2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright 2009, 2010 Jeffrey W. Roberson <jeff@FreeBSD.org>
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
32 #include <sys/param.h>
34 #include <sys/disklabel.h>
35 #include <sys/mount.h>
38 #include <ufs/ufs/ufsmount.h>
39 #include <ufs/ufs/dinode.h>
40 #include <ufs/ufs/dir.h>
41 #include <ufs/ffs/fs.h>
58 #define DOTDOT_OFFSET DIRECTSIZ(1)
59 #define SUJ_HASHSIZE 2048
60 #define SUJ_HASHMASK (SUJ_HASHSIZE - 1)
61 #define SUJ_HASH(x) ((x * 2654435761) & SUJ_HASHMASK)
64 TAILQ_ENTRY(suj_seg) ss_next;
65 struct jsegrec ss_rec;
70 TAILQ_ENTRY(suj_rec) sr_next;
73 TAILQ_HEAD(srechd, suj_rec);
76 LIST_ENTRY(suj_ino) si_next;
77 struct srechd si_recs;
78 struct srechd si_newrecs;
79 struct srechd si_movs;
80 struct jtrncrec *si_trunc;
91 LIST_HEAD(inohd, suj_ino);
94 LIST_ENTRY(suj_blk) sb_next;
95 struct srechd sb_recs;
98 LIST_HEAD(blkhd, suj_blk);
101 LIST_ENTRY(data_blk) db_next;
109 LIST_ENTRY(ino_blk) ib_next;
112 ino_t ib_startinginum;
115 LIST_HEAD(iblkhd, ino_blk);
118 LIST_ENTRY(suj_cg) sc_next;
119 struct blkhd sc_blkhash[SUJ_HASHSIZE];
120 struct inohd sc_inohash[SUJ_HASHSIZE];
121 struct iblkhd sc_iblkhash[SUJ_HASHSIZE];
122 struct ino_blk *sc_lastiblk;
123 struct suj_ino *sc_lastino;
124 struct suj_blk *sc_lastblk;
131 static LIST_HEAD(cghd, suj_cg) cghash[SUJ_HASHSIZE];
132 static LIST_HEAD(dblkhd, data_blk) dbhash[SUJ_HASHSIZE];
133 static struct suj_cg *lastcg;
134 static struct data_blk *lastblk;
136 static TAILQ_HEAD(seghd, suj_seg) allsegs;
137 static uint64_t oldseq;
138 static struct fs *fs = NULL;
142 * Summary statistics.
144 static uint64_t freefrags;
145 static uint64_t freeblocks;
146 static uint64_t freeinos;
147 static uint64_t freedir;
148 static uint64_t jbytes;
149 static uint64_t jrecs;
151 static jmp_buf jmpbuf;
153 typedef void (*ino_visitor)(ino_t, ufs_lbn_t, ufs2_daddr_t, int);
154 static void err_suj(const char *, ...) __dead2;
155 static void ino_trunc(ino_t, off_t);
156 static void ino_decr(ino_t);
157 static void ino_adjust(struct suj_ino *);
158 static void ino_build(struct suj_ino *);
159 static int blk_isfree(ufs2_daddr_t);
160 static void initsuj(void);
161 static void ino_dirty(ino_t);
170 err(EX_OSERR, "malloc(%zu)", n);
175 * When hit a fatal error in journalling check, print out
176 * the error and then offer to fallback to normal fsck.
179 err_suj(const char * restrict fmt, ...)
184 (void)fprintf(stdout, "%s: ", cdevname);
187 (void)vfprintf(stdout, fmt, ap);
194 * Mark file system as clean, write the super-block back, close the disk.
197 closedisk(const char *devnam)
203 * Recompute the fs summary info from correct cs summaries.
205 bzero(&fs->fs_cstotal, sizeof(struct csum_total));
206 for (i = 0; i < fs->fs_ncg; i++) {
207 cgsum = &fs->fs_cs(fs, i);
208 fs->fs_cstotal.cs_nffree += cgsum->cs_nffree;
209 fs->fs_cstotal.cs_nbfree += cgsum->cs_nbfree;
210 fs->fs_cstotal.cs_nifree += cgsum->cs_nifree;
211 fs->fs_cstotal.cs_ndir += cgsum->cs_ndir;
213 fs->fs_pendinginodes = 0;
214 fs->fs_pendingblocks = 0;
216 fs->fs_time = time(NULL);
217 fs->fs_mtime = time(NULL);
218 if (sbput(disk.d_fd, fs, 0) == -1)
219 err(EX_OSERR, "sbput(%s)", devnam);
220 if (ufs_disk_close(&disk) == -1)
221 err(EX_OSERR, "ufs_disk_close(%s)", devnam);
226 * Lookup a cg by number in the hash so we can keep track of which cgs
227 * need stats rebuilt.
229 static struct suj_cg *
235 if (cgx < 0 || cgx >= fs->fs_ncg)
236 err_suj("Bad cg number %d\n", cgx);
237 if (lastcg && lastcg->sc_cgx == cgx)
239 hd = &cghash[SUJ_HASH(cgx)];
240 LIST_FOREACH(sc, hd, sc_next)
241 if (sc->sc_cgx == cgx) {
245 sc = errmalloc(sizeof(*sc));
246 bzero(sc, sizeof(*sc));
247 sc->sc_cgbuf = errmalloc(fs->fs_bsize);
248 sc->sc_cgp = (struct cg *)sc->sc_cgbuf;
250 LIST_INSERT_HEAD(hd, sc, sc_next);
252 * Use bread() here rather than cgget() because the cylinder group
253 * may be corrupted but we want it anyway so we can fix it.
255 if (bread(&disk, fsbtodb(fs, cgtod(fs, sc->sc_cgx)), sc->sc_cgbuf,
257 err_suj("Unable to read cylinder group %d\n", sc->sc_cgx);
263 * Lookup an inode number in the hash and allocate a suj_ino if it does
266 static struct suj_ino *
267 ino_lookup(ino_t ino, int creat)
269 struct suj_ino *sino;
273 sc = cg_lookup(ino_to_cg(fs, ino));
274 if (sc->sc_lastino && sc->sc_lastino->si_ino == ino)
275 return (sc->sc_lastino);
276 hd = &sc->sc_inohash[SUJ_HASH(ino)];
277 LIST_FOREACH(sino, hd, si_next)
278 if (sino->si_ino == ino)
282 sino = errmalloc(sizeof(*sino));
283 bzero(sino, sizeof(*sino));
285 TAILQ_INIT(&sino->si_recs);
286 TAILQ_INIT(&sino->si_newrecs);
287 TAILQ_INIT(&sino->si_movs);
288 LIST_INSERT_HEAD(hd, sino, si_next);
294 * Lookup a block number in the hash and allocate a suj_blk if it does
297 static struct suj_blk *
298 blk_lookup(ufs2_daddr_t blk, int creat)
300 struct suj_blk *sblk;
304 sc = cg_lookup(dtog(fs, blk));
305 if (sc->sc_lastblk && sc->sc_lastblk->sb_blk == blk)
306 return (sc->sc_lastblk);
307 hd = &sc->sc_blkhash[SUJ_HASH(fragstoblks(fs, blk))];
308 LIST_FOREACH(sblk, hd, sb_next)
309 if (sblk->sb_blk == blk)
313 sblk = errmalloc(sizeof(*sblk));
314 bzero(sblk, sizeof(*sblk));
316 TAILQ_INIT(&sblk->sb_recs);
317 LIST_INSERT_HEAD(hd, sblk, sb_next);
322 static struct data_blk *
323 dblk_lookup(ufs2_daddr_t blk)
325 struct data_blk *dblk;
328 hd = &dbhash[SUJ_HASH(fragstoblks(fs, blk))];
329 if (lastblk && lastblk->db_blk == blk)
331 LIST_FOREACH(dblk, hd, db_next)
332 if (dblk->db_blk == blk)
335 * The inode block wasn't located, allocate a new one.
337 dblk = errmalloc(sizeof(*dblk));
338 bzero(dblk, sizeof(*dblk));
339 LIST_INSERT_HEAD(hd, dblk, db_next);
345 dblk_read(ufs2_daddr_t blk, int size)
347 struct data_blk *dblk;
349 dblk = dblk_lookup(blk);
351 * I doubt size mismatches can happen in practice but it is trivial
354 if (size != dblk->db_size) {
357 dblk->db_buf = errmalloc(size);
358 dblk->db_size = size;
359 if (bread(&disk, fsbtodb(fs, blk), dblk->db_buf, size) == -1)
360 err_suj("Failed to read data block %jd\n", blk);
362 return (dblk->db_buf);
366 dblk_dirty(ufs2_daddr_t blk)
368 struct data_blk *dblk;
370 dblk = dblk_lookup(blk);
377 struct data_blk *dblk;
380 for (i = 0; i < SUJ_HASHSIZE; i++) {
381 LIST_FOREACH(dblk, &dbhash[i], db_next) {
382 if (dblk->db_dirty == 0 || dblk->db_size == 0)
384 if (bwrite(&disk, fsbtodb(fs, dblk->db_blk),
385 dblk->db_buf, dblk->db_size) == -1)
386 err_suj("Unable to write block %jd\n",
392 static union dinode *
395 struct ino_blk *iblk;
402 blk = ino_to_fsba(fs, ino);
403 sc = cg_lookup(ino_to_cg(fs, ino));
404 iblk = sc->sc_lastiblk;
405 if (iblk && iblk->ib_blk == blk)
407 hd = &sc->sc_iblkhash[SUJ_HASH(fragstoblks(fs, blk))];
408 LIST_FOREACH(iblk, hd, ib_next)
409 if (iblk->ib_blk == blk)
412 * The inode block wasn't located, allocate a new one.
414 iblk = errmalloc(sizeof(*iblk));
415 bzero(iblk, sizeof(*iblk));
416 iblk->ib_buf = errmalloc(fs->fs_bsize);
418 iblk->ib_startinginum = rounddown(ino, INOPB(fs));
419 LIST_INSERT_HEAD(hd, iblk, ib_next);
420 if (bread(&disk, fsbtodb(fs, blk), iblk->ib_buf, fs->fs_bsize) == -1)
421 err_suj("Failed to read inode block %jd\n", blk);
423 sc->sc_lastiblk = iblk;
424 off = ino_to_fsbo(fs, ino);
425 if (fs->fs_magic == FS_UFS1_MAGIC)
426 return (union dinode *)&((struct ufs1_dinode *)iblk->ib_buf)[off];
427 dp = (union dinode *)&((struct ufs2_dinode *)iblk->ib_buf)[off];
429 ffs_verify_dinode_ckhash(fs, (struct ufs2_dinode *)dp) != 0) {
430 pwarn("ino_read: INODE CHECK-HASH FAILED");
432 if (preen || reply("FIX") != 0) {
434 printf(" (FIXED)\n");
444 struct ino_blk *iblk;
450 blk = ino_to_fsba(fs, ino);
451 sc = cg_lookup(ino_to_cg(fs, ino));
452 iblk = sc->sc_lastiblk;
453 if (iblk && iblk->ib_blk == blk) {
454 if (fs->fs_magic == FS_UFS2_MAGIC) {
455 off = ino_to_fsbo(fs, ino);
456 ffs_update_dinode_ckhash(fs,
457 &((struct ufs2_dinode *)iblk->ib_buf)[off]);
462 hd = &sc->sc_iblkhash[SUJ_HASH(fragstoblks(fs, blk))];
463 LIST_FOREACH(iblk, hd, ib_next) {
464 if (iblk->ib_blk == blk) {
465 if (fs->fs_magic == FS_UFS2_MAGIC) {
466 off = ino_to_fsbo(fs, ino);
467 ffs_update_dinode_ckhash(fs,
468 &((struct ufs2_dinode *)iblk->ib_buf)[off]);
479 iblk_write(struct ino_blk *iblk)
481 struct ufs2_dinode *dp;
484 if (iblk->ib_dirty == 0)
486 if (debug && fs->fs_magic == FS_UFS2_MAGIC) {
487 dp = (struct ufs2_dinode *)iblk->ib_buf;
488 for (i = 0; i < INOPB(fs); dp++, i++) {
489 if (ffs_verify_dinode_ckhash(fs, dp) == 0)
491 pwarn("iblk_write: INODE CHECK-HASH FAILED");
492 prtinode(iblk->ib_startinginum + i, (union dinode *)dp);
493 if (preen || reply("FIX") != 0) {
495 printf(" (FIXED)\n");
496 ino_dirty(iblk->ib_startinginum + i);
500 if (bwrite(&disk, fsbtodb(fs, iblk->ib_blk), iblk->ib_buf,
502 err_suj("Failed to write inode block %jd\n", iblk->ib_blk);
506 blk_overlaps(struct jblkrec *brec, ufs2_daddr_t start, int frags)
513 bstart = brec->jb_blkno + brec->jb_oldfrags;
514 bend = bstart + brec->jb_frags;
515 if (start < bend && end > bstart)
521 blk_equals(struct jblkrec *brec, ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t start,
525 if (brec->jb_ino != ino || brec->jb_lbn != lbn)
527 if (brec->jb_blkno + brec->jb_oldfrags != start)
529 if (brec->jb_frags < frags)
535 blk_setmask(struct jblkrec *brec, int *mask)
539 for (i = brec->jb_oldfrags; i < brec->jb_oldfrags + brec->jb_frags; i++)
544 * Determine whether a given block has been reallocated to a new location.
545 * Returns a mask of overlapping bits if any frags have been reused or
546 * zero if the block has not been re-used and the contents can be trusted.
548 * This is used to ensure that an orphaned pointer due to truncate is safe
549 * to be freed. The mask value can be used to free partial blocks.
552 blk_freemask(ufs2_daddr_t blk, ino_t ino, ufs_lbn_t lbn, int frags)
554 struct suj_blk *sblk;
555 struct suj_rec *srec;
556 struct jblkrec *brec;
561 * To be certain we're not freeing a reallocated block we lookup
562 * this block in the blk hash and see if there is an allocation
563 * journal record that overlaps with any fragments in the block
564 * we're concerned with. If any fragments have ben reallocated
565 * the block has already been freed and re-used for another purpose.
568 sblk = blk_lookup(blknum(fs, blk), 0);
571 off = blk - sblk->sb_blk;
572 TAILQ_FOREACH(srec, &sblk->sb_recs, sr_next) {
573 brec = (struct jblkrec *)srec->sr_rec;
575 * If the block overlaps but does not match
576 * exactly this record refers to the current
579 if (blk_overlaps(brec, blk, frags) == 0)
581 if (blk_equals(brec, ino, lbn, blk, frags) == 1)
584 blk_setmask(brec, &mask);
587 printf("blk_freemask: blk %jd sblk %jd off %d mask 0x%X\n",
588 blk, sblk->sb_blk, off, mask);
589 return (mask >> off);
593 * Determine whether it is safe to follow an indirect. It is not safe
594 * if any part of the indirect has been reallocated or the last journal
595 * entry was an allocation. Just allocated indirects may not have valid
596 * pointers yet and all of their children will have their own records.
597 * It is also not safe to follow an indirect if the cg bitmap has been
598 * cleared as a new allocation may write to the block prior to the journal
601 * Returns 1 if it's safe to follow the indirect and 0 otherwise.
604 blk_isindir(ufs2_daddr_t blk, ino_t ino, ufs_lbn_t lbn)
606 struct suj_blk *sblk;
607 struct jblkrec *brec;
609 sblk = blk_lookup(blk, 0);
612 if (TAILQ_EMPTY(&sblk->sb_recs))
614 brec = (struct jblkrec *)TAILQ_LAST(&sblk->sb_recs, srechd)->sr_rec;
615 if (blk_equals(brec, ino, lbn, blk, fs->fs_frag))
616 if (brec->jb_op == JOP_FREEBLK)
617 return (!blk_isfree(blk));
622 * Clear an inode from the cg bitmap. If the inode was already clear return
623 * 0 so the caller knows it does not have to check the inode contents.
626 ino_free(ino_t ino, int mode)
633 cg = ino_to_cg(fs, ino);
634 ino = ino % fs->fs_ipg;
637 inosused = cg_inosused(cgp);
639 * The bitmap may never have made it to the disk so we have to
640 * conditionally clear. We can avoid writing the cg in this case.
642 if (isclr(inosused, ino))
645 clrbit(inosused, ino);
646 if (ino < cgp->cg_irotor)
647 cgp->cg_irotor = ino;
648 cgp->cg_cs.cs_nifree++;
649 if ((mode & IFMT) == IFDIR) {
651 cgp->cg_cs.cs_ndir--;
659 * Free 'frags' frags starting at filesystem block 'bno' skipping any frags
663 blk_free(ufs2_daddr_t bno, int mask, int frags)
665 ufs1_daddr_t fragno, cgbno;
672 printf("Freeing %d frags at blk %jd mask 0x%x\n",
677 cgbno = dtogd(fs, bno);
678 blksfree = cg_blksfree(cgp);
681 * If it's not allocated we only wrote the journal entry
682 * and never the bitmaps. Here we unconditionally clear and
683 * resolve the cg summary later.
685 if (frags == fs->fs_frag && mask == 0) {
686 fragno = fragstoblks(fs, cgbno);
687 ffs_setblock(fs, blksfree, fragno);
691 * deallocate the fragment
693 for (i = 0; i < frags; i++)
694 if ((mask & (1 << i)) == 0 && isclr(blksfree, cgbno +i)) {
696 setbit(blksfree, cgbno + i);
703 * Returns 1 if the whole block starting at 'bno' is marked free and 0
707 blk_isfree(ufs2_daddr_t bno)
711 sc = cg_lookup(dtog(fs, bno));
712 return ffs_isblock(fs, cg_blksfree(sc->sc_cgp), dtogd(fs, bno));
716 * Fetch an indirect block to find the block at a given lbn. The lbn
717 * may be negative to fetch a specific indirect block pointer or positive
718 * to fetch a specific block.
721 indir_blkatoff(ufs2_daddr_t blk, ino_t ino, ufs_lbn_t cur, ufs_lbn_t lbn)
732 level = lbn_level(cur);
734 err_suj("Invalid indir lbn %jd\n", lbn);
735 if (level == 0 && lbn < 0)
736 err_suj("Invalid lbn %jd\n", lbn);
737 bap2 = (void *)dblk_read(blk, fs->fs_bsize);
740 base = -(cur + level);
741 for (i = level; i > 0; i--)
742 lbnadd *= NINDIR(fs);
744 i = (lbn - base) / lbnadd;
746 i = (-lbn - base) / lbnadd;
747 if (i < 0 || i >= NINDIR(fs))
748 err_suj("Invalid indirect index %d produced by lbn %jd\n",
751 cur = base + (i * lbnadd);
753 cur = -(base + (i * lbnadd)) - (level - 1);
754 if (fs->fs_magic == FS_UFS1_MAGIC)
761 err_suj("Invalid lbn %jd at level 0\n", lbn);
762 return indir_blkatoff(blk, ino, cur, lbn);
766 * Finds the disk block address at the specified lbn within the inode
767 * specified by ip. This follows the whole tree and honors di_size and
768 * di_extsize so it is a true test of reachability. The lbn may be
769 * negative if an extattr or indirect block is requested.
772 ino_blkatoff(union dinode *ip, ino_t ino, ufs_lbn_t lbn, int *frags)
780 * Handle extattr blocks first.
782 if (lbn < 0 && lbn >= -UFS_NXADDR) {
784 if (lbn > lblkno(fs, ip->dp2.di_extsize - 1))
786 *frags = numfrags(fs, sblksize(fs, ip->dp2.di_extsize, lbn));
787 return (ip->dp2.di_extb[lbn]);
790 * Now direct and indirect.
792 if (DIP(ip, di_mode) == IFLNK &&
793 DIP(ip, di_size) < fs->fs_maxsymlinklen)
795 if (lbn >= 0 && lbn < UFS_NDADDR) {
796 *frags = numfrags(fs, sblksize(fs, DIP(ip, di_size), lbn));
797 return (DIP(ip, di_db[lbn]));
799 *frags = fs->fs_frag;
801 for (i = 0, tmpval = NINDIR(fs), cur = UFS_NDADDR; i < UFS_NIADDR; i++,
802 tmpval *= NINDIR(fs), cur = next) {
805 return (DIP(ip, di_ib[i]));
807 * Determine whether the lbn in question is within this tree.
809 if (lbn < 0 && -lbn >= next)
811 if (lbn > 0 && lbn >= next)
813 return indir_blkatoff(DIP(ip, di_ib[i]), ino, -cur - i, lbn);
815 err_suj("lbn %jd not in ino\n", lbn);
820 * Determine whether a block exists at a particular lbn in an inode.
821 * Returns 1 if found, 0 if not. lbn may be negative for indirects
825 blk_isat(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, int *frags)
832 if (DIP(ip, di_nlink) == 0 || DIP(ip, di_mode) == 0)
834 nblk = ino_blkatoff(ip, ino, lbn, frags);
836 return (nblk == blk);
840 * Clear the directory entry at diroff that should point to child. Minimal
841 * checking is done and it is assumed that this path was verified with isat.
844 ino_clrat(ino_t parent, off_t diroff, ino_t child)
856 printf("Clearing inode %ju from parent %ju at offset %jd\n",
857 (uintmax_t)child, (uintmax_t)parent, diroff);
859 lbn = lblkno(fs, diroff);
860 doff = blkoff(fs, diroff);
861 dip = ino_read(parent);
862 blk = ino_blkatoff(dip, parent, lbn, &frags);
863 blksize = sblksize(fs, DIP(dip, di_size), lbn);
864 block = dblk_read(blk, blksize);
865 dp = (struct direct *)&block[doff];
866 if (dp->d_ino != child)
867 errx(1, "Inode %ju does not exist in %ju at %jd",
868 (uintmax_t)child, (uintmax_t)parent, diroff);
872 * The actual .. reference count will already have been removed
873 * from the parent by the .. remref record.
878 * Determines whether a pointer to an inode exists within a directory
879 * at a specified offset. Returns the mode of the found entry.
882 ino_isat(ino_t parent, off_t diroff, ino_t child, int *mode, int *isdot)
895 dip = ino_read(parent);
896 *mode = DIP(dip, di_mode);
897 if ((*mode & IFMT) != IFDIR) {
900 * This can happen if the parent inode
904 printf("Directory %ju has bad mode %o\n",
905 (uintmax_t)parent, *mode);
907 printf("Directory %ju has zero mode\n",
912 lbn = lblkno(fs, diroff);
913 doff = blkoff(fs, diroff);
914 blksize = sblksize(fs, DIP(dip, di_size), lbn);
915 if (diroff + DIRECTSIZ(1) > DIP(dip, di_size) || doff >= blksize) {
917 printf("ino %ju absent from %ju due to offset %jd"
918 " exceeding size %jd\n",
919 (uintmax_t)child, (uintmax_t)parent, diroff,
923 blk = ino_blkatoff(dip, parent, lbn, &frags);
926 printf("Sparse directory %ju", (uintmax_t)parent);
929 block = dblk_read(blk, blksize);
931 * Walk through the records from the start of the block to be
932 * certain we hit a valid record and not some junk in the middle
933 * of a file name. Stop when we reach or pass the expected offset.
935 dpoff = rounddown(doff, DIRBLKSIZ);
937 dp = (struct direct *)&block[dpoff];
940 if (dp->d_reclen == 0)
942 dpoff += dp->d_reclen;
943 } while (dpoff <= doff);
944 if (dpoff > fs->fs_bsize)
945 err_suj("Corrupt directory block in dir ino %ju\n",
950 printf("ino %ju not found in %ju, lbn %jd, dpoff %d\n",
951 (uintmax_t)child, (uintmax_t)parent, lbn, dpoff);
955 * We found the item in question. Record the mode and whether it's
956 * a . or .. link for the caller.
958 if (dp->d_ino == child) {
961 else if (dp->d_namlen == 2 &&
962 dp->d_name[0] == '.' && dp->d_name[1] == '.')
964 *mode = DTTOIF(dp->d_type);
968 printf("ino %ju doesn't match dirent ino %ju in parent %ju\n",
969 (uintmax_t)child, (uintmax_t)dp->d_ino, (uintmax_t)parent);
973 #define VISIT_INDIR 0x0001
974 #define VISIT_EXT 0x0002
975 #define VISIT_ROOT 0x0004 /* Operation came via root & valid pointers. */
978 * Read an indirect level which may or may not be linked into an inode.
981 indir_visit(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, uint64_t *frags,
982 ino_visitor visitor, int flags)
993 * Don't visit indirect blocks with contents we can't trust. This
994 * should only happen when indir_visit() is called to complete a
995 * truncate that never finished and not when a pointer is found via
1000 level = lbn_level(lbn);
1002 err_suj("Invalid level for lbn %jd\n", lbn);
1003 if ((flags & VISIT_ROOT) == 0 && blk_isindir(blk, ino, lbn) == 0) {
1005 printf("blk %jd ino %ju lbn %jd(%d) is not indir.\n",
1006 blk, (uintmax_t)ino, lbn, level);
1010 for (i = level; i > 0; i--)
1011 lbnadd *= NINDIR(fs);
1012 bap1 = (void *)dblk_read(blk, fs->fs_bsize);
1013 bap2 = (void *)bap1;
1014 for (i = 0; i < NINDIR(fs); i++) {
1015 if (fs->fs_magic == FS_UFS1_MAGIC)
1022 nlbn = -lbn + i * lbnadd;
1023 (*frags) += fs->fs_frag;
1024 visitor(ino, nlbn, nblk, fs->fs_frag);
1026 nlbn = (lbn + 1) - (i * lbnadd);
1027 indir_visit(ino, nlbn, nblk, frags, visitor, flags);
1031 if (flags & VISIT_INDIR) {
1032 (*frags) += fs->fs_frag;
1033 visitor(ino, lbn, blk, fs->fs_frag);
1038 * Visit each block in an inode as specified by 'flags' and call a
1039 * callback function. The callback may inspect or free blocks. The
1040 * count of frags found according to the size in the file is returned.
1041 * This is not valid for sparse files but may be used to determine
1042 * the correct di_blocks for a file.
1045 ino_visit(union dinode *ip, ino_t ino, ino_visitor visitor, int flags)
1056 size = DIP(ip, di_size);
1057 mode = DIP(ip, di_mode) & IFMT;
1059 if ((flags & VISIT_EXT) &&
1060 fs->fs_magic == FS_UFS2_MAGIC && ip->dp2.di_extsize) {
1061 for (i = 0; i < UFS_NXADDR; i++) {
1062 if (ip->dp2.di_extb[i] == 0)
1064 frags = sblksize(fs, ip->dp2.di_extsize, i);
1065 frags = numfrags(fs, frags);
1067 visitor(ino, -1 - i, ip->dp2.di_extb[i], frags);
1070 /* Skip datablocks for short links and devices. */
1071 if (mode == IFBLK || mode == IFCHR ||
1072 (mode == IFLNK && size < fs->fs_maxsymlinklen))
1074 for (i = 0; i < UFS_NDADDR; i++) {
1075 if (DIP(ip, di_db[i]) == 0)
1077 frags = sblksize(fs, size, i);
1078 frags = numfrags(fs, frags);
1080 visitor(ino, i, DIP(ip, di_db[i]), frags);
1083 * We know the following indirects are real as we're following
1084 * real pointers to them.
1086 flags |= VISIT_ROOT;
1087 for (i = 0, tmpval = NINDIR(fs), lbn = UFS_NDADDR; i < UFS_NIADDR; i++,
1089 nextlbn = lbn + tmpval;
1090 tmpval *= NINDIR(fs);
1091 if (DIP(ip, di_ib[i]) == 0)
1093 indir_visit(ino, -lbn - i, DIP(ip, di_ib[i]), &fragcnt, visitor,
1100 * Null visitor function used when we just want to count blocks and
1105 null_visit(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, int frags)
1112 * Recalculate di_blocks when we discover that a block allocation or
1113 * free was not successfully completed. The kernel does not roll this back
1114 * because it would be too expensive to compute which indirects were
1115 * reachable at the time the inode was written.
1118 ino_adjblks(struct suj_ino *sino)
1129 /* No need to adjust zero'd inodes. */
1130 if (DIP(ip, di_mode) == 0)
1133 * Visit all blocks and count them as well as recording the last
1134 * valid lbn in the file. If the file size doesn't agree with the
1135 * last lbn we need to truncate to fix it. Otherwise just adjust
1139 frags = ino_visit(ip, ino, null_visit, VISIT_INDIR | VISIT_EXT);
1140 blocks = fsbtodb(fs, frags);
1142 * We assume the size and direct block list is kept coherent by
1143 * softdep. For files that have extended into indirects we truncate
1144 * to the size in the inode or the maximum size permitted by
1145 * populated indirects.
1147 if (visitlbn >= UFS_NDADDR) {
1148 isize = DIP(ip, di_size);
1149 size = lblktosize(fs, visitlbn + 1);
1152 /* Always truncate to free any unpopulated indirects. */
1153 ino_trunc(sino->si_ino, isize);
1156 if (blocks == DIP(ip, di_blocks))
1159 printf("ino %ju adjusting block count from %jd to %jd\n",
1160 (uintmax_t)ino, DIP(ip, di_blocks), blocks);
1161 DIP_SET(ip, di_blocks, blocks);
1166 blk_free_visit(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, int frags)
1169 blk_free(blk, blk_freemask(blk, ino, lbn, frags), frags);
1173 * Free a block or tree of blocks that was previously rooted in ino at
1174 * the given lbn. If the lbn is an indirect all children are freed
1178 blk_free_lbn(ufs2_daddr_t blk, ino_t ino, ufs_lbn_t lbn, int frags, int follow)
1183 mask = blk_freemask(blk, ino, lbn, frags);
1185 if (lbn <= -UFS_NDADDR && follow && mask == 0)
1186 indir_visit(ino, lbn, blk, &resid, blk_free_visit, VISIT_INDIR);
1188 blk_free(blk, mask, frags);
1192 ino_setskip(struct suj_ino *sino, ino_t parent)
1197 if (ino_isat(sino->si_ino, DOTDOT_OFFSET, parent, &mode, &isdot))
1198 sino->si_skipparent = 1;
1202 ino_remref(ino_t parent, ino_t child, uint64_t diroff, int isdotdot)
1204 struct suj_ino *sino;
1205 struct suj_rec *srec;
1206 struct jrefrec *rrec;
1209 * Lookup this inode to see if we have a record for it.
1211 sino = ino_lookup(child, 0);
1213 * Tell any child directories we've already removed their
1214 * parent link cnt. Don't try to adjust our link down again.
1216 if (sino != NULL && isdotdot == 0)
1217 ino_setskip(sino, parent);
1219 * No valid record for this inode. Just drop the on-disk
1222 if (sino == NULL || sino->si_hasrecs == 0) {
1227 * Use ino_adjust() if ino_check() has already processed this
1228 * child. If we lose the last non-dot reference to a
1229 * directory it will be discarded.
1231 if (sino->si_linkadj) {
1234 sino->si_dotlinks--;
1239 * If we haven't yet processed this inode we need to make
1240 * sure we will successfully discover the lost path. If not
1241 * use nlinkadj to remember.
1243 TAILQ_FOREACH(srec, &sino->si_recs, sr_next) {
1244 rrec = (struct jrefrec *)srec->sr_rec;
1245 if (rrec->jr_parent == parent &&
1246 rrec->jr_diroff == diroff)
1249 sino->si_nlinkadj++;
1253 * Free the children of a directory when the directory is discarded.
1256 ino_free_children(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, int frags)
1258 struct suj_ino *sino;
1267 sino = ino_lookup(ino, 0);
1269 skipparent = sino->si_skipparent;
1272 size = lfragtosize(fs, frags);
1273 block = dblk_read(blk, size);
1274 dp = (struct direct *)&block[0];
1275 for (dpoff = 0; dpoff < size && dp->d_reclen; dpoff += dp->d_reclen) {
1276 dp = (struct direct *)&block[dpoff];
1277 if (dp->d_ino == 0 || dp->d_ino == UFS_WINO)
1279 if (dp->d_namlen == 1 && dp->d_name[0] == '.')
1281 isdotdot = dp->d_namlen == 2 && dp->d_name[0] == '.' &&
1282 dp->d_name[1] == '.';
1283 if (isdotdot && skipparent == 1)
1286 printf("Directory %ju removing ino %ju name %s\n",
1287 (uintmax_t)ino, (uintmax_t)dp->d_ino, dp->d_name);
1288 diroff = lblktosize(fs, lbn) + dpoff;
1289 ino_remref(ino, dp->d_ino, diroff, isdotdot);
1294 * Reclaim an inode, freeing all blocks and decrementing all children's
1295 * link counts. Free the inode back to the cg.
1298 ino_reclaim(union dinode *ip, ino_t ino, int mode)
1302 if (ino == UFS_ROOTINO)
1303 err_suj("Attempting to free UFS_ROOTINO\n");
1305 printf("Truncating and freeing ino %ju, nlink %d, mode %o\n",
1306 (uintmax_t)ino, DIP(ip, di_nlink), DIP(ip, di_mode));
1308 /* We are freeing an inode or directory. */
1309 if ((DIP(ip, di_mode) & IFMT) == IFDIR)
1310 ino_visit(ip, ino, ino_free_children, 0);
1311 DIP_SET(ip, di_nlink, 0);
1312 ino_visit(ip, ino, blk_free_visit, VISIT_EXT | VISIT_INDIR);
1313 /* Here we have to clear the inode and release any blocks it holds. */
1314 gen = DIP(ip, di_gen);
1315 if (fs->fs_magic == FS_UFS1_MAGIC)
1316 bzero(ip, sizeof(struct ufs1_dinode));
1318 bzero(ip, sizeof(struct ufs2_dinode));
1319 DIP_SET(ip, di_gen, gen);
1321 ino_free(ino, mode);
1326 * Adjust an inode's link count down by one when a directory goes away.
1337 nlink = DIP(ip, di_nlink);
1338 mode = DIP(ip, di_mode);
1340 err_suj("Inode %d link count %d invalid\n", ino, nlink);
1342 err_suj("Inode %d has a link of %d with 0 mode\n", ino, nlink);
1344 if ((mode & IFMT) == IFDIR)
1348 if (nlink < reqlink) {
1350 printf("ino %ju not enough links to live %d < %d\n",
1351 (uintmax_t)ino, nlink, reqlink);
1352 ino_reclaim(ip, ino, mode);
1355 DIP_SET(ip, di_nlink, nlink);
1360 * Adjust the inode link count to 'nlink'. If the count reaches zero
1364 ino_adjust(struct suj_ino *sino)
1366 struct jrefrec *rrec;
1367 struct suj_rec *srec;
1368 struct suj_ino *stmp;
1377 nlink = sino->si_nlink;
1379 mode = sino->si_mode & IFMT;
1381 * If it's a directory with no dot links, it was truncated before
1382 * the name was cleared. We need to clear the dirent that
1385 if (mode == IFDIR && nlink == 1 && sino->si_dotlinks == 0) {
1386 sino->si_nlink = nlink = 0;
1387 TAILQ_FOREACH(srec, &sino->si_recs, sr_next) {
1388 rrec = (struct jrefrec *)srec->sr_rec;
1389 if (ino_isat(rrec->jr_parent, rrec->jr_diroff, ino,
1390 &recmode, &isdot) == 0)
1392 ino_clrat(rrec->jr_parent, rrec->jr_diroff, ino);
1396 errx(1, "Directory %ju name not found", (uintmax_t)ino);
1399 * If it's a directory with no real names pointing to it go ahead
1400 * and truncate it. This will free any children.
1402 if (mode == IFDIR && nlink - sino->si_dotlinks == 0) {
1403 sino->si_nlink = nlink = 0;
1405 * Mark any .. links so they know not to free this inode
1406 * when they are removed.
1408 TAILQ_FOREACH(srec, &sino->si_recs, sr_next) {
1409 rrec = (struct jrefrec *)srec->sr_rec;
1410 if (rrec->jr_diroff == DOTDOT_OFFSET) {
1411 stmp = ino_lookup(rrec->jr_parent, 0);
1413 ino_setskip(stmp, ino);
1418 mode = DIP(ip, di_mode) & IFMT;
1419 if (nlink > UFS_LINK_MAX)
1420 err_suj("ino %ju nlink manipulation error, new %ju, old %d\n",
1421 (uintmax_t)ino, (uintmax_t)nlink, DIP(ip, di_nlink));
1423 printf("Adjusting ino %ju, nlink %ju, old link %d lastmode %o\n",
1424 (uintmax_t)ino, (uintmax_t)nlink, DIP(ip, di_nlink),
1428 printf("ino %ju, zero inode freeing bitmap\n",
1430 ino_free(ino, sino->si_mode);
1433 /* XXX Should be an assert? */
1434 if (mode != sino->si_mode && debug)
1435 printf("ino %ju, mode %o != %o\n",
1436 (uintmax_t)ino, mode, sino->si_mode);
1437 if ((mode & IFMT) == IFDIR)
1441 /* If the inode doesn't have enough links to live, free it. */
1442 if (nlink < reqlink) {
1444 printf("ino %ju not enough links to live %ju < %ju\n",
1445 (uintmax_t)ino, (uintmax_t)nlink,
1446 (uintmax_t)reqlink);
1447 ino_reclaim(ip, ino, mode);
1450 /* If required write the updated link count. */
1451 if (DIP(ip, di_nlink) == nlink) {
1453 printf("ino %ju, link matches, skipping.\n",
1457 DIP_SET(ip, di_nlink, nlink);
1462 * Truncate some or all blocks in an indirect, freeing any that are required
1463 * and zeroing the indirect.
1466 indir_trunc(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, ufs_lbn_t lastlbn)
1481 level = lbn_level(lbn);
1483 err_suj("Invalid level for lbn %jd\n", lbn);
1485 for (i = level; i > 0; i--)
1486 lbnadd *= NINDIR(fs);
1487 bap1 = (void *)dblk_read(blk, fs->fs_bsize);
1488 bap2 = (void *)bap1;
1489 for (i = 0; i < NINDIR(fs); i++) {
1490 if (fs->fs_magic == FS_UFS1_MAGIC)
1497 nlbn = (lbn + 1) - (i * lbnadd);
1499 * Calculate the lbn of the next indirect to
1500 * determine if any of this indirect must be
1503 next = -(lbn + level) + ((i+1) * lbnadd);
1504 if (next <= lastlbn)
1506 indir_trunc(ino, nlbn, nblk, lastlbn);
1507 /* If all of this indirect was reclaimed, free it. */
1508 nlbn = next - lbnadd;
1512 nlbn = -lbn + i * lbnadd;
1517 blk_free(nblk, 0, fs->fs_frag);
1518 if (fs->fs_magic == FS_UFS1_MAGIC)
1528 * Truncate an inode to the minimum of the given size or the last populated
1529 * block after any over size have been discarded. The kernel would allocate
1530 * the last block in the file but fsck does not and neither do we. This
1531 * code never extends files, only shrinks them.
1534 ino_trunc(ino_t ino, off_t size)
1538 uint64_t totalfrags;
1550 mode = DIP(ip, di_mode) & IFMT;
1551 cursize = DIP(ip, di_size);
1553 printf("Truncating ino %ju, mode %o to size %jd from size %jd\n",
1554 (uintmax_t)ino, mode, size, cursize);
1556 /* Skip datablocks for short links and devices. */
1557 if (mode == 0 || mode == IFBLK || mode == IFCHR ||
1558 (mode == IFLNK && cursize < fs->fs_maxsymlinklen))
1563 lastlbn = lblkno(fs, blkroundup(fs, size));
1564 for (i = lastlbn; i < UFS_NDADDR; i++) {
1565 if (DIP(ip, di_db[i]) == 0)
1567 frags = sblksize(fs, cursize, i);
1568 frags = numfrags(fs, frags);
1569 blk_free(DIP(ip, di_db[i]), 0, frags);
1570 DIP_SET(ip, di_db[i], 0);
1573 * Follow indirect blocks, freeing anything required.
1575 for (i = 0, tmpval = NINDIR(fs), lbn = UFS_NDADDR; i < UFS_NIADDR; i++,
1577 nextlbn = lbn + tmpval;
1578 tmpval *= NINDIR(fs);
1579 /* If we're not freeing any in this indirect range skip it. */
1580 if (lastlbn >= nextlbn)
1582 if (DIP(ip, di_ib[i]) == 0)
1584 indir_trunc(ino, -lbn - i, DIP(ip, di_ib[i]), lastlbn);
1585 /* If we freed everything in this indirect free the indir. */
1588 blk_free(DIP(ip, di_ib[i]), 0, fs->fs_frag);
1589 DIP_SET(ip, di_ib[i], 0);
1593 * Now that we've freed any whole blocks that exceed the desired
1594 * truncation size, figure out how many blocks remain and what the
1595 * last populated lbn is. We will set the size to this last lbn
1596 * rather than worrying about allocating the final lbn as the kernel
1597 * would've done. This is consistent with normal fsck behavior.
1600 totalfrags = ino_visit(ip, ino, null_visit, VISIT_INDIR | VISIT_EXT);
1601 if (size > lblktosize(fs, visitlbn + 1))
1602 size = lblktosize(fs, visitlbn + 1);
1604 * If we're truncating direct blocks we have to adjust frags
1607 if (visitlbn < UFS_NDADDR && totalfrags) {
1608 long oldspace, newspace;
1610 bn = DIP(ip, di_db[visitlbn]);
1612 err_suj("Bad blk at ino %ju lbn %jd\n",
1613 (uintmax_t)ino, visitlbn);
1614 oldspace = sblksize(fs, cursize, visitlbn);
1615 newspace = sblksize(fs, size, visitlbn);
1616 if (oldspace != newspace) {
1617 bn += numfrags(fs, newspace);
1618 frags = numfrags(fs, oldspace - newspace);
1619 blk_free(bn, 0, frags);
1620 totalfrags -= frags;
1623 DIP_SET(ip, di_blocks, fsbtodb(fs, totalfrags));
1624 DIP_SET(ip, di_size, size);
1626 * If we've truncated into the middle of a block or frag we have
1627 * to zero it here. Otherwise the file could extend into
1628 * uninitialized space later.
1630 off = blkoff(fs, size);
1631 if (off && DIP(ip, di_mode) != IFDIR) {
1635 bn = ino_blkatoff(ip, ino, visitlbn, &frags);
1637 err_suj("Block missing from ino %ju at lbn %jd\n",
1638 (uintmax_t)ino, visitlbn);
1639 clrsize = frags * fs->fs_fsize;
1640 buf = dblk_read(bn, clrsize);
1643 bzero(buf, clrsize);
1650 * Process records available for one inode and determine whether the
1651 * link count is correct or needs adjusting.
1654 ino_check(struct suj_ino *sino)
1656 struct suj_rec *srec;
1657 struct jrefrec *rrec;
1667 if (sino->si_hasrecs == 0)
1670 rrec = (struct jrefrec *)TAILQ_FIRST(&sino->si_recs)->sr_rec;
1671 nlink = rrec->jr_nlink;
1674 removes = sino->si_nlinkadj;
1675 TAILQ_FOREACH(srec, &sino->si_recs, sr_next) {
1676 rrec = (struct jrefrec *)srec->sr_rec;
1677 isat = ino_isat(rrec->jr_parent, rrec->jr_diroff,
1678 rrec->jr_ino, &mode, &isdot);
1679 if (isat && (mode & IFMT) != (rrec->jr_mode & IFMT))
1680 err_suj("Inode mode/directory type mismatch %o != %o\n",
1681 mode, rrec->jr_mode);
1683 printf("jrefrec: op %d ino %ju, nlink %ju, parent %ju, "
1684 "diroff %jd, mode %o, isat %d, isdot %d\n",
1685 rrec->jr_op, (uintmax_t)rrec->jr_ino,
1686 (uintmax_t)rrec->jr_nlink,
1687 (uintmax_t)rrec->jr_parent,
1688 (uintmax_t)rrec->jr_diroff,
1689 rrec->jr_mode, isat, isdot);
1690 mode = rrec->jr_mode & IFMT;
1691 if (rrec->jr_op == JOP_REMREF)
1698 * The number of links that remain are the starting link count
1699 * subtracted by the total number of removes with the total
1700 * links discovered back in. An incomplete remove thus
1701 * makes no change to the link count but an add increases
1706 "ino %ju nlink %ju newlinks %ju removes %ju dotlinks %ju\n",
1707 (uintmax_t)ino, (uintmax_t)nlink, (uintmax_t)newlinks,
1708 (uintmax_t)removes, (uintmax_t)dotlinks);
1711 sino->si_linkadj = 1;
1712 sino->si_nlink = nlink;
1713 sino->si_dotlinks = dotlinks;
1714 sino->si_mode = mode;
1719 * Process records available for one block and determine whether it is
1720 * still allocated and whether the owning inode needs to be updated or
1724 blk_check(struct suj_blk *sblk)
1726 struct suj_rec *srec;
1727 struct jblkrec *brec;
1728 struct suj_ino *sino;
1735 * Each suj_blk actually contains records for any fragments in that
1736 * block. As a result we must evaluate each record individually.
1739 TAILQ_FOREACH(srec, &sblk->sb_recs, sr_next) {
1740 brec = (struct jblkrec *)srec->sr_rec;
1741 frags = brec->jb_frags;
1742 blk = brec->jb_blkno + brec->jb_oldfrags;
1743 isat = blk_isat(brec->jb_ino, brec->jb_lbn, blk, &frags);
1744 if (sino == NULL || sino->si_ino != brec->jb_ino) {
1745 sino = ino_lookup(brec->jb_ino, 1);
1746 sino->si_blkadj = 1;
1749 printf("op %d blk %jd ino %ju lbn %jd frags %d isat %d (%d)\n",
1750 brec->jb_op, blk, (uintmax_t)brec->jb_ino,
1751 brec->jb_lbn, brec->jb_frags, isat, frags);
1753 * If we found the block at this address we still have to
1754 * determine if we need to free the tail end that was
1755 * added by adding contiguous fragments from the same block.
1758 if (frags == brec->jb_frags)
1760 mask = blk_freemask(blk, brec->jb_ino, brec->jb_lbn,
1764 frags = brec->jb_frags - frags;
1765 blk_free(blk, mask, frags);
1769 * The block wasn't found, attempt to free it. It won't be
1770 * freed if it was actually reallocated. If this was an
1771 * allocation we don't want to follow indirects as they
1772 * may not be written yet. Any children of the indirect will
1773 * have their own records. If it's a free we need to
1774 * recursively free children.
1776 blk_free_lbn(blk, brec->jb_ino, brec->jb_lbn, brec->jb_frags,
1777 brec->jb_op == JOP_FREEBLK);
1782 * Walk the list of inode records for this cg and resolve moved and duplicate
1783 * inode references now that we have a complete picture.
1786 cg_build(struct suj_cg *sc)
1788 struct suj_ino *sino;
1791 for (i = 0; i < SUJ_HASHSIZE; i++)
1792 LIST_FOREACH(sino, &sc->sc_inohash[i], si_next)
1797 * Handle inodes requiring truncation. This must be done prior to
1798 * looking up any inodes in directories.
1801 cg_trunc(struct suj_cg *sc)
1803 struct suj_ino *sino;
1806 for (i = 0; i < SUJ_HASHSIZE; i++) {
1807 LIST_FOREACH(sino, &sc->sc_inohash[i], si_next) {
1808 if (sino->si_trunc) {
1809 ino_trunc(sino->si_ino,
1810 sino->si_trunc->jt_size);
1811 sino->si_blkadj = 0;
1812 sino->si_trunc = NULL;
1814 if (sino->si_blkadj)
1821 cg_adj_blk(struct suj_cg *sc)
1823 struct suj_ino *sino;
1826 for (i = 0; i < SUJ_HASHSIZE; i++) {
1827 LIST_FOREACH(sino, &sc->sc_inohash[i], si_next) {
1828 if (sino->si_blkadj)
1835 * Free any partially allocated blocks and then resolve inode block
1839 cg_check_blk(struct suj_cg *sc)
1841 struct suj_blk *sblk;
1845 for (i = 0; i < SUJ_HASHSIZE; i++)
1846 LIST_FOREACH(sblk, &sc->sc_blkhash[i], sb_next)
1851 * Walk the list of inode records for this cg, recovering any
1852 * changes which were not complete at the time of crash.
1855 cg_check_ino(struct suj_cg *sc)
1857 struct suj_ino *sino;
1860 for (i = 0; i < SUJ_HASHSIZE; i++)
1861 LIST_FOREACH(sino, &sc->sc_inohash[i], si_next)
1866 * Write a potentially dirty cg. Recalculate the summary information and
1867 * update the superblock summary.
1870 cg_write(struct suj_cg *sc)
1872 ufs1_daddr_t fragno, cgbno, maxbno;
1878 if (sc->sc_dirty == 0)
1881 * Fix the frag and cluster summary.
1884 cgp->cg_cs.cs_nbfree = 0;
1885 cgp->cg_cs.cs_nffree = 0;
1886 bzero(&cgp->cg_frsum, sizeof(cgp->cg_frsum));
1887 maxbno = fragstoblks(fs, fs->fs_fpg);
1888 if (fs->fs_contigsumsize > 0) {
1889 for (i = 1; i <= fs->fs_contigsumsize; i++)
1890 cg_clustersum(cgp)[i] = 0;
1891 bzero(cg_clustersfree(cgp), howmany(maxbno, CHAR_BIT));
1893 blksfree = cg_blksfree(cgp);
1894 for (cgbno = 0; cgbno < maxbno; cgbno++) {
1895 if (ffs_isfreeblock(fs, blksfree, cgbno))
1897 if (ffs_isblock(fs, blksfree, cgbno)) {
1898 ffs_clusteracct(fs, cgp, cgbno, 1);
1899 cgp->cg_cs.cs_nbfree++;
1902 fragno = blkstofrags(fs, cgbno);
1903 blk = blkmap(fs, blksfree, fragno);
1904 ffs_fragacct(fs, blk, cgp->cg_frsum, 1);
1905 for (i = 0; i < fs->fs_frag; i++)
1906 if (isset(blksfree, fragno + i))
1907 cgp->cg_cs.cs_nffree++;
1910 * Update the superblock cg summary from our now correct values
1911 * before writing the block.
1913 fs->fs_cs(fs, sc->sc_cgx) = cgp->cg_cs;
1914 if (cgput(&disk, cgp) == -1)
1915 err_suj("Unable to write cylinder group %d\n", sc->sc_cgx);
1919 * Write out any modified inodes.
1922 cg_write_inos(struct suj_cg *sc)
1924 struct ino_blk *iblk;
1927 for (i = 0; i < SUJ_HASHSIZE; i++)
1928 LIST_FOREACH(iblk, &sc->sc_iblkhash[i], ib_next)
1934 cg_apply(void (*apply)(struct suj_cg *))
1939 for (i = 0; i < SUJ_HASHSIZE; i++)
1940 LIST_FOREACH(scg, &cghash[i], sc_next)
1945 * Process the unlinked but referenced file list. Freeing all inodes.
1955 ino = fs->fs_sujfree;
1959 mode = DIP(ip, di_mode) & IFMT;
1960 inon = DIP(ip, di_freelink);
1961 DIP_SET(ip, di_freelink, 0);
1963 * XXX Should this be an errx?
1965 if (DIP(ip, di_nlink) == 0) {
1967 printf("Freeing unlinked ino %ju mode %o\n",
1968 (uintmax_t)ino, mode);
1969 ino_reclaim(ip, ino, mode);
1971 printf("Skipping ino %ju mode %o with link %d\n",
1972 (uintmax_t)ino, mode, DIP(ip, di_nlink));
1978 * Append a new record to the list of records requiring processing.
1981 ino_append(union jrec *rec)
1983 struct jrefrec *refrec;
1984 struct jmvrec *mvrec;
1985 struct suj_ino *sino;
1986 struct suj_rec *srec;
1988 mvrec = &rec->rec_jmvrec;
1989 refrec = &rec->rec_jrefrec;
1990 if (debug && mvrec->jm_op == JOP_MVREF)
1991 printf("ino move: ino %ju, parent %ju, "
1992 "diroff %jd, oldoff %jd\n",
1993 (uintmax_t)mvrec->jm_ino, (uintmax_t)mvrec->jm_parent,
1994 (uintmax_t)mvrec->jm_newoff, (uintmax_t)mvrec->jm_oldoff);
1996 (refrec->jr_op == JOP_ADDREF || refrec->jr_op == JOP_REMREF))
1997 printf("ino ref: op %d, ino %ju, nlink %ju, "
1998 "parent %ju, diroff %jd\n",
1999 refrec->jr_op, (uintmax_t)refrec->jr_ino,
2000 (uintmax_t)refrec->jr_nlink,
2001 (uintmax_t)refrec->jr_parent, (uintmax_t)refrec->jr_diroff);
2002 sino = ino_lookup(((struct jrefrec *)rec)->jr_ino, 1);
2003 sino->si_hasrecs = 1;
2004 srec = errmalloc(sizeof(*srec));
2006 TAILQ_INSERT_TAIL(&sino->si_newrecs, srec, sr_next);
2010 * Add a reference adjustment to the sino list and eliminate dups. The
2011 * primary loop in ino_build_ref() checks for dups but new ones may be
2012 * created as a result of offset adjustments.
2015 ino_add_ref(struct suj_ino *sino, struct suj_rec *srec)
2017 struct jrefrec *refrec;
2018 struct suj_rec *srn;
2019 struct jrefrec *rrn;
2021 refrec = (struct jrefrec *)srec->sr_rec;
2023 * We walk backwards so that the oldest link count is preserved. If
2024 * an add record conflicts with a remove keep the remove. Redundant
2025 * removes are eliminated in ino_build_ref. Otherwise we keep the
2026 * oldest record at a given location.
2028 for (srn = TAILQ_LAST(&sino->si_recs, srechd); srn;
2029 srn = TAILQ_PREV(srn, srechd, sr_next)) {
2030 rrn = (struct jrefrec *)srn->sr_rec;
2031 if (rrn->jr_parent != refrec->jr_parent ||
2032 rrn->jr_diroff != refrec->jr_diroff)
2034 if (rrn->jr_op == JOP_REMREF || refrec->jr_op == JOP_ADDREF) {
2035 rrn->jr_mode = refrec->jr_mode;
2041 * Replace the record in place with the old nlink in case
2042 * we replace the head of the list. Abandon srec as a dup.
2044 refrec->jr_nlink = rrn->jr_nlink;
2045 srn->sr_rec = srec->sr_rec;
2048 TAILQ_INSERT_TAIL(&sino->si_recs, srec, sr_next);
2052 * Create a duplicate of a reference at a previous location.
2055 ino_dup_ref(struct suj_ino *sino, struct jrefrec *refrec, off_t diroff)
2057 struct jrefrec *rrn;
2058 struct suj_rec *srn;
2060 rrn = errmalloc(sizeof(*refrec));
2062 rrn->jr_op = JOP_ADDREF;
2063 rrn->jr_diroff = diroff;
2064 srn = errmalloc(sizeof(*srn));
2065 srn->sr_rec = (union jrec *)rrn;
2066 ino_add_ref(sino, srn);
2070 * Add a reference to the list at all known locations. We follow the offset
2071 * changes for a single instance and create duplicate add refs at each so
2072 * that we can tolerate any version of the directory block. Eliminate
2073 * removes which collide with adds that are seen in the journal. They should
2074 * not adjust the link count down.
2077 ino_build_ref(struct suj_ino *sino, struct suj_rec *srec)
2079 struct jrefrec *refrec;
2080 struct jmvrec *mvrec;
2081 struct suj_rec *srp;
2082 struct suj_rec *srn;
2083 struct jrefrec *rrn;
2086 refrec = (struct jrefrec *)srec->sr_rec;
2088 * Search for a mvrec that matches this offset. Whether it's an add
2089 * or a remove we can delete the mvref after creating a dup record in
2092 if (!TAILQ_EMPTY(&sino->si_movs)) {
2093 diroff = refrec->jr_diroff;
2094 for (srn = TAILQ_LAST(&sino->si_movs, srechd); srn; srn = srp) {
2095 srp = TAILQ_PREV(srn, srechd, sr_next);
2096 mvrec = (struct jmvrec *)srn->sr_rec;
2097 if (mvrec->jm_parent != refrec->jr_parent ||
2098 mvrec->jm_newoff != diroff)
2100 diroff = mvrec->jm_oldoff;
2101 TAILQ_REMOVE(&sino->si_movs, srn, sr_next);
2103 ino_dup_ref(sino, refrec, diroff);
2107 * If a remove wasn't eliminated by an earlier add just append it to
2110 if (refrec->jr_op == JOP_REMREF) {
2111 ino_add_ref(sino, srec);
2115 * Walk the list of records waiting to be added to the list. We
2116 * must check for moves that apply to our current offset and remove
2117 * them from the list. Remove any duplicates to eliminate removes
2118 * with corresponding adds.
2120 TAILQ_FOREACH_SAFE(srn, &sino->si_newrecs, sr_next, srp) {
2121 switch (srn->sr_rec->rec_jrefrec.jr_op) {
2124 * This should actually be an error we should
2125 * have a remove for every add journaled.
2127 rrn = (struct jrefrec *)srn->sr_rec;
2128 if (rrn->jr_parent != refrec->jr_parent ||
2129 rrn->jr_diroff != refrec->jr_diroff)
2131 TAILQ_REMOVE(&sino->si_newrecs, srn, sr_next);
2135 * Once we remove the current iteration of the
2136 * record at this address we're done.
2138 rrn = (struct jrefrec *)srn->sr_rec;
2139 if (rrn->jr_parent != refrec->jr_parent ||
2140 rrn->jr_diroff != refrec->jr_diroff)
2142 TAILQ_REMOVE(&sino->si_newrecs, srn, sr_next);
2143 ino_add_ref(sino, srec);
2147 * Update our diroff based on any moves that match
2148 * and remove the move.
2150 mvrec = (struct jmvrec *)srn->sr_rec;
2151 if (mvrec->jm_parent != refrec->jr_parent ||
2152 mvrec->jm_oldoff != refrec->jr_diroff)
2154 ino_dup_ref(sino, refrec, mvrec->jm_oldoff);
2155 refrec->jr_diroff = mvrec->jm_newoff;
2156 TAILQ_REMOVE(&sino->si_newrecs, srn, sr_next);
2159 err_suj("ino_build_ref: Unknown op %d\n",
2160 srn->sr_rec->rec_jrefrec.jr_op);
2163 ino_add_ref(sino, srec);
2167 * Walk the list of new records and add them in-order resolving any
2168 * dups and adjusted offsets.
2171 ino_build(struct suj_ino *sino)
2173 struct suj_rec *srec;
2175 while ((srec = TAILQ_FIRST(&sino->si_newrecs)) != NULL) {
2176 TAILQ_REMOVE(&sino->si_newrecs, srec, sr_next);
2177 switch (srec->sr_rec->rec_jrefrec.jr_op) {
2180 ino_build_ref(sino, srec);
2184 * Add this mvrec to the queue of pending mvs.
2186 TAILQ_INSERT_TAIL(&sino->si_movs, srec, sr_next);
2189 err_suj("ino_build: Unknown op %d\n",
2190 srec->sr_rec->rec_jrefrec.jr_op);
2193 if (TAILQ_EMPTY(&sino->si_recs))
2194 sino->si_hasrecs = 0;
2198 * Modify journal records so they refer to the base block number
2199 * and a start and end frag range. This is to facilitate the discovery
2200 * of overlapping fragment allocations.
2203 blk_build(struct jblkrec *blkrec)
2205 struct suj_rec *srec;
2206 struct suj_blk *sblk;
2207 struct jblkrec *blkrn;
2212 printf("blk_build: op %d blkno %jd frags %d oldfrags %d "
2213 "ino %ju lbn %jd\n",
2214 blkrec->jb_op, (uintmax_t)blkrec->jb_blkno,
2215 blkrec->jb_frags, blkrec->jb_oldfrags,
2216 (uintmax_t)blkrec->jb_ino, (uintmax_t)blkrec->jb_lbn);
2218 blk = blknum(fs, blkrec->jb_blkno);
2219 frag = fragnum(fs, blkrec->jb_blkno);
2220 sblk = blk_lookup(blk, 1);
2222 * Rewrite the record using oldfrags to indicate the offset into
2223 * the block. Leave jb_frags as the actual allocated count.
2225 blkrec->jb_blkno -= frag;
2226 blkrec->jb_oldfrags = frag;
2227 if (blkrec->jb_oldfrags + blkrec->jb_frags > fs->fs_frag)
2228 err_suj("Invalid fragment count %d oldfrags %d\n",
2229 blkrec->jb_frags, frag);
2231 * Detect dups. If we detect a dup we always discard the oldest
2232 * record as it is superseded by the new record. This speeds up
2233 * later stages but also eliminates free records which are used
2234 * to indicate that the contents of indirects can be trusted.
2236 TAILQ_FOREACH(srec, &sblk->sb_recs, sr_next) {
2237 blkrn = (struct jblkrec *)srec->sr_rec;
2238 if (blkrn->jb_ino != blkrec->jb_ino ||
2239 blkrn->jb_lbn != blkrec->jb_lbn ||
2240 blkrn->jb_blkno != blkrec->jb_blkno ||
2241 blkrn->jb_frags != blkrec->jb_frags ||
2242 blkrn->jb_oldfrags != blkrec->jb_oldfrags)
2245 printf("Removed dup.\n");
2246 /* Discard the free which is a dup with an alloc. */
2247 if (blkrec->jb_op == JOP_FREEBLK)
2249 TAILQ_REMOVE(&sblk->sb_recs, srec, sr_next);
2253 srec = errmalloc(sizeof(*srec));
2254 srec->sr_rec = (union jrec *)blkrec;
2255 TAILQ_INSERT_TAIL(&sblk->sb_recs, srec, sr_next);
2259 ino_build_trunc(struct jtrncrec *rec)
2261 struct suj_ino *sino;
2264 printf("ino_build_trunc: op %d ino %ju, size %jd\n",
2265 rec->jt_op, (uintmax_t)rec->jt_ino,
2266 (uintmax_t)rec->jt_size);
2267 sino = ino_lookup(rec->jt_ino, 1);
2268 if (rec->jt_op == JOP_SYNC) {
2269 sino->si_trunc = NULL;
2272 if (sino->si_trunc == NULL || sino->si_trunc->jt_size > rec->jt_size)
2273 sino->si_trunc = rec;
2277 * Build up tables of the operations we need to recover.
2282 struct suj_seg *seg;
2287 TAILQ_FOREACH(seg, &allsegs, ss_next) {
2289 printf("seg %jd has %d records, oldseq %jd.\n",
2290 seg->ss_rec.jsr_seq, seg->ss_rec.jsr_cnt,
2291 seg->ss_rec.jsr_oldest);
2293 rec = (union jrec *)seg->ss_blk;
2294 for (i = 0; i < seg->ss_rec.jsr_cnt; off += JREC_SIZE, rec++) {
2295 /* skip the segrec. */
2296 if ((off % real_dev_bsize) == 0)
2298 switch (rec->rec_jrefrec.jr_op) {
2306 blk_build((struct jblkrec *)rec);
2310 ino_build_trunc((struct jtrncrec *)rec);
2313 err_suj("Unknown journal operation %d (%d)\n",
2314 rec->rec_jrefrec.jr_op, off);
2322 * Prune the journal segments to those we care about based on the
2323 * oldest sequence in the newest segment. Order the segment list
2324 * based on sequence number.
2329 struct suj_seg *seg;
2330 struct suj_seg *segn;
2335 printf("Pruning up to %jd\n", oldseq);
2336 /* First free the expired segments. */
2337 TAILQ_FOREACH_SAFE(seg, &allsegs, ss_next, segn) {
2338 if (seg->ss_rec.jsr_seq >= oldseq)
2340 TAILQ_REMOVE(&allsegs, seg, ss_next);
2344 /* Next ensure that segments are ordered properly. */
2345 seg = TAILQ_FIRST(&allsegs);
2348 printf("Empty journal\n");
2351 newseq = seg->ss_rec.jsr_seq;
2353 seg = TAILQ_LAST(&allsegs, seghd);
2354 if (seg->ss_rec.jsr_seq >= newseq)
2356 TAILQ_REMOVE(&allsegs, seg, ss_next);
2357 TAILQ_INSERT_HEAD(&allsegs, seg, ss_next);
2358 newseq = seg->ss_rec.jsr_seq;
2361 if (newseq != oldseq) {
2362 TAILQ_FOREACH(seg, &allsegs, ss_next) {
2363 printf("%jd, ", seg->ss_rec.jsr_seq);
2366 err_suj("Journal file sequence mismatch %jd != %jd\n",
2370 * The kernel may asynchronously write segments which can create
2371 * gaps in the sequence space. Throw away any segments after the
2372 * gap as the kernel guarantees only those that are contiguously
2373 * reachable are marked as completed.
2376 TAILQ_FOREACH_SAFE(seg, &allsegs, ss_next, segn) {
2377 if (!discard && newseq++ == seg->ss_rec.jsr_seq) {
2378 jrecs += seg->ss_rec.jsr_cnt;
2379 jbytes += seg->ss_rec.jsr_blocks * real_dev_bsize;
2384 printf("Journal order mismatch %jd != %jd pruning\n",
2385 newseq-1, seg->ss_rec.jsr_seq);
2386 TAILQ_REMOVE(&allsegs, seg, ss_next);
2391 printf("Processing journal segments from %jd to %jd\n",
2396 * Verify the journal inode before attempting to read records.
2399 suj_verifyino(union dinode *ip)
2402 if (DIP(ip, di_nlink) != 1) {
2403 printf("Invalid link count %d for journal inode %ju\n",
2404 DIP(ip, di_nlink), (uintmax_t)sujino);
2408 if ((DIP(ip, di_flags) & (SF_IMMUTABLE | SF_NOUNLINK)) !=
2409 (SF_IMMUTABLE | SF_NOUNLINK)) {
2410 printf("Invalid flags 0x%X for journal inode %ju\n",
2411 DIP(ip, di_flags), (uintmax_t)sujino);
2415 if (DIP(ip, di_mode) != (IFREG | IREAD)) {
2416 printf("Invalid mode %o for journal inode %ju\n",
2417 DIP(ip, di_mode), (uintmax_t)sujino);
2421 if (DIP(ip, di_size) < SUJ_MIN) {
2422 printf("Invalid size %jd for journal inode %ju\n",
2423 DIP(ip, di_size), (uintmax_t)sujino);
2427 if (DIP(ip, di_modrev) != fs->fs_mtime) {
2428 printf("Journal timestamp does not match fs mount time\n");
2436 struct jextent *jb_extent; /* Extent array. */
2437 int jb_avail; /* Available extents. */
2438 int jb_used; /* Last used extent. */
2439 int jb_head; /* Allocator head. */
2440 int jb_off; /* Allocator extent offset. */
2443 ufs2_daddr_t je_daddr; /* Disk block address. */
2444 int je_blocks; /* Disk block count. */
2447 static struct jblocks *suj_jblocks;
2449 static struct jblocks *
2450 jblocks_create(void)
2452 struct jblocks *jblocks;
2455 jblocks = errmalloc(sizeof(*jblocks));
2456 jblocks->jb_avail = 10;
2457 jblocks->jb_used = 0;
2458 jblocks->jb_head = 0;
2459 jblocks->jb_off = 0;
2460 size = sizeof(struct jextent) * jblocks->jb_avail;
2461 jblocks->jb_extent = errmalloc(size);
2462 bzero(jblocks->jb_extent, size);
2468 * Return the next available disk block and the amount of contiguous
2469 * free space it contains.
2472 jblocks_next(struct jblocks *jblocks, int bytes, int *actual)
2474 struct jextent *jext;
2479 blocks = bytes / disk.d_bsize;
2480 jext = &jblocks->jb_extent[jblocks->jb_head];
2481 freecnt = jext->je_blocks - jblocks->jb_off;
2483 jblocks->jb_off = 0;
2484 if (++jblocks->jb_head > jblocks->jb_used)
2486 jext = &jblocks->jb_extent[jblocks->jb_head];
2487 freecnt = jext->je_blocks;
2489 if (freecnt > blocks)
2491 *actual = freecnt * disk.d_bsize;
2492 daddr = jext->je_daddr + jblocks->jb_off;
2498 * Advance the allocation head by a specified number of bytes, consuming
2499 * one journal segment.
2502 jblocks_advance(struct jblocks *jblocks, int bytes)
2505 jblocks->jb_off += bytes / disk.d_bsize;
2509 jblocks_destroy(struct jblocks *jblocks)
2512 free(jblocks->jb_extent);
2517 jblocks_add(struct jblocks *jblocks, ufs2_daddr_t daddr, int blocks)
2519 struct jextent *jext;
2522 jext = &jblocks->jb_extent[jblocks->jb_used];
2523 /* Adding the first block. */
2524 if (jext->je_daddr == 0) {
2525 jext->je_daddr = daddr;
2526 jext->je_blocks = blocks;
2529 /* Extending the last extent. */
2530 if (jext->je_daddr + jext->je_blocks == daddr) {
2531 jext->je_blocks += blocks;
2534 /* Adding a new extent. */
2535 if (++jblocks->jb_used == jblocks->jb_avail) {
2536 jblocks->jb_avail *= 2;
2537 size = sizeof(struct jextent) * jblocks->jb_avail;
2538 jext = errmalloc(size);
2540 bcopy(jblocks->jb_extent, jext,
2541 sizeof(struct jextent) * jblocks->jb_used);
2542 free(jblocks->jb_extent);
2543 jblocks->jb_extent = jext;
2545 jext = &jblocks->jb_extent[jblocks->jb_used];
2546 jext->je_daddr = daddr;
2547 jext->je_blocks = blocks;
2553 * Add a file block from the journal to the extent map. We can't read
2554 * each file block individually because the kernel treats it as a circular
2555 * buffer and segments may span mutliple contiguous blocks.
2558 suj_add_block(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, int frags)
2561 jblocks_add(suj_jblocks, fsbtodb(fs, blk), fsbtodb(fs, frags));
2567 uint8_t block[1 * 1024 * 1024];
2568 struct suj_seg *seg;
2569 struct jsegrec *recn;
2570 struct jsegrec *rec;
2579 * Read records until we exhaust the journal space. If we find
2580 * an invalid record we start searching for a valid segment header
2581 * at the next block. This is because we don't have a head/tail
2582 * pointer and must recover the information indirectly. At the gap
2583 * between the head and tail we won't necessarily have a valid
2588 size = sizeof(block);
2589 blk = jblocks_next(suj_jblocks, size, &readsize);
2594 * Read 1MB at a time and scan for records within this block.
2596 if (bread(&disk, blk, &block, size) == -1) {
2597 err_suj("Error reading journal block %jd\n",
2600 for (rec = (void *)block; size; size -= recsize,
2601 rec = (struct jsegrec *)((uintptr_t)rec + recsize)) {
2602 recsize = real_dev_bsize;
2603 if (rec->jsr_time != fs->fs_mtime) {
2605 printf("Rec time %jd != fs mtime %jd\n",
2606 rec->jsr_time, fs->fs_mtime);
2607 jblocks_advance(suj_jblocks, recsize);
2610 if (rec->jsr_cnt == 0) {
2612 printf("Found illegal count %d\n",
2614 jblocks_advance(suj_jblocks, recsize);
2617 blocks = rec->jsr_blocks;
2618 recsize = blocks * real_dev_bsize;
2619 if (recsize > size) {
2621 * We may just have run out of buffer, restart
2622 * the loop to re-read from this spot.
2624 if (size < fs->fs_bsize &&
2626 recsize <= fs->fs_bsize)
2629 printf("Found invalid segsize %d > %d\n",
2631 recsize = real_dev_bsize;
2632 jblocks_advance(suj_jblocks, recsize);
2636 * Verify that all blocks in the segment are present.
2638 for (i = 1; i < blocks; i++) {
2639 recn = (void *)((uintptr_t)rec) + i *
2641 if (recn->jsr_seq == rec->jsr_seq &&
2642 recn->jsr_time == rec->jsr_time)
2645 printf("Incomplete record %jd (%d)\n",
2647 recsize = i * real_dev_bsize;
2648 jblocks_advance(suj_jblocks, recsize);
2651 seg = errmalloc(sizeof(*seg));
2652 seg->ss_blk = errmalloc(recsize);
2654 bcopy((void *)rec, seg->ss_blk, recsize);
2655 if (rec->jsr_oldest > oldseq)
2656 oldseq = rec->jsr_oldest;
2657 TAILQ_INSERT_TAIL(&allsegs, seg, ss_next);
2658 jblocks_advance(suj_jblocks, recsize);
2664 * Search a directory block for the SUJ_FILE.
2667 suj_find(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, int frags)
2669 char block[MAXBSIZE];
2676 bytes = lfragtosize(fs, frags);
2677 if (bread(&disk, fsbtodb(fs, blk), block, bytes) <= 0)
2678 err_suj("Failed to read UFS_ROOTINO directory block %jd\n",
2680 for (off = 0; off < bytes; off += dp->d_reclen) {
2681 dp = (struct direct *)&block[off];
2682 if (dp->d_reclen == 0)
2686 if (dp->d_namlen != strlen(SUJ_FILE))
2688 if (bcmp(dp->d_name, SUJ_FILE, dp->d_namlen) != 0)
2696 * Orchestrate the verification of a filesystem via the softupdates journal.
2699 suj_check(const char *filesys)
2705 struct suj_seg *seg;
2706 struct suj_seg *segn;
2710 if (real_dev_bsize == 0 && ioctl(disk.d_fd, DIOCGSECTORSIZE,
2711 &real_dev_bsize) == -1)
2712 real_dev_bsize = secsize;
2714 printf("dev_bsize %u\n", real_dev_bsize);
2717 * Set an exit point when SUJ check failed
2719 retval = setjmp(jmpbuf);
2721 pwarn("UNEXPECTED SU+J INCONSISTENCY\n");
2722 TAILQ_FOREACH_SAFE(seg, &allsegs, ss_next, segn) {
2723 TAILQ_REMOVE(&allsegs, seg, ss_next);
2727 if (reply("FALLBACK TO FULL FSCK") == 0) {
2735 * Find the journal inode.
2737 ip = ino_read(UFS_ROOTINO);
2739 ino_visit(ip, UFS_ROOTINO, suj_find, 0);
2741 printf("Journal inode removed. Use tunefs to re-create.\n");
2742 sblock.fs_flags &= ~FS_SUJ;
2743 sblock.fs_sujfree = 0;
2747 * Fetch the journal inode and verify it.
2749 jip = ino_read(sujino);
2750 printf("** SU+J Recovering %s\n", filesys);
2751 if (suj_verifyino(jip) != 0)
2753 if (!preen && !reply("USE JOURNAL"))
2756 * Build a list of journal blocks in jblocks before parsing the
2757 * available journal blocks in with suj_read().
2759 printf("** Reading %jd byte journal from inode %ju.\n",
2760 DIP(jip, di_size), (uintmax_t)sujino);
2761 suj_jblocks = jblocks_create();
2762 blocks = ino_visit(jip, sujino, suj_add_block, 0);
2763 if (blocks != numfrags(fs, DIP(jip, di_size))) {
2764 printf("Sparse journal inode %ju.\n", (uintmax_t)sujino);
2768 jblocks_destroy(suj_jblocks);
2770 if (preen || reply("RECOVER")) {
2771 printf("** Building recovery table.\n");
2775 printf("** Resolving unreferenced inode list.\n");
2777 printf("** Processing journal entries.\n");
2779 cg_apply(cg_check_blk);
2780 cg_apply(cg_adj_blk);
2781 cg_apply(cg_check_ino);
2783 if (preen == 0 && (jrecs > 0 || jbytes > 0) && reply("WRITE CHANGES") == 0)
2786 * To remain idempotent with partial truncations the free bitmaps
2787 * must be written followed by indirect blocks and lastly inode
2788 * blocks. This preserves access to the modified pointers until
2793 cg_apply(cg_write_inos);
2794 /* Write back superblock. */
2796 if (jrecs > 0 || jbytes > 0) {
2797 printf("** %jd journal records in %jd bytes for %.2f%% utilization\n",
2798 jrecs, jbytes, ((float)jrecs / (float)(jbytes / JREC_SIZE)) * 100);
2799 printf("** Freed %jd inodes (%jd dirs) %jd blocks, and %jd frags.\n",
2800 freeinos, freedir, freeblocks, freefrags);
2811 for (i = 0; i < SUJ_HASHSIZE; i++) {
2812 LIST_INIT(&cghash[i]);
2813 LIST_INIT(&dbhash[i]);
2817 TAILQ_INIT(&allsegs);