 */ 109 110 int 111 lfs_vop_strategy(struct ubuf * bp) 112 { 113 int count; 114 115 if (bp->b_flags & B_READ) { 116 count = kops.ko_pread(bp->b_vp->v_fd, bp->b_data, bp->b_bcount, 117 bp->b_blkno * dev_bsize); 118 if (count == bp->b_bcount) 119 bp->b_flags |= B_DONE; 120 } else { 121 count = kops.ko_pwrite(bp->b_vp->v_fd, bp->b_data, bp->b_bcount, 122 bp->b_blkno * dev_bsize); 123 if (count == 0) { 124 perror("pwrite"); 125 return -1; 126 } 127 bp->b_flags &= ~B_DELWRI; 128 reassignbuf(bp, bp->b_vp); 129 } 130 return 0; 131 } 132 133 int 134 lfs_vop_bwrite(struct ubuf * bp) 135 { 136 struct lfs *fs; 137 138 fs = bp->b_vp->v_fs; 139 if (!(bp->b_flags & B_DELWRI)) { 140 lfs_sb_subavail(fs, lfs_btofsb(fs, bp->b_bcount)); 141 } 142 bp->b_flags |= B_DELWRI | B_LOCKED; 143 reassignbuf(bp, bp->b_vp); 144 brelse(bp, 0); 145 return 0; 146 } 147 148 /* 149 * ulfs_bmaparray does the bmap conversion, and if requested returns the 150 * array of logical blocks which must be traversed to get to a block. 151 * Each entry contains the offset into that block that gets you to the 152 * next block and the disk address of the block (if it is assigned). 153 */ 154 int 155 ulfs_bmaparray(struct lfs * fs, struct uvnode * vp, daddr_t bn, daddr_t * bnp, struct indir * ap, int *nump) 156 { 157 struct inode *ip; 158 struct ubuf *bp; 159 struct indir a[ULFS_NIADDR + 1], *xap; 160 daddr_t daddr; 161 daddr_t metalbn; 162 int error, num; 163 164 ip = VTOI(vp); 165 166 if (bn >= 0 && bn < ULFS_NDADDR) { 167 if (nump != NULL) 168 *nump = 0; 169 *bnp = LFS_FSBTODB(fs, lfs_dino_getdb(fs, ip->i_din, bn)); 170 if (*bnp == 0) 171 *bnp = -1; 172 return (0); 173 } 174 xap = ap == NULL ? a : ap; 175 if (!nump) 176 nump = # 177 if ((error = ulfs_getlbns(fs, vp, bn, xap, nump)) != 0) 178 return (error); 179 180 num = *nump; 181 182 /* Get disk address out of indirect block array */ 183 daddr = lfs_dino_getib(fs, ip->i_din, xap->in_off); 184 185 for (bp = NULL, ++xap; --num; ++xap) { 186 /* Exit the loop if there is no disk address assigned yet and 187 * the indirect block isn't in the cache, or if we were 188 * looking for an indirect block and we've found it. */ 189 190 metalbn = xap->in_lbn; 191 if ((daddr == 0 && !incore(vp, metalbn)) || metalbn == bn) 192 break; 193 /* 194 * If we get here, we've either got the block in the cache 195 * or we have a disk address for it, go fetch it. 196 */ 197 if (bp) 198 brelse(bp, 0); 199 200 xap->in_exists = 1; 201 bp = getblk(vp, metalbn, lfs_sb_getbsize(fs)); 202 203 if (!(bp->b_flags & (B_DONE | B_DELWRI))) { 204 bp->b_blkno = LFS_FSBTODB(fs, daddr); 205 bp->b_flags |= B_READ; 206 VOP_STRATEGY(bp); 207 } 208 daddr = lfs_iblock_get(fs, bp->b_data, xap->in_off); 209 } 210 if (bp) 211 brelse(bp, 0); 212 213 daddr = LFS_FSBTODB(fs, daddr); 214 *bnp = daddr == 0 ? -1 : daddr; 215 return (0); 216 } 217 218 /* 219 * Create an array of logical block number/offset pairs which represent the 220 * path of indirect blocks required to access a data block. The first "pair" 221 * contains the logical block number of the appropriate single, double or 222 * triple indirect block and the offset into the inode indirect block array. 223 * Note, the logical block number of the inode single/double/triple indirect 224 * block appears twice in the array, once with the offset into di_ib and 225 * once with the offset into the page itself. 226 */ 227 int 228 ulfs_getlbns(struct lfs * fs, struct uvnode * vp, daddr_t bn, struct indir * ap, int *nump) 229 { 230 daddr_t metalbn, realbn; 231 int64_t blockcnt; 232 int lbc; 233 int i, numlevels, off; 234 int lognindir, indir; 235 236 metalbn = 0; /* XXXGCC -Wuninitialized [sh3] */ 237 238 if (nump) 239 *nump = 0; 240 numlevels = 0; 241 realbn = bn; 242 if (bn < 0) 243 bn = -bn; 244 245 lognindir = -1; 246 for (indir = lfs_sb_getnindir(fs); indir; indir >>= 1) 247 ++lognindir; 248 249 /* Determine the number of levels of indirection. After this loop is 250 * done, blockcnt indicates the number of data blocks possible at the 251 * given level of indirection, and ULFS_NIADDR - i is the number of levels 252 * of indirection needed to locate the requested block. */ 253 254 bn -= ULFS_NDADDR; 255 for (lbc = 0, i = ULFS_NIADDR;; i--, bn -= blockcnt) { 256 if (i == 0) 257 return (EFBIG); 258 259 lbc += lognindir; 260 blockcnt = (int64_t) 1 << lbc; 261 262 if (bn < blockcnt) 263 break; 264 } 265 266 /* Calculate the address of the first meta-block. */ 267 metalbn = -((realbn >= 0 ? realbn : -realbn) - bn + ULFS_NIADDR - i); 268 269 /* At each iteration, off is the offset into the bap array which is an 270 * array of disk addresses at the current level of indirection. The 271 * logical block number and the offset in that block are stored into 272 * the argument array. */ 273 ap->in_lbn = metalbn; 274 ap->in_off = off = ULFS_NIADDR - i; 275 ap->in_exists = 0; 276 ap++; 277 for (++numlevels; i <= ULFS_NIADDR; i++) { 278 /* If searching for a meta-data block, quit when found. */ 279 if (metalbn == realbn) 280 break; 281 282 lbc -= lognindir; 283 /*blockcnt = (int64_t) 1 << lbc;*/ 284 off = (bn >> lbc) & (lfs_sb_getnindir(fs) - 1); 285 286 ++numlevels; 287 ap->in_lbn = metalbn; 288 ap->in_off = off; 289 ap->in_exists = 0; 290 ++ap; 291 292 metalbn -= -1 + (off << lbc); 293 } 294 if (nump) 295 *nump = numlevels; 296 return (0); 297 } 298 299 int 300 lfs_vop_bmap(struct uvnode * vp, daddr_t lbn, daddr_t * daddrp) 301 { 302 return ulfs_bmaparray(vp->v_fs, vp, lbn, daddrp, NULL, NULL); 303 } 304 305 /* Search a block for a specific dinode. */ 306 union lfs_dinode * 307 lfs_ifind(struct lfs *fs, ino_t ino, struct ubuf *bp) 308 { 309 union lfs_dinode *ldip; 310 unsigned i, num; 311 312 num = LFS_INOPB(fs); 313 314 /* 315 * Read the inode block backwards, since later versions of the 316 * inode will supercede earlier ones. Though it is unlikely, it is 317 * possible that the same inode will appear in the same inode block. 318 */ 319 for (i = num; i-- > 0; ) { 320 ldip = DINO_IN_BLOCK(fs, bp->b_data, i); 321 if (lfs_dino_getinumber(fs, ldip) == ino) 322 return (ldip); 323 } 324 return NULL; 325 } 326 327 /* 328 * lfs_raw_vget makes us a new vnode from the inode at the given disk address. 329 * XXX it currently loses atime information. 330 */ 331 struct uvnode * 332 lfs_raw_vget(struct lfs * fs, ino_t ino, int fd, daddr_t daddr) 333 { 334 struct uvnode *vp; 335 struct inode *ip; 336 union lfs_dinode *dip; 337 struct ubuf *bp; 338 int i, hash; 339 340 vp = ecalloc(1, sizeof(*vp)); 341 vp->v_fd = fd; 342 vp->v_fs = fs; 343 vp->v_usecount = 0; 344 vp->v_strategy_op = lfs_vop_strategy; 345 vp->v_bwrite_op = lfs_vop_bwrite; 346 vp->v_bmap_op = lfs_vop_bmap; 347 LIST_INIT(&vp->v_cleanblkhd); 348 LIST_INIT(&vp->v_dirtyblkhd); 349 350 ip = ecalloc(1, sizeof(*ip)); 351 352 ip->i_din = dip = ecalloc(1, sizeof(*dip)); 353 354 /* Initialize the inode -- from lfs_vcreate. */ 355 ip->inode_ext.lfs = ecalloc(1, sizeof(*ip->inode_ext.lfs)); 356 vp->v_data = ip; 357 /* ip->i_vnode = vp; */ 358 ip->i_lockf = 0; 359 ip->i_state = 0; 360 361 /* Load inode block and find inode */ 362 if (daddr > 0) { 363 bread(fs->lfs_devvp, LFS_FSBTODB(fs, daddr), lfs_sb_getibsize(fs), 364 0, &bp); 365 bp->b_flags |= B_AGE; 366 dip = lfs_ifind(fs, ino, bp); 367 if (dip == NULL) { 368 brelse(bp, 0); 369 free(ip->i_din); 370 free(ip->inode_ext.lfs); 371 free(ip); 372 free(vp); 373 return NULL; 374 } 375 lfs_copy_dinode(fs, ip->i_din, dip); 376 brelse(bp, 0); 377 } 378 ip->i_number = ino; 379 /* ip->i_devvp = fs->lfs_devvp; */ 380 ip->i_lfs = fs; 381 382 ip->i_lfs_effnblks = lfs_dino_getblocks(fs, ip->i_din); 383 ip->i_lfs_osize = lfs_dino_getsize(fs, ip->i_din); 384 #if 0 385 if (lfs_sb_getversion(fs) > 1) { 386 lfs_dino_setatime(fs, ip->i_din, ts.tv_sec); 387 lfs_dino_setatimensec(fs, ip->i_din, ts.tv_nsec); 388 } 389 #endif 390 391 memset(ip->i_lfs_fragsize, 0, ULFS_NDADDR * sizeof(*ip->i_lfs_fragsize)); 392 for (i = 0; i < ULFS_NDADDR; i++) 393 if (lfs_dino_getdb(fs, ip->i_din, i) != 0) 394 ip->i_lfs_fragsize[i] = lfs_blksize(fs, ip, i); 395 396 ++nvnodes; 397 hash = ((int)(intptr_t)fs + ino) & (VNODE_HASH_MAX - 1); 398 LIST_INSERT_HEAD(&getvnodelist[hash], vp, v_getvnodes); 399 LIST_INSERT_HEAD(&vnodelist, vp, v_mntvnodes); 400 401 return vp; 402 } 403 404 static struct uvnode * 405 lfs_vget(void *vfs, ino_t ino) 406 { 407 struct lfs *fs = (struct lfs *)vfs; 408 daddr_t daddr; 409 struct ubuf *bp; 410 IFILE *ifp; 411 412 LFS_IENTRY(ifp, fs, ino, bp); 413 daddr = lfs_if_getdaddr(fs, ifp); 414 brelse(bp, 0); 415 if (daddr <= 0 || lfs_dtosn(fs, daddr) >= lfs_sb_getnseg(fs)) 416 return NULL; 417 return lfs_raw_vget(fs, ino, fs->lfs_ivnode->v_fd, daddr); 418 } 419 420 /* 421 * Check superblock magic number and checksum. 422 * Sets lfs_is64 and lfs_dobyteswap. 423 */ 424 static int 425 check_sb(struct lfs *fs) 426 { 427 u_int32_t checksum; 428 u_int32_t magic; 429 430 /* we can read the magic out of either the 32-bit or 64-bit dlfs */ 431 magic = fs->lfs_dlfs_u.u_32.dlfs_magic; 432 433 switch (magic) { 434 case LFS_MAGIC: 435 fs->lfs_is64 = false; 436 fs->lfs_dobyteswap = false; 437 break; 438 case LFS_MAGIC_SWAPPED: 439 fs->lfs_is64 = false; 440 fs->lfs_dobyteswap = true; 441 break; 442 case LFS64_MAGIC: 443 fs->lfs_is64 = true; 444 fs->lfs_dobyteswap = false; 445 break; 446 case LFS64_MAGIC_SWAPPED: 447 fs->lfs_is64 = true; 448 fs->lfs_dobyteswap = true; 449 break; 450 default: 451 printf("Superblock magic number (0x%lx) does not match " 452 "any of the expected 0x%lx, 0x%lx, 0x%lx or 0x%lx\n", 453 (unsigned long)magic, 454 (unsigned long)LFS_MAGIC, 455 (unsigned long)LFS_MAGIC_SWAPPED, 456 (unsigned long)LFS64_MAGIC, 457 (unsigned long)LFS64_MAGIC_SWAPPED); 458 return 1; 459 } 460 461 /* checksum */ 462 checksum = lfs_sb_cksum(fs); 463 if (lfs_sb_getcksum(fs) != checksum) { 464 printf("Superblock checksum (%lx) does not match computed checksum (%lx)\n", 465 (unsigned long) lfs_sb_getcksum(fs), (unsigned long) checksum); 466 return 1; 467 } 468 return 0; 469 } 470 471 /* Initialize LFS library; load superblocks and choose which to use. */ 472 struct lfs * 473 lfs_init(int devfd, daddr_t sblkno, daddr_t idaddr, int dummy_read, int debug) 474 { 475 struct uvnode *devvp; 476 struct ubuf *bp; 477 int tryalt; 478 struct lfs *fs, *altfs; 479 480 vfs_init(); 481 482 devvp = ecalloc(1, sizeof(*devvp)); 483 devvp->v_fs = NULL; 484 devvp->v_fd = devfd; 485 devvp->v_strategy_op = raw_vop_strategy; 486 devvp->v_bwrite_op = raw_vop_bwrite; 487 devvp->v_bmap_op = raw_vop_bmap; 488 LIST_INIT(&devvp->v_cleanblkhd); 489 LIST_INIT(&devvp->v_dirtyblkhd); 490 491 tryalt = 0; 492 if (dummy_read) { 493 if (sblkno == 0) 494 sblkno = LFS_LABELPAD / dev_bsize; 495 fs = ecalloc(1, sizeof(*fs)); 496 fs->lfs_devvp = devvp; 497 } else { 498 if (sblkno == 0) { 499 sblkno = LFS_LABELPAD / dev_bsize; 500 tryalt = 1; 501 } else if (debug) { 502 printf("No -b flag given, not attempting to verify checkpoint\n"); 503 } 504 505 dev_bsize = DEV_BSIZE; 506 507 (void)bread(devvp, sblkno, LFS_SBPAD, 0, &bp); 508 fs = ecalloc(1, sizeof(*fs)); 509 __CTASSERT(sizeof(struct dlfs) == sizeof(struct dlfs64)); 510 memcpy(&fs->lfs_dlfs_u, bp->b_data, sizeof(struct dlfs)); 511 fs->lfs_devvp = devvp; 512 bp->b_flags |= B_INVAL; 513 brelse(bp, 0); 514 515 /* 516 * Look at the magic number before validating the rest 517 * of the superblock. If the magic number is bad, too, 518 * we don't know where to look for an alternate superblock 519 * either; so bail out now. 520 */ 521 switch (fs->lfs_dlfs_u.u_32.dlfs_magic) { 522 case LFS_MAGIC: 523 case LFS_MAGIC_SWAPPED: 524 break; 525 526 case LFS64_MAGIC: 527 case LFS64_MAGIC_SWAPPED: 528 fs->lfs_is64 = true; 529 break; 530 531 default: 532 return NULL; 533 } 534 535 536 dev_bsize = lfs_sb_getfsize(fs) >> lfs_sb_getfsbtodb(fs); 537 538 if (tryalt) { 539 (void)bread(devvp, LFS_FSBTODB(fs, lfs_sb_getsboff(fs, 1)), 540 LFS_SBPAD, 0, &bp); 541 altfs = ecalloc(1, sizeof(*altfs)); 542 memcpy(&altfs->lfs_dlfs_u, bp->b_data, 543 sizeof(altfs->lfs_dlfs_u)); 544 altfs->lfs_devvp = devvp; 545 bp->b_flags |= B_INVAL; 546 brelse(bp, 0); 547 548 if (check_sb(fs) || lfs_sb_getidaddr(fs) <= 0) { 549 if (debug) 550 printf("Primary superblock is no good, using first alternate\n"); 551 free(fs); 552 fs = altfs; 553 } else { 554 /* If both superblocks check out, try verification */ 555 if (check_sb(altfs)) { 556 if (debug) 557 printf("First alternate superblock is no good, using primary\n"); 558 free(altfs); 559 } else { 560 if (lfs_verify(fs, altfs, devvp, debug) == fs) { 561 free(altfs); 562 } else { 563 free(fs); 564 fs = altfs; 565 } 566 } 567 } 568 } 569 if (check_sb(fs)) { 570 free(fs); 571 return NULL; 572 } 573 } 574 575 /* Compatibility */ 576 if (lfs_sb_getversion(fs) < 2) { 577 lfs_sb_setsumsize(fs, LFS_V1_SUMMARY_SIZE); 578 lfs_sb_setibsize(fs, lfs_sb_getbsize(fs)); 579 lfs_sb_sets0addr(fs, lfs_sb_getsboff(fs, 0)); 580 lfs_sb_settstamp(fs, lfs_sb_getotstamp(fs)); 581 lfs_sb_setfsbtodb(fs, 0); 582 } 583 584 if (idaddr == 0) 585 idaddr = lfs_sb_getidaddr(fs); 586 else 587 lfs_sb_setidaddr(fs, idaddr); 588 /* NB: If dummy_read!=0, idaddr==0 here so we get a fake inode. */ 589 fs->lfs_ivnode = lfs_raw_vget(fs, LFS_IFILE_INUM, 590 devvp->v_fd, idaddr); 591 if (fs->lfs_ivnode == NULL) 592 return NULL; 593 594 register_vget((void *)fs, lfs_vget); 595 596 return fs; 597 } 598 599 /* 600 * Check partial segment validity between fs->lfs_offset and the given goal. 601 * 602 * If goal == 0, just keep on going until the segments stop making sense, 603 * and return the address of the last valid partial segment. 604 * 605 * If goal != 0, return the address of the first partial segment that failed, 606 * or "goal" if we reached it without failure (the partial segment *at* goal 607 * need not be valid). 608 */ 609 daddr_t 610 try_verify(struct lfs *osb, struct uvnode *devvp, daddr_t goal, int debug) 611 { 612 daddr_t daddr, odaddr; 613 SEGSUM *sp; 614 int i, bc, hitclean; 615 struct ubuf *bp; 616 daddr_t nodirop_daddr; 617 u_int64_t serial; 618 619 bc = 0; 620 hitclean = 0; 621 odaddr = -1; 622 daddr = lfs_sb_getoffset(osb); 623 nodirop_daddr = daddr; 624 serial = lfs_sb_getserial(osb); 625 while (daddr != goal) { 626 /* 627 * Don't mistakenly read a superblock, if there is one here. 628 */ 629 if (lfs_sntod(osb, lfs_dtosn(osb, daddr)) == daddr) { 630 if (daddr == lfs_sb_gets0addr(osb)) 631 daddr += lfs_btofsb(osb, LFS_LABELPAD); 632 for (i = 0; i < LFS_MAXNUMSB; i++) { 633 /* XXX dholland 20150828 I think this is wrong */ 634 if (lfs_sb_getsboff(osb, i) < daddr) 635 break; 636 if (lfs_sb_getsboff(osb, i) == daddr) 637 daddr += lfs_btofsb(osb, LFS_SBPAD); 638 } 639 } 640 641 /* Read in summary block */ 642 bread(devvp, LFS_FSBTODB(osb, daddr), lfs_sb_getsumsize(osb), 643 0, &bp); 644 sp = (SEGSUM *)bp->b_data; 645 646 /* 647 * Check for a valid segment summary belonging to our fs. 648 */ 649 if (lfs_ss_getmagic(osb, sp) != SS_MAGIC || 650 lfs_ss_getident(osb, sp) != lfs_sb_getident(osb) || 651 lfs_ss_getserial(osb, sp) < serial || /* XXX strengthen this */ 652 lfs_ss_getsumsum(osb, sp) != 653 cksum((char *)sp + lfs_ss_getsumstart(osb), 654 lfs_sb_getsumsize(osb) - lfs_ss_getsumstart(osb))) { 655 brelse(bp, 0); 656 if (debug) { 657 if (lfs_ss_getmagic(osb, sp) != SS_MAGIC) 658 pwarn("pseg at 0x%jx: " 659 "wrong magic number\n", 660 (uintmax_t)daddr); 661 else if (lfs_ss_getident(osb, sp) != lfs_sb_getident(osb)) 662 pwarn("pseg at 0x%jx: " 663 "expected ident %jx, got %jx\n", 664 (uintmax_t)daddr, 665 (uintmax_t)lfs_ss_getident(osb, sp), 666 (uintmax_t)lfs_sb_getident(osb)); 667 else if (lfs_ss_getserial(osb, sp) >= serial) 668 pwarn("pseg at 0x%jx: " 669 "serial %d < %d\n", 670 (uintmax_t)daddr, 671 (int)lfs_ss_getserial(osb, sp), (int)serial); 672 else 673 pwarn("pseg at 0x%jx: " 674 "summary checksum wrong\n", 675 (uintmax_t)daddr); 676 } 677 break; 678 } 679 if (debug && lfs_ss_getserial(osb, sp) != serial) 680 pwarn("warning, serial=%d ss_serial=%d\n", 681 (int)serial, (int)lfs_ss_getserial(osb, sp)); 682 ++serial; 683 bc = check_summary(osb, sp, daddr, debug, devvp, NULL); 684 if (bc == 0) { 685 brelse(bp, 0); 686 break; 687 } 688 if (debug) 689 pwarn("summary good: 0x%jx/%d\n", (uintmax_t)daddr, 690 (int)lfs_ss_getserial(osb, sp)); 691 assert (bc > 0); 692 odaddr = daddr; 693 daddr += lfs_btofsb(osb, lfs_sb_getsumsize(osb) + bc); 694 if (lfs_dtosn(osb, odaddr) != lfs_dtosn(osb, daddr) || 695 lfs_dtosn(osb, daddr) != lfs_dtosn(osb, daddr + 696 lfs_btofsb(osb, lfs_sb_getsumsize(osb) + lfs_sb_getbsize(osb)) - 1)) { 697 daddr = lfs_ss_getnext(osb, sp); 698 } 699 700 /* 701 * Check for the beginning and ending of a sequence of 702 * dirops. Writes from the cleaner never involve new 703 * information, and are always checkpoints; so don't try 704 * to roll forward through them. Likewise, psegs written 705 * by a previous roll-forward attempt are not interesting. 706 */ 707 if (lfs_ss_getflags(osb, sp) & (SS_CLEAN | SS_RFW)) 708 hitclean = 1; 709 if (hitclean == 0 && (lfs_ss_getflags(osb, sp) & SS_CONT) == 0) 710 nodirop_daddr = daddr; 711 712 brelse(bp, 0); 713 } 714 715 if (goal == 0) 716 return nodirop_daddr; 717 else 718 return daddr; 719 } 720 721 /* Use try_verify to check whether the newer superblock is valid. */ 722 struct lfs * 723 lfs_verify(struct lfs *sb0, struct lfs *sb1, struct uvnode *devvp, int debug) 724 { 725 daddr_t daddr; 726 struct lfs *osb, *nsb; 727 728 /* 729 * Verify the checkpoint of the newer superblock, 730 * if the timestamp/serial number of the two superblocks is 731 * different. 732 */ 733 734 osb = NULL; 735 if (debug) 736 pwarn("sb0 %ju, sb1 %ju", 737 (uintmax_t) lfs_sb_getserial(sb0), 738 (uintmax_t) lfs_sb_getserial(sb1)); 739 740 if ((lfs_sb_getversion(sb0) == 1 && 741 lfs_sb_getotstamp(sb0) != lfs_sb_getotstamp(sb1)) || 742 (lfs_sb_getversion(sb0) > 1 && 743 lfs_sb_getserial(sb0) != lfs_sb_getserial(sb1))) { 744 if (lfs_sb_getversion(sb0) == 1) { 745 if (lfs_sb_getotstamp(sb0) > lfs_sb_getotstamp(sb1)) { 746 osb = sb1; 747 nsb = sb0; 748 } else { 749 osb = sb0; 750 nsb = sb1; 751 } 752 } else { 753 if (lfs_sb_getserial(sb0) > lfs_sb_getserial(sb1)) { 754 osb = sb1; 755 nsb = sb0; 756 } else { 757 osb = sb0; 758 nsb = sb1; 759 } 760 } 761 if (debug) { 762 printf("Attempting to verify newer checkpoint..."); 763 fflush(stdout); 764 } 765 daddr = try_verify(osb, devvp, lfs_sb_getoffset(nsb), debug); 766 767 if (debug) 768 printf("done.\n"); 769 if (daddr == lfs_sb_getoffset(nsb)) { 770 pwarn("** Newer checkpoint verified; recovered %jd seconds of data\n", 771 (intmax_t)(lfs_sb_gettstamp(nsb) - lfs_sb_gettstamp(osb))); 772 sbdirty(); 773 } else { 774 pwarn("** Newer checkpoint invalid; lost %jd seconds of data\n", (intmax_t)(lfs_sb_gettstamp(nsb) - lfs_sb_gettstamp(osb))); 775 } 776 return (daddr == lfs_sb_getoffset(nsb) ? nsb : osb); 777 } 778 /* Nothing to check */ 779 return osb; 780 } 781 782 /* Verify a partial-segment summary; return the number of bytes on disk. */ 783 int 784 check_summary(struct lfs *fs, SEGSUM *sp, daddr_t pseg_addr, int debug, 785 struct uvnode *devvp, void (func(daddr_t, FINFO *))) 786 { 787 FINFO *fp; 788 int bc; /* Bytes in partial segment */ 789 int nblocks; 790 daddr_t daddr; 791 IINFO *iibase, *iip; 792 struct ubuf *bp; 793 int i, j, k, datac, len; 794 lfs_checkword *datap; 795 u_int32_t ccksum; 796 797 /* We've already checked the sumsum, just do the data bounds and sum */ 798 799 /* Count the blocks. */ 800 nblocks = howmany(lfs_ss_getninos(fs, sp), LFS_INOPB(fs)); 801 bc = nblocks << (lfs_sb_getversion(fs) > 1 ? lfs_sb_getffshift(fs) : lfs_sb_getbshift(fs)); 802 assert(bc >= 0); 803 804 fp = SEGSUM_FINFOBASE(fs, sp); 805 for (i = 0; i < lfs_ss_getnfinfo(fs, sp); i++) { 806 nblocks += lfs_fi_getnblocks(fs, fp); 807 bc += lfs_fi_getlastlength(fs, fp) + ((lfs_fi_getnblocks(fs, fp) - 1) 808 << lfs_sb_getbshift(fs)); 809 assert(bc >= 0); 810 fp = NEXT_FINFO(fs, fp); 811 if (((char *)fp) - (char *)sp > lfs_sb_getsumsize(fs)) 812 return 0; 813 } 814 datap = emalloc(nblocks * sizeof(*datap)); 815 datac = 0; 816 817 iibase = SEGSUM_IINFOSTART(fs, sp); 818 819 iip = iibase; 820 daddr = pseg_addr + lfs_btofsb(fs, lfs_sb_getsumsize(fs)); 821 fp = SEGSUM_FINFOBASE(fs, sp); 822 for (i = 0, j = 0; 823 i < lfs_ss_getnfinfo(fs, sp) || j < howmany(lfs_ss_getninos(fs, sp), LFS_INOPB(fs)); i++) { 824 if (i >= lfs_ss_getnfinfo(fs, sp) && lfs_ii_getblock(fs, iip) != daddr) { 825 pwarn("Not enough inode blocks in pseg at 0x%jx: " 826 "found %d, wanted %d\n", 827 pseg_addr, j, howmany(lfs_ss_getninos(fs, sp), 828 LFS_INOPB(fs))); 829 if (debug) 830 pwarn("iip=0x%jx, daddr=0x%jx\n", 831 (uintmax_t)lfs_ii_getblock(fs, iip), 832 (intmax_t)daddr); 833 break; 834 } 835 while (j < howmany(lfs_ss_getninos(fs, sp), LFS_INOPB(fs)) && lfs_ii_getblock(fs, iip) == daddr) { 836 bread(devvp, LFS_FSBTODB(fs, daddr), lfs_sb_getibsize(fs), 837 0, &bp); 838 datap[datac++] = ((lfs_checkword *)bp->b_data)[0]; 839 brelse(bp, 0); 840 841 ++j; 842 daddr += lfs_btofsb(fs, lfs_sb_getibsize(fs)); 843 iip = NEXTLOWER_IINFO(fs, iip); 844 } 845 if (i < lfs_ss_getnfinfo(fs, sp)) { 846 if (func) 847 func(daddr, fp); 848 for (k = 0; k < lfs_fi_getnblocks(fs, fp); k++) { 849 len = (k == lfs_fi_getnblocks(fs, fp) - 1 ? 850 lfs_fi_getlastlength(fs, fp) 851 : lfs_sb_getbsize(fs)); 852 bread(devvp, LFS_FSBTODB(fs, daddr), len, 853 0, &bp); 854 datap[datac++] = ((lfs_checkword *)bp->b_data)[0]; 855 brelse(bp, 0); 856 daddr += lfs_btofsb(fs, len); 857 } 858 fp = NEXT_FINFO(fs, fp); 859 } 860 } 861 862 if (datac != nblocks) { 863 pwarn("Partial segment at 0x%jx expected %d blocks counted %d\n", 864 (intmax_t)pseg_addr, nblocks, datac); 865 } 866 ccksum = cksum(datap, nblocks * sizeof(datap[0])); 867 /* Check the data checksum */ 868 if (ccksum != lfs_ss_getdatasum(fs, sp)) { 869 pwarn("Partial segment at 0x%jx data checksum" 870 " mismatch: given 0x%x, computed 0x%x\n", 871 (uintmax_t)pseg_addr, lfs_ss_getdatasum(fs, sp), ccksum); 872 free(datap); 873 return 0; 874 } 875 free(datap); 876 assert(bc >= 0); 877 return bc; 878 } 879 880 /* print message and exit */ 881 void 882 my_vpanic(int fatal, const char *fmt, va_list ap) 883 { 884 (void) vprintf(fmt, ap); 885 exit(8); 886 } 887 888 void 889 call_panic(const char *fmt, ...) 890 { 891 va_list ap; 892 893 va_start(ap, fmt); 894 panic_func(1, fmt, ap); 895 va_end(ap); 896 } 897 898 /* Allocate a new inode. */ 899 struct uvnode * 900 lfs_valloc(struct lfs *fs, ino_t ino) 901 { 902 struct ubuf *bp, *cbp; 903 IFILE *ifp; 904 ino_t new_ino; 905 int error; 906 CLEANERINFO *cip; 907 908 /* Get the head of the freelist. */ 909 LFS_GET_HEADFREE(fs, cip, cbp, &new_ino); 910 911 /* 912 * Remove the inode from the free list and write the new start 913 * of the free list into the superblock. 914 */ 915 LFS_IENTRY(ifp, fs, new_ino, bp); 916 if (lfs_if_getdaddr(fs, ifp) != LFS_UNUSED_DADDR) 917 panic("lfs_valloc: inuse inode %d on the free list", new_ino); 918 LFS_PUT_HEADFREE(fs, cip, cbp, lfs_if_getnextfree(fs, ifp)); 919 920 brelse(bp, 0); 921 922 /* Extend IFILE so that the next lfs_valloc will succeed. */ 923 if (lfs_sb_getfreehd(fs) == LFS_UNUSED_INUM) { 924 if ((error = extend_ifile(fs)) != 0) { 925 LFS_PUT_HEADFREE(fs, cip, cbp, new_ino); 926 return NULL; 927 } 928 } 929 930 /* Set superblock modified bit and increment file count. */ 931 sbdirty(); 932 lfs_sb_addnfiles(fs, 1); 933 934 return lfs_raw_vget(fs, ino, fs->lfs_devvp->v_fd, 0x0); 935 } 936 937 #ifdef IN_FSCK_LFS 938 void reset_maxino(ino_t); 939 #endif 940 941 /* 942 * Add a new block to the Ifile, to accommodate future file creations. 943 */ 944 int 945 extend_ifile(struct lfs *fs) 946 { 947 struct uvnode *vp; 948 struct inode *ip; 949 IFILE64 *ifp64; 950 IFILE32 *ifp32; 951 IFILE_V1 *ifp_v1; 952 struct ubuf *bp, *cbp; 953 daddr_t i, blkno, max; 954 ino_t oldlast; 955 CLEANERINFO *cip; 956 957 vp = fs->lfs_ivnode; 958 ip = VTOI(vp); 959 blkno = lfs_lblkno(fs, lfs_dino_getsize(fs, ip->i_din)); 960 961 lfs_balloc(vp, lfs_dino_getsize(fs, ip->i_din), lfs_sb_getbsize(fs), &bp); 962 lfs_dino_setsize(fs, ip->i_din, 963 lfs_dino_getsize(fs, ip->i_din) + lfs_sb_getbsize(fs)); 964 ip->i_state |= IN_MODIFIED; 965 966 i = (blkno - lfs_sb_getsegtabsz(fs) - lfs_sb_getcleansz(fs)) * 967 lfs_sb_getifpb(fs); 968 LFS_GET_HEADFREE(fs, cip, cbp, &oldlast); 969 LFS_PUT_HEADFREE(fs, cip, cbp, i); 970 max = i + lfs_sb_getifpb(fs); 971 lfs_sb_subbfree(fs, lfs_btofsb(fs, lfs_sb_getbsize(fs))); 972 973 if (fs->lfs_is64) { 974 for (ifp64 = (IFILE64 *)bp->b_data; i < max; ++ifp64) { 975 ifp64->if_version = 1; 976 ifp64->if_daddr = LFS_UNUSED_DADDR; 977 ifp64->if_nextfree = ++i; 978 } 979 ifp64--; 980 ifp64->if_nextfree = oldlast; 981 } else if (lfs_sb_getversion(fs) > 1) { 982 for (ifp32 = (IFILE32 *)bp->b_data; i < max; ++ifp32) { 983 ifp32->if_version = 1; 984 ifp32->if_daddr = LFS_UNUSED_DADDR; 985 ifp32->if_nextfree = ++i; 986 } 987 ifp32--; 988 ifp32->if_nextfree = oldlast; 989 } else { 990 for (ifp_v1 = (IFILE_V1 *)bp->b_data; i < max; ++ifp_v1) { 991 ifp_v1->if_version = 1; 992 ifp_v1->if_daddr = LFS_UNUSED_DADDR; 993 ifp_v1->if_nextfree = ++i; 994 } 995 ifp_v1--; 996 ifp_v1->if_nextfree = oldlast; 997 } 998 LFS_PUT_TAILFREE(fs, cip, cbp, max - 1); 999 1000 LFS_BWRITE_LOG(bp); 1001 1002 #ifdef IN_FSCK_LFS 1003 reset_maxino(((lfs_dino_getsize(fs, ip->i_din) >> lfs_sb_getbshift(fs)) 1004 - lfs_sb_getsegtabsz(fs) 1005 - lfs_sb_getcleansz(fs)) * lfs_sb_getifpb(fs)); 1006 #endif 1007 return 0; 1008 } 1009 1010 /* 1011 * Allocate a block, and to inode and filesystem block accounting for it 1012 * and for any indirect blocks the may need to be created in order for 1013 * this block to be created. 1014 * 1015 * Blocks which have never been accounted for (i.e., which "do not exist") 1016 * have disk address 0, which is translated by ulfs_bmap to the special value 1017 * UNASSIGNED == -1, as in the historical ULFS. 1018 * 1019 * Blocks which have been accounted for but which have not yet been written 1020 * to disk are given the new special disk address UNWRITTEN == -2, so that 1021 * they can be differentiated from completely new blocks. 1022 */ 1023 int 1024 lfs_balloc(struct uvnode *vp, off_t startoffset, int iosize, struct ubuf **bpp) 1025 { 1026 int offset; 1027 daddr_t daddr, idaddr; 1028 struct ubuf *ibp, *bp; 1029 struct inode *ip; 1030 struct lfs *fs; 1031 struct indir indirs[ULFS_NIADDR+2], *idp; 1032 daddr_t lbn, lastblock; 1033 int bcount; 1034 int error, frags, i, nsize, osize, num; 1035 1036 ip = VTOI(vp); 1037 fs = ip->i_lfs; 1038 offset = lfs_blkoff(fs, startoffset); 1039 lbn = lfs_lblkno(fs, startoffset); 1040 1041 /* 1042 * Three cases: it's a block beyond the end of file, it's a block in 1043 * the file that may or may not have been assigned a disk address or 1044 * we're writing an entire block. 1045 * 1046 * Note, if the daddr is UNWRITTEN, the block already exists in 1047 * the cache (it was read or written earlier). If so, make sure 1048 * we don't count it as a new block or zero out its contents. If 1049 * it did not, make sure we allocate any necessary indirect 1050 * blocks. 1051 * 1052 * If we are writing a block beyond the end of the file, we need to 1053 * check if the old last block was a fragment. If it was, we need 1054 * to rewrite it. 1055 */ 1056 1057 if (bpp) 1058 *bpp = NULL; 1059 1060 /* Check for block beyond end of file and fragment extension needed. */ 1061 lastblock = lfs_lblkno(fs, lfs_dino_getsize(fs, ip->i_din)); 1062 if (lastblock < ULFS_NDADDR && lastblock < lbn) { 1063 osize = lfs_blksize(fs, ip, lastblock); 1064 if (osize < lfs_sb_getbsize(fs) && osize > 0) { 1065 if ((error = lfs_fragextend(vp, osize, lfs_sb_getbsize(fs), 1066 lastblock, 1067 (bpp ? &bp : NULL)))) 1068 return (error); 1069 lfs_dino_setsize(fs, ip->i_din, (lastblock + 1) * lfs_sb_getbsize(fs)); 1070 ip->i_state |= IN_CHANGE | IN_UPDATE; 1071 if (bpp) 1072 (void) VOP_BWRITE(bp); 1073 } 1074 } 1075 1076 /* 1077 * If the block we are writing is a direct block, it's the last 1078 * block in the file, and offset + iosize is less than a full 1079 * block, we can write one or more fragments. There are two cases: 1080 * the block is brand new and we should allocate it the correct 1081 * size or it already exists and contains some fragments and 1082 * may need to extend it. 1083 */ 1084 if (lbn < ULFS_NDADDR && lfs_lblkno(fs, lfs_dino_getsize(fs, ip->i_din)) <= lbn) { 1085 osize = lfs_blksize(fs, ip, lbn); 1086 nsize = lfs_fragroundup(fs, offset + iosize); 1087 if (lfs_lblktosize(fs, lbn) >= lfs_dino_getsize(fs, ip->i_din)) { 1088 /* Brand new block or fragment */ 1089 frags = lfs_numfrags(fs, nsize); 1090 if (bpp) { 1091 *bpp = bp = getblk(vp, lbn, nsize); 1092 bp->b_blkno = UNWRITTEN; 1093 } 1094 ip->i_lfs_effnblks += frags; 1095 lfs_sb_subbfree(fs, frags); 1096 lfs_dino_setdb(fs, ip->i_din, lbn, UNWRITTEN); 1097 } else { 1098 if (nsize <= osize) { 1099 /* No need to extend */ 1100 if (bpp && (error = bread(vp, lbn, osize, 1101 0, &bp))) 1102 return error; 1103 } else { 1104 /* Extend existing block */ 1105 if ((error = 1106 lfs_fragextend(vp, osize, nsize, lbn, 1107 (bpp ? &bp : NULL)))) 1108 return error; 1109 } 1110 if (bpp) 1111 *bpp = bp; 1112 } 1113 return 0; 1114 } 1115 1116 error = ulfs_bmaparray(fs, vp, lbn, &daddr, &indirs[0], &num); 1117 if (error) 1118 return (error); 1119 1120 /* 1121 * Do byte accounting all at once, so we can gracefully fail *before* 1122 * we start assigning blocks. 1123 */ 1124 frags = LFS_FSBTODB(fs, 1); /* frags = VFSTOULFS(vp->v_mount)->um_seqinc; */ 1125 bcount = 0; 1126 if (daddr == UNASSIGNED) { 1127 bcount = frags; 1128 } 1129 for (i = 1; i < num; ++i) { 1130 if (!indirs[i].in_exists) { 1131 bcount += frags; 1132 } 1133 } 1134 lfs_sb_subbfree(fs, bcount); 1135 ip->i_lfs_effnblks += bcount; 1136 1137 if (daddr == UNASSIGNED) { 1138 if (num > 0 && lfs_dino_getib(fs, ip->i_din, indirs[0].in_off) == 0) { 1139 lfs_dino_setib(fs, ip->i_din, indirs[0].in_off, 1140 UNWRITTEN); 1141 } 1142 1143 /* 1144 * Create new indirect blocks if necessary 1145 */ 1146 if (num > 1) { 1147 idaddr = lfs_dino_getib(fs, ip->i_din, indirs[0].in_off); 1148 for (i = 1; i < num; ++i) { 1149 ibp = getblk(vp, indirs[i].in_lbn, 1150 lfs_sb_getbsize(fs)); 1151 if (!indirs[i].in_exists) { 1152 memset(ibp->b_data, 0, ibp->b_bufsize); 1153 ibp->b_blkno = UNWRITTEN; 1154 } else if (!(ibp->b_flags & (B_DELWRI | B_DONE))) { 1155 ibp->b_blkno = LFS_FSBTODB(fs, idaddr); 1156 ibp->b_flags |= B_READ; 1157 VOP_STRATEGY(ibp); 1158 } 1159 /* 1160 * This block exists, but the next one may not. 1161 * If that is the case mark it UNWRITTEN to 1162 * keep the accounting straight. 1163 */ 1164 if (lfs_iblock_get(fs, ibp->b_data, 1165 indirs[i].in_off) == 0) 1166 lfs_iblock_set(fs, ibp->b_data, 1167 indirs[i].in_off, UNWRITTEN); 1168 idaddr = lfs_iblock_get(fs, ibp->b_data, 1169 indirs[i].in_off); 1170 if ((error = VOP_BWRITE(ibp))) 1171 return error; 1172 } 1173 } 1174 } 1175 1176 1177 /* 1178 * Get the existing block from the cache, if requested. 1179 */ 1180 if (bpp) 1181 *bpp = bp = getblk(vp, lbn, lfs_blksize(fs, ip, lbn)); 1182 1183 /* 1184 * The block we are writing may be a brand new block 1185 * in which case we need to do accounting. 1186 * 1187 * We can tell a truly new block because ulfs_bmaparray will say 1188 * it is UNASSIGNED. Once we allocate it we will assign it the 1189 * disk address UNWRITTEN. 1190 */ 1191 if (daddr == UNASSIGNED) { 1192 if (bpp) { 1193 /* Note the new address */ 1194 bp->b_blkno = UNWRITTEN; 1195 } 1196 1197 switch (num) { 1198 case 0: 1199 lfs_dino_setdb(fs, ip->i_din, lbn, UNWRITTEN); 1200 break; 1201 case 1: 1202 lfs_dino_setib(fs, ip->i_din, indirs[0].in_off, 1203 UNWRITTEN); 1204 break; 1205 default: 1206 idp = &indirs[num - 1]; 1207 if (bread(vp, idp->in_lbn, lfs_sb_getbsize(fs), 0, &ibp)) 1208 panic("lfs_balloc: bread bno %lld", 1209 (long long)idp->in_lbn); 1210 lfs_iblock_set(fs, ibp->b_data, idp->in_off, 1211 UNWRITTEN); 1212 VOP_BWRITE(ibp); 1213 } 1214 } else if (bpp && !(bp->b_flags & (B_DONE|B_DELWRI))) { 1215 /* 1216 * Not a brand new block, also not in the cache; 1217 * read it in from disk. 1218 */ 1219 if (iosize == lfs_sb_getbsize(fs)) 1220 /* Optimization: I/O is unnecessary. */ 1221 bp->b_blkno = daddr; 1222 else { 1223 /* 1224 * We need to read the block to preserve the 1225 * existing bytes. 1226 */ 1227 bp->b_blkno = daddr; 1228 bp->b_flags |= B_READ; 1229 VOP_STRATEGY(bp); 1230 return 0; 1231 } 1232 } 1233 1234 return (0); 1235 } 1236 1237 int 1238 lfs_fragextend(struct uvnode *vp, int osize, int nsize, daddr_t lbn, 1239 struct ubuf **bpp) 1240 { 1241 struct inode *ip; 1242 struct lfs *fs; 1243 int frags; 1244 int error; 1245 1246 ip = VTOI(vp); 1247 fs = ip->i_lfs; 1248 frags = (long)lfs_numfrags(fs, nsize - osize); 1249 error = 0; 1250 1251 /* 1252 * If we are not asked to actually return the block, all we need 1253 * to do is allocate space for it. UBC will handle dirtying the 1254 * appropriate things and making sure it all goes to disk. 1255 * Don't bother to read in that case. 1256 */ 1257 if (bpp && (error = bread(vp, lbn, osize, 0, bpp))) { 1258 brelse(*bpp, 0); 1259 goto out; 1260 } 1261 1262 lfs_sb_subbfree(fs, frags); 1263 ip->i_lfs_effnblks += frags; 1264 ip->i_state |= IN_CHANGE | IN_UPDATE; 1265 1266 if (bpp) { 1267 (*bpp)->b_data = erealloc((*bpp)->b_data, nsize); 1268 (void)memset((*bpp)->b_data + osize, 0, nsize - osize); 1269 } 1270 1271 out: 1272 return (error); 1273 } 1274
Indexes created Thu Nov 06 02:09:54 GMT 2025