1/* $NetBSD: lfs.c,v 1.34 2011/06/21 09:36:46 mrg 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#include <ufs/ufs/inode.h> 74#include <ufs/ufs/ufsmount.h> 75#define vnode uvnode 76#include <ufs/lfs/lfs.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 dlfs *); 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 fs->lfs_avail -= 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 * ufs_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 162ufs_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[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 < NDADDR) { 174 if (nump != NULL) 175 *nump = 0; 176 *bnp = fsbtodb(fs, ip->i_ffs1_db[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 = ufs_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 = ip->i_ffs1_ib[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, fs->lfs_bsize); 209 210 if (!(bp->b_flags & (B_DONE | B_DELWRI))) { 211 bp->b_blkno = fsbtodb(fs, daddr); 212 bp->b_flags |= B_READ; 213 VOP_STRATEGY(bp); 214 } 215 daddr = ((ufs_daddr_t *) bp->b_data)[xap->in_off]; 216 } 217 if (bp) 218 brelse(bp, 0); 219 220 daddr = fsbtodb(fs, (ufs_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 the i_ffs1_ib and 232 * once with the offset into the page itself. 233 */ 234int 235ufs_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 = fs->lfs_nindir; 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 NIADDR - i is the number of levels 259 * of indirection needed to locate the requested block. */ 260 261 bn -= NDADDR; 262 for (lbc = 0, i = 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 + 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 = NIADDR - i; 282 ap->in_exists = 0; 283 ap++; 284 for (++numlevels; i <= 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) & (fs->lfs_nindir - 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 ufs_bmaparray(vp->v_fs, vp, lbn, daddrp, NULL, NULL); 310} 311 312/* Search a block for a specific dinode. */ 313struct ufs1_dinode * 314lfs_ifind(struct lfs * fs, ino_t ino, struct ubuf * bp) 315{ 316 struct ufs1_dinode *dip = (struct ufs1_dinode *) bp->b_data; 317 struct ufs1_dinode *ldip, *fin; 318 319 fin = dip + 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 (ldip = fin - 1; ldip >= dip; --ldip) 327 if (ldip->di_inumber == ino) 328 return (ldip); 329 return NULL; 330} 331 332/* 333 * lfs_raw_vget makes us a new vnode from the inode at the given disk address. 334 * XXX it currently loses atime information. 335 */ 336struct uvnode * 337lfs_raw_vget(struct lfs * fs, ino_t ino, int fd, ufs_daddr_t daddr) 338{ 339 struct uvnode *vp; 340 struct inode *ip; 341 struct ufs1_dinode *dip; 342 struct ubuf *bp; 343 int i, hash; 344 345 vp = ecalloc(1, sizeof(*vp)); 346 vp->v_fd = fd; 347 vp->v_fs = fs; 348 vp->v_usecount = 0; 349 vp->v_strategy_op = lfs_vop_strategy; 350 vp->v_bwrite_op = lfs_vop_bwrite; 351 vp->v_bmap_op = lfs_vop_bmap; 352 LIST_INIT(&vp->v_cleanblkhd); 353 LIST_INIT(&vp->v_dirtyblkhd); 354 355 ip = ecalloc(1, sizeof(*ip)); 356 357 ip->i_din.ffs1_din = ecalloc(1, sizeof(*ip->i_din.ffs1_din)); 358 359 /* Initialize the inode -- from lfs_vcreate. */ 360 ip->inode_ext.lfs = ecalloc(1, sizeof(*ip->inode_ext.lfs)); 361 vp->v_data = ip; 362 /* ip->i_vnode = vp; */ 363 ip->i_number = ino; 364 ip->i_lockf = 0; 365 ip->i_lfs_effnblks = 0; 366 ip->i_flag = 0; 367 368 /* Load inode block and find inode */ 369 if (daddr > 0) { 370 bread(fs->lfs_devvp, fsbtodb(fs, daddr), fs->lfs_ibsize, 371 NULL, 0, &bp); 372 bp->b_flags |= B_AGE; 373 dip = lfs_ifind(fs, ino, bp); 374 if (dip == NULL) { 375 brelse(bp, 0); 376 free(ip); 377 free(vp); 378 return NULL; 379 } 380 memcpy(ip->i_din.ffs1_din, dip, sizeof(*dip)); 381 brelse(bp, 0); 382 } 383 ip->i_number = ino; 384 /* ip->i_devvp = fs->lfs_devvp; */ 385 ip->i_lfs = fs; 386 387 ip->i_lfs_effnblks = ip->i_ffs1_blocks; 388 ip->i_lfs_osize = ip->i_ffs1_size; 389#if 0 390 if (fs->lfs_version > 1) { 391 ip->i_ffs1_atime = ts.tv_sec; 392 ip->i_ffs1_atimensec = ts.tv_nsec; 393 } 394#endif 395 396 memset(ip->i_lfs_fragsize, 0, NDADDR * sizeof(*ip->i_lfs_fragsize)); 397 for (i = 0; i < NDADDR; i++) 398 if (ip->i_ffs1_db[i] != 0) 399 ip->i_lfs_fragsize[i] = blksize(fs, ip, i); 400 401 ++nvnodes; 402 hash = ((int)(intptr_t)fs + ino) & (VNODE_HASH_MAX - 1); 403 LIST_INSERT_HEAD(&getvnodelist[hash], vp, v_getvnodes); 404 LIST_INSERT_HEAD(&vnodelist, vp, v_mntvnodes); 405 406 return vp; 407} 408 409static struct uvnode * 410lfs_vget(void *vfs, ino_t ino) 411{ 412 struct lfs *fs = (struct lfs *)vfs; 413 ufs_daddr_t daddr; 414 struct ubuf *bp; 415 IFILE *ifp; 416 417 LFS_IENTRY(ifp, fs, ino, bp); 418 daddr = ifp->if_daddr; 419 brelse(bp, 0); 420 if (daddr <= 0 || dtosn(fs, daddr) >= fs->lfs_nseg) 421 return NULL; 422 return lfs_raw_vget(fs, ino, fs->lfs_ivnode->v_fd, daddr); 423} 424 425/* Check superblock magic number and checksum */ 426static int 427check_sb(struct lfs *fs) 428{ 429 u_int32_t checksum; 430 431 if (fs->lfs_magic != LFS_MAGIC) { 432 printf("Superblock magic number (0x%lx) does not match " 433 "expected 0x%lx\n", (unsigned long) fs->lfs_magic, 434 (unsigned long) LFS_MAGIC); 435 return 1; 436 } 437 /* checksum */ 438 checksum = lfs_sb_cksum(&(fs->lfs_dlfs)); 439 if (fs->lfs_cksum != checksum) { 440 printf("Superblock checksum (%lx) does not match computed checksum (%lx)\n", 441 (unsigned long) fs->lfs_cksum, (unsigned long) checksum); 442 return 1; 443 } 444 return 0; 445} 446 447/* Initialize LFS library; load superblocks and choose which to use. */ 448struct lfs * 449lfs_init(int devfd, daddr_t sblkno, daddr_t idaddr, int dummy_read, int debug) 450{ 451 struct uvnode *devvp; 452 struct ubuf *bp; 453 int tryalt; 454 struct lfs *fs, *altfs; 455 int error; 456 457 vfs_init(); 458 459 devvp = ecalloc(1, sizeof(*devvp)); 460 devvp->v_fs = NULL; 461 devvp->v_fd = devfd; 462 devvp->v_strategy_op = raw_vop_strategy; 463 devvp->v_bwrite_op = raw_vop_bwrite; 464 devvp->v_bmap_op = raw_vop_bmap; 465 LIST_INIT(&devvp->v_cleanblkhd); 466 LIST_INIT(&devvp->v_dirtyblkhd); 467 468 tryalt = 0; 469 if (dummy_read) { 470 if (sblkno == 0) 471 sblkno = LFS_LABELPAD / dev_bsize; 472 fs = ecalloc(1, sizeof(*fs)); 473 fs->lfs_devvp = devvp; 474 } else { 475 if (sblkno == 0) { 476 sblkno = LFS_LABELPAD / dev_bsize; 477 tryalt = 1; 478 } else if (debug) { 479 printf("No -b flag given, not attempting to verify checkpoint\n"); 480 } 481 482 dev_bsize = DEV_BSIZE; 483 484 error = bread(devvp, sblkno, LFS_SBPAD, NOCRED, 0, &bp); 485 fs = ecalloc(1, sizeof(*fs)); 486 fs->lfs_dlfs = *((struct dlfs *) bp->b_data); 487 fs->lfs_devvp = devvp; 488 bp->b_flags |= B_INVAL; 489 brelse(bp, 0); 490 491 dev_bsize = fs->lfs_fsize >> fs->lfs_fsbtodb; 492 493 if (tryalt) { 494 error = bread(devvp, fsbtodb(fs, fs->lfs_sboffs[1]), 495 LFS_SBPAD, NOCRED, 0, &bp); 496 altfs = ecalloc(1, sizeof(*altfs)); 497 altfs->lfs_dlfs = *((struct dlfs *) bp->b_data); 498 altfs->lfs_devvp = devvp; 499 bp->b_flags |= B_INVAL; 500 brelse(bp, 0); 501 502 if (check_sb(fs) || fs->lfs_idaddr <= 0) { 503 if (debug) 504 printf("Primary superblock is no good, using first alternate\n"); 505 free(fs); 506 fs = altfs; 507 } else { 508 /* If both superblocks check out, try verification */ 509 if (check_sb(altfs)) { 510 if (debug) 511 printf("First alternate superblock is no good, using primary\n"); 512 free(altfs); 513 } else { 514 if (lfs_verify(fs, altfs, devvp, debug) == fs) { 515 free(altfs); 516 } else { 517 free(fs); 518 fs = altfs; 519 } 520 } 521 } 522 } 523 if (check_sb(fs)) { 524 free(fs); 525 return NULL; 526 } 527 } 528 529 /* Compatibility */ 530 if (fs->lfs_version < 2) { 531 fs->lfs_sumsize = LFS_V1_SUMMARY_SIZE; 532 fs->lfs_ibsize = fs->lfs_bsize; 533 fs->lfs_start = fs->lfs_sboffs[0]; 534 fs->lfs_tstamp = fs->lfs_otstamp; 535 fs->lfs_fsbtodb = 0; 536 } 537 538 if (!dummy_read) { 539 fs->lfs_suflags = emalloc(2 * sizeof(u_int32_t *)); 540 fs->lfs_suflags[0] = emalloc(fs->lfs_nseg * sizeof(u_int32_t)); 541 fs->lfs_suflags[1] = emalloc(fs->lfs_nseg * sizeof(u_int32_t)); 542 } 543 544 if (idaddr == 0) 545 idaddr = fs->lfs_idaddr; 546 else 547 fs->lfs_idaddr = idaddr; 548 /* NB: If dummy_read!=0, idaddr==0 here so we get a fake inode. */ 549 fs->lfs_ivnode = lfs_raw_vget(fs, 550 (dummy_read ? LFS_IFILE_INUM : fs->lfs_ifile), devvp->v_fd, 551 idaddr); 552 if (fs->lfs_ivnode == NULL) 553 return NULL; 554 555 register_vget((void *)fs, lfs_vget); 556 557 return fs; 558} 559 560/* 561 * Check partial segment validity between fs->lfs_offset and the given goal. 562 * 563 * If goal == 0, just keep on going until the segments stop making sense, 564 * and return the address of the last valid partial segment. 565 * 566 * If goal != 0, return the address of the first partial segment that failed, 567 * or "goal" if we reached it without failure (the partial segment *at* goal 568 * need not be valid). 569 */ 570ufs_daddr_t 571try_verify(struct lfs *osb, struct uvnode *devvp, ufs_daddr_t goal, int debug) 572{ 573 ufs_daddr_t daddr, odaddr; 574 SEGSUM *sp; 575 int i, bc, hitclean; 576 struct ubuf *bp; 577 ufs_daddr_t nodirop_daddr; 578 u_int64_t serial; 579 580 bc = 0; 581 hitclean = 0; 582 odaddr = -1; 583 daddr = osb->lfs_offset; 584 nodirop_daddr = daddr; 585 serial = osb->lfs_serial; 586 while (daddr != goal) { 587 /* 588 * Don't mistakenly read a superblock, if there is one here. 589 */ 590 if (sntod(osb, dtosn(osb, daddr)) == daddr) { 591 if (daddr == osb->lfs_start) 592 daddr += btofsb(osb, LFS_LABELPAD); 593 for (i = 0; i < LFS_MAXNUMSB; i++) { 594 if (osb->lfs_sboffs[i] < daddr) 595 break; 596 if (osb->lfs_sboffs[i] == daddr) 597 daddr += btofsb(osb, LFS_SBPAD); 598 } 599 } 600 601 /* Read in summary block */ 602 bread(devvp, fsbtodb(osb, daddr), osb->lfs_sumsize, 603 NULL, 0, &bp); 604 sp = (SEGSUM *)bp->b_data; 605 606 /* 607 * Check for a valid segment summary belonging to our fs. 608 */ 609 if (sp->ss_magic != SS_MAGIC || 610 sp->ss_ident != osb->lfs_ident || 611 sp->ss_serial < serial || /* XXX strengthen this */ 612 sp->ss_sumsum != cksum(&sp->ss_datasum, osb->lfs_sumsize - 613 sizeof(sp->ss_sumsum))) { 614 brelse(bp, 0); 615 if (debug) { 616 if (sp->ss_magic != SS_MAGIC) 617 pwarn("pseg at 0x%x: " 618 "wrong magic number\n", 619 (int)daddr); 620 else if (sp->ss_ident != osb->lfs_ident) 621 pwarn("pseg at 0x%x: " 622 "expected ident %llx, got %llx\n", 623 (int)daddr, 624 (long long)sp->ss_ident, 625 (long long)osb->lfs_ident); 626 else if (sp->ss_serial >= serial) 627 pwarn("pseg at 0x%x: " 628 "serial %d < %d\n", (int)daddr, 629 (int)sp->ss_serial, (int)serial); 630 else 631 pwarn("pseg at 0x%x: " 632 "summary checksum wrong\n", 633 (int)daddr); 634 } 635 break; 636 } 637 if (debug && sp->ss_serial != serial) 638 pwarn("warning, serial=%d ss_serial=%d\n", 639 (int)serial, (int)sp->ss_serial); 640 ++serial; 641 bc = check_summary(osb, sp, daddr, debug, devvp, NULL); 642 if (bc == 0) { 643 brelse(bp, 0); 644 break; 645 } 646 if (debug) 647 pwarn("summary good: 0x%x/%d\n", (int)daddr, 648 (int)sp->ss_serial); 649 assert (bc > 0); 650 odaddr = daddr; 651 daddr += btofsb(osb, osb->lfs_sumsize + bc); 652 if (dtosn(osb, odaddr) != dtosn(osb, daddr) || 653 dtosn(osb, daddr) != dtosn(osb, daddr + 654 btofsb(osb, osb->lfs_sumsize + osb->lfs_bsize) - 1)) { 655 daddr = sp->ss_next; 656 } 657 658 /* 659 * Check for the beginning and ending of a sequence of 660 * dirops. Writes from the cleaner never involve new 661 * information, and are always checkpoints; so don't try 662 * to roll forward through them. Likewise, psegs written 663 * by a previous roll-forward attempt are not interesting. 664 */ 665 if (sp->ss_flags & (SS_CLEAN | SS_RFW)) 666 hitclean = 1; 667 if (hitclean == 0 && (sp->ss_flags & SS_CONT) == 0) 668 nodirop_daddr = daddr; 669 670 brelse(bp, 0); 671 } 672 673 if (goal == 0) 674 return nodirop_daddr; 675 else 676 return daddr; 677} 678 679/* Use try_verify to check whether the newer superblock is valid. */ 680struct lfs * 681lfs_verify(struct lfs *sb0, struct lfs *sb1, struct uvnode *devvp, int debug) 682{ 683 ufs_daddr_t daddr; 684 struct lfs *osb, *nsb; 685 686 /* 687 * Verify the checkpoint of the newer superblock, 688 * if the timestamp/serial number of the two superblocks is 689 * different. 690 */ 691 692 osb = NULL; 693 if (debug) 694 pwarn("sb0 %lld, sb1 %lld", 695 (long long) sb0->lfs_serial, 696 (long long) sb1->lfs_serial); 697 698 if ((sb0->lfs_version == 1 && 699 sb0->lfs_otstamp != sb1->lfs_otstamp) || 700 (sb0->lfs_version > 1 && 701 sb0->lfs_serial != sb1->lfs_serial)) { 702 if (sb0->lfs_version == 1) { 703 if (sb0->lfs_otstamp > sb1->lfs_otstamp) { 704 osb = sb1; 705 nsb = sb0; 706 } else { 707 osb = sb0; 708 nsb = sb1; 709 } 710 } else { 711 if (sb0->lfs_serial > sb1->lfs_serial) { 712 osb = sb1; 713 nsb = sb0; 714 } else { 715 osb = sb0; 716 nsb = sb1; 717 } 718 } 719 if (debug) { 720 printf("Attempting to verify newer checkpoint..."); 721 fflush(stdout); 722 } 723 daddr = try_verify(osb, devvp, nsb->lfs_offset, debug); 724 725 if (debug) 726 printf("done.\n"); 727 if (daddr == nsb->lfs_offset) { 728 pwarn("** Newer checkpoint verified, recovered %lld seconds of data\n", 729 (long long) nsb->lfs_tstamp - (long long) osb->lfs_tstamp); 730 sbdirty(); 731 } else { 732 pwarn("** Newer checkpoint invalid, lost %lld seconds of data\n", (long long) nsb->lfs_tstamp - (long long) osb->lfs_tstamp); 733 } 734 return (daddr == nsb->lfs_offset ? nsb : osb); 735 } 736 /* Nothing to check */ 737 return osb; 738} 739 740/* Verify a partial-segment summary; return the number of bytes on disk. */ 741int 742check_summary(struct lfs *fs, SEGSUM *sp, ufs_daddr_t pseg_addr, int debug, 743 struct uvnode *devvp, void (func(ufs_daddr_t, FINFO *))) 744{ 745 FINFO *fp; 746 int bc; /* Bytes in partial segment */ 747 int nblocks; 748 ufs_daddr_t seg_addr, daddr; 749 ufs_daddr_t *dp, *idp; 750 struct ubuf *bp; 751 int i, j, k, datac, len; 752 long sn; 753 u_int32_t *datap; 754 u_int32_t ccksum; 755 756 sn = dtosn(fs, pseg_addr); 757 seg_addr = sntod(fs, sn); 758 759 /* We've already checked the sumsum, just do the data bounds and sum */ 760 761 /* Count the blocks. */ 762 nblocks = howmany(sp->ss_ninos, INOPB(fs)); 763 bc = nblocks << (fs->lfs_version > 1 ? fs->lfs_ffshift : fs->lfs_bshift); 764 assert(bc >= 0); 765 766 fp = (FINFO *) (sp + 1); 767 for (i = 0; i < sp->ss_nfinfo; i++) { 768 nblocks += fp->fi_nblocks; 769 bc += fp->fi_lastlength + ((fp->fi_nblocks - 1) 770 << fs->lfs_bshift); 771 assert(bc >= 0); 772 fp = (FINFO *) (fp->fi_blocks + fp->fi_nblocks); 773 if (((char *)fp) - (char *)sp > fs->lfs_sumsize) 774 return 0; 775 } 776 datap = emalloc(nblocks * sizeof(*datap)); 777 datac = 0; 778 779 dp = (ufs_daddr_t *) sp; 780 dp += fs->lfs_sumsize / sizeof(ufs_daddr_t); 781 dp--; 782 783 idp = dp; 784 daddr = pseg_addr + btofsb(fs, fs->lfs_sumsize); 785 fp = (FINFO *) (sp + 1); 786 for (i = 0, j = 0; 787 i < sp->ss_nfinfo || j < howmany(sp->ss_ninos, INOPB(fs)); i++) { 788 if (i >= sp->ss_nfinfo && *idp != daddr) { 789 pwarn("Not enough inode blocks in pseg at 0x%" PRIx32 790 ": found %d, wanted %d\n", 791 pseg_addr, j, howmany(sp->ss_ninos, INOPB(fs))); 792 if (debug) 793 pwarn("*idp=%x, daddr=%" PRIx32 "\n", *idp, 794 daddr); 795 break; 796 } 797 while (j < howmany(sp->ss_ninos, INOPB(fs)) && *idp == daddr) { 798 bread(devvp, fsbtodb(fs, daddr), fs->lfs_ibsize, 799 NOCRED, 0, &bp); 800 datap[datac++] = ((u_int32_t *) (bp->b_data))[0]; 801 brelse(bp, 0); 802 803 ++j; 804 daddr += btofsb(fs, fs->lfs_ibsize); 805 --idp; 806 } 807 if (i < sp->ss_nfinfo) { 808 if (func) 809 func(daddr, fp); 810 for (k = 0; k < fp->fi_nblocks; k++) { 811 len = (k == fp->fi_nblocks - 1 ? 812 fp->fi_lastlength 813 : fs->lfs_bsize); 814 bread(devvp, fsbtodb(fs, daddr), len, 815 NOCRED, 0, &bp); 816 datap[datac++] = ((u_int32_t *) (bp->b_data))[0]; 817 brelse(bp, 0); 818 daddr += btofsb(fs, len); 819 } 820 fp = (FINFO *) (fp->fi_blocks + fp->fi_nblocks); 821 } 822 } 823 824 if (datac != nblocks) { 825 pwarn("Partial segment at 0x%llx expected %d blocks counted %d\n", 826 (long long) pseg_addr, nblocks, datac); 827 } 828 ccksum = cksum(datap, nblocks * sizeof(u_int32_t)); 829 /* Check the data checksum */ 830 if (ccksum != sp->ss_datasum) { 831 pwarn("Partial segment at 0x%" PRIx32 " data checksum" 832 " mismatch: given 0x%x, computed 0x%x\n", 833 pseg_addr, sp->ss_datasum, ccksum); 834 free(datap); 835 return 0; 836 } 837 free(datap); 838 assert(bc >= 0); 839 return bc; 840} 841 842/* print message and exit */ 843void 844my_vpanic(int fatal, const char *fmt, va_list ap) 845{ 846 (void) vprintf(fmt, ap); 847 exit(8); 848} 849 850void 851call_panic(const char *fmt, ...) 852{ 853 va_list ap; 854 855 va_start(ap, fmt); 856 panic_func(1, fmt, ap); 857 va_end(ap); 858} 859 860/* Allocate a new inode. */ 861struct uvnode * 862lfs_valloc(struct lfs *fs, ino_t ino) 863{ 864 struct ubuf *bp, *cbp; 865 struct ifile *ifp; 866 ino_t new_ino; 867 int error; 868 int new_gen; 869 CLEANERINFO *cip; 870 871 /* Get the head of the freelist. */ 872 LFS_GET_HEADFREE(fs, cip, cbp, &new_ino); 873 874 /* 875 * Remove the inode from the free list and write the new start 876 * of the free list into the superblock. 877 */ 878 LFS_IENTRY(ifp, fs, new_ino, bp); 879 if (ifp->if_daddr != LFS_UNUSED_DADDR) 880 panic("lfs_valloc: inuse inode %d on the free list", new_ino); 881 LFS_PUT_HEADFREE(fs, cip, cbp, ifp->if_nextfree); 882 883 new_gen = ifp->if_version; /* version was updated by vfree */ 884 brelse(bp, 0); 885 886 /* Extend IFILE so that the next lfs_valloc will succeed. */ 887 if (fs->lfs_freehd == LFS_UNUSED_INUM) { 888 if ((error = extend_ifile(fs)) != 0) { 889 LFS_PUT_HEADFREE(fs, cip, cbp, new_ino); 890 return NULL; 891 } 892 } 893 894 /* Set superblock modified bit and increment file count. */ 895 sbdirty(); 896 ++fs->lfs_nfiles; 897 898 return lfs_raw_vget(fs, ino, fs->lfs_devvp->v_fd, 0x0); 899} 900 901#ifdef IN_FSCK_LFS 902void reset_maxino(ino_t); 903#endif 904 905/* 906 * Add a new block to the Ifile, to accommodate future file creations. 907 */ 908int 909extend_ifile(struct lfs *fs) 910{ 911 struct uvnode *vp; 912 struct inode *ip; 913 IFILE *ifp; 914 IFILE_V1 *ifp_v1; 915 struct ubuf *bp, *cbp; 916 daddr_t i, blkno, max; 917 ino_t oldlast; 918 CLEANERINFO *cip; 919 920 vp = fs->lfs_ivnode; 921 ip = VTOI(vp); 922 blkno = lblkno(fs, ip->i_ffs1_size); 923 924 lfs_balloc(vp, ip->i_ffs1_size, fs->lfs_bsize, &bp); 925 ip->i_ffs1_size += fs->lfs_bsize; 926 ip->i_flag |= IN_MODIFIED; 927 928 i = (blkno - fs->lfs_segtabsz - fs->lfs_cleansz) * 929 fs->lfs_ifpb; 930 LFS_GET_HEADFREE(fs, cip, cbp, &oldlast); 931 LFS_PUT_HEADFREE(fs, cip, cbp, i); 932 max = i + fs->lfs_ifpb; 933 fs->lfs_bfree -= btofsb(fs, fs->lfs_bsize); 934 935 if (fs->lfs_version == 1) { 936 for (ifp_v1 = (IFILE_V1 *)bp->b_data; i < max; ++ifp_v1) { 937 ifp_v1->if_version = 1; 938 ifp_v1->if_daddr = LFS_UNUSED_DADDR; 939 ifp_v1->if_nextfree = ++i; 940 } 941 ifp_v1--; 942 ifp_v1->if_nextfree = oldlast; 943 } else { 944 for (ifp = (IFILE *)bp->b_data; i < max; ++ifp) { 945 ifp->if_version = 1; 946 ifp->if_daddr = LFS_UNUSED_DADDR; 947 ifp->if_nextfree = ++i; 948 } 949 ifp--; 950 ifp->if_nextfree = oldlast; 951 } 952 LFS_PUT_TAILFREE(fs, cip, cbp, max - 1); 953 954 LFS_BWRITE_LOG(bp); 955 956#ifdef IN_FSCK_LFS 957 reset_maxino(((ip->i_ffs1_size >> fs->lfs_bshift) - fs->lfs_segtabsz - 958 fs->lfs_cleansz) * fs->lfs_ifpb); 959#endif 960 return 0; 961} 962 963/* 964 * Allocate a block, and to inode and filesystem block accounting for it 965 * and for any indirect blocks the may need to be created in order for 966 * this block to be created. 967 * 968 * Blocks which have never been accounted for (i.e., which "do not exist") 969 * have disk address 0, which is translated by ufs_bmap to the special value 970 * UNASSIGNED == -1, as in the historical UFS. 971 * 972 * Blocks which have been accounted for but which have not yet been written 973 * to disk are given the new special disk address UNWRITTEN == -2, so that 974 * they can be differentiated from completely new blocks. 975 */ 976int 977lfs_balloc(struct uvnode *vp, off_t startoffset, int iosize, struct ubuf **bpp) 978{ 979 int offset; 980 daddr_t daddr, idaddr; 981 struct ubuf *ibp, *bp; 982 struct inode *ip; 983 struct lfs *fs; 984 struct indir indirs[NIADDR+2], *idp; 985 daddr_t lbn, lastblock; 986 int bcount; 987 int error, frags, i, nsize, osize, num; 988 989 ip = VTOI(vp); 990 fs = ip->i_lfs; 991 offset = blkoff(fs, startoffset); 992 lbn = lblkno(fs, startoffset); 993 994 /* 995 * Three cases: it's a block beyond the end of file, it's a block in 996 * the file that may or may not have been assigned a disk address or 997 * we're writing an entire block. 998 * 999 * Note, if the daddr is UNWRITTEN, the block already exists in 1000 * the cache (it was read or written earlier). If so, make sure 1001 * we don't count it as a new block or zero out its contents. If 1002 * it did not, make sure we allocate any necessary indirect 1003 * blocks. 1004 * 1005 * If we are writing a block beyond the end of the file, we need to 1006 * check if the old last block was a fragment. If it was, we need 1007 * to rewrite it. 1008 */ 1009 1010 if (bpp) 1011 *bpp = NULL; 1012 1013 /* Check for block beyond end of file and fragment extension needed. */ 1014 lastblock = lblkno(fs, ip->i_ffs1_size); 1015 if (lastblock < NDADDR && lastblock < lbn) { 1016 osize = blksize(fs, ip, lastblock); 1017 if (osize < fs->lfs_bsize && osize > 0) { 1018 if ((error = lfs_fragextend(vp, osize, fs->lfs_bsize, 1019 lastblock, 1020 (bpp ? &bp : NULL)))) 1021 return (error); 1022 ip->i_ffs1_size = (lastblock + 1) * fs->lfs_bsize; 1023 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1024 if (bpp) 1025 (void) VOP_BWRITE(bp); 1026 } 1027 } 1028 1029 /* 1030 * If the block we are writing is a direct block, it's the last 1031 * block in the file, and offset + iosize is less than a full 1032 * block, we can write one or more fragments. There are two cases: 1033 * the block is brand new and we should allocate it the correct 1034 * size or it already exists and contains some fragments and 1035 * may need to extend it. 1036 */ 1037 if (lbn < NDADDR && lblkno(fs, ip->i_ffs1_size) <= lbn) { 1038 osize = blksize(fs, ip, lbn); 1039 nsize = fragroundup(fs, offset + iosize); 1040 if (lblktosize(fs, lbn) >= ip->i_ffs1_size) { 1041 /* Brand new block or fragment */ 1042 frags = numfrags(fs, nsize); 1043 if (bpp) { 1044 *bpp = bp = getblk(vp, lbn, nsize); 1045 bp->b_blkno = UNWRITTEN; 1046 } 1047 ip->i_lfs_effnblks += frags; 1048 fs->lfs_bfree -= frags; 1049 ip->i_ffs1_db[lbn] = UNWRITTEN; 1050 } else { 1051 if (nsize <= osize) { 1052 /* No need to extend */ 1053 if (bpp && (error = bread(vp, lbn, osize, 1054 NOCRED, 0, &bp))) 1055 return error; 1056 } else { 1057 /* Extend existing block */ 1058 if ((error = 1059 lfs_fragextend(vp, osize, nsize, lbn, 1060 (bpp ? &bp : NULL)))) 1061 return error; 1062 } 1063 if (bpp) 1064 *bpp = bp; 1065 } 1066 return 0; 1067 } 1068 1069 error = ufs_bmaparray(fs, vp, lbn, &daddr, &indirs[0], &num); 1070 if (error) 1071 return (error); 1072 1073 daddr = (daddr_t)((int32_t)daddr); /* XXX ondisk32 */ 1074 1075 /* 1076 * Do byte accounting all at once, so we can gracefully fail *before* 1077 * we start assigning blocks. 1078 */ 1079 frags = fsbtodb(fs, 1); /* frags = VFSTOUFS(vp->v_mount)->um_seqinc; */ 1080 bcount = 0; 1081 if (daddr == UNASSIGNED) { 1082 bcount = frags; 1083 } 1084 for (i = 1; i < num; ++i) { 1085 if (!indirs[i].in_exists) { 1086 bcount += frags; 1087 } 1088 } 1089 fs->lfs_bfree -= bcount; 1090 ip->i_lfs_effnblks += bcount; 1091 1092 if (daddr == UNASSIGNED) { 1093 if (num > 0 && ip->i_ffs1_ib[indirs[0].in_off] == 0) { 1094 ip->i_ffs1_ib[indirs[0].in_off] = UNWRITTEN; 1095 } 1096 1097 /* 1098 * Create new indirect blocks if necessary 1099 */ 1100 if (num > 1) { 1101 idaddr = ip->i_ffs1_ib[indirs[0].in_off]; 1102 for (i = 1; i < num; ++i) { 1103 ibp = getblk(vp, indirs[i].in_lbn, 1104 fs->lfs_bsize); 1105 if (!indirs[i].in_exists) { 1106 memset(ibp->b_data, 0, ibp->b_bufsize); 1107 ibp->b_blkno = UNWRITTEN; 1108 } else if (!(ibp->b_flags & (B_DELWRI | B_DONE))) { 1109 ibp->b_blkno = fsbtodb(fs, idaddr); 1110 ibp->b_flags |= B_READ; 1111 VOP_STRATEGY(ibp); 1112 } 1113 /* 1114 * This block exists, but the next one may not. 1115 * If that is the case mark it UNWRITTEN to 1116 * keep the accounting straight. 1117 */ 1118 /* XXX ondisk32 */ 1119 if (((int32_t *)ibp->b_data)[indirs[i].in_off] == 0) 1120 ((int32_t *)ibp->b_data)[indirs[i].in_off] = 1121 UNWRITTEN; 1122 /* XXX ondisk32 */ 1123 idaddr = ((int32_t *)ibp->b_data)[indirs[i].in_off]; 1124 if ((error = VOP_BWRITE(ibp))) 1125 return error; 1126 } 1127 } 1128 } 1129 1130 1131 /* 1132 * Get the existing block from the cache, if requested. 1133 */ 1134 if (bpp) 1135 *bpp = bp = getblk(vp, lbn, blksize(fs, ip, lbn)); 1136 1137 /* 1138 * The block we are writing may be a brand new block 1139 * in which case we need to do accounting. 1140 * 1141 * We can tell a truly new block because ufs_bmaparray will say 1142 * it is UNASSIGNED. Once we allocate it we will assign it the 1143 * disk address UNWRITTEN. 1144 */ 1145 if (daddr == UNASSIGNED) { 1146 if (bpp) { 1147 /* Note the new address */ 1148 bp->b_blkno = UNWRITTEN; 1149 } 1150 1151 switch (num) { 1152 case 0: 1153 ip->i_ffs1_db[lbn] = UNWRITTEN; 1154 break; 1155 case 1: 1156 ip->i_ffs1_ib[indirs[0].in_off] = UNWRITTEN; 1157 break; 1158 default: 1159 idp = &indirs[num - 1]; 1160 if (bread(vp, idp->in_lbn, fs->lfs_bsize, NOCRED, 1161 0, &ibp)) 1162 panic("lfs_balloc: bread bno %lld", 1163 (long long)idp->in_lbn); 1164 /* XXX ondisk32 */ 1165 ((int32_t *)ibp->b_data)[idp->in_off] = UNWRITTEN; 1166 VOP_BWRITE(ibp); 1167 } 1168 } else if (bpp && !(bp->b_flags & (B_DONE|B_DELWRI))) { 1169 /* 1170 * Not a brand new block, also not in the cache; 1171 * read it in from disk. 1172 */ 1173 if (iosize == fs->lfs_bsize) 1174 /* Optimization: I/O is unnecessary. */ 1175 bp->b_blkno = daddr; 1176 else { 1177 /* 1178 * We need to read the block to preserve the 1179 * existing bytes. 1180 */ 1181 bp->b_blkno = daddr; 1182 bp->b_flags |= B_READ; 1183 VOP_STRATEGY(bp); 1184 return 0; 1185 } 1186 } 1187 1188 return (0); 1189} 1190 1191int 1192lfs_fragextend(struct uvnode *vp, int osize, int nsize, daddr_t lbn, 1193 struct ubuf **bpp) 1194{ 1195 struct inode *ip; 1196 struct lfs *fs; 1197 int frags; 1198 int error; 1199 size_t obufsize; 1200 1201 ip = VTOI(vp); 1202 fs = ip->i_lfs; 1203 frags = (long)numfrags(fs, nsize - osize); 1204 error = 0; 1205 1206 /* 1207 * If we are not asked to actually return the block, all we need 1208 * to do is allocate space for it. UBC will handle dirtying the 1209 * appropriate things and making sure it all goes to disk. 1210 * Don't bother to read in that case. 1211 */ 1212 if (bpp && (error = bread(vp, lbn, osize, NOCRED, 0, bpp))) { 1213 brelse(*bpp, 0); 1214 goto out; 1215 } 1216 1217 fs->lfs_bfree -= frags; 1218 ip->i_lfs_effnblks += frags; 1219 ip->i_flag |= IN_CHANGE | IN_UPDATE; 1220 1221 if (bpp) { 1222 obufsize = (*bpp)->b_bufsize; 1223 (*bpp)->b_data = erealloc((*bpp)->b_data, nsize); 1224 (void)memset((*bpp)->b_data + osize, 0, nsize - osize); 1225 } 1226 1227 out: 1228 return (error); 1229} 1230