lfs.c revision 1.61
1/* $NetBSD: lfs.c,v 1.61 2015/09/01 06:13:33 dholland 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@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 <stdio.h> 84#include <stdlib.h> 85#include <string.h> 86#include <unistd.h> 87#include <util.h> 88 89#include "bufcache.h" 90#include "vnode.h" 91#include "lfs_user.h" 92#include "segwrite.h" 93#include "kernelops.h" 94 95#define panic call_panic 96 97extern u_int32_t cksum(void *, size_t); 98extern u_int32_t lfs_sb_cksum(struct lfs *); 99extern void pwarn(const char *, ...); 100 101extern struct uvnodelst vnodelist; 102extern struct uvnodelst getvnodelist[VNODE_HASH_MAX]; 103extern int nvnodes; 104 105long dev_bsize = DEV_BSIZE; 106 107static int 108lfs_fragextend(struct uvnode *, int, int, daddr_t, struct ubuf **); 109 110int fsdirty = 0; 111void (*panic_func)(int, const char *, va_list) = my_vpanic; 112 113/* 114 * LFS buffer and uvnode operations 115 */ 116 117int 118lfs_vop_strategy(struct ubuf * bp) 119{ 120 int count; 121 122 if (bp->b_flags & B_READ) { 123 count = kops.ko_pread(bp->b_vp->v_fd, bp->b_data, bp->b_bcount, 124 bp->b_blkno * dev_bsize); 125 if (count == bp->b_bcount) 126 bp->b_flags |= B_DONE; 127 } else { 128 count = kops.ko_pwrite(bp->b_vp->v_fd, bp->b_data, bp->b_bcount, 129 bp->b_blkno * dev_bsize); 130 if (count == 0) { 131 perror("pwrite"); 132 return -1; 133 } 134 bp->b_flags &= ~B_DELWRI; 135 reassignbuf(bp, bp->b_vp); 136 } 137 return 0; 138} 139 140int 141lfs_vop_bwrite(struct ubuf * bp) 142{ 143 struct lfs *fs; 144 145 fs = bp->b_vp->v_fs; 146 if (!(bp->b_flags & B_DELWRI)) { 147 lfs_sb_subavail(fs, lfs_btofsb(fs, bp->b_bcount)); 148 } 149 bp->b_flags |= B_DELWRI | B_LOCKED; 150 reassignbuf(bp, bp->b_vp); 151 brelse(bp, 0); 152 return 0; 153} 154 155/* 156 * ulfs_bmaparray does the bmap conversion, and if requested returns the 157 * array of logical blocks which must be traversed to get to a block. 158 * Each entry contains the offset into that block that gets you to the 159 * next block and the disk address of the block (if it is assigned). 160 */ 161int 162ulfs_bmaparray(struct lfs * fs, struct uvnode * vp, daddr_t bn, daddr_t * bnp, struct indir * ap, int *nump) 163{ 164 struct inode *ip; 165 struct ubuf *bp; 166 struct indir a[ULFS_NIADDR + 1], *xap; 167 daddr_t daddr; 168 daddr_t metalbn; 169 int error, num; 170 171 ip = VTOI(vp); 172 173 if (bn >= 0 && bn < ULFS_NDADDR) { 174 if (nump != NULL) 175 *nump = 0; 176 *bnp = LFS_FSBTODB(fs, lfs_dino_getdb(fs, ip->i_din, bn)); 177 if (*bnp == 0) 178 *bnp = -1; 179 return (0); 180 } 181 xap = ap == NULL ? a : ap; 182 if (!nump) 183 nump = # 184 if ((error = ulfs_getlbns(fs, vp, bn, xap, nump)) != 0) 185 return (error); 186 187 num = *nump; 188 189 /* Get disk address out of indirect block array */ 190 daddr = lfs_dino_getib(fs, ip->i_din, xap->in_off); 191 192 for (bp = NULL, ++xap; --num; ++xap) { 193 /* Exit the loop if there is no disk address assigned yet and 194 * the indirect block isn't in the cache, or if we were 195 * looking for an indirect block and we've found it. */ 196 197 metalbn = xap->in_lbn; 198 if ((daddr == 0 && !incore(vp, metalbn)) || metalbn == bn) 199 break; 200 /* 201 * If we get here, we've either got the block in the cache 202 * or we have a disk address for it, go fetch it. 203 */ 204 if (bp) 205 brelse(bp, 0); 206 207 xap->in_exists = 1; 208 bp = getblk(vp, metalbn, lfs_sb_getbsize(fs)); 209 210 if (!(bp->b_flags & (B_DONE | B_DELWRI))) { 211 bp->b_blkno = LFS_FSBTODB(fs, daddr); 212 bp->b_flags |= B_READ; 213 VOP_STRATEGY(bp); 214 } 215 daddr = ((ulfs_daddr_t *) bp->b_data)[xap->in_off]; 216 } 217 if (bp) 218 brelse(bp, 0); 219 220 daddr = LFS_FSBTODB(fs, (ulfs_daddr_t) daddr); 221 *bnp = daddr == 0 ? -1 : daddr; 222 return (0); 223} 224 225/* 226 * Create an array of logical block number/offset pairs which represent the 227 * path of indirect blocks required to access a data block. The first "pair" 228 * contains the logical block number of the appropriate single, double or 229 * triple indirect block and the offset into the inode indirect block array. 230 * Note, the logical block number of the inode single/double/triple indirect 231 * block appears twice in the array, once with the offset into di_ib and 232 * once with the offset into the page itself. 233 */ 234int 235ulfs_getlbns(struct lfs * fs, struct uvnode * vp, daddr_t bn, struct indir * ap, int *nump) 236{ 237 daddr_t metalbn, realbn; 238 int64_t blockcnt; 239 int lbc; 240 int i, numlevels, off; 241 int lognindir, indir; 242 243 metalbn = 0; /* XXXGCC -Wuninitialized [sh3] */ 244 245 if (nump) 246 *nump = 0; 247 numlevels = 0; 248 realbn = bn; 249 if (bn < 0) 250 bn = -bn; 251 252 lognindir = -1; 253 for (indir = lfs_sb_getnindir(fs); indir; indir >>= 1) 254 ++lognindir; 255 256 /* Determine the number of levels of indirection. After this loop is 257 * done, blockcnt indicates the number of data blocks possible at the 258 * given level of indirection, and ULFS_NIADDR - i is the number of levels 259 * of indirection needed to locate the requested block. */ 260 261 bn -= ULFS_NDADDR; 262 for (lbc = 0, i = ULFS_NIADDR;; i--, bn -= blockcnt) { 263 if (i == 0) 264 return (EFBIG); 265 266 lbc += lognindir; 267 blockcnt = (int64_t) 1 << lbc; 268 269 if (bn < blockcnt) 270 break; 271 } 272 273 /* Calculate the address of the first meta-block. */ 274 metalbn = -((realbn >= 0 ? realbn : -realbn) - bn + ULFS_NIADDR - i); 275 276 /* At each iteration, off is the offset into the bap array which is an 277 * array of disk addresses at the current level of indirection. The 278 * logical block number and the offset in that block are stored into 279 * the argument array. */ 280 ap->in_lbn = metalbn; 281 ap->in_off = off = ULFS_NIADDR - i; 282 ap->in_exists = 0; 283 ap++; 284 for (++numlevels; i <= ULFS_NIADDR; i++) { 285 /* If searching for a meta-data block, quit when found. */ 286 if (metalbn == realbn) 287 break; 288 289 lbc -= lognindir; 290 blockcnt = (int64_t) 1 << lbc; 291 off = (bn >> lbc) & (lfs_sb_getnindir(fs) - 1); 292 293 ++numlevels; 294 ap->in_lbn = metalbn; 295 ap->in_off = off; 296 ap->in_exists = 0; 297 ++ap; 298 299 metalbn -= -1 + (off << lbc); 300 } 301 if (nump) 302 *nump = numlevels; 303 return (0); 304} 305 306int 307lfs_vop_bmap(struct uvnode * vp, daddr_t lbn, daddr_t * daddrp) 308{ 309 return ulfs_bmaparray(vp->v_fs, vp, lbn, daddrp, NULL, NULL); 310} 311 312/* Search a block for a specific dinode. */ 313union lfs_dinode * 314lfs_ifind(struct lfs *fs, ino_t ino, struct ubuf *bp) 315{ 316 union lfs_dinode *ldip; 317 unsigned i, num; 318 319 num = LFS_INOPB(fs); 320 321 /* 322 * Read the inode block backwards, since later versions of the 323 * inode will supercede earlier ones. Though it is unlikely, it is 324 * possible that the same inode will appear in the same inode block. 325 */ 326 for (i = num; i-- > 0; ) { 327 ldip = DINO_IN_BLOCK(fs, bp->b_data, i); 328 if (lfs_dino_getinumber(fs, ldip) == ino) 329 return (ldip); 330 } 331 return NULL; 332} 333 334/* 335 * lfs_raw_vget makes us a new vnode from the inode at the given disk address. 336 * XXX it currently loses atime information. 337 */ 338struct uvnode * 339lfs_raw_vget(struct lfs * fs, ino_t ino, int fd, daddr_t daddr) 340{ 341 struct uvnode *vp; 342 struct inode *ip; 343 union lfs_dinode *dip; 344 struct ubuf *bp; 345 int i, hash; 346 347 vp = ecalloc(1, sizeof(*vp)); 348 vp->v_fd = fd; 349 vp->v_fs = fs; 350 vp->v_usecount = 0; 351 vp->v_strategy_op = lfs_vop_strategy; 352 vp->v_bwrite_op = lfs_vop_bwrite; 353 vp->v_bmap_op = lfs_vop_bmap; 354 LIST_INIT(&vp->v_cleanblkhd); 355 LIST_INIT(&vp->v_dirtyblkhd); 356 357 ip = ecalloc(1, sizeof(*ip)); 358 359 ip->i_din = dip = ecalloc(1, sizeof(*dip)); 360 361 /* Initialize the inode -- from lfs_vcreate. */ 362 ip->inode_ext.lfs = ecalloc(1, sizeof(*ip->inode_ext.lfs)); 363 vp->v_data = ip; 364 /* ip->i_vnode = vp; */ 365 ip->i_number = ino; 366 ip->i_lockf = 0; 367 ip->i_lfs_effnblks = 0; 368 ip->i_flag = 0; 369 370 /* Load inode block and find inode */ 371 if (daddr > 0) { 372 bread(fs->lfs_devvp, LFS_FSBTODB(fs, daddr), lfs_sb_getibsize(fs), 373 0, &bp); 374 bp->b_flags |= B_AGE; 375 dip = lfs_ifind(fs, ino, bp); 376 if (dip == NULL) { 377 brelse(bp, 0); 378 free(ip); 379 free(vp); 380 return NULL; 381 } 382 lfs_copy_dinode(fs, ip->i_din, dip); 383 brelse(bp, 0); 384 } 385 ip->i_number = ino; 386 /* ip->i_devvp = fs->lfs_devvp; */ 387 ip->i_lfs = fs; 388 389 ip->i_lfs_effnblks = lfs_dino_getblocks(fs, ip->i_din); 390 ip->i_lfs_osize = lfs_dino_getsize(fs, ip->i_din); 391#if 0 392 if (lfs_sb_getversion(fs) > 1) { 393 lfs_dino_setatime(fs, ip->i_din, ts.tv_sec); 394 lfs_dino_setatimensec(fs, ip->i_din, ts.tv_nsec); 395 } 396#endif 397 398 memset(ip->i_lfs_fragsize, 0, ULFS_NDADDR * sizeof(*ip->i_lfs_fragsize)); 399 for (i = 0; i < ULFS_NDADDR; i++) 400 if (lfs_dino_getdb(fs, ip->i_din, i) != 0) 401 ip->i_lfs_fragsize[i] = lfs_blksize(fs, ip, i); 402 403 ++nvnodes; 404 hash = ((int)(intptr_t)fs + ino) & (VNODE_HASH_MAX - 1); 405 LIST_INSERT_HEAD(&getvnodelist[hash], vp, v_getvnodes); 406 LIST_INSERT_HEAD(&vnodelist, vp, v_mntvnodes); 407 408 return vp; 409} 410 411static struct uvnode * 412lfs_vget(void *vfs, ino_t ino) 413{ 414 struct lfs *fs = (struct lfs *)vfs; 415 daddr_t daddr; 416 struct ubuf *bp; 417 IFILE *ifp; 418 419 LFS_IENTRY(ifp, fs, ino, bp); 420 daddr = lfs_if_getdaddr(fs, ifp); 421 brelse(bp, 0); 422 if (daddr <= 0 || lfs_dtosn(fs, daddr) >= lfs_sb_getnseg(fs)) 423 return NULL; 424 return lfs_raw_vget(fs, ino, fs->lfs_ivnode->v_fd, daddr); 425} 426 427/* 428 * Check superblock magic number and checksum. 429 * Sets lfs_is64 and lfs_dobyteswap. 430 */ 431static int 432check_sb(struct lfs *fs) 433{ 434 u_int32_t checksum; 435 u_int32_t magic; 436 437 /* we can read the magic out of either the 32-bit or 64-bit dlfs */ 438 magic = fs->lfs_dlfs_u.u_32.dlfs_magic; 439 440 if (magic != LFS_MAGIC) { 441 printf("Superblock magic number (0x%lx) does not match " 442 "expected 0x%lx\n", (unsigned long) magic, 443 (unsigned long) LFS_MAGIC); 444 return 1; 445 } 446 fs->lfs_is64 = 0; /* XXX notyet */ 447 fs->lfs_dobyteswap = 0; /* XXX notyet */ 448 449 /* checksum */ 450 checksum = lfs_sb_cksum(fs); 451 if (lfs_sb_getcksum(fs) != checksum) { 452 printf("Superblock checksum (%lx) does not match computed checksum (%lx)\n", 453 (unsigned long) lfs_sb_getcksum(fs), (unsigned long) checksum); 454 return 1; 455 } 456 return 0; 457} 458 459/* Initialize LFS library; load superblocks and choose which to use. */ 460struct lfs * 461lfs_init(int devfd, daddr_t sblkno, daddr_t idaddr, int dummy_read, int debug) 462{ 463 struct uvnode *devvp; 464 struct ubuf *bp; 465 int tryalt; 466 struct lfs *fs, *altfs; 467 468 vfs_init(); 469 470 devvp = ecalloc(1, sizeof(*devvp)); 471 devvp->v_fs = NULL; 472 devvp->v_fd = devfd; 473 devvp->v_strategy_op = raw_vop_strategy; 474 devvp->v_bwrite_op = raw_vop_bwrite; 475 devvp->v_bmap_op = raw_vop_bmap; 476 LIST_INIT(&devvp->v_cleanblkhd); 477 LIST_INIT(&devvp->v_dirtyblkhd); 478 479 tryalt = 0; 480 if (dummy_read) { 481 if (sblkno == 0) 482 sblkno = LFS_LABELPAD / dev_bsize; 483 fs = ecalloc(1, sizeof(*fs)); 484 fs->lfs_devvp = devvp; 485 } else { 486 if (sblkno == 0) { 487 sblkno = LFS_LABELPAD / dev_bsize; 488 tryalt = 1; 489 } else if (debug) { 490 printf("No -b flag given, not attempting to verify checkpoint\n"); 491 } 492 493 dev_bsize = DEV_BSIZE; 494 495 (void)bread(devvp, sblkno, LFS_SBPAD, 0, &bp); 496 fs = ecalloc(1, sizeof(*fs)); 497 __CTASSERT(sizeof(struct dlfs) == sizeof(struct dlfs64)); 498 memcpy(&fs->lfs_dlfs_u, bp->b_data, sizeof(struct dlfs)); 499 fs->lfs_devvp = devvp; 500 bp->b_flags |= B_INVAL; 501 brelse(bp, 0); 502 503 dev_bsize = lfs_sb_getfsize(fs) >> lfs_sb_getfsbtodb(fs); 504 505 if (tryalt) { 506 (void)bread(devvp, LFS_FSBTODB(fs, lfs_sb_getsboff(fs, 1)), 507 LFS_SBPAD, 0, &bp); 508 altfs = ecalloc(1, sizeof(*altfs)); 509 memcpy(&altfs->lfs_dlfs_u, bp->b_data, 510 sizeof(struct dlfs)); 511 altfs->lfs_devvp = devvp; 512 bp->b_flags |= B_INVAL; 513 brelse(bp, 0); 514 515 if (check_sb(fs) || lfs_sb_getidaddr(fs) <= 0) { 516 if (debug) 517 printf("Primary superblock is no good, using first alternate\n"); 518 free(fs); 519 fs = altfs; 520 } else { 521 /* If both superblocks check out, try verification */ 522 if (check_sb(altfs)) { 523 if (debug) 524 printf("First alternate superblock is no good, using primary\n"); 525 free(altfs); 526 } else { 527 if (lfs_verify(fs, altfs, devvp, debug) == fs) { 528 free(altfs); 529 } else { 530 free(fs); 531 fs = altfs; 532 } 533 } 534 } 535 } 536 if (check_sb(fs)) { 537 free(fs); 538 return NULL; 539 } 540 } 541 542 /* Compatibility */ 543 if (lfs_sb_getversion(fs) < 2) { 544 lfs_sb_setsumsize(fs, LFS_V1_SUMMARY_SIZE); 545 lfs_sb_setibsize(fs, lfs_sb_getbsize(fs)); 546 lfs_sb_sets0addr(fs, lfs_sb_getsboff(fs, 0)); 547 lfs_sb_settstamp(fs, lfs_sb_getotstamp(fs)); 548 lfs_sb_setfsbtodb(fs, 0); 549 } 550 551 if (!dummy_read) { 552 fs->lfs_suflags = emalloc(2 * sizeof(u_int32_t *)); 553 fs->lfs_suflags[0] = emalloc(lfs_sb_getnseg(fs) * sizeof(u_int32_t)); 554 fs->lfs_suflags[1] = emalloc(lfs_sb_getnseg(fs) * sizeof(u_int32_t)); 555 } 556 557 if (idaddr == 0) 558 idaddr = lfs_sb_getidaddr(fs); 559 else 560 lfs_sb_setidaddr(fs, idaddr); 561 /* NB: If dummy_read!=0, idaddr==0 here so we get a fake inode. */ 562 fs->lfs_ivnode = lfs_raw_vget(fs, LFS_IFILE_INUM, 563 devvp->v_fd, idaddr); 564 if (fs->lfs_ivnode == NULL) 565 return NULL; 566 567 register_vget((void *)fs, lfs_vget); 568 569 return fs; 570} 571 572/* 573 * Check partial segment validity between fs->lfs_offset and the given goal. 574 * 575 * If goal == 0, just keep on going until the segments stop making sense, 576 * and return the address of the last valid partial segment. 577 * 578 * If goal != 0, return the address of the first partial segment that failed, 579 * or "goal" if we reached it without failure (the partial segment *at* goal 580 * need not be valid). 581 */ 582ulfs_daddr_t 583try_verify(struct lfs *osb, struct uvnode *devvp, ulfs_daddr_t goal, int debug) 584{ 585 ulfs_daddr_t daddr, odaddr; 586 SEGSUM *sp; 587 int i, bc, hitclean; 588 struct ubuf *bp; 589 ulfs_daddr_t nodirop_daddr; 590 u_int64_t serial; 591 592 bc = 0; 593 hitclean = 0; 594 odaddr = -1; 595 daddr = lfs_sb_getoffset(osb); 596 nodirop_daddr = daddr; 597 serial = lfs_sb_getserial(osb); 598 while (daddr != goal) { 599 /* 600 * Don't mistakenly read a superblock, if there is one here. 601 */ 602 if (lfs_sntod(osb, lfs_dtosn(osb, daddr)) == daddr) { 603 if (daddr == lfs_sb_gets0addr(osb)) 604 daddr += lfs_btofsb(osb, LFS_LABELPAD); 605 for (i = 0; i < LFS_MAXNUMSB; i++) { 606 if (lfs_sb_getsboff(osb, i) < daddr) 607 break; 608 if (lfs_sb_getsboff(osb, i) == daddr) 609 daddr += lfs_btofsb(osb, LFS_SBPAD); 610 } 611 } 612 613 /* Read in summary block */ 614 bread(devvp, LFS_FSBTODB(osb, daddr), lfs_sb_getsumsize(osb), 615 0, &bp); 616 sp = (SEGSUM *)bp->b_data; 617 618 /* 619 * Check for a valid segment summary belonging to our fs. 620 */ 621 if (lfs_ss_getmagic(osb, sp) != SS_MAGIC || 622 lfs_ss_getident(osb, sp) != lfs_sb_getident(osb) || 623 lfs_ss_getserial(osb, sp) < serial || /* XXX strengthen this */ 624 lfs_ss_getsumsum(osb, sp) != 625 cksum((char *)sp + lfs_ss_getsumstart(osb), 626 lfs_sb_getsumsize(osb) - lfs_ss_getsumstart(osb))) { 627 brelse(bp, 0); 628 if (debug) { 629 if (lfs_ss_getmagic(osb, sp) != SS_MAGIC) 630 pwarn("pseg at 0x%jx: " 631 "wrong magic number\n", 632 (uintmax_t)daddr); 633 else if (lfs_ss_getident(osb, sp) != lfs_sb_getident(osb)) 634 pwarn("pseg at 0x%jx: " 635 "expected ident %jx, got %jx\n", 636 (uintmax_t)daddr, 637 (uintmax_t)lfs_ss_getident(osb, sp), 638 (uintmax_t)lfs_sb_getident(osb)); 639 else if (lfs_ss_getserial(osb, sp) >= serial) 640 pwarn("pseg at 0x%jx: " 641 "serial %d < %d\n", 642 (uintmax_t)daddr, 643 (int)lfs_ss_getserial(osb, sp), (int)serial); 644 else 645 pwarn("pseg at 0x%jx: " 646 "summary checksum wrong\n", 647 (uintmax_t)daddr); 648 } 649 break; 650 } 651 if (debug && lfs_ss_getserial(osb, sp) != serial) 652 pwarn("warning, serial=%d ss_serial=%d\n", 653 (int)serial, (int)lfs_ss_getserial(osb, sp)); 654 ++serial; 655 bc = check_summary(osb, sp, daddr, debug, devvp, NULL); 656 if (bc == 0) { 657 brelse(bp, 0); 658 break; 659 } 660 if (debug) 661 pwarn("summary good: 0x%x/%d\n", (uintmax_t)daddr, 662 (int)lfs_ss_getserial(osb, sp)); 663 assert (bc > 0); 664 odaddr = daddr; 665 daddr += lfs_btofsb(osb, lfs_sb_getsumsize(osb) + bc); 666 if (lfs_dtosn(osb, odaddr) != lfs_dtosn(osb, daddr) || 667 lfs_dtosn(osb, daddr) != lfs_dtosn(osb, daddr + 668 lfs_btofsb(osb, lfs_sb_getsumsize(osb) + lfs_sb_getbsize(osb)) - 1)) { 669 daddr = lfs_ss_getnext(osb, sp); 670 } 671 672 /* 673 * Check for the beginning and ending of a sequence of 674 * dirops. Writes from the cleaner never involve new 675 * information, and are always checkpoints; so don't try 676 * to roll forward through them. Likewise, psegs written 677 * by a previous roll-forward attempt are not interesting. 678 */ 679 if (lfs_ss_getflags(osb, sp) & (SS_CLEAN | SS_RFW)) 680 hitclean = 1; 681 if (hitclean == 0 && (lfs_ss_getflags(osb, sp) & SS_CONT) == 0) 682 nodirop_daddr = daddr; 683 684 brelse(bp, 0); 685 } 686 687 if (goal == 0) 688 return nodirop_daddr; 689 else 690 return daddr; 691} 692 693/* Use try_verify to check whether the newer superblock is valid. */ 694struct lfs * 695lfs_verify(struct lfs *sb0, struct lfs *sb1, struct uvnode *devvp, int debug) 696{ 697 ulfs_daddr_t daddr; 698 struct lfs *osb, *nsb; 699 700 /* 701 * Verify the checkpoint of the newer superblock, 702 * if the timestamp/serial number of the two superblocks is 703 * different. 704 */ 705 706 osb = NULL; 707 if (debug) 708 pwarn("sb0 %ju, sb1 %ju", 709 (uintmax_t) lfs_sb_getserial(sb0), 710 (uintmax_t) lfs_sb_getserial(sb1)); 711 712 if ((lfs_sb_getversion(sb0) == 1 && 713 lfs_sb_getotstamp(sb0) != lfs_sb_getotstamp(sb1)) || 714 (lfs_sb_getversion(sb0) > 1 && 715 lfs_sb_getserial(sb0) != lfs_sb_getserial(sb1))) { 716 if (lfs_sb_getversion(sb0) == 1) { 717 if (lfs_sb_getotstamp(sb0) > lfs_sb_getotstamp(sb1)) { 718 osb = sb1; 719 nsb = sb0; 720 } else { 721 osb = sb0; 722 nsb = sb1; 723 } 724 } else { 725 if (lfs_sb_getserial(sb0) > lfs_sb_getserial(sb1)) { 726 osb = sb1; 727 nsb = sb0; 728 } else { 729 osb = sb0; 730 nsb = sb1; 731 } 732 } 733 if (debug) { 734 printf("Attempting to verify newer checkpoint..."); 735 fflush(stdout); 736 } 737 daddr = try_verify(osb, devvp, lfs_sb_getoffset(nsb), debug); 738 739 if (debug) 740 printf("done.\n"); 741 if (daddr == lfs_sb_getoffset(nsb)) { 742 pwarn("** Newer checkpoint verified; recovered %jd seconds of data\n", 743 (intmax_t)(lfs_sb_gettstamp(nsb) - lfs_sb_gettstamp(osb))); 744 sbdirty(); 745 } else { 746 pwarn("** Newer checkpoint invalid; lost %jd seconds of data\n", (intmax_t)(lfs_sb_gettstamp(nsb) - lfs_sb_gettstamp(osb))); 747 } 748 return (daddr == lfs_sb_getoffset(nsb) ? nsb : osb); 749 } 750 /* Nothing to check */ 751 return osb; 752} 753 754/* Verify a partial-segment summary; return the number of bytes on disk. */ 755int 756check_summary(struct lfs *fs, SEGSUM *sp, ulfs_daddr_t pseg_addr, int debug, 757 struct uvnode *devvp, void (func(ulfs_daddr_t, FINFO *))) 758{ 759 FINFO *fp; 760 int bc; /* Bytes in partial segment */ 761 int nblocks; 762 ulfs_daddr_t daddr; 763 ulfs_daddr_t *dp, *idp; 764 struct ubuf *bp; 765 int i, j, k, datac, len; 766 u_int32_t *datap; 767 u_int32_t ccksum; 768 769 /* We've already checked the sumsum, just do the data bounds and sum */ 770 771 /* Count the blocks. */ 772 nblocks = howmany(lfs_ss_getninos(fs, sp), LFS_INOPB(fs)); 773 bc = nblocks << (lfs_sb_getversion(fs) > 1 ? lfs_sb_getffshift(fs) : lfs_sb_getbshift(fs)); 774 assert(bc >= 0); 775 776 fp = SEGSUM_FINFOBASE(fs, sp); 777 for (i = 0; i < lfs_ss_getnfinfo(fs, sp); i++) { 778 nblocks += lfs_fi_getnblocks(fs, fp); 779 bc += lfs_fi_getlastlength(fs, fp) + ((lfs_fi_getnblocks(fs, fp) - 1) 780 << lfs_sb_getbshift(fs)); 781 assert(bc >= 0); 782 fp = NEXT_FINFO(fs, fp); 783 if (((char *)fp) - (char *)sp > lfs_sb_getsumsize(fs)) 784 return 0; 785 } 786 datap = emalloc(nblocks * sizeof(*datap)); 787 datac = 0; 788 789 dp = (ulfs_daddr_t *) sp; 790 dp += lfs_sb_getsumsize(fs) / sizeof(ulfs_daddr_t); 791 dp--; 792 793 idp = dp; 794 daddr = pseg_addr + lfs_btofsb(fs, lfs_sb_getsumsize(fs)); 795 fp = (FINFO *) (sp + 1); 796 for (i = 0, j = 0; 797 i < lfs_ss_getnfinfo(fs, sp) || j < howmany(lfs_ss_getninos(fs, sp), LFS_INOPB(fs)); i++) { 798 if (i >= lfs_ss_getnfinfo(fs, sp) && *idp != daddr) { 799 pwarn("Not enough inode blocks in pseg at 0x%" PRIx32 800 ": found %d, wanted %d\n", 801 pseg_addr, j, howmany(lfs_ss_getninos(fs, sp), 802 LFS_INOPB(fs))); 803 if (debug) 804 pwarn("*idp=%x, daddr=%" PRIx32 "\n", *idp, 805 daddr); 806 break; 807 } 808 while (j < howmany(lfs_ss_getninos(fs, sp), LFS_INOPB(fs)) && *idp == daddr) { 809 bread(devvp, LFS_FSBTODB(fs, daddr), lfs_sb_getibsize(fs), 810 0, &bp); 811 datap[datac++] = ((u_int32_t *) (bp->b_data))[0]; 812 brelse(bp, 0); 813 814 ++j; 815 daddr += lfs_btofsb(fs, lfs_sb_getibsize(fs)); 816 --idp; 817 } 818 if (i < lfs_ss_getnfinfo(fs, sp)) { 819 if (func) 820 func(daddr, fp); 821 for (k = 0; k < lfs_fi_getnblocks(fs, fp); k++) { 822 len = (k == lfs_fi_getnblocks(fs, fp) - 1 ? 823 lfs_fi_getlastlength(fs, fp) 824 : lfs_sb_getbsize(fs)); 825 bread(devvp, LFS_FSBTODB(fs, daddr), len, 826 0, &bp); 827 datap[datac++] = ((u_int32_t *) (bp->b_data))[0]; 828 brelse(bp, 0); 829 daddr += lfs_btofsb(fs, len); 830 } 831 fp = NEXT_FINFO(fs, fp); 832 } 833 } 834 835 if (datac != nblocks) { 836 pwarn("Partial segment at 0x%jx expected %d blocks counted %d\n", 837 (intmax_t)pseg_addr, nblocks, datac); 838 } 839 /* XXX ondisk32 */ 840 ccksum = cksum(datap, nblocks * sizeof(u_int32_t)); 841 /* Check the data checksum */ 842 if (ccksum != lfs_ss_getdatasum(fs, sp)) { 843 pwarn("Partial segment at 0x%jx data checksum" 844 " mismatch: given 0x%x, computed 0x%x\n", 845 (uintmax_t)pseg_addr, lfs_ss_getdatasum(fs, sp), ccksum); 846 free(datap); 847 return 0; 848 } 849 free(datap); 850 assert(bc >= 0); 851 return bc; 852} 853 854/* print message and exit */ 855void 856my_vpanic(int fatal, const char *fmt, va_list ap) 857{ 858 (void) vprintf(fmt, ap); 859 exit(8); 860} 861 862void 863call_panic(const char *fmt, ...) 864{ 865 va_list ap; 866 867 va_start(ap, fmt); 868 panic_func(1, fmt, ap); 869 va_end(ap); 870} 871 872/* Allocate a new inode. */ 873struct uvnode * 874lfs_valloc(struct lfs *fs, ino_t ino) 875{ 876 struct ubuf *bp, *cbp; 877 IFILE *ifp; 878 ino_t new_ino; 879 int error; 880 CLEANERINFO *cip; 881 882 /* Get the head of the freelist. */ 883 LFS_GET_HEADFREE(fs, cip, cbp, &new_ino); 884 885 /* 886 * Remove the inode from the free list and write the new start 887 * of the free list into the superblock. 888 */ 889 LFS_IENTRY(ifp, fs, new_ino, bp); 890 if (lfs_if_getdaddr(fs, ifp) != LFS_UNUSED_DADDR) 891 panic("lfs_valloc: inuse inode %d on the free list", new_ino); 892 LFS_PUT_HEADFREE(fs, cip, cbp, lfs_if_getnextfree(fs, ifp)); 893 894 brelse(bp, 0); 895 896 /* Extend IFILE so that the next lfs_valloc will succeed. */ 897 if (lfs_sb_getfreehd(fs) == LFS_UNUSED_INUM) { 898 if ((error = extend_ifile(fs)) != 0) { 899 LFS_PUT_HEADFREE(fs, cip, cbp, new_ino); 900 return NULL; 901 } 902 } 903 904 /* Set superblock modified bit and increment file count. */ 905 sbdirty(); 906 lfs_sb_addnfiles(fs, 1); 907 908 return lfs_raw_vget(fs, ino, fs->lfs_devvp->v_fd, 0x0); 909} 910 911#ifdef IN_FSCK_LFS 912void reset_maxino(ino_t); 913#endif 914 915/* 916 * Add a new block to the Ifile, to accommodate future file creations. 917 */ 918int 919extend_ifile(struct lfs *fs) 920{ 921 struct uvnode *vp; 922 struct inode *ip; 923 IFILE64 *ifp64; 924 IFILE32 *ifp32; 925 IFILE_V1 *ifp_v1; 926 struct ubuf *bp, *cbp; 927 daddr_t i, blkno, max; 928 ino_t oldlast; 929 CLEANERINFO *cip; 930 931 vp = fs->lfs_ivnode; 932 ip = VTOI(vp); 933 blkno = lfs_lblkno(fs, lfs_dino_getsize(fs, ip->i_din)); 934 935 lfs_balloc(vp, lfs_dino_getsize(fs, ip->i_din), lfs_sb_getbsize(fs), &bp); 936 lfs_dino_setsize(fs, ip->i_din, 937 lfs_dino_getsize(fs, ip->i_din) + lfs_sb_getbsize(fs)); 938 ip->i_flag |= IN_MODIFIED; 939 940 i = (blkno - lfs_sb_getsegtabsz(fs) - lfs_sb_getcleansz(fs)) * 941 lfs_sb_getifpb(fs); 942 LFS_GET_HEADFREE(fs, cip, cbp, &oldlast); 943 LFS_PUT_HEADFREE(fs, cip, cbp, i); 944 max = i + lfs_sb_getifpb(fs); 945 lfs_sb_subbfree(fs, lfs_btofsb(fs, lfs_sb_getbsize(fs))); 946 947 if (fs->lfs_is64) { 948 for (ifp64 = (IFILE64 *)bp->b_data; i < max; ++ifp64) { 949 ifp64->if_version = 1; 950 ifp64->if_daddr = LFS_UNUSED_DADDR; 951 ifp64->if_nextfree = ++i; 952 } 953 ifp64--; 954 ifp64->if_nextfree = oldlast; 955 } else if (lfs_sb_getversion(fs) > 1) { 956 for (ifp32 = (IFILE32 *)bp->b_data; i < max; ++ifp32) { 957 ifp32->if_version = 1; 958 ifp32->if_daddr = LFS_UNUSED_DADDR; 959 ifp32->if_nextfree = ++i; 960 } 961 ifp32--; 962 ifp32->if_nextfree = oldlast; 963 } else { 964 for (ifp_v1 = (IFILE_V1 *)bp->b_data; i < max; ++ifp_v1) { 965 ifp_v1->if_version = 1; 966 ifp_v1->if_daddr = LFS_UNUSED_DADDR; 967 ifp_v1->if_nextfree = ++i; 968 } 969 ifp_v1--; 970 ifp_v1->if_nextfree = oldlast; 971 } 972 LFS_PUT_TAILFREE(fs, cip, cbp, max - 1); 973 974 LFS_BWRITE_LOG(bp); 975 976#ifdef IN_FSCK_LFS 977 reset_maxino(((lfs_dino_getsize(fs, ip->i_din) >> lfs_sb_getbshift(fs)) 978 - lfs_sb_getsegtabsz(fs) 979 - lfs_sb_getcleansz(fs)) * lfs_sb_getifpb(fs)); 980#endif 981 return 0; 982} 983 984/* 985 * Allocate a block, and to inode and filesystem block accounting for it 986 * and for any indirect blocks the may need to be created in order for 987 * this block to be created. 988 * 989 * Blocks which have never been accounted for (i.e., which "do not exist") 990 * have disk address 0, which is translated by ulfs_bmap to the special value 991 * UNASSIGNED == -1, as in the historical ULFS. 992 * 993 * Blocks which have been accounted for but which have not yet been written 994 * to disk are given the new special disk address UNWRITTEN == -2, so that 995 * they can be differentiated from completely new blocks. 996 */ 997int 998lfs_balloc(struct uvnode *vp, off_t startoffset, int iosize, struct ubuf **bpp) 999{ 1000 int offset; 1001 daddr_t daddr, idaddr; 1002 struct ubuf *ibp, *bp; 1003 struct inode *ip; 1004 struct lfs *fs; 1005 struct indir indirs[ULFS_NIADDR+2], *idp; 1006 daddr_t lbn, lastblock; 1007 int bcount; 1008 int error, frags, i, nsize, osize, num; 1009 1010 ip = VTOI(vp); 1011 fs = ip->i_lfs; 1012 offset = lfs_blkoff(fs, startoffset); 1013 lbn = lfs_lblkno(fs, startoffset); 1014 1015 /* 1016 * Three cases: it's a block beyond the end of file, it's a block in 1017 * the file that may or may not have been assigned a disk address or 1018 * we're writing an entire block. 1019 * 1020 * Note, if the daddr is UNWRITTEN, the block already exists in 1021 * the cache (it was read or written earlier). If so, make sure 1022 * we don't count it as a new block or zero out its contents. If 1023 * it did not, make sure we allocate any necessary indirect 1024 * blocks. 1025 * 1026 * If we are writing a block beyond the end of the file, we need to 1027 * check if the old last block was a fragment. If it was, we need 1028 * to rewrite it. 1029 */ 1030 1031 if (bpp) 1032 *bpp = NULL; 1033 1034 /* Check for block beyond end of file and fragment extension needed. */ 1035 lastblock = lfs_lblkno(fs, lfs_dino_getsize(fs, ip->i_din)); 1036 if (lastblock < ULFS_NDADDR && lastblock < lbn) { 1037 osize = lfs_blksize(fs, ip, lastblock); 1038 if (osize < lfs_sb_getbsize(fs) && osize > 0) { 1039 if ((error = lfs_fragextend(vp, osize, lfs_sb_getbsize(fs), 1040 lastblock, 1041 (bpp ? &bp : NULL)))) 1042 return (error); 1043 lfs_dino_setsize(fs, ip->i_din, (lastblock + 1) * lfs_sb_getbsize(fs)); 1044 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1045 if (bpp) 1046 (void) VOP_BWRITE(bp); 1047 } 1048 } 1049 1050 /* 1051 * If the block we are writing is a direct block, it's the last 1052 * block in the file, and offset + iosize is less than a full 1053 * block, we can write one or more fragments. There are two cases: 1054 * the block is brand new and we should allocate it the correct 1055 * size or it already exists and contains some fragments and 1056 * may need to extend it. 1057 */ 1058 if (lbn < ULFS_NDADDR && lfs_lblkno(fs, lfs_dino_getsize(fs, ip->i_din)) <= lbn) { 1059 osize = lfs_blksize(fs, ip, lbn); 1060 nsize = lfs_fragroundup(fs, offset + iosize); 1061 if (lfs_lblktosize(fs, lbn) >= lfs_dino_getsize(fs, ip->i_din)) { 1062 /* Brand new block or fragment */ 1063 frags = lfs_numfrags(fs, nsize); 1064 if (bpp) { 1065 *bpp = bp = getblk(vp, lbn, nsize); 1066 bp->b_blkno = UNWRITTEN; 1067 } 1068 ip->i_lfs_effnblks += frags; 1069 lfs_sb_subbfree(fs, frags); 1070 lfs_dino_setdb(fs, ip->i_din, lbn, UNWRITTEN); 1071 } else { 1072 if (nsize <= osize) { 1073 /* No need to extend */ 1074 if (bpp && (error = bread(vp, lbn, osize, 1075 0, &bp))) 1076 return error; 1077 } else { 1078 /* Extend existing block */ 1079 if ((error = 1080 lfs_fragextend(vp, osize, nsize, lbn, 1081 (bpp ? &bp : NULL)))) 1082 return error; 1083 } 1084 if (bpp) 1085 *bpp = bp; 1086 } 1087 return 0; 1088 } 1089 1090 error = ulfs_bmaparray(fs, vp, lbn, &daddr, &indirs[0], &num); 1091 if (error) 1092 return (error); 1093 1094 daddr = (daddr_t)((int32_t)daddr); /* XXX ondisk32 */ 1095 1096 /* 1097 * Do byte accounting all at once, so we can gracefully fail *before* 1098 * we start assigning blocks. 1099 */ 1100 frags = LFS_FSBTODB(fs, 1); /* frags = VFSTOULFS(vp->v_mount)->um_seqinc; */ 1101 bcount = 0; 1102 if (daddr == UNASSIGNED) { 1103 bcount = frags; 1104 } 1105 for (i = 1; i < num; ++i) { 1106 if (!indirs[i].in_exists) { 1107 bcount += frags; 1108 } 1109 } 1110 lfs_sb_subbfree(fs, bcount); 1111 ip->i_lfs_effnblks += bcount; 1112 1113 if (daddr == UNASSIGNED) { 1114 if (num > 0 && lfs_dino_getib(fs, ip->i_din, indirs[0].in_off) == 0) { 1115 lfs_dino_setib(fs, ip->i_din, indirs[0].in_off, 1116 UNWRITTEN); 1117 } 1118 1119 /* 1120 * Create new indirect blocks if necessary 1121 */ 1122 if (num > 1) { 1123 idaddr = lfs_dino_getib(fs, ip->i_din, indirs[0].in_off); 1124 for (i = 1; i < num; ++i) { 1125 ibp = getblk(vp, indirs[i].in_lbn, 1126 lfs_sb_getbsize(fs)); 1127 if (!indirs[i].in_exists) { 1128 memset(ibp->b_data, 0, ibp->b_bufsize); 1129 ibp->b_blkno = UNWRITTEN; 1130 } else if (!(ibp->b_flags & (B_DELWRI | B_DONE))) { 1131 ibp->b_blkno = LFS_FSBTODB(fs, idaddr); 1132 ibp->b_flags |= B_READ; 1133 VOP_STRATEGY(ibp); 1134 } 1135 /* 1136 * This block exists, but the next one may not. 1137 * If that is the case mark it UNWRITTEN to 1138 * keep the accounting straight. 1139 */ 1140 /* XXX ondisk32 */ 1141 if (((int32_t *)ibp->b_data)[indirs[i].in_off] == 0) 1142 ((int32_t *)ibp->b_data)[indirs[i].in_off] = 1143 UNWRITTEN; 1144 /* XXX ondisk32 */ 1145 idaddr = ((int32_t *)ibp->b_data)[indirs[i].in_off]; 1146 if ((error = VOP_BWRITE(ibp))) 1147 return error; 1148 } 1149 } 1150 } 1151 1152 1153 /* 1154 * Get the existing block from the cache, if requested. 1155 */ 1156 if (bpp) 1157 *bpp = bp = getblk(vp, lbn, lfs_blksize(fs, ip, lbn)); 1158 1159 /* 1160 * The block we are writing may be a brand new block 1161 * in which case we need to do accounting. 1162 * 1163 * We can tell a truly new block because ulfs_bmaparray will say 1164 * it is UNASSIGNED. Once we allocate it we will assign it the 1165 * disk address UNWRITTEN. 1166 */ 1167 if (daddr == UNASSIGNED) { 1168 if (bpp) { 1169 /* Note the new address */ 1170 bp->b_blkno = UNWRITTEN; 1171 } 1172 1173 switch (num) { 1174 case 0: 1175 lfs_dino_setdb(fs, ip->i_din, lbn, UNWRITTEN); 1176 break; 1177 case 1: 1178 lfs_dino_setib(fs, ip->i_din, indirs[0].in_off, 1179 UNWRITTEN); 1180 break; 1181 default: 1182 idp = &indirs[num - 1]; 1183 if (bread(vp, idp->in_lbn, lfs_sb_getbsize(fs), 0, &ibp)) 1184 panic("lfs_balloc: bread bno %lld", 1185 (long long)idp->in_lbn); 1186 /* XXX ondisk32 */ 1187 lfs_iblock_set(fs, ibp->b_data, idp->in_off, 1188 UNWRITTEN); 1189 VOP_BWRITE(ibp); 1190 } 1191 } else if (bpp && !(bp->b_flags & (B_DONE|B_DELWRI))) { 1192 /* 1193 * Not a brand new block, also not in the cache; 1194 * read it in from disk. 1195 */ 1196 if (iosize == lfs_sb_getbsize(fs)) 1197 /* Optimization: I/O is unnecessary. */ 1198 bp->b_blkno = daddr; 1199 else { 1200 /* 1201 * We need to read the block to preserve the 1202 * existing bytes. 1203 */ 1204 bp->b_blkno = daddr; 1205 bp->b_flags |= B_READ; 1206 VOP_STRATEGY(bp); 1207 return 0; 1208 } 1209 } 1210 1211 return (0); 1212} 1213 1214int 1215lfs_fragextend(struct uvnode *vp, int osize, int nsize, daddr_t lbn, 1216 struct ubuf **bpp) 1217{ 1218 struct inode *ip; 1219 struct lfs *fs; 1220 int frags; 1221 int error; 1222 1223 ip = VTOI(vp); 1224 fs = ip->i_lfs; 1225 frags = (long)lfs_numfrags(fs, nsize - osize); 1226 error = 0; 1227 1228 /* 1229 * If we are not asked to actually return the block, all we need 1230 * to do is allocate space for it. UBC will handle dirtying the 1231 * appropriate things and making sure it all goes to disk. 1232 * Don't bother to read in that case. 1233 */ 1234 if (bpp && (error = bread(vp, lbn, osize, 0, bpp))) { 1235 brelse(*bpp, 0); 1236 goto out; 1237 } 1238 1239 lfs_sb_subbfree(fs, frags); 1240 ip->i_lfs_effnblks += frags; 1241 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1242 1243 if (bpp) { 1244 (*bpp)->b_data = erealloc((*bpp)->b_data, nsize); 1245 (void)memset((*bpp)->b_data + osize, 0, nsize - osize); 1246 } 1247 1248 out: 1249 return (error); 1250} 1251