suj.c revision 249788
1/*- 2 * Copyright 2009, 2010 Jeffrey W. Roberson <jeff@FreeBSD.org> 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 14 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND 15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE 18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24 * SUCH DAMAGE. 25 */ 26 27#include <sys/cdefs.h> 28__FBSDID("$FreeBSD: stable/9/sbin/fsck_ffs/suj.c 249788 2013-04-23 06:28:49Z mckusick $"); 29 30#include <sys/param.h> 31#include <sys/disk.h> 32#include <sys/disklabel.h> 33#include <sys/mount.h> 34#include <sys/stat.h> 35 36#include <ufs/ufs/ufsmount.h> 37#include <ufs/ufs/dinode.h> 38#include <ufs/ufs/dir.h> 39#include <ufs/ffs/fs.h> 40 41#include <assert.h> 42#include <err.h> 43#include <setjmp.h> 44#include <stdarg.h> 45#include <stdio.h> 46#include <stdlib.h> 47#include <stdint.h> 48#include <libufs.h> 49#include <string.h> 50#include <strings.h> 51#include <sysexits.h> 52#include <time.h> 53 54#include "fsck.h" 55 56#define DOTDOT_OFFSET DIRECTSIZ(1) 57#define SUJ_HASHSIZE 2048 58#define SUJ_HASHMASK (SUJ_HASHSIZE - 1) 59#define SUJ_HASH(x) ((x * 2654435761) & SUJ_HASHMASK) 60 61struct suj_seg { 62 TAILQ_ENTRY(suj_seg) ss_next; 63 struct jsegrec ss_rec; 64 uint8_t *ss_blk; 65}; 66 67struct suj_rec { 68 TAILQ_ENTRY(suj_rec) sr_next; 69 union jrec *sr_rec; 70}; 71TAILQ_HEAD(srechd, suj_rec); 72 73struct suj_ino { 74 LIST_ENTRY(suj_ino) si_next; 75 struct srechd si_recs; 76 struct srechd si_newrecs; 77 struct srechd si_movs; 78 struct jtrncrec *si_trunc; 79 ino_t si_ino; 80 char si_skipparent; 81 char si_hasrecs; 82 char si_blkadj; 83 char si_linkadj; 84 int si_mode; 85 nlink_t si_nlinkadj; 86 nlink_t si_nlink; 87 nlink_t si_dotlinks; 88}; 89LIST_HEAD(inohd, suj_ino); 90 91struct suj_blk { 92 LIST_ENTRY(suj_blk) sb_next; 93 struct srechd sb_recs; 94 ufs2_daddr_t sb_blk; 95}; 96LIST_HEAD(blkhd, suj_blk); 97 98struct data_blk { 99 LIST_ENTRY(data_blk) db_next; 100 uint8_t *db_buf; 101 ufs2_daddr_t db_blk; 102 int db_size; 103 int db_dirty; 104}; 105 106struct ino_blk { 107 LIST_ENTRY(ino_blk) ib_next; 108 uint8_t *ib_buf; 109 int ib_dirty; 110 ufs2_daddr_t ib_blk; 111}; 112LIST_HEAD(iblkhd, ino_blk); 113 114struct suj_cg { 115 LIST_ENTRY(suj_cg) sc_next; 116 struct blkhd sc_blkhash[SUJ_HASHSIZE]; 117 struct inohd sc_inohash[SUJ_HASHSIZE]; 118 struct iblkhd sc_iblkhash[SUJ_HASHSIZE]; 119 struct ino_blk *sc_lastiblk; 120 struct suj_ino *sc_lastino; 121 struct suj_blk *sc_lastblk; 122 uint8_t *sc_cgbuf; 123 struct cg *sc_cgp; 124 int sc_dirty; 125 int sc_cgx; 126}; 127 128LIST_HEAD(cghd, suj_cg) cghash[SUJ_HASHSIZE]; 129LIST_HEAD(dblkhd, data_blk) dbhash[SUJ_HASHSIZE]; 130struct suj_cg *lastcg; 131struct data_blk *lastblk; 132 133TAILQ_HEAD(seghd, suj_seg) allsegs; 134uint64_t oldseq; 135static struct uufsd *disk = NULL; 136static struct fs *fs = NULL; 137ino_t sujino; 138 139/* 140 * Summary statistics. 141 */ 142uint64_t freefrags; 143uint64_t freeblocks; 144uint64_t freeinos; 145uint64_t freedir; 146uint64_t jbytes; 147uint64_t jrecs; 148 149static jmp_buf jmpbuf; 150 151typedef void (*ino_visitor)(ino_t, ufs_lbn_t, ufs2_daddr_t, int); 152static void err_suj(const char *, ...) __dead2; 153static void ino_trunc(ino_t, off_t); 154static void ino_decr(ino_t); 155static void ino_adjust(struct suj_ino *); 156static void ino_build(struct suj_ino *); 157static int blk_isfree(ufs2_daddr_t); 158 159static void * 160errmalloc(size_t n) 161{ 162 void *a; 163 164 a = Malloc(n); 165 if (a == NULL) 166 err(EX_OSERR, "malloc(%zu)", n); 167 return (a); 168} 169 170/* 171 * When hit a fatal error in journalling check, print out 172 * the error and then offer to fallback to normal fsck. 173 */ 174static void 175err_suj(const char * restrict fmt, ...) 176{ 177 va_list ap; 178 179 if (preen) 180 (void)fprintf(stdout, "%s: ", cdevname); 181 182 va_start(ap, fmt); 183 (void)vfprintf(stdout, fmt, ap); 184 va_end(ap); 185 186 longjmp(jmpbuf, -1); 187} 188 189/* 190 * Open the given provider, load superblock. 191 */ 192static void 193opendisk(const char *devnam) 194{ 195 if (disk != NULL) 196 return; 197 disk = Malloc(sizeof(*disk)); 198 if (disk == NULL) 199 err(EX_OSERR, "malloc(%zu)", sizeof(*disk)); 200 if (ufs_disk_fillout(disk, devnam) == -1) { 201 err(EX_OSERR, "ufs_disk_fillout(%s) failed: %s", devnam, 202 disk->d_error); 203 } 204 fs = &disk->d_fs; 205 if (real_dev_bsize == 0 && ioctl(disk->d_fd, DIOCGSECTORSIZE, 206 &real_dev_bsize) == -1) 207 real_dev_bsize = secsize; 208 if (debug) 209 printf("dev_bsize %u\n", real_dev_bsize); 210} 211 212/* 213 * Mark file system as clean, write the super-block back, close the disk. 214 */ 215static void 216closedisk(const char *devnam) 217{ 218 struct csum *cgsum; 219 int i; 220 221 /* 222 * Recompute the fs summary info from correct cs summaries. 223 */ 224 bzero(&fs->fs_cstotal, sizeof(struct csum_total)); 225 for (i = 0; i < fs->fs_ncg; i++) { 226 cgsum = &fs->fs_cs(fs, i); 227 fs->fs_cstotal.cs_nffree += cgsum->cs_nffree; 228 fs->fs_cstotal.cs_nbfree += cgsum->cs_nbfree; 229 fs->fs_cstotal.cs_nifree += cgsum->cs_nifree; 230 fs->fs_cstotal.cs_ndir += cgsum->cs_ndir; 231 } 232 fs->fs_pendinginodes = 0; 233 fs->fs_pendingblocks = 0; 234 fs->fs_clean = 1; 235 fs->fs_time = time(NULL); 236 fs->fs_mtime = time(NULL); 237 if (sbwrite(disk, 0) == -1) 238 err(EX_OSERR, "sbwrite(%s)", devnam); 239 if (ufs_disk_close(disk) == -1) 240 err(EX_OSERR, "ufs_disk_close(%s)", devnam); 241 free(disk); 242 disk = NULL; 243 fs = NULL; 244} 245 246/* 247 * Lookup a cg by number in the hash so we can keep track of which cgs 248 * need stats rebuilt. 249 */ 250static struct suj_cg * 251cg_lookup(int cgx) 252{ 253 struct cghd *hd; 254 struct suj_cg *sc; 255 256 if (cgx < 0 || cgx >= fs->fs_ncg) 257 err_suj("Bad cg number %d\n", cgx); 258 if (lastcg && lastcg->sc_cgx == cgx) 259 return (lastcg); 260 hd = &cghash[SUJ_HASH(cgx)]; 261 LIST_FOREACH(sc, hd, sc_next) 262 if (sc->sc_cgx == cgx) { 263 lastcg = sc; 264 return (sc); 265 } 266 sc = errmalloc(sizeof(*sc)); 267 bzero(sc, sizeof(*sc)); 268 sc->sc_cgbuf = errmalloc(fs->fs_bsize); 269 sc->sc_cgp = (struct cg *)sc->sc_cgbuf; 270 sc->sc_cgx = cgx; 271 LIST_INSERT_HEAD(hd, sc, sc_next); 272 if (bread(disk, fsbtodb(fs, cgtod(fs, sc->sc_cgx)), sc->sc_cgbuf, 273 fs->fs_bsize) == -1) 274 err_suj("Unable to read cylinder group %d\n", sc->sc_cgx); 275 276 return (sc); 277} 278 279/* 280 * Lookup an inode number in the hash and allocate a suj_ino if it does 281 * not exist. 282 */ 283static struct suj_ino * 284ino_lookup(ino_t ino, int creat) 285{ 286 struct suj_ino *sino; 287 struct inohd *hd; 288 struct suj_cg *sc; 289 290 sc = cg_lookup(ino_to_cg(fs, ino)); 291 if (sc->sc_lastino && sc->sc_lastino->si_ino == ino) 292 return (sc->sc_lastino); 293 hd = &sc->sc_inohash[SUJ_HASH(ino)]; 294 LIST_FOREACH(sino, hd, si_next) 295 if (sino->si_ino == ino) 296 return (sino); 297 if (creat == 0) 298 return (NULL); 299 sino = errmalloc(sizeof(*sino)); 300 bzero(sino, sizeof(*sino)); 301 sino->si_ino = ino; 302 TAILQ_INIT(&sino->si_recs); 303 TAILQ_INIT(&sino->si_newrecs); 304 TAILQ_INIT(&sino->si_movs); 305 LIST_INSERT_HEAD(hd, sino, si_next); 306 307 return (sino); 308} 309 310/* 311 * Lookup a block number in the hash and allocate a suj_blk if it does 312 * not exist. 313 */ 314static struct suj_blk * 315blk_lookup(ufs2_daddr_t blk, int creat) 316{ 317 struct suj_blk *sblk; 318 struct suj_cg *sc; 319 struct blkhd *hd; 320 321 sc = cg_lookup(dtog(fs, blk)); 322 if (sc->sc_lastblk && sc->sc_lastblk->sb_blk == blk) 323 return (sc->sc_lastblk); 324 hd = &sc->sc_blkhash[SUJ_HASH(fragstoblks(fs, blk))]; 325 LIST_FOREACH(sblk, hd, sb_next) 326 if (sblk->sb_blk == blk) 327 return (sblk); 328 if (creat == 0) 329 return (NULL); 330 sblk = errmalloc(sizeof(*sblk)); 331 bzero(sblk, sizeof(*sblk)); 332 sblk->sb_blk = blk; 333 TAILQ_INIT(&sblk->sb_recs); 334 LIST_INSERT_HEAD(hd, sblk, sb_next); 335 336 return (sblk); 337} 338 339static struct data_blk * 340dblk_lookup(ufs2_daddr_t blk) 341{ 342 struct data_blk *dblk; 343 struct dblkhd *hd; 344 345 hd = &dbhash[SUJ_HASH(fragstoblks(fs, blk))]; 346 if (lastblk && lastblk->db_blk == blk) 347 return (lastblk); 348 LIST_FOREACH(dblk, hd, db_next) 349 if (dblk->db_blk == blk) 350 return (dblk); 351 /* 352 * The inode block wasn't located, allocate a new one. 353 */ 354 dblk = errmalloc(sizeof(*dblk)); 355 bzero(dblk, sizeof(*dblk)); 356 LIST_INSERT_HEAD(hd, dblk, db_next); 357 dblk->db_blk = blk; 358 return (dblk); 359} 360 361static uint8_t * 362dblk_read(ufs2_daddr_t blk, int size) 363{ 364 struct data_blk *dblk; 365 366 dblk = dblk_lookup(blk); 367 /* 368 * I doubt size mismatches can happen in practice but it is trivial 369 * to handle. 370 */ 371 if (size != dblk->db_size) { 372 if (dblk->db_buf) 373 free(dblk->db_buf); 374 dblk->db_buf = errmalloc(size); 375 dblk->db_size = size; 376 if (bread(disk, fsbtodb(fs, blk), dblk->db_buf, size) == -1) 377 err_suj("Failed to read data block %jd\n", blk); 378 } 379 return (dblk->db_buf); 380} 381 382static void 383dblk_dirty(ufs2_daddr_t blk) 384{ 385 struct data_blk *dblk; 386 387 dblk = dblk_lookup(blk); 388 dblk->db_dirty = 1; 389} 390 391static void 392dblk_write(void) 393{ 394 struct data_blk *dblk; 395 int i; 396 397 for (i = 0; i < SUJ_HASHSIZE; i++) { 398 LIST_FOREACH(dblk, &dbhash[i], db_next) { 399 if (dblk->db_dirty == 0 || dblk->db_size == 0) 400 continue; 401 if (bwrite(disk, fsbtodb(fs, dblk->db_blk), 402 dblk->db_buf, dblk->db_size) == -1) 403 err_suj("Unable to write block %jd\n", 404 dblk->db_blk); 405 } 406 } 407} 408 409static union dinode * 410ino_read(ino_t ino) 411{ 412 struct ino_blk *iblk; 413 struct iblkhd *hd; 414 struct suj_cg *sc; 415 ufs2_daddr_t blk; 416 int off; 417 418 blk = ino_to_fsba(fs, ino); 419 sc = cg_lookup(ino_to_cg(fs, ino)); 420 iblk = sc->sc_lastiblk; 421 if (iblk && iblk->ib_blk == blk) 422 goto found; 423 hd = &sc->sc_iblkhash[SUJ_HASH(fragstoblks(fs, blk))]; 424 LIST_FOREACH(iblk, hd, ib_next) 425 if (iblk->ib_blk == blk) 426 goto found; 427 /* 428 * The inode block wasn't located, allocate a new one. 429 */ 430 iblk = errmalloc(sizeof(*iblk)); 431 bzero(iblk, sizeof(*iblk)); 432 iblk->ib_buf = errmalloc(fs->fs_bsize); 433 iblk->ib_blk = blk; 434 LIST_INSERT_HEAD(hd, iblk, ib_next); 435 if (bread(disk, fsbtodb(fs, blk), iblk->ib_buf, fs->fs_bsize) == -1) 436 err_suj("Failed to read inode block %jd\n", blk); 437found: 438 sc->sc_lastiblk = iblk; 439 off = ino_to_fsbo(fs, ino); 440 if (fs->fs_magic == FS_UFS1_MAGIC) 441 return (union dinode *)&((struct ufs1_dinode *)iblk->ib_buf)[off]; 442 else 443 return (union dinode *)&((struct ufs2_dinode *)iblk->ib_buf)[off]; 444} 445 446static void 447ino_dirty(ino_t ino) 448{ 449 struct ino_blk *iblk; 450 struct iblkhd *hd; 451 struct suj_cg *sc; 452 ufs2_daddr_t blk; 453 454 blk = ino_to_fsba(fs, ino); 455 sc = cg_lookup(ino_to_cg(fs, ino)); 456 iblk = sc->sc_lastiblk; 457 if (iblk && iblk->ib_blk == blk) { 458 iblk->ib_dirty = 1; 459 return; 460 } 461 hd = &sc->sc_iblkhash[SUJ_HASH(fragstoblks(fs, blk))]; 462 LIST_FOREACH(iblk, hd, ib_next) { 463 if (iblk->ib_blk == blk) { 464 iblk->ib_dirty = 1; 465 return; 466 } 467 } 468 ino_read(ino); 469 ino_dirty(ino); 470} 471 472static void 473iblk_write(struct ino_blk *iblk) 474{ 475 476 if (iblk->ib_dirty == 0) 477 return; 478 if (bwrite(disk, fsbtodb(fs, iblk->ib_blk), iblk->ib_buf, 479 fs->fs_bsize) == -1) 480 err_suj("Failed to write inode block %jd\n", iblk->ib_blk); 481} 482 483static int 484blk_overlaps(struct jblkrec *brec, ufs2_daddr_t start, int frags) 485{ 486 ufs2_daddr_t bstart; 487 ufs2_daddr_t bend; 488 ufs2_daddr_t end; 489 490 end = start + frags; 491 bstart = brec->jb_blkno + brec->jb_oldfrags; 492 bend = bstart + brec->jb_frags; 493 if (start < bend && end > bstart) 494 return (1); 495 return (0); 496} 497 498static int 499blk_equals(struct jblkrec *brec, ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t start, 500 int frags) 501{ 502 503 if (brec->jb_ino != ino || brec->jb_lbn != lbn) 504 return (0); 505 if (brec->jb_blkno + brec->jb_oldfrags != start) 506 return (0); 507 if (brec->jb_frags < frags) 508 return (0); 509 return (1); 510} 511 512static void 513blk_setmask(struct jblkrec *brec, int *mask) 514{ 515 int i; 516 517 for (i = brec->jb_oldfrags; i < brec->jb_oldfrags + brec->jb_frags; i++) 518 *mask |= 1 << i; 519} 520 521/* 522 * Determine whether a given block has been reallocated to a new location. 523 * Returns a mask of overlapping bits if any frags have been reused or 524 * zero if the block has not been re-used and the contents can be trusted. 525 * 526 * This is used to ensure that an orphaned pointer due to truncate is safe 527 * to be freed. The mask value can be used to free partial blocks. 528 */ 529static int 530blk_freemask(ufs2_daddr_t blk, ino_t ino, ufs_lbn_t lbn, int frags) 531{ 532 struct suj_blk *sblk; 533 struct suj_rec *srec; 534 struct jblkrec *brec; 535 int mask; 536 int off; 537 538 /* 539 * To be certain we're not freeing a reallocated block we lookup 540 * this block in the blk hash and see if there is an allocation 541 * journal record that overlaps with any fragments in the block 542 * we're concerned with. If any fragments have ben reallocated 543 * the block has already been freed and re-used for another purpose. 544 */ 545 mask = 0; 546 sblk = blk_lookup(blknum(fs, blk), 0); 547 if (sblk == NULL) 548 return (0); 549 off = blk - sblk->sb_blk; 550 TAILQ_FOREACH(srec, &sblk->sb_recs, sr_next) { 551 brec = (struct jblkrec *)srec->sr_rec; 552 /* 553 * If the block overlaps but does not match 554 * exactly this record refers to the current 555 * location. 556 */ 557 if (blk_overlaps(brec, blk, frags) == 0) 558 continue; 559 if (blk_equals(brec, ino, lbn, blk, frags) == 1) 560 mask = 0; 561 else 562 blk_setmask(brec, &mask); 563 } 564 if (debug) 565 printf("blk_freemask: blk %jd sblk %jd off %d mask 0x%X\n", 566 blk, sblk->sb_blk, off, mask); 567 return (mask >> off); 568} 569 570/* 571 * Determine whether it is safe to follow an indirect. It is not safe 572 * if any part of the indirect has been reallocated or the last journal 573 * entry was an allocation. Just allocated indirects may not have valid 574 * pointers yet and all of their children will have their own records. 575 * It is also not safe to follow an indirect if the cg bitmap has been 576 * cleared as a new allocation may write to the block prior to the journal 577 * being written. 578 * 579 * Returns 1 if it's safe to follow the indirect and 0 otherwise. 580 */ 581static int 582blk_isindir(ufs2_daddr_t blk, ino_t ino, ufs_lbn_t lbn) 583{ 584 struct suj_blk *sblk; 585 struct jblkrec *brec; 586 587 sblk = blk_lookup(blk, 0); 588 if (sblk == NULL) 589 return (1); 590 if (TAILQ_EMPTY(&sblk->sb_recs)) 591 return (1); 592 brec = (struct jblkrec *)TAILQ_LAST(&sblk->sb_recs, srechd)->sr_rec; 593 if (blk_equals(brec, ino, lbn, blk, fs->fs_frag)) 594 if (brec->jb_op == JOP_FREEBLK) 595 return (!blk_isfree(blk)); 596 return (0); 597} 598 599/* 600 * Clear an inode from the cg bitmap. If the inode was already clear return 601 * 0 so the caller knows it does not have to check the inode contents. 602 */ 603static int 604ino_free(ino_t ino, int mode) 605{ 606 struct suj_cg *sc; 607 uint8_t *inosused; 608 struct cg *cgp; 609 int cg; 610 611 cg = ino_to_cg(fs, ino); 612 ino = ino % fs->fs_ipg; 613 sc = cg_lookup(cg); 614 cgp = sc->sc_cgp; 615 inosused = cg_inosused(cgp); 616 /* 617 * The bitmap may never have made it to the disk so we have to 618 * conditionally clear. We can avoid writing the cg in this case. 619 */ 620 if (isclr(inosused, ino)) 621 return (0); 622 freeinos++; 623 clrbit(inosused, ino); 624 if (ino < cgp->cg_irotor) 625 cgp->cg_irotor = ino; 626 cgp->cg_cs.cs_nifree++; 627 if ((mode & IFMT) == IFDIR) { 628 freedir++; 629 cgp->cg_cs.cs_ndir--; 630 } 631 sc->sc_dirty = 1; 632 633 return (1); 634} 635 636/* 637 * Free 'frags' frags starting at filesystem block 'bno' skipping any frags 638 * set in the mask. 639 */ 640static void 641blk_free(ufs2_daddr_t bno, int mask, int frags) 642{ 643 ufs1_daddr_t fragno, cgbno; 644 struct suj_cg *sc; 645 struct cg *cgp; 646 int i, cg; 647 uint8_t *blksfree; 648 649 if (debug) 650 printf("Freeing %d frags at blk %jd mask 0x%x\n", 651 frags, bno, mask); 652 cg = dtog(fs, bno); 653 sc = cg_lookup(cg); 654 cgp = sc->sc_cgp; 655 cgbno = dtogd(fs, bno); 656 blksfree = cg_blksfree(cgp); 657 658 /* 659 * If it's not allocated we only wrote the journal entry 660 * and never the bitmaps. Here we unconditionally clear and 661 * resolve the cg summary later. 662 */ 663 if (frags == fs->fs_frag && mask == 0) { 664 fragno = fragstoblks(fs, cgbno); 665 ffs_setblock(fs, blksfree, fragno); 666 freeblocks++; 667 } else { 668 /* 669 * deallocate the fragment 670 */ 671 for (i = 0; i < frags; i++) 672 if ((mask & (1 << i)) == 0 && isclr(blksfree, cgbno +i)) { 673 freefrags++; 674 setbit(blksfree, cgbno + i); 675 } 676 } 677 sc->sc_dirty = 1; 678} 679 680/* 681 * Returns 1 if the whole block starting at 'bno' is marked free and 0 682 * otherwise. 683 */ 684static int 685blk_isfree(ufs2_daddr_t bno) 686{ 687 struct suj_cg *sc; 688 689 sc = cg_lookup(dtog(fs, bno)); 690 return ffs_isblock(fs, cg_blksfree(sc->sc_cgp), dtogd(fs, bno)); 691} 692 693/* 694 * Fetch an indirect block to find the block at a given lbn. The lbn 695 * may be negative to fetch a specific indirect block pointer or positive 696 * to fetch a specific block. 697 */ 698static ufs2_daddr_t 699indir_blkatoff(ufs2_daddr_t blk, ino_t ino, ufs_lbn_t cur, ufs_lbn_t lbn) 700{ 701 ufs2_daddr_t *bap2; 702 ufs2_daddr_t *bap1; 703 ufs_lbn_t lbnadd; 704 ufs_lbn_t base; 705 int level; 706 int i; 707 708 if (blk == 0) 709 return (0); 710 level = lbn_level(cur); 711 if (level == -1) 712 err_suj("Invalid indir lbn %jd\n", lbn); 713 if (level == 0 && lbn < 0) 714 err_suj("Invalid lbn %jd\n", lbn); 715 bap2 = (void *)dblk_read(blk, fs->fs_bsize); 716 bap1 = (void *)bap2; 717 lbnadd = 1; 718 base = -(cur + level); 719 for (i = level; i > 0; i--) 720 lbnadd *= NINDIR(fs); 721 if (lbn > 0) 722 i = (lbn - base) / lbnadd; 723 else 724 i = (-lbn - base) / lbnadd; 725 if (i < 0 || i >= NINDIR(fs)) 726 err_suj("Invalid indirect index %d produced by lbn %jd\n", 727 i, lbn); 728 if (level == 0) 729 cur = base + (i * lbnadd); 730 else 731 cur = -(base + (i * lbnadd)) - (level - 1); 732 if (fs->fs_magic == FS_UFS1_MAGIC) 733 blk = bap1[i]; 734 else 735 blk = bap2[i]; 736 if (cur == lbn) 737 return (blk); 738 if (level == 0) 739 err_suj("Invalid lbn %jd at level 0\n", lbn); 740 return indir_blkatoff(blk, ino, cur, lbn); 741} 742 743/* 744 * Finds the disk block address at the specified lbn within the inode 745 * specified by ip. This follows the whole tree and honors di_size and 746 * di_extsize so it is a true test of reachability. The lbn may be 747 * negative if an extattr or indirect block is requested. 748 */ 749static ufs2_daddr_t 750ino_blkatoff(union dinode *ip, ino_t ino, ufs_lbn_t lbn, int *frags) 751{ 752 ufs_lbn_t tmpval; 753 ufs_lbn_t cur; 754 ufs_lbn_t next; 755 int i; 756 757 /* 758 * Handle extattr blocks first. 759 */ 760 if (lbn < 0 && lbn >= -NXADDR) { 761 lbn = -1 - lbn; 762 if (lbn > lblkno(fs, ip->dp2.di_extsize - 1)) 763 return (0); 764 *frags = numfrags(fs, sblksize(fs, ip->dp2.di_extsize, lbn)); 765 return (ip->dp2.di_extb[lbn]); 766 } 767 /* 768 * Now direct and indirect. 769 */ 770 if (DIP(ip, di_mode) == IFLNK && 771 DIP(ip, di_size) < fs->fs_maxsymlinklen) 772 return (0); 773 if (lbn >= 0 && lbn < NDADDR) { 774 *frags = numfrags(fs, sblksize(fs, DIP(ip, di_size), lbn)); 775 return (DIP(ip, di_db[lbn])); 776 } 777 *frags = fs->fs_frag; 778 779 for (i = 0, tmpval = NINDIR(fs), cur = NDADDR; i < NIADDR; i++, 780 tmpval *= NINDIR(fs), cur = next) { 781 next = cur + tmpval; 782 if (lbn == -cur - i) 783 return (DIP(ip, di_ib[i])); 784 /* 785 * Determine whether the lbn in question is within this tree. 786 */ 787 if (lbn < 0 && -lbn >= next) 788 continue; 789 if (lbn > 0 && lbn >= next) 790 continue; 791 return indir_blkatoff(DIP(ip, di_ib[i]), ino, -cur - i, lbn); 792 } 793 err_suj("lbn %jd not in ino\n", lbn); 794 /* NOTREACHED */ 795} 796 797/* 798 * Determine whether a block exists at a particular lbn in an inode. 799 * Returns 1 if found, 0 if not. lbn may be negative for indirects 800 * or ext blocks. 801 */ 802static int 803blk_isat(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, int *frags) 804{ 805 union dinode *ip; 806 ufs2_daddr_t nblk; 807 808 ip = ino_read(ino); 809 810 if (DIP(ip, di_nlink) == 0 || DIP(ip, di_mode) == 0) 811 return (0); 812 nblk = ino_blkatoff(ip, ino, lbn, frags); 813 814 return (nblk == blk); 815} 816 817/* 818 * Clear the directory entry at diroff that should point to child. Minimal 819 * checking is done and it is assumed that this path was verified with isat. 820 */ 821static void 822ino_clrat(ino_t parent, off_t diroff, ino_t child) 823{ 824 union dinode *dip; 825 struct direct *dp; 826 ufs2_daddr_t blk; 827 uint8_t *block; 828 ufs_lbn_t lbn; 829 int blksize; 830 int frags; 831 int doff; 832 833 if (debug) 834 printf("Clearing inode %d from parent %d at offset %jd\n", 835 child, parent, diroff); 836 837 lbn = lblkno(fs, diroff); 838 doff = blkoff(fs, diroff); 839 dip = ino_read(parent); 840 blk = ino_blkatoff(dip, parent, lbn, &frags); 841 blksize = sblksize(fs, DIP(dip, di_size), lbn); 842 block = dblk_read(blk, blksize); 843 dp = (struct direct *)&block[doff]; 844 if (dp->d_ino != child) 845 errx(1, "Inode %d does not exist in %d at %jd", 846 child, parent, diroff); 847 dp->d_ino = 0; 848 dblk_dirty(blk); 849 /* 850 * The actual .. reference count will already have been removed 851 * from the parent by the .. remref record. 852 */ 853} 854 855/* 856 * Determines whether a pointer to an inode exists within a directory 857 * at a specified offset. Returns the mode of the found entry. 858 */ 859static int 860ino_isat(ino_t parent, off_t diroff, ino_t child, int *mode, int *isdot) 861{ 862 union dinode *dip; 863 struct direct *dp; 864 ufs2_daddr_t blk; 865 uint8_t *block; 866 ufs_lbn_t lbn; 867 int blksize; 868 int frags; 869 int dpoff; 870 int doff; 871 872 *isdot = 0; 873 dip = ino_read(parent); 874 *mode = DIP(dip, di_mode); 875 if ((*mode & IFMT) != IFDIR) { 876 if (debug) { 877 /* 878 * This can happen if the parent inode 879 * was reallocated. 880 */ 881 if (*mode != 0) 882 printf("Directory %d has bad mode %o\n", 883 parent, *mode); 884 else 885 printf("Directory %d has zero mode\n", parent); 886 } 887 return (0); 888 } 889 lbn = lblkno(fs, diroff); 890 doff = blkoff(fs, diroff); 891 blksize = sblksize(fs, DIP(dip, di_size), lbn); 892 if (diroff + DIRECTSIZ(1) > DIP(dip, di_size) || doff >= blksize) { 893 if (debug) 894 printf("ino %d absent from %d due to offset %jd" 895 " exceeding size %jd\n", 896 child, parent, diroff, DIP(dip, di_size)); 897 return (0); 898 } 899 blk = ino_blkatoff(dip, parent, lbn, &frags); 900 if (blk <= 0) { 901 if (debug) 902 printf("Sparse directory %d", parent); 903 return (0); 904 } 905 block = dblk_read(blk, blksize); 906 /* 907 * Walk through the records from the start of the block to be 908 * certain we hit a valid record and not some junk in the middle 909 * of a file name. Stop when we reach or pass the expected offset. 910 */ 911 dpoff = (doff / DIRBLKSIZ) * DIRBLKSIZ; 912 do { 913 dp = (struct direct *)&block[dpoff]; 914 if (dpoff == doff) 915 break; 916 if (dp->d_reclen == 0) 917 break; 918 dpoff += dp->d_reclen; 919 } while (dpoff <= doff); 920 if (dpoff > fs->fs_bsize) 921 err_suj("Corrupt directory block in dir ino %d\n", parent); 922 /* Not found. */ 923 if (dpoff != doff) { 924 if (debug) 925 printf("ino %d not found in %d, lbn %jd, dpoff %d\n", 926 child, parent, lbn, dpoff); 927 return (0); 928 } 929 /* 930 * We found the item in question. Record the mode and whether it's 931 * a . or .. link for the caller. 932 */ 933 if (dp->d_ino == child) { 934 if (child == parent) 935 *isdot = 1; 936 else if (dp->d_namlen == 2 && 937 dp->d_name[0] == '.' && dp->d_name[1] == '.') 938 *isdot = 1; 939 *mode = DTTOIF(dp->d_type); 940 return (1); 941 } 942 if (debug) 943 printf("ino %d doesn't match dirent ino %d in parent %d\n", 944 child, dp->d_ino, parent); 945 return (0); 946} 947 948#define VISIT_INDIR 0x0001 949#define VISIT_EXT 0x0002 950#define VISIT_ROOT 0x0004 /* Operation came via root & valid pointers. */ 951 952/* 953 * Read an indirect level which may or may not be linked into an inode. 954 */ 955static void 956indir_visit(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, uint64_t *frags, 957 ino_visitor visitor, int flags) 958{ 959 ufs2_daddr_t *bap2; 960 ufs1_daddr_t *bap1; 961 ufs_lbn_t lbnadd; 962 ufs2_daddr_t nblk; 963 ufs_lbn_t nlbn; 964 int level; 965 int i; 966 967 /* 968 * Don't visit indirect blocks with contents we can't trust. This 969 * should only happen when indir_visit() is called to complete a 970 * truncate that never finished and not when a pointer is found via 971 * an inode. 972 */ 973 if (blk == 0) 974 return; 975 level = lbn_level(lbn); 976 if (level == -1) 977 err_suj("Invalid level for lbn %jd\n", lbn); 978 if ((flags & VISIT_ROOT) == 0 && blk_isindir(blk, ino, lbn) == 0) { 979 if (debug) 980 printf("blk %jd ino %d lbn %jd(%d) is not indir.\n", 981 blk, ino, lbn, level); 982 goto out; 983 } 984 lbnadd = 1; 985 for (i = level; i > 0; i--) 986 lbnadd *= NINDIR(fs); 987 bap1 = (void *)dblk_read(blk, fs->fs_bsize); 988 bap2 = (void *)bap1; 989 for (i = 0; i < NINDIR(fs); i++) { 990 if (fs->fs_magic == FS_UFS1_MAGIC) 991 nblk = *bap1++; 992 else 993 nblk = *bap2++; 994 if (nblk == 0) 995 continue; 996 if (level == 0) { 997 nlbn = -lbn + i * lbnadd; 998 (*frags) += fs->fs_frag; 999 visitor(ino, nlbn, nblk, fs->fs_frag); 1000 } else { 1001 nlbn = (lbn + 1) - (i * lbnadd); 1002 indir_visit(ino, nlbn, nblk, frags, visitor, flags); 1003 } 1004 } 1005out: 1006 if (flags & VISIT_INDIR) { 1007 (*frags) += fs->fs_frag; 1008 visitor(ino, lbn, blk, fs->fs_frag); 1009 } 1010} 1011 1012/* 1013 * Visit each block in an inode as specified by 'flags' and call a 1014 * callback function. The callback may inspect or free blocks. The 1015 * count of frags found according to the size in the file is returned. 1016 * This is not valid for sparse files but may be used to determine 1017 * the correct di_blocks for a file. 1018 */ 1019static uint64_t 1020ino_visit(union dinode *ip, ino_t ino, ino_visitor visitor, int flags) 1021{ 1022 ufs_lbn_t nextlbn; 1023 ufs_lbn_t tmpval; 1024 ufs_lbn_t lbn; 1025 uint64_t size; 1026 uint64_t fragcnt; 1027 int mode; 1028 int frags; 1029 int i; 1030 1031 size = DIP(ip, di_size); 1032 mode = DIP(ip, di_mode) & IFMT; 1033 fragcnt = 0; 1034 if ((flags & VISIT_EXT) && 1035 fs->fs_magic == FS_UFS2_MAGIC && ip->dp2.di_extsize) { 1036 for (i = 0; i < NXADDR; i++) { 1037 if (ip->dp2.di_extb[i] == 0) 1038 continue; 1039 frags = sblksize(fs, ip->dp2.di_extsize, i); 1040 frags = numfrags(fs, frags); 1041 fragcnt += frags; 1042 visitor(ino, -1 - i, ip->dp2.di_extb[i], frags); 1043 } 1044 } 1045 /* Skip datablocks for short links and devices. */ 1046 if (mode == IFBLK || mode == IFCHR || 1047 (mode == IFLNK && size < fs->fs_maxsymlinklen)) 1048 return (fragcnt); 1049 for (i = 0; i < NDADDR; i++) { 1050 if (DIP(ip, di_db[i]) == 0) 1051 continue; 1052 frags = sblksize(fs, size, i); 1053 frags = numfrags(fs, frags); 1054 fragcnt += frags; 1055 visitor(ino, i, DIP(ip, di_db[i]), frags); 1056 } 1057 /* 1058 * We know the following indirects are real as we're following 1059 * real pointers to them. 1060 */ 1061 flags |= VISIT_ROOT; 1062 for (i = 0, tmpval = NINDIR(fs), lbn = NDADDR; i < NIADDR; i++, 1063 lbn = nextlbn) { 1064 nextlbn = lbn + tmpval; 1065 tmpval *= NINDIR(fs); 1066 if (DIP(ip, di_ib[i]) == 0) 1067 continue; 1068 indir_visit(ino, -lbn - i, DIP(ip, di_ib[i]), &fragcnt, visitor, 1069 flags); 1070 } 1071 return (fragcnt); 1072} 1073 1074/* 1075 * Null visitor function used when we just want to count blocks and 1076 * record the lbn. 1077 */ 1078ufs_lbn_t visitlbn; 1079static void 1080null_visit(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, int frags) 1081{ 1082 if (lbn > 0) 1083 visitlbn = lbn; 1084} 1085 1086/* 1087 * Recalculate di_blocks when we discover that a block allocation or 1088 * free was not successfully completed. The kernel does not roll this back 1089 * because it would be too expensive to compute which indirects were 1090 * reachable at the time the inode was written. 1091 */ 1092static void 1093ino_adjblks(struct suj_ino *sino) 1094{ 1095 union dinode *ip; 1096 uint64_t blocks; 1097 uint64_t frags; 1098 off_t isize; 1099 off_t size; 1100 ino_t ino; 1101 1102 ino = sino->si_ino; 1103 ip = ino_read(ino); 1104 /* No need to adjust zero'd inodes. */ 1105 if (DIP(ip, di_mode) == 0) 1106 return; 1107 /* 1108 * Visit all blocks and count them as well as recording the last 1109 * valid lbn in the file. If the file size doesn't agree with the 1110 * last lbn we need to truncate to fix it. Otherwise just adjust 1111 * the blocks count. 1112 */ 1113 visitlbn = 0; 1114 frags = ino_visit(ip, ino, null_visit, VISIT_INDIR | VISIT_EXT); 1115 blocks = fsbtodb(fs, frags); 1116 /* 1117 * We assume the size and direct block list is kept coherent by 1118 * softdep. For files that have extended into indirects we truncate 1119 * to the size in the inode or the maximum size permitted by 1120 * populated indirects. 1121 */ 1122 if (visitlbn >= NDADDR) { 1123 isize = DIP(ip, di_size); 1124 size = lblktosize(fs, visitlbn + 1); 1125 if (isize > size) 1126 isize = size; 1127 /* Always truncate to free any unpopulated indirects. */ 1128 ino_trunc(sino->si_ino, isize); 1129 return; 1130 } 1131 if (blocks == DIP(ip, di_blocks)) 1132 return; 1133 if (debug) 1134 printf("ino %d adjusting block count from %jd to %jd\n", 1135 ino, DIP(ip, di_blocks), blocks); 1136 DIP_SET(ip, di_blocks, blocks); 1137 ino_dirty(ino); 1138} 1139 1140static void 1141blk_free_visit(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, int frags) 1142{ 1143 1144 blk_free(blk, blk_freemask(blk, ino, lbn, frags), frags); 1145} 1146 1147/* 1148 * Free a block or tree of blocks that was previously rooted in ino at 1149 * the given lbn. If the lbn is an indirect all children are freed 1150 * recursively. 1151 */ 1152static void 1153blk_free_lbn(ufs2_daddr_t blk, ino_t ino, ufs_lbn_t lbn, int frags, int follow) 1154{ 1155 uint64_t resid; 1156 int mask; 1157 1158 mask = blk_freemask(blk, ino, lbn, frags); 1159 resid = 0; 1160 if (lbn <= -NDADDR && follow && mask == 0) 1161 indir_visit(ino, lbn, blk, &resid, blk_free_visit, VISIT_INDIR); 1162 else 1163 blk_free(blk, mask, frags); 1164} 1165 1166static void 1167ino_setskip(struct suj_ino *sino, ino_t parent) 1168{ 1169 int isdot; 1170 int mode; 1171 1172 if (ino_isat(sino->si_ino, DOTDOT_OFFSET, parent, &mode, &isdot)) 1173 sino->si_skipparent = 1; 1174} 1175 1176static void 1177ino_remref(ino_t parent, ino_t child, uint64_t diroff, int isdotdot) 1178{ 1179 struct suj_ino *sino; 1180 struct suj_rec *srec; 1181 struct jrefrec *rrec; 1182 1183 /* 1184 * Lookup this inode to see if we have a record for it. 1185 */ 1186 sino = ino_lookup(child, 0); 1187 /* 1188 * Tell any child directories we've already removed their 1189 * parent link cnt. Don't try to adjust our link down again. 1190 */ 1191 if (sino != NULL && isdotdot == 0) 1192 ino_setskip(sino, parent); 1193 /* 1194 * No valid record for this inode. Just drop the on-disk 1195 * link by one. 1196 */ 1197 if (sino == NULL || sino->si_hasrecs == 0) { 1198 ino_decr(child); 1199 return; 1200 } 1201 /* 1202 * Use ino_adjust() if ino_check() has already processed this 1203 * child. If we lose the last non-dot reference to a 1204 * directory it will be discarded. 1205 */ 1206 if (sino->si_linkadj) { 1207 sino->si_nlink--; 1208 if (isdotdot) 1209 sino->si_dotlinks--; 1210 ino_adjust(sino); 1211 return; 1212 } 1213 /* 1214 * If we haven't yet processed this inode we need to make 1215 * sure we will successfully discover the lost path. If not 1216 * use nlinkadj to remember. 1217 */ 1218 TAILQ_FOREACH(srec, &sino->si_recs, sr_next) { 1219 rrec = (struct jrefrec *)srec->sr_rec; 1220 if (rrec->jr_parent == parent && 1221 rrec->jr_diroff == diroff) 1222 return; 1223 } 1224 sino->si_nlinkadj++; 1225} 1226 1227/* 1228 * Free the children of a directory when the directory is discarded. 1229 */ 1230static void 1231ino_free_children(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, int frags) 1232{ 1233 struct suj_ino *sino; 1234 struct direct *dp; 1235 off_t diroff; 1236 uint8_t *block; 1237 int skipparent; 1238 int isdotdot; 1239 int dpoff; 1240 int size; 1241 1242 sino = ino_lookup(ino, 0); 1243 if (sino) 1244 skipparent = sino->si_skipparent; 1245 else 1246 skipparent = 0; 1247 size = lfragtosize(fs, frags); 1248 block = dblk_read(blk, size); 1249 dp = (struct direct *)&block[0]; 1250 for (dpoff = 0; dpoff < size && dp->d_reclen; dpoff += dp->d_reclen) { 1251 dp = (struct direct *)&block[dpoff]; 1252 if (dp->d_ino == 0 || dp->d_ino == WINO) 1253 continue; 1254 if (dp->d_namlen == 1 && dp->d_name[0] == '.') 1255 continue; 1256 isdotdot = dp->d_namlen == 2 && dp->d_name[0] == '.' && 1257 dp->d_name[1] == '.'; 1258 if (isdotdot && skipparent == 1) 1259 continue; 1260 if (debug) 1261 printf("Directory %d removing ino %d name %s\n", 1262 ino, dp->d_ino, dp->d_name); 1263 diroff = lblktosize(fs, lbn) + dpoff; 1264 ino_remref(ino, dp->d_ino, diroff, isdotdot); 1265 } 1266} 1267 1268/* 1269 * Reclaim an inode, freeing all blocks and decrementing all children's 1270 * link counts. Free the inode back to the cg. 1271 */ 1272static void 1273ino_reclaim(union dinode *ip, ino_t ino, int mode) 1274{ 1275 uint32_t gen; 1276 1277 if (ino == ROOTINO) 1278 err_suj("Attempting to free ROOTINO\n"); 1279 if (debug) 1280 printf("Truncating and freeing ino %d, nlink %d, mode %o\n", 1281 ino, DIP(ip, di_nlink), DIP(ip, di_mode)); 1282 1283 /* We are freeing an inode or directory. */ 1284 if ((DIP(ip, di_mode) & IFMT) == IFDIR) 1285 ino_visit(ip, ino, ino_free_children, 0); 1286 DIP_SET(ip, di_nlink, 0); 1287 ino_visit(ip, ino, blk_free_visit, VISIT_EXT | VISIT_INDIR); 1288 /* Here we have to clear the inode and release any blocks it holds. */ 1289 gen = DIP(ip, di_gen); 1290 if (fs->fs_magic == FS_UFS1_MAGIC) 1291 bzero(ip, sizeof(struct ufs1_dinode)); 1292 else 1293 bzero(ip, sizeof(struct ufs2_dinode)); 1294 DIP_SET(ip, di_gen, gen); 1295 ino_dirty(ino); 1296 ino_free(ino, mode); 1297 return; 1298} 1299 1300/* 1301 * Adjust an inode's link count down by one when a directory goes away. 1302 */ 1303static void 1304ino_decr(ino_t ino) 1305{ 1306 union dinode *ip; 1307 int reqlink; 1308 int nlink; 1309 int mode; 1310 1311 ip = ino_read(ino); 1312 nlink = DIP(ip, di_nlink); 1313 mode = DIP(ip, di_mode); 1314 if (nlink < 1) 1315 err_suj("Inode %d link count %d invalid\n", ino, nlink); 1316 if (mode == 0) 1317 err_suj("Inode %d has a link of %d with 0 mode\n", ino, nlink); 1318 nlink--; 1319 if ((mode & IFMT) == IFDIR) 1320 reqlink = 2; 1321 else 1322 reqlink = 1; 1323 if (nlink < reqlink) { 1324 if (debug) 1325 printf("ino %d not enough links to live %d < %d\n", 1326 ino, nlink, reqlink); 1327 ino_reclaim(ip, ino, mode); 1328 return; 1329 } 1330 DIP_SET(ip, di_nlink, nlink); 1331 ino_dirty(ino); 1332} 1333 1334/* 1335 * Adjust the inode link count to 'nlink'. If the count reaches zero 1336 * free it. 1337 */ 1338static void 1339ino_adjust(struct suj_ino *sino) 1340{ 1341 struct jrefrec *rrec; 1342 struct suj_rec *srec; 1343 struct suj_ino *stmp; 1344 union dinode *ip; 1345 nlink_t nlink; 1346 int recmode; 1347 int reqlink; 1348 int isdot; 1349 int mode; 1350 ino_t ino; 1351 1352 nlink = sino->si_nlink; 1353 ino = sino->si_ino; 1354 mode = sino->si_mode & IFMT; 1355 /* 1356 * If it's a directory with no dot links, it was truncated before 1357 * the name was cleared. We need to clear the dirent that 1358 * points at it. 1359 */ 1360 if (mode == IFDIR && nlink == 1 && sino->si_dotlinks == 0) { 1361 sino->si_nlink = nlink = 0; 1362 TAILQ_FOREACH(srec, &sino->si_recs, sr_next) { 1363 rrec = (struct jrefrec *)srec->sr_rec; 1364 if (ino_isat(rrec->jr_parent, rrec->jr_diroff, ino, 1365 &recmode, &isdot) == 0) 1366 continue; 1367 ino_clrat(rrec->jr_parent, rrec->jr_diroff, ino); 1368 break; 1369 } 1370 if (srec == NULL) 1371 errx(1, "Directory %d name not found", ino); 1372 } 1373 /* 1374 * If it's a directory with no real names pointing to it go ahead 1375 * and truncate it. This will free any children. 1376 */ 1377 if (mode == IFDIR && nlink - sino->si_dotlinks == 0) { 1378 sino->si_nlink = nlink = 0; 1379 /* 1380 * Mark any .. links so they know not to free this inode 1381 * when they are removed. 1382 */ 1383 TAILQ_FOREACH(srec, &sino->si_recs, sr_next) { 1384 rrec = (struct jrefrec *)srec->sr_rec; 1385 if (rrec->jr_diroff == DOTDOT_OFFSET) { 1386 stmp = ino_lookup(rrec->jr_parent, 0); 1387 if (stmp) 1388 ino_setskip(stmp, ino); 1389 } 1390 } 1391 } 1392 ip = ino_read(ino); 1393 mode = DIP(ip, di_mode) & IFMT; 1394 if (nlink > LINK_MAX) 1395 err_suj( 1396 "ino %d nlink manipulation error, new link %d, old link %d\n", 1397 ino, nlink, DIP(ip, di_nlink)); 1398 if (debug) 1399 printf("Adjusting ino %d, nlink %d, old link %d lastmode %o\n", 1400 ino, nlink, DIP(ip, di_nlink), sino->si_mode); 1401 if (mode == 0) { 1402 if (debug) 1403 printf("ino %d, zero inode freeing bitmap\n", ino); 1404 ino_free(ino, sino->si_mode); 1405 return; 1406 } 1407 /* XXX Should be an assert? */ 1408 if (mode != sino->si_mode && debug) 1409 printf("ino %d, mode %o != %o\n", ino, mode, sino->si_mode); 1410 if ((mode & IFMT) == IFDIR) 1411 reqlink = 2; 1412 else 1413 reqlink = 1; 1414 /* If the inode doesn't have enough links to live, free it. */ 1415 if (nlink < reqlink) { 1416 if (debug) 1417 printf("ino %d not enough links to live %d < %d\n", 1418 ino, nlink, reqlink); 1419 ino_reclaim(ip, ino, mode); 1420 return; 1421 } 1422 /* If required write the updated link count. */ 1423 if (DIP(ip, di_nlink) == nlink) { 1424 if (debug) 1425 printf("ino %d, link matches, skipping.\n", ino); 1426 return; 1427 } 1428 DIP_SET(ip, di_nlink, nlink); 1429 ino_dirty(ino); 1430} 1431 1432/* 1433 * Truncate some or all blocks in an indirect, freeing any that are required 1434 * and zeroing the indirect. 1435 */ 1436static void 1437indir_trunc(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, ufs_lbn_t lastlbn) 1438{ 1439 ufs2_daddr_t *bap2; 1440 ufs1_daddr_t *bap1; 1441 ufs_lbn_t lbnadd; 1442 ufs2_daddr_t nblk; 1443 ufs_lbn_t next; 1444 ufs_lbn_t nlbn; 1445 int dirty; 1446 int level; 1447 int i; 1448 1449 if (blk == 0) 1450 return; 1451 dirty = 0; 1452 level = lbn_level(lbn); 1453 if (level == -1) 1454 err_suj("Invalid level for lbn %jd\n", lbn); 1455 lbnadd = 1; 1456 for (i = level; i > 0; i--) 1457 lbnadd *= NINDIR(fs); 1458 bap1 = (void *)dblk_read(blk, fs->fs_bsize); 1459 bap2 = (void *)bap1; 1460 for (i = 0; i < NINDIR(fs); i++) { 1461 if (fs->fs_magic == FS_UFS1_MAGIC) 1462 nblk = *bap1++; 1463 else 1464 nblk = *bap2++; 1465 if (nblk == 0) 1466 continue; 1467 if (level != 0) { 1468 nlbn = (lbn + 1) - (i * lbnadd); 1469 /* 1470 * Calculate the lbn of the next indirect to 1471 * determine if any of this indirect must be 1472 * reclaimed. 1473 */ 1474 next = -(lbn + level) + ((i+1) * lbnadd); 1475 if (next <= lastlbn) 1476 continue; 1477 indir_trunc(ino, nlbn, nblk, lastlbn); 1478 /* If all of this indirect was reclaimed, free it. */ 1479 nlbn = next - lbnadd; 1480 if (nlbn < lastlbn) 1481 continue; 1482 } else { 1483 nlbn = -lbn + i * lbnadd; 1484 if (nlbn < lastlbn) 1485 continue; 1486 } 1487 dirty = 1; 1488 blk_free(nblk, 0, fs->fs_frag); 1489 if (fs->fs_magic == FS_UFS1_MAGIC) 1490 *(bap1 - 1) = 0; 1491 else 1492 *(bap2 - 1) = 0; 1493 } 1494 if (dirty) 1495 dblk_dirty(blk); 1496} 1497 1498/* 1499 * Truncate an inode to the minimum of the given size or the last populated 1500 * block after any over size have been discarded. The kernel would allocate 1501 * the last block in the file but fsck does not and neither do we. This 1502 * code never extends files, only shrinks them. 1503 */ 1504static void 1505ino_trunc(ino_t ino, off_t size) 1506{ 1507 union dinode *ip; 1508 ufs2_daddr_t bn; 1509 uint64_t totalfrags; 1510 ufs_lbn_t nextlbn; 1511 ufs_lbn_t lastlbn; 1512 ufs_lbn_t tmpval; 1513 ufs_lbn_t lbn; 1514 ufs_lbn_t i; 1515 int frags; 1516 off_t cursize; 1517 off_t off; 1518 int mode; 1519 1520 ip = ino_read(ino); 1521 mode = DIP(ip, di_mode) & IFMT; 1522 cursize = DIP(ip, di_size); 1523 if (debug) 1524 printf("Truncating ino %d, mode %o to size %jd from size %jd\n", 1525 ino, mode, size, cursize); 1526 1527 /* Skip datablocks for short links and devices. */ 1528 if (mode == 0 || mode == IFBLK || mode == IFCHR || 1529 (mode == IFLNK && cursize < fs->fs_maxsymlinklen)) 1530 return; 1531 /* Don't extend. */ 1532 if (size > cursize) 1533 size = cursize; 1534 lastlbn = lblkno(fs, blkroundup(fs, size)); 1535 for (i = lastlbn; i < NDADDR; i++) { 1536 if (DIP(ip, di_db[i]) == 0) 1537 continue; 1538 frags = sblksize(fs, cursize, i); 1539 frags = numfrags(fs, frags); 1540 blk_free(DIP(ip, di_db[i]), 0, frags); 1541 DIP_SET(ip, di_db[i], 0); 1542 } 1543 /* 1544 * Follow indirect blocks, freeing anything required. 1545 */ 1546 for (i = 0, tmpval = NINDIR(fs), lbn = NDADDR; i < NIADDR; i++, 1547 lbn = nextlbn) { 1548 nextlbn = lbn + tmpval; 1549 tmpval *= NINDIR(fs); 1550 /* If we're not freeing any in this indirect range skip it. */ 1551 if (lastlbn >= nextlbn) 1552 continue; 1553 if (DIP(ip, di_ib[i]) == 0) 1554 continue; 1555 indir_trunc(ino, -lbn - i, DIP(ip, di_ib[i]), lastlbn); 1556 /* If we freed everything in this indirect free the indir. */ 1557 if (lastlbn > lbn) 1558 continue; 1559 blk_free(DIP(ip, di_ib[i]), 0, frags); 1560 DIP_SET(ip, di_ib[i], 0); 1561 } 1562 ino_dirty(ino); 1563 /* 1564 * Now that we've freed any whole blocks that exceed the desired 1565 * truncation size, figure out how many blocks remain and what the 1566 * last populated lbn is. We will set the size to this last lbn 1567 * rather than worrying about allocating the final lbn as the kernel 1568 * would've done. This is consistent with normal fsck behavior. 1569 */ 1570 visitlbn = 0; 1571 totalfrags = ino_visit(ip, ino, null_visit, VISIT_INDIR | VISIT_EXT); 1572 if (size > lblktosize(fs, visitlbn + 1)) 1573 size = lblktosize(fs, visitlbn + 1); 1574 /* 1575 * If we're truncating direct blocks we have to adjust frags 1576 * accordingly. 1577 */ 1578 if (visitlbn < NDADDR && totalfrags) { 1579 long oldspace, newspace; 1580 1581 bn = DIP(ip, di_db[visitlbn]); 1582 if (bn == 0) 1583 err_suj("Bad blk at ino %d lbn %jd\n", ino, visitlbn); 1584 oldspace = sblksize(fs, cursize, visitlbn); 1585 newspace = sblksize(fs, size, visitlbn); 1586 if (oldspace != newspace) { 1587 bn += numfrags(fs, newspace); 1588 frags = numfrags(fs, oldspace - newspace); 1589 blk_free(bn, 0, frags); 1590 totalfrags -= frags; 1591 } 1592 } 1593 DIP_SET(ip, di_blocks, fsbtodb(fs, totalfrags)); 1594 DIP_SET(ip, di_size, size); 1595 /* 1596 * If we've truncated into the middle of a block or frag we have 1597 * to zero it here. Otherwise the file could extend into 1598 * uninitialized space later. 1599 */ 1600 off = blkoff(fs, size); 1601 if (off && DIP(ip, di_mode) != IFDIR) { 1602 uint8_t *buf; 1603 long clrsize; 1604 1605 bn = ino_blkatoff(ip, ino, visitlbn, &frags); 1606 if (bn == 0) 1607 err_suj("Block missing from ino %d at lbn %jd\n", 1608 ino, visitlbn); 1609 clrsize = frags * fs->fs_fsize; 1610 buf = dblk_read(bn, clrsize); 1611 clrsize -= off; 1612 buf += off; 1613 bzero(buf, clrsize); 1614 dblk_dirty(bn); 1615 } 1616 return; 1617} 1618 1619/* 1620 * Process records available for one inode and determine whether the 1621 * link count is correct or needs adjusting. 1622 */ 1623static void 1624ino_check(struct suj_ino *sino) 1625{ 1626 struct suj_rec *srec; 1627 struct jrefrec *rrec; 1628 nlink_t dotlinks; 1629 int newlinks; 1630 int removes; 1631 int nlink; 1632 ino_t ino; 1633 int isdot; 1634 int isat; 1635 int mode; 1636 1637 if (sino->si_hasrecs == 0) 1638 return; 1639 ino = sino->si_ino; 1640 rrec = (struct jrefrec *)TAILQ_FIRST(&sino->si_recs)->sr_rec; 1641 nlink = rrec->jr_nlink; 1642 newlinks = 0; 1643 dotlinks = 0; 1644 removes = sino->si_nlinkadj; 1645 TAILQ_FOREACH(srec, &sino->si_recs, sr_next) { 1646 rrec = (struct jrefrec *)srec->sr_rec; 1647 isat = ino_isat(rrec->jr_parent, rrec->jr_diroff, 1648 rrec->jr_ino, &mode, &isdot); 1649 if (isat && (mode & IFMT) != (rrec->jr_mode & IFMT)) 1650 err_suj("Inode mode/directory type mismatch %o != %o\n", 1651 mode, rrec->jr_mode); 1652 if (debug) 1653 printf("jrefrec: op %d ino %d, nlink %d, parent %d, " 1654 "diroff %jd, mode %o, isat %d, isdot %d\n", 1655 rrec->jr_op, rrec->jr_ino, rrec->jr_nlink, 1656 rrec->jr_parent, rrec->jr_diroff, rrec->jr_mode, 1657 isat, isdot); 1658 mode = rrec->jr_mode & IFMT; 1659 if (rrec->jr_op == JOP_REMREF) 1660 removes++; 1661 newlinks += isat; 1662 if (isdot) 1663 dotlinks += isat; 1664 } 1665 /* 1666 * The number of links that remain are the starting link count 1667 * subtracted by the total number of removes with the total 1668 * links discovered back in. An incomplete remove thus 1669 * makes no change to the link count but an add increases 1670 * by one. 1671 */ 1672 if (debug) 1673 printf("ino %d nlink %d newlinks %d removes %d dotlinks %d\n", 1674 ino, nlink, newlinks, removes, dotlinks); 1675 nlink += newlinks; 1676 nlink -= removes; 1677 sino->si_linkadj = 1; 1678 sino->si_nlink = nlink; 1679 sino->si_dotlinks = dotlinks; 1680 sino->si_mode = mode; 1681 ino_adjust(sino); 1682} 1683 1684/* 1685 * Process records available for one block and determine whether it is 1686 * still allocated and whether the owning inode needs to be updated or 1687 * a free completed. 1688 */ 1689static void 1690blk_check(struct suj_blk *sblk) 1691{ 1692 struct suj_rec *srec; 1693 struct jblkrec *brec; 1694 struct suj_ino *sino; 1695 ufs2_daddr_t blk; 1696 int mask; 1697 int frags; 1698 int isat; 1699 1700 /* 1701 * Each suj_blk actually contains records for any fragments in that 1702 * block. As a result we must evaluate each record individually. 1703 */ 1704 sino = NULL; 1705 TAILQ_FOREACH(srec, &sblk->sb_recs, sr_next) { 1706 brec = (struct jblkrec *)srec->sr_rec; 1707 frags = brec->jb_frags; 1708 blk = brec->jb_blkno + brec->jb_oldfrags; 1709 isat = blk_isat(brec->jb_ino, brec->jb_lbn, blk, &frags); 1710 if (sino == NULL || sino->si_ino != brec->jb_ino) { 1711 sino = ino_lookup(brec->jb_ino, 1); 1712 sino->si_blkadj = 1; 1713 } 1714 if (debug) 1715 printf("op %d blk %jd ino %d lbn %jd frags %d isat %d (%d)\n", 1716 brec->jb_op, blk, brec->jb_ino, brec->jb_lbn, 1717 brec->jb_frags, isat, frags); 1718 /* 1719 * If we found the block at this address we still have to 1720 * determine if we need to free the tail end that was 1721 * added by adding contiguous fragments from the same block. 1722 */ 1723 if (isat == 1) { 1724 if (frags == brec->jb_frags) 1725 continue; 1726 mask = blk_freemask(blk, brec->jb_ino, brec->jb_lbn, 1727 brec->jb_frags); 1728 mask >>= frags; 1729 blk += frags; 1730 frags = brec->jb_frags - frags; 1731 blk_free(blk, mask, frags); 1732 continue; 1733 } 1734 /* 1735 * The block wasn't found, attempt to free it. It won't be 1736 * freed if it was actually reallocated. If this was an 1737 * allocation we don't want to follow indirects as they 1738 * may not be written yet. Any children of the indirect will 1739 * have their own records. If it's a free we need to 1740 * recursively free children. 1741 */ 1742 blk_free_lbn(blk, brec->jb_ino, brec->jb_lbn, brec->jb_frags, 1743 brec->jb_op == JOP_FREEBLK); 1744 } 1745} 1746 1747/* 1748 * Walk the list of inode records for this cg and resolve moved and duplicate 1749 * inode references now that we have a complete picture. 1750 */ 1751static void 1752cg_build(struct suj_cg *sc) 1753{ 1754 struct suj_ino *sino; 1755 int i; 1756 1757 for (i = 0; i < SUJ_HASHSIZE; i++) 1758 LIST_FOREACH(sino, &sc->sc_inohash[i], si_next) 1759 ino_build(sino); 1760} 1761 1762/* 1763 * Handle inodes requiring truncation. This must be done prior to 1764 * looking up any inodes in directories. 1765 */ 1766static void 1767cg_trunc(struct suj_cg *sc) 1768{ 1769 struct suj_ino *sino; 1770 int i; 1771 1772 for (i = 0; i < SUJ_HASHSIZE; i++) { 1773 LIST_FOREACH(sino, &sc->sc_inohash[i], si_next) { 1774 if (sino->si_trunc) { 1775 ino_trunc(sino->si_ino, 1776 sino->si_trunc->jt_size); 1777 sino->si_blkadj = 0; 1778 sino->si_trunc = NULL; 1779 } 1780 if (sino->si_blkadj) 1781 ino_adjblks(sino); 1782 } 1783 } 1784} 1785 1786static void 1787cg_adj_blk(struct suj_cg *sc) 1788{ 1789 struct suj_ino *sino; 1790 int i; 1791 1792 for (i = 0; i < SUJ_HASHSIZE; i++) { 1793 LIST_FOREACH(sino, &sc->sc_inohash[i], si_next) { 1794 if (sino->si_blkadj) 1795 ino_adjblks(sino); 1796 } 1797 } 1798} 1799 1800/* 1801 * Free any partially allocated blocks and then resolve inode block 1802 * counts. 1803 */ 1804static void 1805cg_check_blk(struct suj_cg *sc) 1806{ 1807 struct suj_blk *sblk; 1808 int i; 1809 1810 1811 for (i = 0; i < SUJ_HASHSIZE; i++) 1812 LIST_FOREACH(sblk, &sc->sc_blkhash[i], sb_next) 1813 blk_check(sblk); 1814} 1815 1816/* 1817 * Walk the list of inode records for this cg, recovering any 1818 * changes which were not complete at the time of crash. 1819 */ 1820static void 1821cg_check_ino(struct suj_cg *sc) 1822{ 1823 struct suj_ino *sino; 1824 int i; 1825 1826 for (i = 0; i < SUJ_HASHSIZE; i++) 1827 LIST_FOREACH(sino, &sc->sc_inohash[i], si_next) 1828 ino_check(sino); 1829} 1830 1831/* 1832 * Write a potentially dirty cg. Recalculate the summary information and 1833 * update the superblock summary. 1834 */ 1835static void 1836cg_write(struct suj_cg *sc) 1837{ 1838 ufs1_daddr_t fragno, cgbno, maxbno; 1839 u_int8_t *blksfree; 1840 struct cg *cgp; 1841 int blk; 1842 int i; 1843 1844 if (sc->sc_dirty == 0) 1845 return; 1846 /* 1847 * Fix the frag and cluster summary. 1848 */ 1849 cgp = sc->sc_cgp; 1850 cgp->cg_cs.cs_nbfree = 0; 1851 cgp->cg_cs.cs_nffree = 0; 1852 bzero(&cgp->cg_frsum, sizeof(cgp->cg_frsum)); 1853 maxbno = fragstoblks(fs, fs->fs_fpg); 1854 if (fs->fs_contigsumsize > 0) { 1855 for (i = 1; i <= fs->fs_contigsumsize; i++) 1856 cg_clustersum(cgp)[i] = 0; 1857 bzero(cg_clustersfree(cgp), howmany(maxbno, CHAR_BIT)); 1858 } 1859 blksfree = cg_blksfree(cgp); 1860 for (cgbno = 0; cgbno < maxbno; cgbno++) { 1861 if (ffs_isfreeblock(fs, blksfree, cgbno)) 1862 continue; 1863 if (ffs_isblock(fs, blksfree, cgbno)) { 1864 ffs_clusteracct(fs, cgp, cgbno, 1); 1865 cgp->cg_cs.cs_nbfree++; 1866 continue; 1867 } 1868 fragno = blkstofrags(fs, cgbno); 1869 blk = blkmap(fs, blksfree, fragno); 1870 ffs_fragacct(fs, blk, cgp->cg_frsum, 1); 1871 for (i = 0; i < fs->fs_frag; i++) 1872 if (isset(blksfree, fragno + i)) 1873 cgp->cg_cs.cs_nffree++; 1874 } 1875 /* 1876 * Update the superblock cg summary from our now correct values 1877 * before writing the block. 1878 */ 1879 fs->fs_cs(fs, sc->sc_cgx) = cgp->cg_cs; 1880 if (bwrite(disk, fsbtodb(fs, cgtod(fs, sc->sc_cgx)), sc->sc_cgbuf, 1881 fs->fs_bsize) == -1) 1882 err_suj("Unable to write cylinder group %d\n", sc->sc_cgx); 1883} 1884 1885/* 1886 * Write out any modified inodes. 1887 */ 1888static void 1889cg_write_inos(struct suj_cg *sc) 1890{ 1891 struct ino_blk *iblk; 1892 int i; 1893 1894 for (i = 0; i < SUJ_HASHSIZE; i++) 1895 LIST_FOREACH(iblk, &sc->sc_iblkhash[i], ib_next) 1896 if (iblk->ib_dirty) 1897 iblk_write(iblk); 1898} 1899 1900static void 1901cg_apply(void (*apply)(struct suj_cg *)) 1902{ 1903 struct suj_cg *scg; 1904 int i; 1905 1906 for (i = 0; i < SUJ_HASHSIZE; i++) 1907 LIST_FOREACH(scg, &cghash[i], sc_next) 1908 apply(scg); 1909} 1910 1911/* 1912 * Process the unlinked but referenced file list. Freeing all inodes. 1913 */ 1914static void 1915ino_unlinked(void) 1916{ 1917 union dinode *ip; 1918 uint16_t mode; 1919 ino_t inon; 1920 ino_t ino; 1921 1922 ino = fs->fs_sujfree; 1923 fs->fs_sujfree = 0; 1924 while (ino != 0) { 1925 ip = ino_read(ino); 1926 mode = DIP(ip, di_mode) & IFMT; 1927 inon = DIP(ip, di_freelink); 1928 DIP_SET(ip, di_freelink, 0); 1929 /* 1930 * XXX Should this be an errx? 1931 */ 1932 if (DIP(ip, di_nlink) == 0) { 1933 if (debug) 1934 printf("Freeing unlinked ino %d mode %o\n", 1935 ino, mode); 1936 ino_reclaim(ip, ino, mode); 1937 } else if (debug) 1938 printf("Skipping ino %d mode %o with link %d\n", 1939 ino, mode, DIP(ip, di_nlink)); 1940 ino = inon; 1941 } 1942} 1943 1944/* 1945 * Append a new record to the list of records requiring processing. 1946 */ 1947static void 1948ino_append(union jrec *rec) 1949{ 1950 struct jrefrec *refrec; 1951 struct jmvrec *mvrec; 1952 struct suj_ino *sino; 1953 struct suj_rec *srec; 1954 1955 mvrec = &rec->rec_jmvrec; 1956 refrec = &rec->rec_jrefrec; 1957 if (debug && mvrec->jm_op == JOP_MVREF) 1958 printf("ino move: ino %d, parent %d, diroff %jd, oldoff %jd\n", 1959 mvrec->jm_ino, mvrec->jm_parent, mvrec->jm_newoff, 1960 mvrec->jm_oldoff); 1961 else if (debug && 1962 (refrec->jr_op == JOP_ADDREF || refrec->jr_op == JOP_REMREF)) 1963 printf("ino ref: op %d, ino %d, nlink %d, " 1964 "parent %d, diroff %jd\n", 1965 refrec->jr_op, refrec->jr_ino, refrec->jr_nlink, 1966 refrec->jr_parent, refrec->jr_diroff); 1967 sino = ino_lookup(((struct jrefrec *)rec)->jr_ino, 1); 1968 sino->si_hasrecs = 1; 1969 srec = errmalloc(sizeof(*srec)); 1970 srec->sr_rec = rec; 1971 TAILQ_INSERT_TAIL(&sino->si_newrecs, srec, sr_next); 1972} 1973 1974/* 1975 * Add a reference adjustment to the sino list and eliminate dups. The 1976 * primary loop in ino_build_ref() checks for dups but new ones may be 1977 * created as a result of offset adjustments. 1978 */ 1979static void 1980ino_add_ref(struct suj_ino *sino, struct suj_rec *srec) 1981{ 1982 struct jrefrec *refrec; 1983 struct suj_rec *srn; 1984 struct jrefrec *rrn; 1985 1986 refrec = (struct jrefrec *)srec->sr_rec; 1987 /* 1988 * We walk backwards so that the oldest link count is preserved. If 1989 * an add record conflicts with a remove keep the remove. Redundant 1990 * removes are eliminated in ino_build_ref. Otherwise we keep the 1991 * oldest record at a given location. 1992 */ 1993 for (srn = TAILQ_LAST(&sino->si_recs, srechd); srn; 1994 srn = TAILQ_PREV(srn, srechd, sr_next)) { 1995 rrn = (struct jrefrec *)srn->sr_rec; 1996 if (rrn->jr_parent != refrec->jr_parent || 1997 rrn->jr_diroff != refrec->jr_diroff) 1998 continue; 1999 if (rrn->jr_op == JOP_REMREF || refrec->jr_op == JOP_ADDREF) { 2000 rrn->jr_mode = refrec->jr_mode; 2001 return; 2002 } 2003 /* 2004 * Adding a remove. 2005 * 2006 * Replace the record in place with the old nlink in case 2007 * we replace the head of the list. Abandon srec as a dup. 2008 */ 2009 refrec->jr_nlink = rrn->jr_nlink; 2010 srn->sr_rec = srec->sr_rec; 2011 return; 2012 } 2013 TAILQ_INSERT_TAIL(&sino->si_recs, srec, sr_next); 2014} 2015 2016/* 2017 * Create a duplicate of a reference at a previous location. 2018 */ 2019static void 2020ino_dup_ref(struct suj_ino *sino, struct jrefrec *refrec, off_t diroff) 2021{ 2022 struct jrefrec *rrn; 2023 struct suj_rec *srn; 2024 2025 rrn = errmalloc(sizeof(*refrec)); 2026 *rrn = *refrec; 2027 rrn->jr_op = JOP_ADDREF; 2028 rrn->jr_diroff = diroff; 2029 srn = errmalloc(sizeof(*srn)); 2030 srn->sr_rec = (union jrec *)rrn; 2031 ino_add_ref(sino, srn); 2032} 2033 2034/* 2035 * Add a reference to the list at all known locations. We follow the offset 2036 * changes for a single instance and create duplicate add refs at each so 2037 * that we can tolerate any version of the directory block. Eliminate 2038 * removes which collide with adds that are seen in the journal. They should 2039 * not adjust the link count down. 2040 */ 2041static void 2042ino_build_ref(struct suj_ino *sino, struct suj_rec *srec) 2043{ 2044 struct jrefrec *refrec; 2045 struct jmvrec *mvrec; 2046 struct suj_rec *srp; 2047 struct suj_rec *srn; 2048 struct jrefrec *rrn; 2049 off_t diroff; 2050 2051 refrec = (struct jrefrec *)srec->sr_rec; 2052 /* 2053 * Search for a mvrec that matches this offset. Whether it's an add 2054 * or a remove we can delete the mvref after creating a dup record in 2055 * the old location. 2056 */ 2057 if (!TAILQ_EMPTY(&sino->si_movs)) { 2058 diroff = refrec->jr_diroff; 2059 for (srn = TAILQ_LAST(&sino->si_movs, srechd); srn; srn = srp) { 2060 srp = TAILQ_PREV(srn, srechd, sr_next); 2061 mvrec = (struct jmvrec *)srn->sr_rec; 2062 if (mvrec->jm_parent != refrec->jr_parent || 2063 mvrec->jm_newoff != diroff) 2064 continue; 2065 diroff = mvrec->jm_oldoff; 2066 TAILQ_REMOVE(&sino->si_movs, srn, sr_next); 2067 free(srn); 2068 ino_dup_ref(sino, refrec, diroff); 2069 } 2070 } 2071 /* 2072 * If a remove wasn't eliminated by an earlier add just append it to 2073 * the list. 2074 */ 2075 if (refrec->jr_op == JOP_REMREF) { 2076 ino_add_ref(sino, srec); 2077 return; 2078 } 2079 /* 2080 * Walk the list of records waiting to be added to the list. We 2081 * must check for moves that apply to our current offset and remove 2082 * them from the list. Remove any duplicates to eliminate removes 2083 * with corresponding adds. 2084 */ 2085 TAILQ_FOREACH_SAFE(srn, &sino->si_newrecs, sr_next, srp) { 2086 switch (srn->sr_rec->rec_jrefrec.jr_op) { 2087 case JOP_ADDREF: 2088 /* 2089 * This should actually be an error we should 2090 * have a remove for every add journaled. 2091 */ 2092 rrn = (struct jrefrec *)srn->sr_rec; 2093 if (rrn->jr_parent != refrec->jr_parent || 2094 rrn->jr_diroff != refrec->jr_diroff) 2095 break; 2096 TAILQ_REMOVE(&sino->si_newrecs, srn, sr_next); 2097 break; 2098 case JOP_REMREF: 2099 /* 2100 * Once we remove the current iteration of the 2101 * record at this address we're done. 2102 */ 2103 rrn = (struct jrefrec *)srn->sr_rec; 2104 if (rrn->jr_parent != refrec->jr_parent || 2105 rrn->jr_diroff != refrec->jr_diroff) 2106 break; 2107 TAILQ_REMOVE(&sino->si_newrecs, srn, sr_next); 2108 ino_add_ref(sino, srec); 2109 return; 2110 case JOP_MVREF: 2111 /* 2112 * Update our diroff based on any moves that match 2113 * and remove the move. 2114 */ 2115 mvrec = (struct jmvrec *)srn->sr_rec; 2116 if (mvrec->jm_parent != refrec->jr_parent || 2117 mvrec->jm_oldoff != refrec->jr_diroff) 2118 break; 2119 ino_dup_ref(sino, refrec, mvrec->jm_oldoff); 2120 refrec->jr_diroff = mvrec->jm_newoff; 2121 TAILQ_REMOVE(&sino->si_newrecs, srn, sr_next); 2122 break; 2123 default: 2124 err_suj("ino_build_ref: Unknown op %d\n", 2125 srn->sr_rec->rec_jrefrec.jr_op); 2126 } 2127 } 2128 ino_add_ref(sino, srec); 2129} 2130 2131/* 2132 * Walk the list of new records and add them in-order resolving any 2133 * dups and adjusted offsets. 2134 */ 2135static void 2136ino_build(struct suj_ino *sino) 2137{ 2138 struct suj_rec *srec; 2139 2140 while ((srec = TAILQ_FIRST(&sino->si_newrecs)) != NULL) { 2141 TAILQ_REMOVE(&sino->si_newrecs, srec, sr_next); 2142 switch (srec->sr_rec->rec_jrefrec.jr_op) { 2143 case JOP_ADDREF: 2144 case JOP_REMREF: 2145 ino_build_ref(sino, srec); 2146 break; 2147 case JOP_MVREF: 2148 /* 2149 * Add this mvrec to the queue of pending mvs. 2150 */ 2151 TAILQ_INSERT_TAIL(&sino->si_movs, srec, sr_next); 2152 break; 2153 default: 2154 err_suj("ino_build: Unknown op %d\n", 2155 srec->sr_rec->rec_jrefrec.jr_op); 2156 } 2157 } 2158 if (TAILQ_EMPTY(&sino->si_recs)) 2159 sino->si_hasrecs = 0; 2160} 2161 2162/* 2163 * Modify journal records so they refer to the base block number 2164 * and a start and end frag range. This is to facilitate the discovery 2165 * of overlapping fragment allocations. 2166 */ 2167static void 2168blk_build(struct jblkrec *blkrec) 2169{ 2170 struct suj_rec *srec; 2171 struct suj_blk *sblk; 2172 struct jblkrec *blkrn; 2173 ufs2_daddr_t blk; 2174 int frag; 2175 2176 if (debug) 2177 printf("blk_build: op %d blkno %jd frags %d oldfrags %d " 2178 "ino %d lbn %jd\n", 2179 blkrec->jb_op, blkrec->jb_blkno, blkrec->jb_frags, 2180 blkrec->jb_oldfrags, blkrec->jb_ino, blkrec->jb_lbn); 2181 2182 blk = blknum(fs, blkrec->jb_blkno); 2183 frag = fragnum(fs, blkrec->jb_blkno); 2184 sblk = blk_lookup(blk, 1); 2185 /* 2186 * Rewrite the record using oldfrags to indicate the offset into 2187 * the block. Leave jb_frags as the actual allocated count. 2188 */ 2189 blkrec->jb_blkno -= frag; 2190 blkrec->jb_oldfrags = frag; 2191 if (blkrec->jb_oldfrags + blkrec->jb_frags > fs->fs_frag) 2192 err_suj("Invalid fragment count %d oldfrags %d\n", 2193 blkrec->jb_frags, frag); 2194 /* 2195 * Detect dups. If we detect a dup we always discard the oldest 2196 * record as it is superseded by the new record. This speeds up 2197 * later stages but also eliminates free records which are used 2198 * to indicate that the contents of indirects can be trusted. 2199 */ 2200 TAILQ_FOREACH(srec, &sblk->sb_recs, sr_next) { 2201 blkrn = (struct jblkrec *)srec->sr_rec; 2202 if (blkrn->jb_ino != blkrec->jb_ino || 2203 blkrn->jb_lbn != blkrec->jb_lbn || 2204 blkrn->jb_blkno != blkrec->jb_blkno || 2205 blkrn->jb_frags != blkrec->jb_frags || 2206 blkrn->jb_oldfrags != blkrec->jb_oldfrags) 2207 continue; 2208 if (debug) 2209 printf("Removed dup.\n"); 2210 /* Discard the free which is a dup with an alloc. */ 2211 if (blkrec->jb_op == JOP_FREEBLK) 2212 return; 2213 TAILQ_REMOVE(&sblk->sb_recs, srec, sr_next); 2214 free(srec); 2215 break; 2216 } 2217 srec = errmalloc(sizeof(*srec)); 2218 srec->sr_rec = (union jrec *)blkrec; 2219 TAILQ_INSERT_TAIL(&sblk->sb_recs, srec, sr_next); 2220} 2221 2222static void 2223ino_build_trunc(struct jtrncrec *rec) 2224{ 2225 struct suj_ino *sino; 2226 2227 if (debug) 2228 printf("ino_build_trunc: op %d ino %d, size %jd\n", 2229 rec->jt_op, rec->jt_ino, rec->jt_size); 2230 sino = ino_lookup(rec->jt_ino, 1); 2231 if (rec->jt_op == JOP_SYNC) { 2232 sino->si_trunc = NULL; 2233 return; 2234 } 2235 if (sino->si_trunc == NULL || sino->si_trunc->jt_size > rec->jt_size) 2236 sino->si_trunc = rec; 2237} 2238 2239/* 2240 * Build up tables of the operations we need to recover. 2241 */ 2242static void 2243suj_build(void) 2244{ 2245 struct suj_seg *seg; 2246 union jrec *rec; 2247 int off; 2248 int i; 2249 2250 TAILQ_FOREACH(seg, &allsegs, ss_next) { 2251 if (debug) 2252 printf("seg %jd has %d records, oldseq %jd.\n", 2253 seg->ss_rec.jsr_seq, seg->ss_rec.jsr_cnt, 2254 seg->ss_rec.jsr_oldest); 2255 off = 0; 2256 rec = (union jrec *)seg->ss_blk; 2257 for (i = 0; i < seg->ss_rec.jsr_cnt; off += JREC_SIZE, rec++) { 2258 /* skip the segrec. */ 2259 if ((off % real_dev_bsize) == 0) 2260 continue; 2261 switch (rec->rec_jrefrec.jr_op) { 2262 case JOP_ADDREF: 2263 case JOP_REMREF: 2264 case JOP_MVREF: 2265 ino_append(rec); 2266 break; 2267 case JOP_NEWBLK: 2268 case JOP_FREEBLK: 2269 blk_build((struct jblkrec *)rec); 2270 break; 2271 case JOP_TRUNC: 2272 case JOP_SYNC: 2273 ino_build_trunc((struct jtrncrec *)rec); 2274 break; 2275 default: 2276 err_suj("Unknown journal operation %d (%d)\n", 2277 rec->rec_jrefrec.jr_op, off); 2278 } 2279 i++; 2280 } 2281 } 2282} 2283 2284/* 2285 * Prune the journal segments to those we care about based on the 2286 * oldest sequence in the newest segment. Order the segment list 2287 * based on sequence number. 2288 */ 2289static void 2290suj_prune(void) 2291{ 2292 struct suj_seg *seg; 2293 struct suj_seg *segn; 2294 uint64_t newseq; 2295 int discard; 2296 2297 if (debug) 2298 printf("Pruning up to %jd\n", oldseq); 2299 /* First free the expired segments. */ 2300 TAILQ_FOREACH_SAFE(seg, &allsegs, ss_next, segn) { 2301 if (seg->ss_rec.jsr_seq >= oldseq) 2302 continue; 2303 TAILQ_REMOVE(&allsegs, seg, ss_next); 2304 free(seg->ss_blk); 2305 free(seg); 2306 } 2307 /* Next ensure that segments are ordered properly. */ 2308 seg = TAILQ_FIRST(&allsegs); 2309 if (seg == NULL) { 2310 if (debug) 2311 printf("Empty journal\n"); 2312 return; 2313 } 2314 newseq = seg->ss_rec.jsr_seq; 2315 for (;;) { 2316 seg = TAILQ_LAST(&allsegs, seghd); 2317 if (seg->ss_rec.jsr_seq >= newseq) 2318 break; 2319 TAILQ_REMOVE(&allsegs, seg, ss_next); 2320 TAILQ_INSERT_HEAD(&allsegs, seg, ss_next); 2321 newseq = seg->ss_rec.jsr_seq; 2322 2323 } 2324 if (newseq != oldseq) { 2325 TAILQ_FOREACH(seg, &allsegs, ss_next) { 2326 printf("%jd, ", seg->ss_rec.jsr_seq); 2327 } 2328 printf("\n"); 2329 err_suj("Journal file sequence mismatch %jd != %jd\n", 2330 newseq, oldseq); 2331 } 2332 /* 2333 * The kernel may asynchronously write segments which can create 2334 * gaps in the sequence space. Throw away any segments after the 2335 * gap as the kernel guarantees only those that are contiguously 2336 * reachable are marked as completed. 2337 */ 2338 discard = 0; 2339 TAILQ_FOREACH_SAFE(seg, &allsegs, ss_next, segn) { 2340 if (!discard && newseq++ == seg->ss_rec.jsr_seq) { 2341 jrecs += seg->ss_rec.jsr_cnt; 2342 jbytes += seg->ss_rec.jsr_blocks * real_dev_bsize; 2343 continue; 2344 } 2345 discard = 1; 2346 if (debug) 2347 printf("Journal order mismatch %jd != %jd pruning\n", 2348 newseq-1, seg->ss_rec.jsr_seq); 2349 TAILQ_REMOVE(&allsegs, seg, ss_next); 2350 free(seg->ss_blk); 2351 free(seg); 2352 } 2353 if (debug) 2354 printf("Processing journal segments from %jd to %jd\n", 2355 oldseq, newseq-1); 2356} 2357 2358/* 2359 * Verify the journal inode before attempting to read records. 2360 */ 2361static int 2362suj_verifyino(union dinode *ip) 2363{ 2364 2365 if (DIP(ip, di_nlink) != 1) { 2366 printf("Invalid link count %d for journal inode %d\n", 2367 DIP(ip, di_nlink), sujino); 2368 return (-1); 2369 } 2370 2371 if ((DIP(ip, di_flags) & (SF_IMMUTABLE | SF_NOUNLINK)) != 2372 (SF_IMMUTABLE | SF_NOUNLINK)) { 2373 printf("Invalid flags 0x%X for journal inode %d\n", 2374 DIP(ip, di_flags), sujino); 2375 return (-1); 2376 } 2377 2378 if (DIP(ip, di_mode) != (IFREG | IREAD)) { 2379 printf("Invalid mode %o for journal inode %d\n", 2380 DIP(ip, di_mode), sujino); 2381 return (-1); 2382 } 2383 2384 if (DIP(ip, di_size) < SUJ_MIN) { 2385 printf("Invalid size %jd for journal inode %d\n", 2386 DIP(ip, di_size), sujino); 2387 return (-1); 2388 } 2389 2390 if (DIP(ip, di_modrev) != fs->fs_mtime) { 2391 printf("Journal timestamp does not match fs mount time\n"); 2392 return (-1); 2393 } 2394 2395 return (0); 2396} 2397 2398struct jblocks { 2399 struct jextent *jb_extent; /* Extent array. */ 2400 int jb_avail; /* Available extents. */ 2401 int jb_used; /* Last used extent. */ 2402 int jb_head; /* Allocator head. */ 2403 int jb_off; /* Allocator extent offset. */ 2404}; 2405struct jextent { 2406 ufs2_daddr_t je_daddr; /* Disk block address. */ 2407 int je_blocks; /* Disk block count. */ 2408}; 2409 2410struct jblocks *suj_jblocks; 2411 2412static struct jblocks * 2413jblocks_create(void) 2414{ 2415 struct jblocks *jblocks; 2416 int size; 2417 2418 jblocks = errmalloc(sizeof(*jblocks)); 2419 jblocks->jb_avail = 10; 2420 jblocks->jb_used = 0; 2421 jblocks->jb_head = 0; 2422 jblocks->jb_off = 0; 2423 size = sizeof(struct jextent) * jblocks->jb_avail; 2424 jblocks->jb_extent = errmalloc(size); 2425 bzero(jblocks->jb_extent, size); 2426 2427 return (jblocks); 2428} 2429 2430/* 2431 * Return the next available disk block and the amount of contiguous 2432 * free space it contains. 2433 */ 2434static ufs2_daddr_t 2435jblocks_next(struct jblocks *jblocks, int bytes, int *actual) 2436{ 2437 struct jextent *jext; 2438 ufs2_daddr_t daddr; 2439 int freecnt; 2440 int blocks; 2441 2442 blocks = bytes / disk->d_bsize; 2443 jext = &jblocks->jb_extent[jblocks->jb_head]; 2444 freecnt = jext->je_blocks - jblocks->jb_off; 2445 if (freecnt == 0) { 2446 jblocks->jb_off = 0; 2447 if (++jblocks->jb_head > jblocks->jb_used) 2448 return (0); 2449 jext = &jblocks->jb_extent[jblocks->jb_head]; 2450 freecnt = jext->je_blocks; 2451 } 2452 if (freecnt > blocks) 2453 freecnt = blocks; 2454 *actual = freecnt * disk->d_bsize; 2455 daddr = jext->je_daddr + jblocks->jb_off; 2456 2457 return (daddr); 2458} 2459 2460/* 2461 * Advance the allocation head by a specified number of bytes, consuming 2462 * one journal segment. 2463 */ 2464static void 2465jblocks_advance(struct jblocks *jblocks, int bytes) 2466{ 2467 2468 jblocks->jb_off += bytes / disk->d_bsize; 2469} 2470 2471static void 2472jblocks_destroy(struct jblocks *jblocks) 2473{ 2474 2475 free(jblocks->jb_extent); 2476 free(jblocks); 2477} 2478 2479static void 2480jblocks_add(struct jblocks *jblocks, ufs2_daddr_t daddr, int blocks) 2481{ 2482 struct jextent *jext; 2483 int size; 2484 2485 jext = &jblocks->jb_extent[jblocks->jb_used]; 2486 /* Adding the first block. */ 2487 if (jext->je_daddr == 0) { 2488 jext->je_daddr = daddr; 2489 jext->je_blocks = blocks; 2490 return; 2491 } 2492 /* Extending the last extent. */ 2493 if (jext->je_daddr + jext->je_blocks == daddr) { 2494 jext->je_blocks += blocks; 2495 return; 2496 } 2497 /* Adding a new extent. */ 2498 if (++jblocks->jb_used == jblocks->jb_avail) { 2499 jblocks->jb_avail *= 2; 2500 size = sizeof(struct jextent) * jblocks->jb_avail; 2501 jext = errmalloc(size); 2502 bzero(jext, size); 2503 bcopy(jblocks->jb_extent, jext, 2504 sizeof(struct jextent) * jblocks->jb_used); 2505 free(jblocks->jb_extent); 2506 jblocks->jb_extent = jext; 2507 } 2508 jext = &jblocks->jb_extent[jblocks->jb_used]; 2509 jext->je_daddr = daddr; 2510 jext->je_blocks = blocks; 2511 2512 return; 2513} 2514 2515/* 2516 * Add a file block from the journal to the extent map. We can't read 2517 * each file block individually because the kernel treats it as a circular 2518 * buffer and segments may span mutliple contiguous blocks. 2519 */ 2520static void 2521suj_add_block(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, int frags) 2522{ 2523 2524 jblocks_add(suj_jblocks, fsbtodb(fs, blk), fsbtodb(fs, frags)); 2525} 2526 2527static void 2528suj_read(void) 2529{ 2530 uint8_t block[1 * 1024 * 1024]; 2531 struct suj_seg *seg; 2532 struct jsegrec *recn; 2533 struct jsegrec *rec; 2534 ufs2_daddr_t blk; 2535 int readsize; 2536 int blocks; 2537 int recsize; 2538 int size; 2539 int i; 2540 2541 /* 2542 * Read records until we exhaust the journal space. If we find 2543 * an invalid record we start searching for a valid segment header 2544 * at the next block. This is because we don't have a head/tail 2545 * pointer and must recover the information indirectly. At the gap 2546 * between the head and tail we won't necessarily have a valid 2547 * segment. 2548 */ 2549restart: 2550 for (;;) { 2551 size = sizeof(block); 2552 blk = jblocks_next(suj_jblocks, size, &readsize); 2553 if (blk == 0) 2554 return; 2555 size = readsize; 2556 /* 2557 * Read 1MB at a time and scan for records within this block. 2558 */ 2559 if (bread(disk, blk, &block, size) == -1) { 2560 err_suj("Error reading journal block %jd\n", 2561 (intmax_t)blk); 2562 } 2563 for (rec = (void *)block; size; size -= recsize, 2564 rec = (struct jsegrec *)((uintptr_t)rec + recsize)) { 2565 recsize = real_dev_bsize; 2566 if (rec->jsr_time != fs->fs_mtime) { 2567 if (debug) 2568 printf("Rec time %jd != fs mtime %jd\n", 2569 rec->jsr_time, fs->fs_mtime); 2570 jblocks_advance(suj_jblocks, recsize); 2571 continue; 2572 } 2573 if (rec->jsr_cnt == 0) { 2574 if (debug) 2575 printf("Found illegal count %d\n", 2576 rec->jsr_cnt); 2577 jblocks_advance(suj_jblocks, recsize); 2578 continue; 2579 } 2580 blocks = rec->jsr_blocks; 2581 recsize = blocks * real_dev_bsize; 2582 if (recsize > size) { 2583 /* 2584 * We may just have run out of buffer, restart 2585 * the loop to re-read from this spot. 2586 */ 2587 if (size < fs->fs_bsize && 2588 size != readsize && 2589 recsize <= fs->fs_bsize) 2590 goto restart; 2591 if (debug) 2592 printf("Found invalid segsize %d > %d\n", 2593 recsize, size); 2594 recsize = real_dev_bsize; 2595 jblocks_advance(suj_jblocks, recsize); 2596 continue; 2597 } 2598 /* 2599 * Verify that all blocks in the segment are present. 2600 */ 2601 for (i = 1; i < blocks; i++) { 2602 recn = (void *)((uintptr_t)rec) + i * 2603 real_dev_bsize; 2604 if (recn->jsr_seq == rec->jsr_seq && 2605 recn->jsr_time == rec->jsr_time) 2606 continue; 2607 if (debug) 2608 printf("Incomplete record %jd (%d)\n", 2609 rec->jsr_seq, i); 2610 recsize = i * real_dev_bsize; 2611 jblocks_advance(suj_jblocks, recsize); 2612 goto restart; 2613 } 2614 seg = errmalloc(sizeof(*seg)); 2615 seg->ss_blk = errmalloc(recsize); 2616 seg->ss_rec = *rec; 2617 bcopy((void *)rec, seg->ss_blk, recsize); 2618 if (rec->jsr_oldest > oldseq) 2619 oldseq = rec->jsr_oldest; 2620 TAILQ_INSERT_TAIL(&allsegs, seg, ss_next); 2621 jblocks_advance(suj_jblocks, recsize); 2622 } 2623 } 2624} 2625 2626/* 2627 * Search a directory block for the SUJ_FILE. 2628 */ 2629static void 2630suj_find(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, int frags) 2631{ 2632 char block[MAXBSIZE]; 2633 struct direct *dp; 2634 int bytes; 2635 int off; 2636 2637 if (sujino) 2638 return; 2639 bytes = lfragtosize(fs, frags); 2640 if (bread(disk, fsbtodb(fs, blk), block, bytes) <= 0) 2641 err_suj("Failed to read ROOTINO directory block %jd\n", blk); 2642 for (off = 0; off < bytes; off += dp->d_reclen) { 2643 dp = (struct direct *)&block[off]; 2644 if (dp->d_reclen == 0) 2645 break; 2646 if (dp->d_ino == 0) 2647 continue; 2648 if (dp->d_namlen != strlen(SUJ_FILE)) 2649 continue; 2650 if (bcmp(dp->d_name, SUJ_FILE, dp->d_namlen) != 0) 2651 continue; 2652 sujino = dp->d_ino; 2653 return; 2654 } 2655} 2656 2657/* 2658 * Orchestrate the verification of a filesystem via the softupdates journal. 2659 */ 2660int 2661suj_check(const char *filesys) 2662{ 2663 union dinode *jip; 2664 union dinode *ip; 2665 uint64_t blocks; 2666 int retval; 2667 struct suj_seg *seg; 2668 struct suj_seg *segn; 2669 2670 opendisk(filesys); 2671 TAILQ_INIT(&allsegs); 2672 2673 /* 2674 * Set an exit point when SUJ check failed 2675 */ 2676 retval = setjmp(jmpbuf); 2677 if (retval != 0) { 2678 pwarn("UNEXPECTED SU+J INCONSISTENCY\n"); 2679 TAILQ_FOREACH_SAFE(seg, &allsegs, ss_next, segn) { 2680 TAILQ_REMOVE(&allsegs, seg, ss_next); 2681 free(seg->ss_blk); 2682 free(seg); 2683 } 2684 if (reply("FALLBACK TO FULL FSCK") == 0) { 2685 ckfini(0); 2686 exit(EEXIT); 2687 } else 2688 return (-1); 2689 } 2690 2691 /* 2692 * Find the journal inode. 2693 */ 2694 ip = ino_read(ROOTINO); 2695 sujino = 0; 2696 ino_visit(ip, ROOTINO, suj_find, 0); 2697 if (sujino == 0) { 2698 printf("Journal inode removed. Use tunefs to re-create.\n"); 2699 sblock.fs_flags &= ~FS_SUJ; 2700 sblock.fs_sujfree = 0; 2701 return (-1); 2702 } 2703 /* 2704 * Fetch the journal inode and verify it. 2705 */ 2706 jip = ino_read(sujino); 2707 printf("** SU+J Recovering %s\n", filesys); 2708 if (suj_verifyino(jip) != 0) 2709 return (-1); 2710 /* 2711 * Build a list of journal blocks in jblocks before parsing the 2712 * available journal blocks in with suj_read(). 2713 */ 2714 printf("** Reading %jd byte journal from inode %d.\n", 2715 DIP(jip, di_size), sujino); 2716 suj_jblocks = jblocks_create(); 2717 blocks = ino_visit(jip, sujino, suj_add_block, 0); 2718 if (blocks != numfrags(fs, DIP(jip, di_size))) { 2719 printf("Sparse journal inode %d.\n", sujino); 2720 return (-1); 2721 } 2722 suj_read(); 2723 jblocks_destroy(suj_jblocks); 2724 suj_jblocks = NULL; 2725 if (preen || reply("RECOVER")) { 2726 printf("** Building recovery table.\n"); 2727 suj_prune(); 2728 suj_build(); 2729 cg_apply(cg_build); 2730 printf("** Resolving unreferenced inode list.\n"); 2731 ino_unlinked(); 2732 printf("** Processing journal entries.\n"); 2733 cg_apply(cg_trunc); 2734 cg_apply(cg_check_blk); 2735 cg_apply(cg_adj_blk); 2736 cg_apply(cg_check_ino); 2737 } 2738 if (preen == 0 && (jrecs > 0 || jbytes > 0) && reply("WRITE CHANGES") == 0) 2739 return (0); 2740 /* 2741 * To remain idempotent with partial truncations the free bitmaps 2742 * must be written followed by indirect blocks and lastly inode 2743 * blocks. This preserves access to the modified pointers until 2744 * they are freed. 2745 */ 2746 cg_apply(cg_write); 2747 dblk_write(); 2748 cg_apply(cg_write_inos); 2749 /* Write back superblock. */ 2750 closedisk(filesys); 2751 if (jrecs > 0 || jbytes > 0) { 2752 printf("** %jd journal records in %jd bytes for %.2f%% utilization\n", 2753 jrecs, jbytes, ((float)jrecs / (float)(jbytes / JREC_SIZE)) * 100); 2754 printf("** Freed %jd inodes (%jd dirs) %jd blocks, and %jd frags.\n", 2755 freeinos, freedir, freeblocks, freefrags); 2756 } 2757 2758 return (0); 2759} 2760