1 /* $NetBSD: lfs.c,v 1.74 2018/06/12 20:12:21 zafer 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 <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 "vnode.h" 92 #include "lfs_user.h" 93 #include "segwrite.h" 94 #include "kernelops.h" 95 96 #define panic call_panic 97 98 extern u_int32_t cksum(void *, size_t); 99 extern u_int32_t lfs_sb_cksum(struct lfs *); 100 extern void pwarn(const char *, ...); 101 102 extern struct uvnodelst vnodelist; 103 extern struct uvnodelst getvnodelist[VNODE_HASH_MAX]; 104 extern int nvnodes; 105 106 long dev_bsize = DEV_BSIZE; 107 108 static int 109 lfs_fragextend(struct uvnode *, int, int, daddr_t, struct ubuf **); 110 111 int fsdirty = 0; 112 void (*panic_func)(int, const char *, va_list) = my_vpanic; 113 114 /* 115 * LFS buffer and uvnode operations 116 */ 117 118 int 119 lfs_vop_strategy(struct ubuf * bp) 120 { 121 int count; 122 123 if (bp->b_flags & B_READ) { 124 count = kops.ko_pread(bp->b_vp->v_fd, bp->b_data, bp->b_bcount, 125 bp->b_blkno * dev_bsize); 126 if (count == bp->b_bcount) 127 bp->b_flags |= B_DONE; 128 } else { 129 count = kops.ko_pwrite(bp->b_vp->v_fd, bp->b_data, bp->b_bcount, 130 bp->b_blkno * dev_bsize); 131 if (count == 0) { 132 perror("pwrite"); 133 return -1; 134 } 135 bp->b_flags &= ~B_DELWRI; 136 reassignbuf(bp, bp->b_vp); 137 } 138 return 0; 139 } 140 141 int 142 lfs_vop_bwrite(struct ubuf * bp) 143 { 144 struct lfs *fs; 145 146 fs = bp->b_vp->v_fs; 147 if (!(bp->b_flags & B_DELWRI)) { 148 lfs_sb_subavail(fs, lfs_btofsb(fs, bp->b_bcount)); 149 } 150 bp->b_flags |= B_DELWRI | B_LOCKED; 151 reassignbuf(bp, bp->b_vp); 152 brelse(bp, 0); 153 return 0; 154 } 155 156 /* 157 * ulfs_bmaparray does the bmap conversion, and if requested returns the 158 * array of logical blocks which must be traversed to get to a block. 159 * Each entry contains the offset into that block that gets you to the 160 * next block and the disk address of the block (if it is assigned). 161 */ 162 int 163 ulfs_bmaparray(struct lfs * fs, struct uvnode * vp, daddr_t bn, daddr_t * bnp, struct indir * ap, int *nump) 164 { 165 struct inode *ip; 166 struct ubuf *bp; 167 struct indir a[ULFS_NIADDR + 1], *xap; 168 daddr_t daddr; 169 daddr_t metalbn; 170 int error, num; 171 172 ip = VTOI(vp); 173 174 if (bn >= 0 && bn < ULFS_NDADDR) { 175 if (nump != NULL) 176 *nump = 0; 177 *bnp = LFS_FSBTODB(fs, lfs_dino_getdb(fs, ip->i_din, bn)); 178 if (*bnp == 0) 179 *bnp = -1; 180 return (0); 181 } 182 xap = ap == NULL ? a : ap; 183 if (!nump) 184 nump = # 185 if ((error = ulfs_getlbns(fs, vp, bn, xap, nump)) != 0) 186 return (error); 187 188 num = *nump; 189 190 /* Get disk address out of indirect block array */ 191 daddr = lfs_dino_getib(fs, ip->i_din, xap->in_off); 192 193 for (bp = NULL, ++xap; --num; ++xap) { 194 /* Exit the loop if there is no disk address assigned yet and 195 * the indirect block isn't in the cache, or if we were 196 * looking for an indirect block and we've found it. */ 197 198 metalbn = xap->in_lbn; 199 if ((daddr == 0 && !incore(vp, metalbn)) || metalbn == bn) 200 break; 201 /* 202 * If we get here, we've either got the block in the cache 203 * or we have a disk address for it, go fetch it. 204 */ 205 if (bp) 206 brelse(bp, 0); 207 208 xap->in_exists = 1; 209 bp = getblk(vp, metalbn, lfs_sb_getbsize(fs)); 210 211 if (!(bp->b_flags & (B_DONE | B_DELWRI))) { 212 bp->b_blkno = LFS_FSBTODB(fs, daddr); 213 bp->b_flags |= B_READ; 214 VOP_STRATEGY(bp); 215 } 216 daddr = lfs_iblock_get(fs, bp->b_data, xap->in_off); 217 } 218 if (bp) 219 brelse(bp, 0); 220 221 daddr = LFS_FSBTODB(fs, daddr); 222 *bnp = daddr == 0 ? -1 : daddr; 223 return (0); 224 } 225 226 /* 227 * Create an array of logical block number/offset pairs which represent the 228 * path of indirect blocks required to access a data block. The first "pair" 229 * contains the logical block number of the appropriate single, double or 230 * triple indirect block and the offset into the inode indirect block array. 231 * Note, the logical block number of the inode single/double/triple indirect 232 * block appears twice in the array, once with the offset into di_ib and 233 * once with the offset into the page itself. 234 */ 235 int 236 ulfs_getlbns(struct lfs * fs, struct uvnode * vp, daddr_t bn, struct indir * ap, int *nump) 237 { 238 daddr_t metalbn, realbn; 239 int64_t blockcnt; 240 int lbc; 241 int i, numlevels, off; 242 int lognindir, indir; 243 244 metalbn = 0; /* XXXGCC -Wuninitialized [sh3] */ 245 246 if (nump) 247 *nump = 0; 248 numlevels = 0; 249 realbn = bn; 250 if (bn < 0) 251 bn = -bn; 252 253 lognindir = -1; 254 for (indir = lfs_sb_getnindir(fs); indir; indir >>= 1) 255 ++lognindir; 256 257 /* Determine the number of levels of indirection. After this loop is 258 * done, blockcnt indicates the number of data blocks possible at the 259 * given level of indirection, and ULFS_NIADDR - i is the number of levels 260 * of indirection needed to locate the requested block. */ 261 262 bn -= ULFS_NDADDR; 263 for (lbc = 0, i = ULFS_NIADDR;; i--, bn -= blockcnt) { 264 if (i == 0) 265 return (EFBIG); 266 267 lbc += lognindir; 268 blockcnt = (int64_t) 1 << lbc; 269 270 if (bn < blockcnt) 271 break; 272 } 273 274 /* Calculate the address of the first meta-block. */ 275 metalbn = -((realbn >= 0 ? realbn : -realbn) - bn + ULFS_NIADDR - i); 276 277 /* At each iteration, off is the offset into the bap array which is an 278 * array of disk addresses at the current level of indirection. The 279 * logical block number and the offset in that block are stored into 280 * the argument array. */ 281 ap->in_lbn = metalbn; 282 ap->in_off = off = ULFS_NIADDR - i; 283 ap->in_exists = 0; 284 ap++; 285 for (++numlevels; i <= ULFS_NIADDR; i++) { 286 /* If searching for a meta-data block, quit when found. */ 287 if (metalbn == realbn) 288 break; 289 290 lbc -= lognindir; 291 /*blockcnt = (int64_t) 1 << lbc;*/ 292 off = (bn >> lbc) & (lfs_sb_getnindir(fs) - 1); 293 294 ++numlevels; 295 ap->in_lbn = metalbn; 296 ap->in_off = off; 297 ap->in_exists = 0; 298 ++ap; 299 300 metalbn -= -1 + (off << lbc); 301 } 302 if (nump) 303 *nump = numlevels; 304 return (0); 305 } 306 307 int 308 lfs_vop_bmap(struct uvnode * vp, daddr_t lbn, daddr_t * daddrp) 309 { 310 return ulfs_bmaparray(vp->v_fs, vp, lbn, daddrp, NULL, NULL); 311 } 312 313 /* Search a block for a specific dinode. */ 314 union lfs_dinode * 315 lfs_ifind(struct lfs *fs, ino_t ino, struct ubuf *bp) 316 { 317 union lfs_dinode *ldip; 318 unsigned i, num; 319 320 num = LFS_INOPB(fs); 321 322 /* 323 * Read the inode block backwards, since later versions of the 324 * inode will supercede earlier ones. Though it is unlikely, it is 325 * possible that the same inode will appear in the same inode block. 326 */ 327 for (i = num; i-- > 0; ) { 328 ldip = DINO_IN_BLOCK(fs, bp->b_data, i); 329 if (lfs_dino_getinumber(fs, ldip) == ino) 330 return (ldip); 331 } 332 return NULL; 333 } 334 335 /* 336 * lfs_raw_vget makes us a new vnode from the inode at the given disk address. 337 * XXX it currently loses atime information. 338 */ 339 struct uvnode * 340 lfs_raw_vget(struct lfs * fs, ino_t ino, int fd, daddr_t daddr) 341 { 342 struct uvnode *vp; 343 struct inode *ip; 344 union lfs_dinode *dip; 345 struct ubuf *bp; 346 int i, hash; 347 348 vp = ecalloc(1, sizeof(*vp)); 349 vp->v_fd = fd; 350 vp->v_fs = fs; 351 vp->v_usecount = 0; 352 vp->v_strategy_op = lfs_vop_strategy; 353 vp->v_bwrite_op = lfs_vop_bwrite; 354 vp->v_bmap_op = lfs_vop_bmap; 355 LIST_INIT(&vp->v_cleanblkhd); 356 LIST_INIT(&vp->v_dirtyblkhd); 357 358 ip = ecalloc(1, sizeof(*ip)); 359 360 ip->i_din = dip = ecalloc(1, sizeof(*dip)); 361 362 /* Initialize the inode -- from lfs_vcreate. */ 363 ip->inode_ext.lfs = ecalloc(1, sizeof(*ip->inode_ext.lfs)); 364 vp->v_data = ip; 365 /* ip->i_vnode = vp; */ 366 ip->i_lockf = 0; 367 ip->i_state = 0; 368 369 /* Load inode block and find inode */ 370 if (daddr > 0) { 371 bread(fs->lfs_devvp, LFS_FSBTODB(fs, daddr), lfs_sb_getibsize(fs), 372 0, &bp); 373 bp->b_flags |= B_AGE; 374 dip = lfs_ifind(fs, ino, bp); 375 if (dip == NULL) { 376 brelse(bp, 0); 377 free(ip->i_din); 378 free(ip->inode_ext.lfs); 379 free(ip); 380 free(vp); 381 return NULL; 382 } 383 lfs_copy_dinode(fs, ip->i_din, dip); 384 brelse(bp, 0); 385 } 386 ip->i_number = ino; 387 /* ip->i_devvp = fs->lfs_devvp; */ 388 ip->i_lfs = fs; 389 390 ip->i_lfs_effnblks = lfs_dino_getblocks(fs, ip->i_din); 391 ip->i_lfs_osize = lfs_dino_getsize(fs, ip->i_din); 392 #if 0 393 if (lfs_sb_getversion(fs) > 1) { 394 lfs_dino_setatime(fs, ip->i_din, ts.tv_sec); 395 lfs_dino_setatimensec(fs, ip->i_din, ts.tv_nsec); 396 } 397 #endif 398 399 memset(ip->i_lfs_fragsize, 0, ULFS_NDADDR * sizeof(*ip->i_lfs_fragsize)); 400 for (i = 0; i < ULFS_NDADDR; i++) 401 if (lfs_dino_getdb(fs, ip->i_din, i) != 0) 402 ip->i_lfs_fragsize[i] = lfs_blksize(fs, ip, i); 403 404 ++nvnodes; 405 hash = ((int)(intptr_t)fs + ino) & (VNODE_HASH_MAX - 1); 406 LIST_INSERT_HEAD(&getvnodelist[hash], vp, v_getvnodes); 407 LIST_INSERT_HEAD(&vnodelist, vp, v_mntvnodes); 408 409 return vp; 410 } 411 412 static struct uvnode * 413 lfs_vget(void *vfs, ino_t ino) 414 { 415 struct lfs *fs = (struct lfs *)vfs; 416 daddr_t daddr; 417 struct ubuf *bp; 418 IFILE *ifp; 419 420 LFS_IENTRY(ifp, fs, ino, bp); 421 daddr = lfs_if_getdaddr(fs, ifp); 422 brelse(bp, 0); 423 if (daddr <= 0 || lfs_dtosn(fs, daddr) >= lfs_sb_getnseg(fs)) 424 return NULL; 425 return lfs_raw_vget(fs, ino, fs->lfs_ivnode->v_fd, daddr); 426 } 427 428 /* 429 * Check superblock magic number and checksum. 430 * Sets lfs_is64 and lfs_dobyteswap. 431 */ 432 static int 433 check_sb(struct lfs *fs) 434 { 435 u_int32_t checksum; 436 u_int32_t magic; 437 438 /* we can read the magic out of either the 32-bit or 64-bit dlfs */ 439 magic = fs->lfs_dlfs_u.u_32.dlfs_magic; 440 441 switch (magic) { 442 case LFS_MAGIC: 443 fs->lfs_is64 = false; 444 fs->lfs_dobyteswap = false; 445 break; 446 case LFS_MAGIC_SWAPPED: 447 fs->lfs_is64 = false; 448 fs->lfs_dobyteswap = true; 449 break; 450 case LFS64_MAGIC: 451 fs->lfs_is64 = true; 452 fs->lfs_dobyteswap = false; 453 break; 454 case LFS64_MAGIC_SWAPPED: 455 fs->lfs_is64 = true; 456 fs->lfs_dobyteswap = true; 457 break; 458 default: 459 printf("Superblock magic number (0x%lx) does not match " 460 "expected 0x%lx\n", (unsigned long) magic, 461 (unsigned long) LFS_MAGIC); 462 return 1; 463 } 464 465 /* checksum */ 466 checksum = lfs_sb_cksum(fs); 467 if (lfs_sb_getcksum(fs) != checksum) { 468 printf("Superblock checksum (%lx) does not match computed checksum (%lx)\n", 469 (unsigned long) lfs_sb_getcksum(fs), (unsigned long) checksum); 470 return 1; 471 } 472 return 0; 473 } 474 475 /* Initialize LFS library; load superblocks and choose which to use. */ 476 struct lfs * 477 lfs_init(int devfd, daddr_t sblkno, daddr_t idaddr, int dummy_read, int debug) 478 { 479 struct uvnode *devvp; 480 struct ubuf *bp; 481 int tryalt; 482 struct lfs *fs, *altfs; 483 484 vfs_init(); 485 486 devvp = ecalloc(1, sizeof(*devvp)); 487 devvp->v_fs = NULL; 488 devvp->v_fd = devfd; 489 devvp->v_strategy_op = raw_vop_strategy; 490 devvp->v_bwrite_op = raw_vop_bwrite; 491 devvp->v_bmap_op = raw_vop_bmap; 492 LIST_INIT(&devvp->v_cleanblkhd); 493 LIST_INIT(&devvp->v_dirtyblkhd); 494 495 tryalt = 0; 496 if (dummy_read) { 497 if (sblkno == 0) 498 sblkno = LFS_LABELPAD / dev_bsize; 499 fs = ecalloc(1, sizeof(*fs)); 500 fs->lfs_devvp = devvp; 501 } else { 502 if (sblkno == 0) { 503 sblkno = LFS_LABELPAD / dev_bsize; 504 tryalt = 1; 505 } else if (debug) { 506 printf("No -b flag given, not attempting to verify checkpoint\n"); 507 } 508 509 dev_bsize = DEV_BSIZE; 510 511 (void)bread(devvp, sblkno, LFS_SBPAD, 0, &bp); 512 fs = ecalloc(1, sizeof(*fs)); 513 __CTASSERT(sizeof(struct dlfs) == sizeof(struct dlfs64)); 514 memcpy(&fs->lfs_dlfs_u, bp->b_data, sizeof(struct dlfs)); 515 fs->lfs_devvp = devvp; 516 bp->b_flags |= B_INVAL; 517 brelse(bp, 0); 518 519 dev_bsize = lfs_sb_getfsize(fs) >> lfs_sb_getfsbtodb(fs); 520 521 if (tryalt) { 522 (void)bread(devvp, LFS_FSBTODB(fs, lfs_sb_getsboff(fs, 1)), 523 LFS_SBPAD, 0, &bp); 524 altfs = ecalloc(1, sizeof(*altfs)); 525 memcpy(&altfs->lfs_dlfs_u, bp->b_data, 526 sizeof(struct dlfs)); 527 altfs->lfs_devvp = devvp; 528 bp->b_flags |= B_INVAL; 529 brelse(bp, 0); 530 531 if (check_sb(fs) || lfs_sb_getidaddr(fs) <= 0) { 532 if (debug) 533 printf("Primary superblock is no good, using first alternate\n"); 534 free(fs); 535 fs = altfs; 536 } else { 537 /* If both superblocks check out, try verification */ 538 if (check_sb(altfs)) { 539 if (debug) 540 printf("First alternate superblock is no good, using primary\n"); 541 free(altfs); 542 } else { 543 if (lfs_verify(fs, altfs, devvp, debug) == fs) { 544 free(altfs); 545 } else { 546 free(fs); 547 fs = altfs; 548 } 549 } 550 } 551 } 552 if (check_sb(fs)) { 553 free(fs); 554 return NULL; 555 } 556 } 557 558 /* Compatibility */ 559 if (lfs_sb_getversion(fs) < 2) { 560 lfs_sb_setsumsize(fs, LFS_V1_SUMMARY_SIZE); 561 lfs_sb_setibsize(fs, lfs_sb_getbsize(fs)); 562 lfs_sb_sets0addr(fs, lfs_sb_getsboff(fs, 0)); 563 lfs_sb_settstamp(fs, lfs_sb_getotstamp(fs)); 564 lfs_sb_setfsbtodb(fs, 0); 565 } 566 567 if (!dummy_read) { 568 fs->lfs_suflags = emalloc(2 * sizeof(u_int32_t *)); 569 fs->lfs_suflags[0] = emalloc(lfs_sb_getnseg(fs) * sizeof(u_int32_t)); 570 fs->lfs_suflags[1] = emalloc(lfs_sb_getnseg(fs) * sizeof(u_int32_t)); 571 } 572 573 if (idaddr == 0) 574 idaddr = lfs_sb_getidaddr(fs); 575 else 576 lfs_sb_setidaddr(fs, idaddr); 577 /* NB: If dummy_read!=0, idaddr==0 here so we get a fake inode. */ 578 fs->lfs_ivnode = lfs_raw_vget(fs, LFS_IFILE_INUM, 579 devvp->v_fd, idaddr); 580 if (fs->lfs_ivnode == NULL) 581 return NULL; 582 583 register_vget((void *)fs, lfs_vget); 584 585 return fs; 586 } 587 588 /* 589 * Check partial segment validity between fs->lfs_offset and the given goal. 590 * 591 * If goal == 0, just keep on going until the segments stop making sense, 592 * and return the address of the last valid partial segment. 593 * 594 * If goal != 0, return the address of the first partial segment that failed, 595 * or "goal" if we reached it without failure (the partial segment *at* goal 596 * need not be valid). 597 */ 598 daddr_t 599 try_verify(struct lfs *osb, struct uvnode *devvp, daddr_t goal, int debug) 600 { 601 daddr_t daddr, odaddr; 602 SEGSUM *sp; 603 int i, bc, hitclean; 604 struct ubuf *bp; 605 daddr_t nodirop_daddr; 606 u_int64_t serial; 607 608 bc = 0; 609 hitclean = 0; 610 odaddr = -1; 611 daddr = lfs_sb_getoffset(osb); 612 nodirop_daddr = daddr; 613 serial = lfs_sb_getserial(osb); 614 while (daddr != goal) { 615 /* 616 * Don't mistakenly read a superblock, if there is one here. 617 */ 618 if (lfs_sntod(osb, lfs_dtosn(osb, daddr)) == daddr) { 619 if (daddr == lfs_sb_gets0addr(osb)) 620 daddr += lfs_btofsb(osb, LFS_LABELPAD); 621 for (i = 0; i < LFS_MAXNUMSB; i++) { 622 /* XXX dholland 20150828 I think this is wrong */ 623 if (lfs_sb_getsboff(osb, i) < daddr) 624 break; 625 if (lfs_sb_getsboff(osb, i) == daddr) 626 daddr += lfs_btofsb(osb, LFS_SBPAD); 627 } 628 } 629 630 /* Read in summary block */ 631 bread(devvp, LFS_FSBTODB(osb, daddr), lfs_sb_getsumsize(osb), 632 0, &bp); 633 sp = (SEGSUM *)bp->b_data; 634 635 /* 636 * Check for a valid segment summary belonging to our fs. 637 */ 638 if (lfs_ss_getmagic(osb, sp) != SS_MAGIC || 639 lfs_ss_getident(osb, sp) != lfs_sb_getident(osb) || 640 lfs_ss_getserial(osb, sp) < serial || /* XXX strengthen this */ 641 lfs_ss_getsumsum(osb, sp) != 642 cksum((char *)sp + lfs_ss_getsumstart(osb), 643 lfs_sb_getsumsize(osb) - lfs_ss_getsumstart(osb))) { 644 brelse(bp, 0); 645 if (debug) { 646 if (lfs_ss_getmagic(osb, sp) != SS_MAGIC) 647 pwarn("pseg at 0x%jx: " 648 "wrong magic number\n", 649 (uintmax_t)daddr); 650 else if (lfs_ss_getident(osb, sp) != lfs_sb_getident(osb)) 651 pwarn("pseg at 0x%jx: " 652 "expected ident %jx, got %jx\n", 653 (uintmax_t)daddr, 654 (uintmax_t)lfs_ss_getident(osb, sp), 655 (uintmax_t)lfs_sb_getident(osb)); 656 else if (lfs_ss_getserial(osb, sp) >= serial) 657 pwarn("pseg at 0x%jx: " 658 "serial %d < %d\n", 659 (uintmax_t)daddr, 660 (int)lfs_ss_getserial(osb, sp), (int)serial); 661 else 662 pwarn("pseg at 0x%jx: " 663 "summary checksum wrong\n", 664 (uintmax_t)daddr); 665 } 666 break; 667 } 668 if (debug && lfs_ss_getserial(osb, sp) != serial) 669 pwarn("warning, serial=%d ss_serial=%d\n", 670 (int)serial, (int)lfs_ss_getserial(osb, sp)); 671 ++serial; 672 bc = check_summary(osb, sp, daddr, debug, devvp, NULL); 673 if (bc == 0) { 674 brelse(bp, 0); 675 break; 676 } 677 if (debug) 678 pwarn("summary good: 0x%x/%d\n", (uintmax_t)daddr, 679 (int)lfs_ss_getserial(osb, sp)); 680 assert (bc > 0); 681 odaddr = daddr; 682 daddr += lfs_btofsb(osb, lfs_sb_getsumsize(osb) + bc); 683 if (lfs_dtosn(osb, odaddr) != lfs_dtosn(osb, daddr) || 684 lfs_dtosn(osb, daddr) != lfs_dtosn(osb, daddr + 685 lfs_btofsb(osb, lfs_sb_getsumsize(osb) + lfs_sb_getbsize(osb)) - 1)) { 686 daddr = lfs_ss_getnext(osb, sp); 687 } 688 689 /* 690 * Check for the beginning and ending of a sequence of 691 * dirops. Writes from the cleaner never involve new 692 * information, and are always checkpoints; so don't try 693 * to roll forward through them. Likewise, psegs written 694 * by a previous roll-forward attempt are not interesting. 695 */ 696 if (lfs_ss_getflags(osb, sp) & (SS_CLEAN | SS_RFW)) 697 hitclean = 1; 698 if (hitclean == 0 && (lfs_ss_getflags(osb, sp) & SS_CONT) == 0) 699 nodirop_daddr = daddr; 700 701 brelse(bp, 0); 702 } 703 704 if (goal == 0) 705 return nodirop_daddr; 706 else 707 return daddr; 708 } 709 710 /* Use try_verify to check whether the newer superblock is valid. */ 711 struct lfs * 712 lfs_verify(struct lfs *sb0, struct lfs *sb1, struct uvnode *devvp, int debug) 713 { 714 daddr_t daddr; 715 struct lfs *osb, *nsb; 716 717 /* 718 * Verify the checkpoint of the newer superblock, 719 * if the timestamp/serial number of the two superblocks is 720 * different. 721 */ 722 723 osb = NULL; 724 if (debug) 725 pwarn("sb0 %ju, sb1 %ju", 726 (uintmax_t) lfs_sb_getserial(sb0), 727 (uintmax_t) lfs_sb_getserial(sb1)); 728 729 if ((lfs_sb_getversion(sb0) == 1 && 730 lfs_sb_getotstamp(sb0) != lfs_sb_getotstamp(sb1)) || 731 (lfs_sb_getversion(sb0) > 1 && 732 lfs_sb_getserial(sb0) != lfs_sb_getserial(sb1))) { 733 if (lfs_sb_getversion(sb0) == 1) { 734 if (lfs_sb_getotstamp(sb0) > lfs_sb_getotstamp(sb1)) { 735 osb = sb1; 736 nsb = sb0; 737 } else { 738 osb = sb0; 739 nsb = sb1; 740 } 741 } else { 742 if (lfs_sb_getserial(sb0) > lfs_sb_getserial(sb1)) { 743 osb = sb1; 744 nsb = sb0; 745 } else { 746 osb = sb0; 747 nsb = sb1; 748 } 749 } 750 if (debug) { 751 printf("Attempting to verify newer checkpoint..."); 752 fflush(stdout); 753 } 754 daddr = try_verify(osb, devvp, lfs_sb_getoffset(nsb), debug); 755 756 if (debug) 757 printf("done.\n"); 758 if (daddr == lfs_sb_getoffset(nsb)) { 759 pwarn("** Newer checkpoint verified; recovered %jd seconds of data\n", 760 (intmax_t)(lfs_sb_gettstamp(nsb) - lfs_sb_gettstamp(osb))); 761 sbdirty(); 762 } else { 763 pwarn("** Newer checkpoint invalid; lost %jd seconds of data\n", (intmax_t)(lfs_sb_gettstamp(nsb) - lfs_sb_gettstamp(osb))); 764 } 765 return (daddr == lfs_sb_getoffset(nsb) ? nsb : osb); 766 } 767 /* Nothing to check */ 768 return osb; 769 } 770 771 /* Verify a partial-segment summary; return the number of bytes on disk. */ 772 int 773 check_summary(struct lfs *fs, SEGSUM *sp, daddr_t pseg_addr, int debug, 774 struct uvnode *devvp, void (func(daddr_t, FINFO *))) 775 { 776 FINFO *fp; 777 int bc; /* Bytes in partial segment */ 778 int nblocks; 779 daddr_t daddr; 780 IINFO *iibase, *iip; 781 struct ubuf *bp; 782 int i, j, k, datac, len; 783 lfs_checkword *datap; 784 u_int32_t ccksum; 785 786 /* We've already checked the sumsum, just do the data bounds and sum */ 787 788 /* Count the blocks. */ 789 nblocks = howmany(lfs_ss_getninos(fs, sp), LFS_INOPB(fs)); 790 bc = nblocks << (lfs_sb_getversion(fs) > 1 ? lfs_sb_getffshift(fs) : lfs_sb_getbshift(fs)); 791 assert(bc >= 0); 792 793 fp = SEGSUM_FINFOBASE(fs, sp); 794 for (i = 0; i < lfs_ss_getnfinfo(fs, sp); i++) { 795 nblocks += lfs_fi_getnblocks(fs, fp); 796 bc += lfs_fi_getlastlength(fs, fp) + ((lfs_fi_getnblocks(fs, fp) - 1) 797 << lfs_sb_getbshift(fs)); 798 assert(bc >= 0); 799 fp = NEXT_FINFO(fs, fp); 800 if (((char *)fp) - (char *)sp > lfs_sb_getsumsize(fs)) 801 return 0; 802 } 803 datap = emalloc(nblocks * sizeof(*datap)); 804 datac = 0; 805 806 iibase = SEGSUM_IINFOSTART(fs, sp); 807 808 iip = iibase; 809 daddr = pseg_addr + lfs_btofsb(fs, lfs_sb_getsumsize(fs)); 810 fp = SEGSUM_FINFOBASE(fs, sp); 811 for (i = 0, j = 0; 812 i < lfs_ss_getnfinfo(fs, sp) || j < howmany(lfs_ss_getninos(fs, sp), LFS_INOPB(fs)); i++) { 813 if (i >= lfs_ss_getnfinfo(fs, sp) && lfs_ii_getblock(fs, iip) != daddr) { 814 pwarn("Not enough inode blocks in pseg at 0x%jx: " 815 "found %d, wanted %d\n", 816 pseg_addr, j, howmany(lfs_ss_getninos(fs, sp), 817 LFS_INOPB(fs))); 818 if (debug) 819 pwarn("iip=0x%jx, daddr=0x%jx\n", 820 (uintmax_t)lfs_ii_getblock(fs, iip), 821 (intmax_t)daddr); 822 break; 823 } 824 while (j < howmany(lfs_ss_getninos(fs, sp), LFS_INOPB(fs)) && lfs_ii_getblock(fs, iip) == daddr) { 825 bread(devvp, LFS_FSBTODB(fs, daddr), lfs_sb_getibsize(fs), 826 0, &bp); 827 datap[datac++] = ((lfs_checkword *)bp->b_data)[0]; 828 brelse(bp, 0); 829 830 ++j; 831 daddr += lfs_btofsb(fs, lfs_sb_getibsize(fs)); 832 iip = NEXTLOWER_IINFO(fs, iip); 833 } 834 if (i < lfs_ss_getnfinfo(fs, sp)) { 835 if (func) 836 func(daddr, fp); 837 for (k = 0; k < lfs_fi_getnblocks(fs, fp); k++) { 838 len = (k == lfs_fi_getnblocks(fs, fp) - 1 ? 839 lfs_fi_getlastlength(fs, fp) 840 : lfs_sb_getbsize(fs)); 841 bread(devvp, LFS_FSBTODB(fs, daddr), len, 842 0, &bp); 843 datap[datac++] = ((lfs_checkword *)bp->b_data)[0]; 844 brelse(bp, 0); 845 daddr += lfs_btofsb(fs, len); 846 } 847 fp = NEXT_FINFO(fs, fp); 848 } 849 } 850 851 if (datac != nblocks) { 852 pwarn("Partial segment at 0x%jx expected %d blocks counted %d\n", 853 (intmax_t)pseg_addr, nblocks, datac); 854 } 855 ccksum = cksum(datap, nblocks * sizeof(datap[0])); 856 /* Check the data checksum */ 857 if (ccksum != lfs_ss_getdatasum(fs, sp)) { 858 pwarn("Partial segment at 0x%jx data checksum" 859 " mismatch: given 0x%x, computed 0x%x\n", 860 (uintmax_t)pseg_addr, lfs_ss_getdatasum(fs, sp), ccksum); 861 free(datap); 862 return 0; 863 } 864 free(datap); 865 assert(bc >= 0); 866 return bc; 867 } 868 869 /* print message and exit */ 870 void 871 my_vpanic(int fatal, const char *fmt, va_list ap) 872 { 873 (void) vprintf(fmt, ap); 874 exit(8); 875 } 876 877 void 878 call_panic(const char *fmt, ...) 879 { 880 va_list ap; 881 882 va_start(ap, fmt); 883 panic_func(1, fmt, ap); 884 va_end(ap); 885 } 886 887 /* Allocate a new inode. */ 888 struct uvnode * 889 lfs_valloc(struct lfs *fs, ino_t ino) 890 { 891 struct ubuf *bp, *cbp; 892 IFILE *ifp; 893 ino_t new_ino; 894 int error; 895 CLEANERINFO *cip; 896 897 /* Get the head of the freelist. */ 898 LFS_GET_HEADFREE(fs, cip, cbp, &new_ino); 899 900 /* 901 * Remove the inode from the free list and write the new start 902 * of the free list into the superblock. 903 */ 904 LFS_IENTRY(ifp, fs, new_ino, bp); 905 if (lfs_if_getdaddr(fs, ifp) != LFS_UNUSED_DADDR) 906 panic("lfs_valloc: inuse inode %d on the free list", new_ino); 907 LFS_PUT_HEADFREE(fs, cip, cbp, lfs_if_getnextfree(fs, ifp)); 908 909 brelse(bp, 0); 910 911 /* Extend IFILE so that the next lfs_valloc will succeed. */ 912 if (lfs_sb_getfreehd(fs) == LFS_UNUSED_INUM) { 913 if ((error = extend_ifile(fs)) != 0) { 914 LFS_PUT_HEADFREE(fs, cip, cbp, new_ino); 915 return NULL; 916 } 917 } 918 919 /* Set superblock modified bit and increment file count. */ 920 sbdirty(); 921 lfs_sb_addnfiles(fs, 1); 922 923 return lfs_raw_vget(fs, ino, fs->lfs_devvp->v_fd, 0x0); 924 } 925 926 #ifdef IN_FSCK_LFS 927 void reset_maxino(ino_t); 928 #endif 929 930 /* 931 * Add a new block to the Ifile, to accommodate future file creations. 932 */ 933 int 934 extend_ifile(struct lfs *fs) 935 { 936 struct uvnode *vp; 937 struct inode *ip; 938 IFILE64 *ifp64; 939 IFILE32 *ifp32; 940 IFILE_V1 *ifp_v1; 941 struct ubuf *bp, *cbp; 942 daddr_t i, blkno, max; 943 ino_t oldlast; 944 CLEANERINFO *cip; 945 946 vp = fs->lfs_ivnode; 947 ip = VTOI(vp); 948 blkno = lfs_lblkno(fs, lfs_dino_getsize(fs, ip->i_din)); 949 950 lfs_balloc(vp, lfs_dino_getsize(fs, ip->i_din), lfs_sb_getbsize(fs), &bp); 951 lfs_dino_setsize(fs, ip->i_din, 952 lfs_dino_getsize(fs, ip->i_din) + lfs_sb_getbsize(fs)); 953 ip->i_state |= IN_MODIFIED; 954 955 i = (blkno - lfs_sb_getsegtabsz(fs) - lfs_sb_getcleansz(fs)) * 956 lfs_sb_getifpb(fs); 957 LFS_GET_HEADFREE(fs, cip, cbp, &oldlast); 958 LFS_PUT_HEADFREE(fs, cip, cbp, i); 959 max = i + lfs_sb_getifpb(fs); 960 lfs_sb_subbfree(fs, lfs_btofsb(fs, lfs_sb_getbsize(fs))); 961 962 if (fs->lfs_is64) { 963 for (ifp64 = (IFILE64 *)bp->b_data; i < max; ++ifp64) { 964 ifp64->if_version = 1; 965 ifp64->if_daddr = LFS_UNUSED_DADDR; 966 ifp64->if_nextfree = ++i; 967 } 968 ifp64--; 969 ifp64->if_nextfree = oldlast; 970 } else if (lfs_sb_getversion(fs) > 1) { 971 for (ifp32 = (IFILE32 *)bp->b_data; i < max; ++ifp32) { 972 ifp32->if_version = 1; 973 ifp32->if_daddr = LFS_UNUSED_DADDR; 974 ifp32->if_nextfree = ++i; 975 } 976 ifp32--; 977 ifp32->if_nextfree = oldlast; 978 } else { 979 for (ifp_v1 = (IFILE_V1 *)bp->b_data; i < max; ++ifp_v1) { 980 ifp_v1->if_version = 1; 981 ifp_v1->if_daddr = LFS_UNUSED_DADDR; 982 ifp_v1->if_nextfree = ++i; 983 } 984 ifp_v1--; 985 ifp_v1->if_nextfree = oldlast; 986 } 987 LFS_PUT_TAILFREE(fs, cip, cbp, max - 1); 988 989 LFS_BWRITE_LOG(bp); 990 991 #ifdef IN_FSCK_LFS 992 reset_maxino(((lfs_dino_getsize(fs, ip->i_din) >> lfs_sb_getbshift(fs)) 993 - lfs_sb_getsegtabsz(fs) 994 - lfs_sb_getcleansz(fs)) * lfs_sb_getifpb(fs)); 995 #endif 996 return 0; 997 } 998 999 /* 1000 * Allocate a block, and to inode and filesystem block accounting for it 1001 * and for any indirect blocks the may need to be created in order for 1002 * this block to be created. 1003 * 1004 * Blocks which have never been accounted for (i.e., which "do not exist") 1005 * have disk address 0, which is translated by ulfs_bmap to the special value 1006 * UNASSIGNED == -1, as in the historical ULFS. 1007 * 1008 * Blocks which have been accounted for but which have not yet been written 1009 * to disk are given the new special disk address UNWRITTEN == -2, so that 1010 * they can be differentiated from completely new blocks. 1011 */ 1012 int 1013 lfs_balloc(struct uvnode *vp, off_t startoffset, int iosize, struct ubuf **bpp) 1014 { 1015 int offset; 1016 daddr_t daddr, idaddr; 1017 struct ubuf *ibp, *bp; 1018 struct inode *ip; 1019 struct lfs *fs; 1020 struct indir indirs[ULFS_NIADDR+2], *idp; 1021 daddr_t lbn, lastblock; 1022 int bcount; 1023 int error, frags, i, nsize, osize, num; 1024 1025 ip = VTOI(vp); 1026 fs = ip->i_lfs; 1027 offset = lfs_blkoff(fs, startoffset); 1028 lbn = lfs_lblkno(fs, startoffset); 1029 1030 /* 1031 * Three cases: it's a block beyond the end of file, it's a block in 1032 * the file that may or may not have been assigned a disk address or 1033 * we're writing an entire block. 1034 * 1035 * Note, if the daddr is UNWRITTEN, the block already exists in 1036 * the cache (it was read or written earlier). If so, make sure 1037 * we don't count it as a new block or zero out its contents. If 1038 * it did not, make sure we allocate any necessary indirect 1039 * blocks. 1040 * 1041 * If we are writing a block beyond the end of the file, we need to 1042 * check if the old last block was a fragment. If it was, we need 1043 * to rewrite it. 1044 */ 1045 1046 if (bpp) 1047 *bpp = NULL; 1048 1049 /* Check for block beyond end of file and fragment extension needed. */ 1050 lastblock = lfs_lblkno(fs, lfs_dino_getsize(fs, ip->i_din)); 1051 if (lastblock < ULFS_NDADDR && lastblock < lbn) { 1052 osize = lfs_blksize(fs, ip, lastblock); 1053 if (osize < lfs_sb_getbsize(fs) && osize > 0) { 1054 if ((error = lfs_fragextend(vp, osize, lfs_sb_getbsize(fs), 1055 lastblock, 1056 (bpp ? &bp : NULL)))) 1057 return (error); 1058 lfs_dino_setsize(fs, ip->i_din, (lastblock + 1) * lfs_sb_getbsize(fs)); 1059 ip->i_state |= IN_CHANGE | IN_UPDATE; 1060 if (bpp) 1061 (void) VOP_BWRITE(bp); 1062 } 1063 } 1064 1065 /* 1066 * If the block we are writing is a direct block, it's the last 1067 * block in the file, and offset + iosize is less than a full 1068 * block, we can write one or more fragments. There are two cases: 1069 * the block is brand new and we should allocate it the correct 1070 * size or it already exists and contains some fragments and 1071 * may need to extend it. 1072 */ 1073 if (lbn < ULFS_NDADDR && lfs_lblkno(fs, lfs_dino_getsize(fs, ip->i_din)) <= lbn) { 1074 osize = lfs_blksize(fs, ip, lbn); 1075 nsize = lfs_fragroundup(fs, offset + iosize); 1076 if (lfs_lblktosize(fs, lbn) >= lfs_dino_getsize(fs, ip->i_din)) { 1077 /* Brand new block or fragment */ 1078 frags = lfs_numfrags(fs, nsize); 1079 if (bpp) { 1080 *bpp = bp = getblk(vp, lbn, nsize); 1081 bp->b_blkno = UNWRITTEN; 1082 } 1083 ip->i_lfs_effnblks += frags; 1084 lfs_sb_subbfree(fs, frags); 1085 lfs_dino_setdb(fs, ip->i_din, lbn, UNWRITTEN); 1086 } else { 1087 if (nsize <= osize) { 1088 /* No need to extend */ 1089 if (bpp && (error = bread(vp, lbn, osize, 1090 0, &bp))) 1091 return error; 1092 } else { 1093 /* Extend existing block */ 1094 if ((error = 1095 lfs_fragextend(vp, osize, nsize, lbn, 1096 (bpp ? &bp : NULL)))) 1097 return error; 1098 } 1099 if (bpp) 1100 *bpp = bp; 1101 } 1102 return 0; 1103 } 1104 1105 error = ulfs_bmaparray(fs, vp, lbn, &daddr, &indirs[0], &num); 1106 if (error) 1107 return (error); 1108 1109 /* 1110 * Do byte accounting all at once, so we can gracefully fail *before* 1111 * we start assigning blocks. 1112 */ 1113 frags = LFS_FSBTODB(fs, 1); /* frags = VFSTOULFS(vp->v_mount)->um_seqinc; */ 1114 bcount = 0; 1115 if (daddr == UNASSIGNED) { 1116 bcount = frags; 1117 } 1118 for (i = 1; i < num; ++i) { 1119 if (!indirs[i].in_exists) { 1120 bcount += frags; 1121 } 1122 } 1123 lfs_sb_subbfree(fs, bcount); 1124 ip->i_lfs_effnblks += bcount; 1125 1126 if (daddr == UNASSIGNED) { 1127 if (num > 0 && lfs_dino_getib(fs, ip->i_din, indirs[0].in_off) == 0) { 1128 lfs_dino_setib(fs, ip->i_din, indirs[0].in_off, 1129 UNWRITTEN); 1130 } 1131 1132 /* 1133 * Create new indirect blocks if necessary 1134 */ 1135 if (num > 1) { 1136 idaddr = lfs_dino_getib(fs, ip->i_din, indirs[0].in_off); 1137 for (i = 1; i < num; ++i) { 1138 ibp = getblk(vp, indirs[i].in_lbn, 1139 lfs_sb_getbsize(fs)); 1140 if (!indirs[i].in_exists) { 1141 memset(ibp->b_data, 0, ibp->b_bufsize); 1142 ibp->b_blkno = UNWRITTEN; 1143 } else if (!(ibp->b_flags & (B_DELWRI | B_DONE))) { 1144 ibp->b_blkno = LFS_FSBTODB(fs, idaddr); 1145 ibp->b_flags |= B_READ; 1146 VOP_STRATEGY(ibp); 1147 } 1148 /* 1149 * This block exists, but the next one may not. 1150 * If that is the case mark it UNWRITTEN to 1151 * keep the accounting straight. 1152 */ 1153 if (lfs_iblock_get(fs, ibp->b_data, 1154 indirs[i].in_off) == 0) 1155 lfs_iblock_set(fs, ibp->b_data, 1156 indirs[i].in_off, UNWRITTEN); 1157 idaddr = lfs_iblock_get(fs, ibp->b_data, 1158 indirs[i].in_off); 1159 if ((error = VOP_BWRITE(ibp))) 1160 return error; 1161 } 1162 } 1163 } 1164 1165 1166 /* 1167 * Get the existing block from the cache, if requested. 1168 */ 1169 if (bpp) 1170 *bpp = bp = getblk(vp, lbn, lfs_blksize(fs, ip, lbn)); 1171 1172 /* 1173 * The block we are writing may be a brand new block 1174 * in which case we need to do accounting. 1175 * 1176 * We can tell a truly new block because ulfs_bmaparray will say 1177 * it is UNASSIGNED. Once we allocate it we will assign it the 1178 * disk address UNWRITTEN. 1179 */ 1180 if (daddr == UNASSIGNED) { 1181 if (bpp) { 1182 /* Note the new address */ 1183 bp->b_blkno = UNWRITTEN; 1184 } 1185 1186 switch (num) { 1187 case 0: 1188 lfs_dino_setdb(fs, ip->i_din, lbn, UNWRITTEN); 1189 break; 1190 case 1: 1191 lfs_dino_setib(fs, ip->i_din, indirs[0].in_off, 1192 UNWRITTEN); 1193 break; 1194 default: 1195 idp = &indirs[num - 1]; 1196 if (bread(vp, idp->in_lbn, lfs_sb_getbsize(fs), 0, &ibp)) 1197 panic("lfs_balloc: bread bno %lld", 1198 (long long)idp->in_lbn); 1199 lfs_iblock_set(fs, ibp->b_data, idp->in_off, 1200 UNWRITTEN); 1201 VOP_BWRITE(ibp); 1202 } 1203 } else if (bpp && !(bp->b_flags & (B_DONE|B_DELWRI))) { 1204 /* 1205 * Not a brand new block, also not in the cache; 1206 * read it in from disk. 1207 */ 1208 if (iosize == lfs_sb_getbsize(fs)) 1209 /* Optimization: I/O is unnecessary. */ 1210 bp->b_blkno = daddr; 1211 else { 1212 /* 1213 * We need to read the block to preserve the 1214 * existing bytes. 1215 */ 1216 bp->b_blkno = daddr; 1217 bp->b_flags |= B_READ; 1218 VOP_STRATEGY(bp); 1219 return 0; 1220 } 1221 } 1222 1223 return (0); 1224 } 1225 1226 int 1227 lfs_fragextend(struct uvnode *vp, int osize, int nsize, daddr_t lbn, 1228 struct ubuf **bpp) 1229 { 1230 struct inode *ip; 1231 struct lfs *fs; 1232 int frags; 1233 int error; 1234 1235 ip = VTOI(vp); 1236 fs = ip->i_lfs; 1237 frags = (long)lfs_numfrags(fs, nsize - osize); 1238 error = 0; 1239 1240 /* 1241 * If we are not asked to actually return the block, all we need 1242 * to do is allocate space for it. UBC will handle dirtying the 1243 * appropriate things and making sure it all goes to disk. 1244 * Don't bother to read in that case. 1245 */ 1246 if (bpp && (error = bread(vp, lbn, osize, 0, bpp))) { 1247 brelse(*bpp, 0); 1248 goto out; 1249 } 1250 1251 lfs_sb_subbfree(fs, frags); 1252 ip->i_lfs_effnblks += frags; 1253 ip->i_state |= IN_CHANGE | IN_UPDATE; 1254 1255 if (bpp) { 1256 (*bpp)->b_data = erealloc((*bpp)->b_data, nsize); 1257 (void)memset((*bpp)->b_data + osize, 0, nsize - osize); 1258 } 1259 1260 out: 1261 return (error); 1262 } 1263