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;
114 LIST_HEAD(iblkhd, ino_blk);
117 LIST_ENTRY(suj_cg) sc_next;
118 struct blkhd sc_blkhash[SUJ_HASHSIZE];
119 struct inohd sc_inohash[SUJ_HASHSIZE];
120 struct iblkhd sc_iblkhash[SUJ_HASHSIZE];
121 struct ino_blk *sc_lastiblk;
122 struct suj_ino *sc_lastino;
123 struct suj_blk *sc_lastblk;
130 static LIST_HEAD(cghd, suj_cg) cghash[SUJ_HASHSIZE];
131 static LIST_HEAD(dblkhd, data_blk) dbhash[SUJ_HASHSIZE];
132 static struct suj_cg *lastcg;
133 static struct data_blk *lastblk;
135 static TAILQ_HEAD(seghd, suj_seg) allsegs;
136 static uint64_t oldseq;
137 static struct fs *fs = NULL;
141 * Summary statistics.
143 static uint64_t freefrags;
144 static uint64_t freeblocks;
145 static uint64_t freeinos;
146 static uint64_t freedir;
147 static uint64_t jbytes;
148 static uint64_t jrecs;
150 static jmp_buf jmpbuf;
152 typedef void (*ino_visitor)(ino_t, ufs_lbn_t, ufs2_daddr_t, int);
153 static void err_suj(const char *, ...) __dead2;
154 static void ino_trunc(ino_t, off_t);
155 static void ino_decr(ino_t);
156 static void ino_adjust(struct suj_ino *);
157 static void ino_build(struct suj_ino *);
158 static int blk_isfree(ufs2_daddr_t);
159 static void initsuj(void);
168 err(EX_OSERR, "malloc(%zu)", n);
173 * When hit a fatal error in journalling check, print out
174 * the error and then offer to fallback to normal fsck.
177 err_suj(const char * restrict fmt, ...)
182 (void)fprintf(stdout, "%s: ", cdevname);
185 (void)vfprintf(stdout, fmt, ap);
192 * Mark file system as clean, write the super-block back, close the disk.
195 closedisk(const char *devnam)
201 * Recompute the fs summary info from correct cs summaries.
203 bzero(&fs->fs_cstotal, sizeof(struct csum_total));
204 for (i = 0; i < fs->fs_ncg; i++) {
205 cgsum = &fs->fs_cs(fs, i);
206 fs->fs_cstotal.cs_nffree += cgsum->cs_nffree;
207 fs->fs_cstotal.cs_nbfree += cgsum->cs_nbfree;
208 fs->fs_cstotal.cs_nifree += cgsum->cs_nifree;
209 fs->fs_cstotal.cs_ndir += cgsum->cs_ndir;
211 fs->fs_pendinginodes = 0;
212 fs->fs_pendingblocks = 0;
214 fs->fs_time = time(NULL);
215 fs->fs_mtime = time(NULL);
216 if (sbput(disk.d_fd, fs, 0) == -1)
217 err(EX_OSERR, "sbput(%s)", devnam);
218 if (ufs_disk_close(&disk) == -1)
219 err(EX_OSERR, "ufs_disk_close(%s)", devnam);
224 * Lookup a cg by number in the hash so we can keep track of which cgs
225 * need stats rebuilt.
227 static struct suj_cg *
233 if (cgx < 0 || cgx >= fs->fs_ncg)
234 err_suj("Bad cg number %d\n", cgx);
235 if (lastcg && lastcg->sc_cgx == cgx)
237 hd = &cghash[SUJ_HASH(cgx)];
238 LIST_FOREACH(sc, hd, sc_next)
239 if (sc->sc_cgx == cgx) {
243 sc = errmalloc(sizeof(*sc));
244 bzero(sc, sizeof(*sc));
245 sc->sc_cgbuf = errmalloc(fs->fs_bsize);
246 sc->sc_cgp = (struct cg *)sc->sc_cgbuf;
248 LIST_INSERT_HEAD(hd, sc, sc_next);
250 * Use bread() here rather than cgget() because the cylinder group
251 * may be corrupted but we want it anyway so we can fix it.
253 if (bread(&disk, fsbtodb(fs, cgtod(fs, sc->sc_cgx)), sc->sc_cgbuf,
255 err_suj("Unable to read cylinder group %d\n", sc->sc_cgx);
261 * Lookup an inode number in the hash and allocate a suj_ino if it does
264 static struct suj_ino *
265 ino_lookup(ino_t ino, int creat)
267 struct suj_ino *sino;
271 sc = cg_lookup(ino_to_cg(fs, ino));
272 if (sc->sc_lastino && sc->sc_lastino->si_ino == ino)
273 return (sc->sc_lastino);
274 hd = &sc->sc_inohash[SUJ_HASH(ino)];
275 LIST_FOREACH(sino, hd, si_next)
276 if (sino->si_ino == ino)
280 sino = errmalloc(sizeof(*sino));
281 bzero(sino, sizeof(*sino));
283 TAILQ_INIT(&sino->si_recs);
284 TAILQ_INIT(&sino->si_newrecs);
285 TAILQ_INIT(&sino->si_movs);
286 LIST_INSERT_HEAD(hd, sino, si_next);
292 * Lookup a block number in the hash and allocate a suj_blk if it does
295 static struct suj_blk *
296 blk_lookup(ufs2_daddr_t blk, int creat)
298 struct suj_blk *sblk;
302 sc = cg_lookup(dtog(fs, blk));
303 if (sc->sc_lastblk && sc->sc_lastblk->sb_blk == blk)
304 return (sc->sc_lastblk);
305 hd = &sc->sc_blkhash[SUJ_HASH(fragstoblks(fs, blk))];
306 LIST_FOREACH(sblk, hd, sb_next)
307 if (sblk->sb_blk == blk)
311 sblk = errmalloc(sizeof(*sblk));
312 bzero(sblk, sizeof(*sblk));
314 TAILQ_INIT(&sblk->sb_recs);
315 LIST_INSERT_HEAD(hd, sblk, sb_next);
320 static struct data_blk *
321 dblk_lookup(ufs2_daddr_t blk)
323 struct data_blk *dblk;
326 hd = &dbhash[SUJ_HASH(fragstoblks(fs, blk))];
327 if (lastblk && lastblk->db_blk == blk)
329 LIST_FOREACH(dblk, hd, db_next)
330 if (dblk->db_blk == blk)
333 * The inode block wasn't located, allocate a new one.
335 dblk = errmalloc(sizeof(*dblk));
336 bzero(dblk, sizeof(*dblk));
337 LIST_INSERT_HEAD(hd, dblk, db_next);
343 dblk_read(ufs2_daddr_t blk, int size)
345 struct data_blk *dblk;
347 dblk = dblk_lookup(blk);
349 * I doubt size mismatches can happen in practice but it is trivial
352 if (size != dblk->db_size) {
355 dblk->db_buf = errmalloc(size);
356 dblk->db_size = size;
357 if (bread(&disk, fsbtodb(fs, blk), dblk->db_buf, size) == -1)
358 err_suj("Failed to read data block %jd\n", blk);
360 return (dblk->db_buf);
364 dblk_dirty(ufs2_daddr_t blk)
366 struct data_blk *dblk;
368 dblk = dblk_lookup(blk);
375 struct data_blk *dblk;
378 for (i = 0; i < SUJ_HASHSIZE; i++) {
379 LIST_FOREACH(dblk, &dbhash[i], db_next) {
380 if (dblk->db_dirty == 0 || dblk->db_size == 0)
382 if (bwrite(&disk, fsbtodb(fs, dblk->db_blk),
383 dblk->db_buf, dblk->db_size) == -1)
384 err_suj("Unable to write block %jd\n",
390 static union dinode *
393 struct ino_blk *iblk;
399 blk = ino_to_fsba(fs, ino);
400 sc = cg_lookup(ino_to_cg(fs, ino));
401 iblk = sc->sc_lastiblk;
402 if (iblk && iblk->ib_blk == blk)
404 hd = &sc->sc_iblkhash[SUJ_HASH(fragstoblks(fs, blk))];
405 LIST_FOREACH(iblk, hd, ib_next)
406 if (iblk->ib_blk == blk)
409 * The inode block wasn't located, allocate a new one.
411 iblk = errmalloc(sizeof(*iblk));
412 bzero(iblk, sizeof(*iblk));
413 iblk->ib_buf = errmalloc(fs->fs_bsize);
415 LIST_INSERT_HEAD(hd, iblk, ib_next);
416 if (bread(&disk, fsbtodb(fs, blk), iblk->ib_buf, fs->fs_bsize) == -1)
417 err_suj("Failed to read inode block %jd\n", blk);
419 sc->sc_lastiblk = iblk;
420 off = ino_to_fsbo(fs, ino);
421 if (fs->fs_magic == FS_UFS1_MAGIC)
422 return (union dinode *)&((struct ufs1_dinode *)iblk->ib_buf)[off];
424 return (union dinode *)&((struct ufs2_dinode *)iblk->ib_buf)[off];
430 struct ino_blk *iblk;
435 blk = ino_to_fsba(fs, ino);
436 sc = cg_lookup(ino_to_cg(fs, ino));
437 iblk = sc->sc_lastiblk;
438 if (iblk && iblk->ib_blk == blk) {
442 hd = &sc->sc_iblkhash[SUJ_HASH(fragstoblks(fs, blk))];
443 LIST_FOREACH(iblk, hd, ib_next) {
444 if (iblk->ib_blk == blk) {
454 iblk_write(struct ino_blk *iblk)
457 if (iblk->ib_dirty == 0)
459 if (bwrite(&disk, fsbtodb(fs, iblk->ib_blk), iblk->ib_buf,
461 err_suj("Failed to write inode block %jd\n", iblk->ib_blk);
465 blk_overlaps(struct jblkrec *brec, ufs2_daddr_t start, int frags)
472 bstart = brec->jb_blkno + brec->jb_oldfrags;
473 bend = bstart + brec->jb_frags;
474 if (start < bend && end > bstart)
480 blk_equals(struct jblkrec *brec, ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t start,
484 if (brec->jb_ino != ino || brec->jb_lbn != lbn)
486 if (brec->jb_blkno + brec->jb_oldfrags != start)
488 if (brec->jb_frags < frags)
494 blk_setmask(struct jblkrec *brec, int *mask)
498 for (i = brec->jb_oldfrags; i < brec->jb_oldfrags + brec->jb_frags; i++)
503 * Determine whether a given block has been reallocated to a new location.
504 * Returns a mask of overlapping bits if any frags have been reused or
505 * zero if the block has not been re-used and the contents can be trusted.
507 * This is used to ensure that an orphaned pointer due to truncate is safe
508 * to be freed. The mask value can be used to free partial blocks.
511 blk_freemask(ufs2_daddr_t blk, ino_t ino, ufs_lbn_t lbn, int frags)
513 struct suj_blk *sblk;
514 struct suj_rec *srec;
515 struct jblkrec *brec;
520 * To be certain we're not freeing a reallocated block we lookup
521 * this block in the blk hash and see if there is an allocation
522 * journal record that overlaps with any fragments in the block
523 * we're concerned with. If any fragments have ben reallocated
524 * the block has already been freed and re-used for another purpose.
527 sblk = blk_lookup(blknum(fs, blk), 0);
530 off = blk - sblk->sb_blk;
531 TAILQ_FOREACH(srec, &sblk->sb_recs, sr_next) {
532 brec = (struct jblkrec *)srec->sr_rec;
534 * If the block overlaps but does not match
535 * exactly this record refers to the current
538 if (blk_overlaps(brec, blk, frags) == 0)
540 if (blk_equals(brec, ino, lbn, blk, frags) == 1)
543 blk_setmask(brec, &mask);
546 printf("blk_freemask: blk %jd sblk %jd off %d mask 0x%X\n",
547 blk, sblk->sb_blk, off, mask);
548 return (mask >> off);
552 * Determine whether it is safe to follow an indirect. It is not safe
553 * if any part of the indirect has been reallocated or the last journal
554 * entry was an allocation. Just allocated indirects may not have valid
555 * pointers yet and all of their children will have their own records.
556 * It is also not safe to follow an indirect if the cg bitmap has been
557 * cleared as a new allocation may write to the block prior to the journal
560 * Returns 1 if it's safe to follow the indirect and 0 otherwise.
563 blk_isindir(ufs2_daddr_t blk, ino_t ino, ufs_lbn_t lbn)
565 struct suj_blk *sblk;
566 struct jblkrec *brec;
568 sblk = blk_lookup(blk, 0);
571 if (TAILQ_EMPTY(&sblk->sb_recs))
573 brec = (struct jblkrec *)TAILQ_LAST(&sblk->sb_recs, srechd)->sr_rec;
574 if (blk_equals(brec, ino, lbn, blk, fs->fs_frag))
575 if (brec->jb_op == JOP_FREEBLK)
576 return (!blk_isfree(blk));
581 * Clear an inode from the cg bitmap. If the inode was already clear return
582 * 0 so the caller knows it does not have to check the inode contents.
585 ino_free(ino_t ino, int mode)
592 cg = ino_to_cg(fs, ino);
593 ino = ino % fs->fs_ipg;
596 inosused = cg_inosused(cgp);
598 * The bitmap may never have made it to the disk so we have to
599 * conditionally clear. We can avoid writing the cg in this case.
601 if (isclr(inosused, ino))
604 clrbit(inosused, ino);
605 if (ino < cgp->cg_irotor)
606 cgp->cg_irotor = ino;
607 cgp->cg_cs.cs_nifree++;
608 if ((mode & IFMT) == IFDIR) {
610 cgp->cg_cs.cs_ndir--;
618 * Free 'frags' frags starting at filesystem block 'bno' skipping any frags
622 blk_free(ufs2_daddr_t bno, int mask, int frags)
624 ufs1_daddr_t fragno, cgbno;
631 printf("Freeing %d frags at blk %jd mask 0x%x\n",
636 cgbno = dtogd(fs, bno);
637 blksfree = cg_blksfree(cgp);
640 * If it's not allocated we only wrote the journal entry
641 * and never the bitmaps. Here we unconditionally clear and
642 * resolve the cg summary later.
644 if (frags == fs->fs_frag && mask == 0) {
645 fragno = fragstoblks(fs, cgbno);
646 ffs_setblock(fs, blksfree, fragno);
650 * deallocate the fragment
652 for (i = 0; i < frags; i++)
653 if ((mask & (1 << i)) == 0 && isclr(blksfree, cgbno +i)) {
655 setbit(blksfree, cgbno + i);
662 * Returns 1 if the whole block starting at 'bno' is marked free and 0
666 blk_isfree(ufs2_daddr_t bno)
670 sc = cg_lookup(dtog(fs, bno));
671 return ffs_isblock(fs, cg_blksfree(sc->sc_cgp), dtogd(fs, bno));
675 * Fetch an indirect block to find the block at a given lbn. The lbn
676 * may be negative to fetch a specific indirect block pointer or positive
677 * to fetch a specific block.
680 indir_blkatoff(ufs2_daddr_t blk, ino_t ino, ufs_lbn_t cur, ufs_lbn_t lbn)
691 level = lbn_level(cur);
693 err_suj("Invalid indir lbn %jd\n", lbn);
694 if (level == 0 && lbn < 0)
695 err_suj("Invalid lbn %jd\n", lbn);
696 bap2 = (void *)dblk_read(blk, fs->fs_bsize);
699 base = -(cur + level);
700 for (i = level; i > 0; i--)
701 lbnadd *= NINDIR(fs);
703 i = (lbn - base) / lbnadd;
705 i = (-lbn - base) / lbnadd;
706 if (i < 0 || i >= NINDIR(fs))
707 err_suj("Invalid indirect index %d produced by lbn %jd\n",
710 cur = base + (i * lbnadd);
712 cur = -(base + (i * lbnadd)) - (level - 1);
713 if (fs->fs_magic == FS_UFS1_MAGIC)
720 err_suj("Invalid lbn %jd at level 0\n", lbn);
721 return indir_blkatoff(blk, ino, cur, lbn);
725 * Finds the disk block address at the specified lbn within the inode
726 * specified by ip. This follows the whole tree and honors di_size and
727 * di_extsize so it is a true test of reachability. The lbn may be
728 * negative if an extattr or indirect block is requested.
731 ino_blkatoff(union dinode *ip, ino_t ino, ufs_lbn_t lbn, int *frags)
739 * Handle extattr blocks first.
741 if (lbn < 0 && lbn >= -UFS_NXADDR) {
743 if (lbn > lblkno(fs, ip->dp2.di_extsize - 1))
745 *frags = numfrags(fs, sblksize(fs, ip->dp2.di_extsize, lbn));
746 return (ip->dp2.di_extb[lbn]);
749 * Now direct and indirect.
751 if (DIP(ip, di_mode) == IFLNK &&
752 DIP(ip, di_size) < fs->fs_maxsymlinklen)
754 if (lbn >= 0 && lbn < UFS_NDADDR) {
755 *frags = numfrags(fs, sblksize(fs, DIP(ip, di_size), lbn));
756 return (DIP(ip, di_db[lbn]));
758 *frags = fs->fs_frag;
760 for (i = 0, tmpval = NINDIR(fs), cur = UFS_NDADDR; i < UFS_NIADDR; i++,
761 tmpval *= NINDIR(fs), cur = next) {
764 return (DIP(ip, di_ib[i]));
766 * Determine whether the lbn in question is within this tree.
768 if (lbn < 0 && -lbn >= next)
770 if (lbn > 0 && lbn >= next)
772 return indir_blkatoff(DIP(ip, di_ib[i]), ino, -cur - i, lbn);
774 err_suj("lbn %jd not in ino\n", lbn);
779 * Determine whether a block exists at a particular lbn in an inode.
780 * Returns 1 if found, 0 if not. lbn may be negative for indirects
784 blk_isat(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, int *frags)
791 if (DIP(ip, di_nlink) == 0 || DIP(ip, di_mode) == 0)
793 nblk = ino_blkatoff(ip, ino, lbn, frags);
795 return (nblk == blk);
799 * Clear the directory entry at diroff that should point to child. Minimal
800 * checking is done and it is assumed that this path was verified with isat.
803 ino_clrat(ino_t parent, off_t diroff, ino_t child)
815 printf("Clearing inode %ju from parent %ju at offset %jd\n",
816 (uintmax_t)child, (uintmax_t)parent, diroff);
818 lbn = lblkno(fs, diroff);
819 doff = blkoff(fs, diroff);
820 dip = ino_read(parent);
821 blk = ino_blkatoff(dip, parent, lbn, &frags);
822 blksize = sblksize(fs, DIP(dip, di_size), lbn);
823 block = dblk_read(blk, blksize);
824 dp = (struct direct *)&block[doff];
825 if (dp->d_ino != child)
826 errx(1, "Inode %ju does not exist in %ju at %jd",
827 (uintmax_t)child, (uintmax_t)parent, diroff);
831 * The actual .. reference count will already have been removed
832 * from the parent by the .. remref record.
837 * Determines whether a pointer to an inode exists within a directory
838 * at a specified offset. Returns the mode of the found entry.
841 ino_isat(ino_t parent, off_t diroff, ino_t child, int *mode, int *isdot)
854 dip = ino_read(parent);
855 *mode = DIP(dip, di_mode);
856 if ((*mode & IFMT) != IFDIR) {
859 * This can happen if the parent inode
863 printf("Directory %ju has bad mode %o\n",
864 (uintmax_t)parent, *mode);
866 printf("Directory %ju has zero mode\n",
871 lbn = lblkno(fs, diroff);
872 doff = blkoff(fs, diroff);
873 blksize = sblksize(fs, DIP(dip, di_size), lbn);
874 if (diroff + DIRECTSIZ(1) > DIP(dip, di_size) || doff >= blksize) {
876 printf("ino %ju absent from %ju due to offset %jd"
877 " exceeding size %jd\n",
878 (uintmax_t)child, (uintmax_t)parent, diroff,
882 blk = ino_blkatoff(dip, parent, lbn, &frags);
885 printf("Sparse directory %ju", (uintmax_t)parent);
888 block = dblk_read(blk, blksize);
890 * Walk through the records from the start of the block to be
891 * certain we hit a valid record and not some junk in the middle
892 * of a file name. Stop when we reach or pass the expected offset.
894 dpoff = rounddown(doff, DIRBLKSIZ);
896 dp = (struct direct *)&block[dpoff];
899 if (dp->d_reclen == 0)
901 dpoff += dp->d_reclen;
902 } while (dpoff <= doff);
903 if (dpoff > fs->fs_bsize)
904 err_suj("Corrupt directory block in dir ino %ju\n",
909 printf("ino %ju not found in %ju, lbn %jd, dpoff %d\n",
910 (uintmax_t)child, (uintmax_t)parent, lbn, dpoff);
914 * We found the item in question. Record the mode and whether it's
915 * a . or .. link for the caller.
917 if (dp->d_ino == child) {
920 else if (dp->d_namlen == 2 &&
921 dp->d_name[0] == '.' && dp->d_name[1] == '.')
923 *mode = DTTOIF(dp->d_type);
927 printf("ino %ju doesn't match dirent ino %ju in parent %ju\n",
928 (uintmax_t)child, (uintmax_t)dp->d_ino, (uintmax_t)parent);
932 #define VISIT_INDIR 0x0001
933 #define VISIT_EXT 0x0002
934 #define VISIT_ROOT 0x0004 /* Operation came via root & valid pointers. */
937 * Read an indirect level which may or may not be linked into an inode.
940 indir_visit(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, uint64_t *frags,
941 ino_visitor visitor, int flags)
952 * Don't visit indirect blocks with contents we can't trust. This
953 * should only happen when indir_visit() is called to complete a
954 * truncate that never finished and not when a pointer is found via
959 level = lbn_level(lbn);
961 err_suj("Invalid level for lbn %jd\n", lbn);
962 if ((flags & VISIT_ROOT) == 0 && blk_isindir(blk, ino, lbn) == 0) {
964 printf("blk %jd ino %ju lbn %jd(%d) is not indir.\n",
965 blk, (uintmax_t)ino, lbn, level);
969 for (i = level; i > 0; i--)
970 lbnadd *= NINDIR(fs);
971 bap1 = (void *)dblk_read(blk, fs->fs_bsize);
973 for (i = 0; i < NINDIR(fs); i++) {
974 if (fs->fs_magic == FS_UFS1_MAGIC)
981 nlbn = -lbn + i * lbnadd;
982 (*frags) += fs->fs_frag;
983 visitor(ino, nlbn, nblk, fs->fs_frag);
985 nlbn = (lbn + 1) - (i * lbnadd);
986 indir_visit(ino, nlbn, nblk, frags, visitor, flags);
990 if (flags & VISIT_INDIR) {
991 (*frags) += fs->fs_frag;
992 visitor(ino, lbn, blk, fs->fs_frag);
997 * Visit each block in an inode as specified by 'flags' and call a
998 * callback function. The callback may inspect or free blocks. The
999 * count of frags found according to the size in the file is returned.
1000 * This is not valid for sparse files but may be used to determine
1001 * the correct di_blocks for a file.
1004 ino_visit(union dinode *ip, ino_t ino, ino_visitor visitor, int flags)
1015 size = DIP(ip, di_size);
1016 mode = DIP(ip, di_mode) & IFMT;
1018 if ((flags & VISIT_EXT) &&
1019 fs->fs_magic == FS_UFS2_MAGIC && ip->dp2.di_extsize) {
1020 for (i = 0; i < UFS_NXADDR; i++) {
1021 if (ip->dp2.di_extb[i] == 0)
1023 frags = sblksize(fs, ip->dp2.di_extsize, i);
1024 frags = numfrags(fs, frags);
1026 visitor(ino, -1 - i, ip->dp2.di_extb[i], frags);
1029 /* Skip datablocks for short links and devices. */
1030 if (mode == IFBLK || mode == IFCHR ||
1031 (mode == IFLNK && size < fs->fs_maxsymlinklen))
1033 for (i = 0; i < UFS_NDADDR; i++) {
1034 if (DIP(ip, di_db[i]) == 0)
1036 frags = sblksize(fs, size, i);
1037 frags = numfrags(fs, frags);
1039 visitor(ino, i, DIP(ip, di_db[i]), frags);
1042 * We know the following indirects are real as we're following
1043 * real pointers to them.
1045 flags |= VISIT_ROOT;
1046 for (i = 0, tmpval = NINDIR(fs), lbn = UFS_NDADDR; i < UFS_NIADDR; i++,
1048 nextlbn = lbn + tmpval;
1049 tmpval *= NINDIR(fs);
1050 if (DIP(ip, di_ib[i]) == 0)
1052 indir_visit(ino, -lbn - i, DIP(ip, di_ib[i]), &fragcnt, visitor,
1059 * Null visitor function used when we just want to count blocks and
1064 null_visit(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, int frags)
1071 * Recalculate di_blocks when we discover that a block allocation or
1072 * free was not successfully completed. The kernel does not roll this back
1073 * because it would be too expensive to compute which indirects were
1074 * reachable at the time the inode was written.
1077 ino_adjblks(struct suj_ino *sino)
1088 /* No need to adjust zero'd inodes. */
1089 if (DIP(ip, di_mode) == 0)
1092 * Visit all blocks and count them as well as recording the last
1093 * valid lbn in the file. If the file size doesn't agree with the
1094 * last lbn we need to truncate to fix it. Otherwise just adjust
1098 frags = ino_visit(ip, ino, null_visit, VISIT_INDIR | VISIT_EXT);
1099 blocks = fsbtodb(fs, frags);
1101 * We assume the size and direct block list is kept coherent by
1102 * softdep. For files that have extended into indirects we truncate
1103 * to the size in the inode or the maximum size permitted by
1104 * populated indirects.
1106 if (visitlbn >= UFS_NDADDR) {
1107 isize = DIP(ip, di_size);
1108 size = lblktosize(fs, visitlbn + 1);
1111 /* Always truncate to free any unpopulated indirects. */
1112 ino_trunc(sino->si_ino, isize);
1115 if (blocks == DIP(ip, di_blocks))
1118 printf("ino %ju adjusting block count from %jd to %jd\n",
1119 (uintmax_t)ino, DIP(ip, di_blocks), blocks);
1120 DIP_SET(ip, di_blocks, blocks);
1125 blk_free_visit(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, int frags)
1128 blk_free(blk, blk_freemask(blk, ino, lbn, frags), frags);
1132 * Free a block or tree of blocks that was previously rooted in ino at
1133 * the given lbn. If the lbn is an indirect all children are freed
1137 blk_free_lbn(ufs2_daddr_t blk, ino_t ino, ufs_lbn_t lbn, int frags, int follow)
1142 mask = blk_freemask(blk, ino, lbn, frags);
1144 if (lbn <= -UFS_NDADDR && follow && mask == 0)
1145 indir_visit(ino, lbn, blk, &resid, blk_free_visit, VISIT_INDIR);
1147 blk_free(blk, mask, frags);
1151 ino_setskip(struct suj_ino *sino, ino_t parent)
1156 if (ino_isat(sino->si_ino, DOTDOT_OFFSET, parent, &mode, &isdot))
1157 sino->si_skipparent = 1;
1161 ino_remref(ino_t parent, ino_t child, uint64_t diroff, int isdotdot)
1163 struct suj_ino *sino;
1164 struct suj_rec *srec;
1165 struct jrefrec *rrec;
1168 * Lookup this inode to see if we have a record for it.
1170 sino = ino_lookup(child, 0);
1172 * Tell any child directories we've already removed their
1173 * parent link cnt. Don't try to adjust our link down again.
1175 if (sino != NULL && isdotdot == 0)
1176 ino_setskip(sino, parent);
1178 * No valid record for this inode. Just drop the on-disk
1181 if (sino == NULL || sino->si_hasrecs == 0) {
1186 * Use ino_adjust() if ino_check() has already processed this
1187 * child. If we lose the last non-dot reference to a
1188 * directory it will be discarded.
1190 if (sino->si_linkadj) {
1193 sino->si_dotlinks--;
1198 * If we haven't yet processed this inode we need to make
1199 * sure we will successfully discover the lost path. If not
1200 * use nlinkadj to remember.
1202 TAILQ_FOREACH(srec, &sino->si_recs, sr_next) {
1203 rrec = (struct jrefrec *)srec->sr_rec;
1204 if (rrec->jr_parent == parent &&
1205 rrec->jr_diroff == diroff)
1208 sino->si_nlinkadj++;
1212 * Free the children of a directory when the directory is discarded.
1215 ino_free_children(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, int frags)
1217 struct suj_ino *sino;
1226 sino = ino_lookup(ino, 0);
1228 skipparent = sino->si_skipparent;
1231 size = lfragtosize(fs, frags);
1232 block = dblk_read(blk, size);
1233 dp = (struct direct *)&block[0];
1234 for (dpoff = 0; dpoff < size && dp->d_reclen; dpoff += dp->d_reclen) {
1235 dp = (struct direct *)&block[dpoff];
1236 if (dp->d_ino == 0 || dp->d_ino == UFS_WINO)
1238 if (dp->d_namlen == 1 && dp->d_name[0] == '.')
1240 isdotdot = dp->d_namlen == 2 && dp->d_name[0] == '.' &&
1241 dp->d_name[1] == '.';
1242 if (isdotdot && skipparent == 1)
1245 printf("Directory %ju removing ino %ju name %s\n",
1246 (uintmax_t)ino, (uintmax_t)dp->d_ino, dp->d_name);
1247 diroff = lblktosize(fs, lbn) + dpoff;
1248 ino_remref(ino, dp->d_ino, diroff, isdotdot);
1253 * Reclaim an inode, freeing all blocks and decrementing all children's
1254 * link counts. Free the inode back to the cg.
1257 ino_reclaim(union dinode *ip, ino_t ino, int mode)
1261 if (ino == UFS_ROOTINO)
1262 err_suj("Attempting to free UFS_ROOTINO\n");
1264 printf("Truncating and freeing ino %ju, nlink %d, mode %o\n",
1265 (uintmax_t)ino, DIP(ip, di_nlink), DIP(ip, di_mode));
1267 /* We are freeing an inode or directory. */
1268 if ((DIP(ip, di_mode) & IFMT) == IFDIR)
1269 ino_visit(ip, ino, ino_free_children, 0);
1270 DIP_SET(ip, di_nlink, 0);
1271 ino_visit(ip, ino, blk_free_visit, VISIT_EXT | VISIT_INDIR);
1272 /* Here we have to clear the inode and release any blocks it holds. */
1273 gen = DIP(ip, di_gen);
1274 if (fs->fs_magic == FS_UFS1_MAGIC)
1275 bzero(ip, sizeof(struct ufs1_dinode));
1277 bzero(ip, sizeof(struct ufs2_dinode));
1278 DIP_SET(ip, di_gen, gen);
1280 ino_free(ino, mode);
1285 * Adjust an inode's link count down by one when a directory goes away.
1296 nlink = DIP(ip, di_nlink);
1297 mode = DIP(ip, di_mode);
1299 err_suj("Inode %d link count %d invalid\n", ino, nlink);
1301 err_suj("Inode %d has a link of %d with 0 mode\n", ino, nlink);
1303 if ((mode & IFMT) == IFDIR)
1307 if (nlink < reqlink) {
1309 printf("ino %ju not enough links to live %d < %d\n",
1310 (uintmax_t)ino, nlink, reqlink);
1311 ino_reclaim(ip, ino, mode);
1314 DIP_SET(ip, di_nlink, nlink);
1319 * Adjust the inode link count to 'nlink'. If the count reaches zero
1323 ino_adjust(struct suj_ino *sino)
1325 struct jrefrec *rrec;
1326 struct suj_rec *srec;
1327 struct suj_ino *stmp;
1336 nlink = sino->si_nlink;
1338 mode = sino->si_mode & IFMT;
1340 * If it's a directory with no dot links, it was truncated before
1341 * the name was cleared. We need to clear the dirent that
1344 if (mode == IFDIR && nlink == 1 && sino->si_dotlinks == 0) {
1345 sino->si_nlink = nlink = 0;
1346 TAILQ_FOREACH(srec, &sino->si_recs, sr_next) {
1347 rrec = (struct jrefrec *)srec->sr_rec;
1348 if (ino_isat(rrec->jr_parent, rrec->jr_diroff, ino,
1349 &recmode, &isdot) == 0)
1351 ino_clrat(rrec->jr_parent, rrec->jr_diroff, ino);
1355 errx(1, "Directory %ju name not found", (uintmax_t)ino);
1358 * If it's a directory with no real names pointing to it go ahead
1359 * and truncate it. This will free any children.
1361 if (mode == IFDIR && nlink - sino->si_dotlinks == 0) {
1362 sino->si_nlink = nlink = 0;
1364 * Mark any .. links so they know not to free this inode
1365 * when they are removed.
1367 TAILQ_FOREACH(srec, &sino->si_recs, sr_next) {
1368 rrec = (struct jrefrec *)srec->sr_rec;
1369 if (rrec->jr_diroff == DOTDOT_OFFSET) {
1370 stmp = ino_lookup(rrec->jr_parent, 0);
1372 ino_setskip(stmp, ino);
1377 mode = DIP(ip, di_mode) & IFMT;
1378 if (nlink > UFS_LINK_MAX)
1379 err_suj("ino %ju nlink manipulation error, new %ju, old %d\n",
1380 (uintmax_t)ino, (uintmax_t)nlink, DIP(ip, di_nlink));
1382 printf("Adjusting ino %ju, nlink %ju, old link %d lastmode %o\n",
1383 (uintmax_t)ino, (uintmax_t)nlink, DIP(ip, di_nlink),
1387 printf("ino %ju, zero inode freeing bitmap\n",
1389 ino_free(ino, sino->si_mode);
1392 /* XXX Should be an assert? */
1393 if (mode != sino->si_mode && debug)
1394 printf("ino %ju, mode %o != %o\n",
1395 (uintmax_t)ino, mode, sino->si_mode);
1396 if ((mode & IFMT) == IFDIR)
1400 /* If the inode doesn't have enough links to live, free it. */
1401 if (nlink < reqlink) {
1403 printf("ino %ju not enough links to live %ju < %ju\n",
1404 (uintmax_t)ino, (uintmax_t)nlink,
1405 (uintmax_t)reqlink);
1406 ino_reclaim(ip, ino, mode);
1409 /* If required write the updated link count. */
1410 if (DIP(ip, di_nlink) == nlink) {
1412 printf("ino %ju, link matches, skipping.\n",
1416 DIP_SET(ip, di_nlink, nlink);
1421 * Truncate some or all blocks in an indirect, freeing any that are required
1422 * and zeroing the indirect.
1425 indir_trunc(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, ufs_lbn_t lastlbn)
1440 level = lbn_level(lbn);
1442 err_suj("Invalid level for lbn %jd\n", lbn);
1444 for (i = level; i > 0; i--)
1445 lbnadd *= NINDIR(fs);
1446 bap1 = (void *)dblk_read(blk, fs->fs_bsize);
1447 bap2 = (void *)bap1;
1448 for (i = 0; i < NINDIR(fs); i++) {
1449 if (fs->fs_magic == FS_UFS1_MAGIC)
1456 nlbn = (lbn + 1) - (i * lbnadd);
1458 * Calculate the lbn of the next indirect to
1459 * determine if any of this indirect must be
1462 next = -(lbn + level) + ((i+1) * lbnadd);
1463 if (next <= lastlbn)
1465 indir_trunc(ino, nlbn, nblk, lastlbn);
1466 /* If all of this indirect was reclaimed, free it. */
1467 nlbn = next - lbnadd;
1471 nlbn = -lbn + i * lbnadd;
1476 blk_free(nblk, 0, fs->fs_frag);
1477 if (fs->fs_magic == FS_UFS1_MAGIC)
1487 * Truncate an inode to the minimum of the given size or the last populated
1488 * block after any over size have been discarded. The kernel would allocate
1489 * the last block in the file but fsck does not and neither do we. This
1490 * code never extends files, only shrinks them.
1493 ino_trunc(ino_t ino, off_t size)
1497 uint64_t totalfrags;
1509 mode = DIP(ip, di_mode) & IFMT;
1510 cursize = DIP(ip, di_size);
1512 printf("Truncating ino %ju, mode %o to size %jd from size %jd\n",
1513 (uintmax_t)ino, mode, size, cursize);
1515 /* Skip datablocks for short links and devices. */
1516 if (mode == 0 || mode == IFBLK || mode == IFCHR ||
1517 (mode == IFLNK && cursize < fs->fs_maxsymlinklen))
1522 lastlbn = lblkno(fs, blkroundup(fs, size));
1523 for (i = lastlbn; i < UFS_NDADDR; i++) {
1524 if (DIP(ip, di_db[i]) == 0)
1526 frags = sblksize(fs, cursize, i);
1527 frags = numfrags(fs, frags);
1528 blk_free(DIP(ip, di_db[i]), 0, frags);
1529 DIP_SET(ip, di_db[i], 0);
1532 * Follow indirect blocks, freeing anything required.
1534 for (i = 0, tmpval = NINDIR(fs), lbn = UFS_NDADDR; i < UFS_NIADDR; i++,
1536 nextlbn = lbn + tmpval;
1537 tmpval *= NINDIR(fs);
1538 /* If we're not freeing any in this indirect range skip it. */
1539 if (lastlbn >= nextlbn)
1541 if (DIP(ip, di_ib[i]) == 0)
1543 indir_trunc(ino, -lbn - i, DIP(ip, di_ib[i]), lastlbn);
1544 /* If we freed everything in this indirect free the indir. */
1547 blk_free(DIP(ip, di_ib[i]), 0, frags);
1548 DIP_SET(ip, di_ib[i], 0);
1552 * Now that we've freed any whole blocks that exceed the desired
1553 * truncation size, figure out how many blocks remain and what the
1554 * last populated lbn is. We will set the size to this last lbn
1555 * rather than worrying about allocating the final lbn as the kernel
1556 * would've done. This is consistent with normal fsck behavior.
1559 totalfrags = ino_visit(ip, ino, null_visit, VISIT_INDIR | VISIT_EXT);
1560 if (size > lblktosize(fs, visitlbn + 1))
1561 size = lblktosize(fs, visitlbn + 1);
1563 * If we're truncating direct blocks we have to adjust frags
1566 if (visitlbn < UFS_NDADDR && totalfrags) {
1567 long oldspace, newspace;
1569 bn = DIP(ip, di_db[visitlbn]);
1571 err_suj("Bad blk at ino %ju lbn %jd\n",
1572 (uintmax_t)ino, visitlbn);
1573 oldspace = sblksize(fs, cursize, visitlbn);
1574 newspace = sblksize(fs, size, visitlbn);
1575 if (oldspace != newspace) {
1576 bn += numfrags(fs, newspace);
1577 frags = numfrags(fs, oldspace - newspace);
1578 blk_free(bn, 0, frags);
1579 totalfrags -= frags;
1582 DIP_SET(ip, di_blocks, fsbtodb(fs, totalfrags));
1583 DIP_SET(ip, di_size, size);
1585 * If we've truncated into the middle of a block or frag we have
1586 * to zero it here. Otherwise the file could extend into
1587 * uninitialized space later.
1589 off = blkoff(fs, size);
1590 if (off && DIP(ip, di_mode) != IFDIR) {
1594 bn = ino_blkatoff(ip, ino, visitlbn, &frags);
1596 err_suj("Block missing from ino %ju at lbn %jd\n",
1597 (uintmax_t)ino, visitlbn);
1598 clrsize = frags * fs->fs_fsize;
1599 buf = dblk_read(bn, clrsize);
1602 bzero(buf, clrsize);
1609 * Process records available for one inode and determine whether the
1610 * link count is correct or needs adjusting.
1613 ino_check(struct suj_ino *sino)
1615 struct suj_rec *srec;
1616 struct jrefrec *rrec;
1626 if (sino->si_hasrecs == 0)
1629 rrec = (struct jrefrec *)TAILQ_FIRST(&sino->si_recs)->sr_rec;
1630 nlink = rrec->jr_nlink;
1633 removes = sino->si_nlinkadj;
1634 TAILQ_FOREACH(srec, &sino->si_recs, sr_next) {
1635 rrec = (struct jrefrec *)srec->sr_rec;
1636 isat = ino_isat(rrec->jr_parent, rrec->jr_diroff,
1637 rrec->jr_ino, &mode, &isdot);
1638 if (isat && (mode & IFMT) != (rrec->jr_mode & IFMT))
1639 err_suj("Inode mode/directory type mismatch %o != %o\n",
1640 mode, rrec->jr_mode);
1642 printf("jrefrec: op %d ino %ju, nlink %ju, parent %ju, "
1643 "diroff %jd, mode %o, isat %d, isdot %d\n",
1644 rrec->jr_op, (uintmax_t)rrec->jr_ino,
1645 (uintmax_t)rrec->jr_nlink,
1646 (uintmax_t)rrec->jr_parent,
1647 (uintmax_t)rrec->jr_diroff,
1648 rrec->jr_mode, isat, isdot);
1649 mode = rrec->jr_mode & IFMT;
1650 if (rrec->jr_op == JOP_REMREF)
1657 * The number of links that remain are the starting link count
1658 * subtracted by the total number of removes with the total
1659 * links discovered back in. An incomplete remove thus
1660 * makes no change to the link count but an add increases
1665 "ino %ju nlink %ju newlinks %ju removes %ju dotlinks %ju\n",
1666 (uintmax_t)ino, (uintmax_t)nlink, (uintmax_t)newlinks,
1667 (uintmax_t)removes, (uintmax_t)dotlinks);
1670 sino->si_linkadj = 1;
1671 sino->si_nlink = nlink;
1672 sino->si_dotlinks = dotlinks;
1673 sino->si_mode = mode;
1678 * Process records available for one block and determine whether it is
1679 * still allocated and whether the owning inode needs to be updated or
1683 blk_check(struct suj_blk *sblk)
1685 struct suj_rec *srec;
1686 struct jblkrec *brec;
1687 struct suj_ino *sino;
1694 * Each suj_blk actually contains records for any fragments in that
1695 * block. As a result we must evaluate each record individually.
1698 TAILQ_FOREACH(srec, &sblk->sb_recs, sr_next) {
1699 brec = (struct jblkrec *)srec->sr_rec;
1700 frags = brec->jb_frags;
1701 blk = brec->jb_blkno + brec->jb_oldfrags;
1702 isat = blk_isat(brec->jb_ino, brec->jb_lbn, blk, &frags);
1703 if (sino == NULL || sino->si_ino != brec->jb_ino) {
1704 sino = ino_lookup(brec->jb_ino, 1);
1705 sino->si_blkadj = 1;
1708 printf("op %d blk %jd ino %ju lbn %jd frags %d isat %d (%d)\n",
1709 brec->jb_op, blk, (uintmax_t)brec->jb_ino,
1710 brec->jb_lbn, brec->jb_frags, isat, frags);
1712 * If we found the block at this address we still have to
1713 * determine if we need to free the tail end that was
1714 * added by adding contiguous fragments from the same block.
1717 if (frags == brec->jb_frags)
1719 mask = blk_freemask(blk, brec->jb_ino, brec->jb_lbn,
1723 frags = brec->jb_frags - frags;
1724 blk_free(blk, mask, frags);
1728 * The block wasn't found, attempt to free it. It won't be
1729 * freed if it was actually reallocated. If this was an
1730 * allocation we don't want to follow indirects as they
1731 * may not be written yet. Any children of the indirect will
1732 * have their own records. If it's a free we need to
1733 * recursively free children.
1735 blk_free_lbn(blk, brec->jb_ino, brec->jb_lbn, brec->jb_frags,
1736 brec->jb_op == JOP_FREEBLK);
1741 * Walk the list of inode records for this cg and resolve moved and duplicate
1742 * inode references now that we have a complete picture.
1745 cg_build(struct suj_cg *sc)
1747 struct suj_ino *sino;
1750 for (i = 0; i < SUJ_HASHSIZE; i++)
1751 LIST_FOREACH(sino, &sc->sc_inohash[i], si_next)
1756 * Handle inodes requiring truncation. This must be done prior to
1757 * looking up any inodes in directories.
1760 cg_trunc(struct suj_cg *sc)
1762 struct suj_ino *sino;
1765 for (i = 0; i < SUJ_HASHSIZE; i++) {
1766 LIST_FOREACH(sino, &sc->sc_inohash[i], si_next) {
1767 if (sino->si_trunc) {
1768 ino_trunc(sino->si_ino,
1769 sino->si_trunc->jt_size);
1770 sino->si_blkadj = 0;
1771 sino->si_trunc = NULL;
1773 if (sino->si_blkadj)
1780 cg_adj_blk(struct suj_cg *sc)
1782 struct suj_ino *sino;
1785 for (i = 0; i < SUJ_HASHSIZE; i++) {
1786 LIST_FOREACH(sino, &sc->sc_inohash[i], si_next) {
1787 if (sino->si_blkadj)
1794 * Free any partially allocated blocks and then resolve inode block
1798 cg_check_blk(struct suj_cg *sc)
1800 struct suj_blk *sblk;
1804 for (i = 0; i < SUJ_HASHSIZE; i++)
1805 LIST_FOREACH(sblk, &sc->sc_blkhash[i], sb_next)
1810 * Walk the list of inode records for this cg, recovering any
1811 * changes which were not complete at the time of crash.
1814 cg_check_ino(struct suj_cg *sc)
1816 struct suj_ino *sino;
1819 for (i = 0; i < SUJ_HASHSIZE; i++)
1820 LIST_FOREACH(sino, &sc->sc_inohash[i], si_next)
1825 * Write a potentially dirty cg. Recalculate the summary information and
1826 * update the superblock summary.
1829 cg_write(struct suj_cg *sc)
1831 ufs1_daddr_t fragno, cgbno, maxbno;
1837 if (sc->sc_dirty == 0)
1840 * Fix the frag and cluster summary.
1843 cgp->cg_cs.cs_nbfree = 0;
1844 cgp->cg_cs.cs_nffree = 0;
1845 bzero(&cgp->cg_frsum, sizeof(cgp->cg_frsum));
1846 maxbno = fragstoblks(fs, fs->fs_fpg);
1847 if (fs->fs_contigsumsize > 0) {
1848 for (i = 1; i <= fs->fs_contigsumsize; i++)
1849 cg_clustersum(cgp)[i] = 0;
1850 bzero(cg_clustersfree(cgp), howmany(maxbno, CHAR_BIT));
1852 blksfree = cg_blksfree(cgp);
1853 for (cgbno = 0; cgbno < maxbno; cgbno++) {
1854 if (ffs_isfreeblock(fs, blksfree, cgbno))
1856 if (ffs_isblock(fs, blksfree, cgbno)) {
1857 ffs_clusteracct(fs, cgp, cgbno, 1);
1858 cgp->cg_cs.cs_nbfree++;
1861 fragno = blkstofrags(fs, cgbno);
1862 blk = blkmap(fs, blksfree, fragno);
1863 ffs_fragacct(fs, blk, cgp->cg_frsum, 1);
1864 for (i = 0; i < fs->fs_frag; i++)
1865 if (isset(blksfree, fragno + i))
1866 cgp->cg_cs.cs_nffree++;
1869 * Update the superblock cg summary from our now correct values
1870 * before writing the block.
1872 fs->fs_cs(fs, sc->sc_cgx) = cgp->cg_cs;
1873 if (cgput(&disk, cgp) == -1)
1874 err_suj("Unable to write cylinder group %d\n", sc->sc_cgx);
1878 * Write out any modified inodes.
1881 cg_write_inos(struct suj_cg *sc)
1883 struct ino_blk *iblk;
1886 for (i = 0; i < SUJ_HASHSIZE; i++)
1887 LIST_FOREACH(iblk, &sc->sc_iblkhash[i], ib_next)
1893 cg_apply(void (*apply)(struct suj_cg *))
1898 for (i = 0; i < SUJ_HASHSIZE; i++)
1899 LIST_FOREACH(scg, &cghash[i], sc_next)
1904 * Process the unlinked but referenced file list. Freeing all inodes.
1914 ino = fs->fs_sujfree;
1918 mode = DIP(ip, di_mode) & IFMT;
1919 inon = DIP(ip, di_freelink);
1920 DIP_SET(ip, di_freelink, 0);
1922 * XXX Should this be an errx?
1924 if (DIP(ip, di_nlink) == 0) {
1926 printf("Freeing unlinked ino %ju mode %o\n",
1927 (uintmax_t)ino, mode);
1928 ino_reclaim(ip, ino, mode);
1930 printf("Skipping ino %ju mode %o with link %d\n",
1931 (uintmax_t)ino, mode, DIP(ip, di_nlink));
1937 * Append a new record to the list of records requiring processing.
1940 ino_append(union jrec *rec)
1942 struct jrefrec *refrec;
1943 struct jmvrec *mvrec;
1944 struct suj_ino *sino;
1945 struct suj_rec *srec;
1947 mvrec = &rec->rec_jmvrec;
1948 refrec = &rec->rec_jrefrec;
1949 if (debug && mvrec->jm_op == JOP_MVREF)
1950 printf("ino move: ino %ju, parent %ju, "
1951 "diroff %jd, oldoff %jd\n",
1952 (uintmax_t)mvrec->jm_ino, (uintmax_t)mvrec->jm_parent,
1953 (uintmax_t)mvrec->jm_newoff, (uintmax_t)mvrec->jm_oldoff);
1955 (refrec->jr_op == JOP_ADDREF || refrec->jr_op == JOP_REMREF))
1956 printf("ino ref: op %d, ino %ju, nlink %ju, "
1957 "parent %ju, diroff %jd\n",
1958 refrec->jr_op, (uintmax_t)refrec->jr_ino,
1959 (uintmax_t)refrec->jr_nlink,
1960 (uintmax_t)refrec->jr_parent, (uintmax_t)refrec->jr_diroff);
1961 sino = ino_lookup(((struct jrefrec *)rec)->jr_ino, 1);
1962 sino->si_hasrecs = 1;
1963 srec = errmalloc(sizeof(*srec));
1965 TAILQ_INSERT_TAIL(&sino->si_newrecs, srec, sr_next);
1969 * Add a reference adjustment to the sino list and eliminate dups. The
1970 * primary loop in ino_build_ref() checks for dups but new ones may be
1971 * created as a result of offset adjustments.
1974 ino_add_ref(struct suj_ino *sino, struct suj_rec *srec)
1976 struct jrefrec *refrec;
1977 struct suj_rec *srn;
1978 struct jrefrec *rrn;
1980 refrec = (struct jrefrec *)srec->sr_rec;
1982 * We walk backwards so that the oldest link count is preserved. If
1983 * an add record conflicts with a remove keep the remove. Redundant
1984 * removes are eliminated in ino_build_ref. Otherwise we keep the
1985 * oldest record at a given location.
1987 for (srn = TAILQ_LAST(&sino->si_recs, srechd); srn;
1988 srn = TAILQ_PREV(srn, srechd, sr_next)) {
1989 rrn = (struct jrefrec *)srn->sr_rec;
1990 if (rrn->jr_parent != refrec->jr_parent ||
1991 rrn->jr_diroff != refrec->jr_diroff)
1993 if (rrn->jr_op == JOP_REMREF || refrec->jr_op == JOP_ADDREF) {
1994 rrn->jr_mode = refrec->jr_mode;
2000 * Replace the record in place with the old nlink in case
2001 * we replace the head of the list. Abandon srec as a dup.
2003 refrec->jr_nlink = rrn->jr_nlink;
2004 srn->sr_rec = srec->sr_rec;
2007 TAILQ_INSERT_TAIL(&sino->si_recs, srec, sr_next);
2011 * Create a duplicate of a reference at a previous location.
2014 ino_dup_ref(struct suj_ino *sino, struct jrefrec *refrec, off_t diroff)
2016 struct jrefrec *rrn;
2017 struct suj_rec *srn;
2019 rrn = errmalloc(sizeof(*refrec));
2021 rrn->jr_op = JOP_ADDREF;
2022 rrn->jr_diroff = diroff;
2023 srn = errmalloc(sizeof(*srn));
2024 srn->sr_rec = (union jrec *)rrn;
2025 ino_add_ref(sino, srn);
2029 * Add a reference to the list at all known locations. We follow the offset
2030 * changes for a single instance and create duplicate add refs at each so
2031 * that we can tolerate any version of the directory block. Eliminate
2032 * removes which collide with adds that are seen in the journal. They should
2033 * not adjust the link count down.
2036 ino_build_ref(struct suj_ino *sino, struct suj_rec *srec)
2038 struct jrefrec *refrec;
2039 struct jmvrec *mvrec;
2040 struct suj_rec *srp;
2041 struct suj_rec *srn;
2042 struct jrefrec *rrn;
2045 refrec = (struct jrefrec *)srec->sr_rec;
2047 * Search for a mvrec that matches this offset. Whether it's an add
2048 * or a remove we can delete the mvref after creating a dup record in
2051 if (!TAILQ_EMPTY(&sino->si_movs)) {
2052 diroff = refrec->jr_diroff;
2053 for (srn = TAILQ_LAST(&sino->si_movs, srechd); srn; srn = srp) {
2054 srp = TAILQ_PREV(srn, srechd, sr_next);
2055 mvrec = (struct jmvrec *)srn->sr_rec;
2056 if (mvrec->jm_parent != refrec->jr_parent ||
2057 mvrec->jm_newoff != diroff)
2059 diroff = mvrec->jm_oldoff;
2060 TAILQ_REMOVE(&sino->si_movs, srn, sr_next);
2062 ino_dup_ref(sino, refrec, diroff);
2066 * If a remove wasn't eliminated by an earlier add just append it to
2069 if (refrec->jr_op == JOP_REMREF) {
2070 ino_add_ref(sino, srec);
2074 * Walk the list of records waiting to be added to the list. We
2075 * must check for moves that apply to our current offset and remove
2076 * them from the list. Remove any duplicates to eliminate removes
2077 * with corresponding adds.
2079 TAILQ_FOREACH_SAFE(srn, &sino->si_newrecs, sr_next, srp) {
2080 switch (srn->sr_rec->rec_jrefrec.jr_op) {
2083 * This should actually be an error we should
2084 * have a remove for every add journaled.
2086 rrn = (struct jrefrec *)srn->sr_rec;
2087 if (rrn->jr_parent != refrec->jr_parent ||
2088 rrn->jr_diroff != refrec->jr_diroff)
2090 TAILQ_REMOVE(&sino->si_newrecs, srn, sr_next);
2094 * Once we remove the current iteration of the
2095 * record at this address we're done.
2097 rrn = (struct jrefrec *)srn->sr_rec;
2098 if (rrn->jr_parent != refrec->jr_parent ||
2099 rrn->jr_diroff != refrec->jr_diroff)
2101 TAILQ_REMOVE(&sino->si_newrecs, srn, sr_next);
2102 ino_add_ref(sino, srec);
2106 * Update our diroff based on any moves that match
2107 * and remove the move.
2109 mvrec = (struct jmvrec *)srn->sr_rec;
2110 if (mvrec->jm_parent != refrec->jr_parent ||
2111 mvrec->jm_oldoff != refrec->jr_diroff)
2113 ino_dup_ref(sino, refrec, mvrec->jm_oldoff);
2114 refrec->jr_diroff = mvrec->jm_newoff;
2115 TAILQ_REMOVE(&sino->si_newrecs, srn, sr_next);
2118 err_suj("ino_build_ref: Unknown op %d\n",
2119 srn->sr_rec->rec_jrefrec.jr_op);
2122 ino_add_ref(sino, srec);
2126 * Walk the list of new records and add them in-order resolving any
2127 * dups and adjusted offsets.
2130 ino_build(struct suj_ino *sino)
2132 struct suj_rec *srec;
2134 while ((srec = TAILQ_FIRST(&sino->si_newrecs)) != NULL) {
2135 TAILQ_REMOVE(&sino->si_newrecs, srec, sr_next);
2136 switch (srec->sr_rec->rec_jrefrec.jr_op) {
2139 ino_build_ref(sino, srec);
2143 * Add this mvrec to the queue of pending mvs.
2145 TAILQ_INSERT_TAIL(&sino->si_movs, srec, sr_next);
2148 err_suj("ino_build: Unknown op %d\n",
2149 srec->sr_rec->rec_jrefrec.jr_op);
2152 if (TAILQ_EMPTY(&sino->si_recs))
2153 sino->si_hasrecs = 0;
2157 * Modify journal records so they refer to the base block number
2158 * and a start and end frag range. This is to facilitate the discovery
2159 * of overlapping fragment allocations.
2162 blk_build(struct jblkrec *blkrec)
2164 struct suj_rec *srec;
2165 struct suj_blk *sblk;
2166 struct jblkrec *blkrn;
2171 printf("blk_build: op %d blkno %jd frags %d oldfrags %d "
2172 "ino %ju lbn %jd\n",
2173 blkrec->jb_op, (uintmax_t)blkrec->jb_blkno,
2174 blkrec->jb_frags, blkrec->jb_oldfrags,
2175 (uintmax_t)blkrec->jb_ino, (uintmax_t)blkrec->jb_lbn);
2177 blk = blknum(fs, blkrec->jb_blkno);
2178 frag = fragnum(fs, blkrec->jb_blkno);
2179 sblk = blk_lookup(blk, 1);
2181 * Rewrite the record using oldfrags to indicate the offset into
2182 * the block. Leave jb_frags as the actual allocated count.
2184 blkrec->jb_blkno -= frag;
2185 blkrec->jb_oldfrags = frag;
2186 if (blkrec->jb_oldfrags + blkrec->jb_frags > fs->fs_frag)
2187 err_suj("Invalid fragment count %d oldfrags %d\n",
2188 blkrec->jb_frags, frag);
2190 * Detect dups. If we detect a dup we always discard the oldest
2191 * record as it is superseded by the new record. This speeds up
2192 * later stages but also eliminates free records which are used
2193 * to indicate that the contents of indirects can be trusted.
2195 TAILQ_FOREACH(srec, &sblk->sb_recs, sr_next) {
2196 blkrn = (struct jblkrec *)srec->sr_rec;
2197 if (blkrn->jb_ino != blkrec->jb_ino ||
2198 blkrn->jb_lbn != blkrec->jb_lbn ||
2199 blkrn->jb_blkno != blkrec->jb_blkno ||
2200 blkrn->jb_frags != blkrec->jb_frags ||
2201 blkrn->jb_oldfrags != blkrec->jb_oldfrags)
2204 printf("Removed dup.\n");
2205 /* Discard the free which is a dup with an alloc. */
2206 if (blkrec->jb_op == JOP_FREEBLK)
2208 TAILQ_REMOVE(&sblk->sb_recs, srec, sr_next);
2212 srec = errmalloc(sizeof(*srec));
2213 srec->sr_rec = (union jrec *)blkrec;
2214 TAILQ_INSERT_TAIL(&sblk->sb_recs, srec, sr_next);
2218 ino_build_trunc(struct jtrncrec *rec)
2220 struct suj_ino *sino;
2223 printf("ino_build_trunc: op %d ino %ju, size %jd\n",
2224 rec->jt_op, (uintmax_t)rec->jt_ino,
2225 (uintmax_t)rec->jt_size);
2226 sino = ino_lookup(rec->jt_ino, 1);
2227 if (rec->jt_op == JOP_SYNC) {
2228 sino->si_trunc = NULL;
2231 if (sino->si_trunc == NULL || sino->si_trunc->jt_size > rec->jt_size)
2232 sino->si_trunc = rec;
2236 * Build up tables of the operations we need to recover.
2241 struct suj_seg *seg;
2246 TAILQ_FOREACH(seg, &allsegs, ss_next) {
2248 printf("seg %jd has %d records, oldseq %jd.\n",
2249 seg->ss_rec.jsr_seq, seg->ss_rec.jsr_cnt,
2250 seg->ss_rec.jsr_oldest);
2252 rec = (union jrec *)seg->ss_blk;
2253 for (i = 0; i < seg->ss_rec.jsr_cnt; off += JREC_SIZE, rec++) {
2254 /* skip the segrec. */
2255 if ((off % real_dev_bsize) == 0)
2257 switch (rec->rec_jrefrec.jr_op) {
2265 blk_build((struct jblkrec *)rec);
2269 ino_build_trunc((struct jtrncrec *)rec);
2272 err_suj("Unknown journal operation %d (%d)\n",
2273 rec->rec_jrefrec.jr_op, off);
2281 * Prune the journal segments to those we care about based on the
2282 * oldest sequence in the newest segment. Order the segment list
2283 * based on sequence number.
2288 struct suj_seg *seg;
2289 struct suj_seg *segn;
2294 printf("Pruning up to %jd\n", oldseq);
2295 /* First free the expired segments. */
2296 TAILQ_FOREACH_SAFE(seg, &allsegs, ss_next, segn) {
2297 if (seg->ss_rec.jsr_seq >= oldseq)
2299 TAILQ_REMOVE(&allsegs, seg, ss_next);
2303 /* Next ensure that segments are ordered properly. */
2304 seg = TAILQ_FIRST(&allsegs);
2307 printf("Empty journal\n");
2310 newseq = seg->ss_rec.jsr_seq;
2312 seg = TAILQ_LAST(&allsegs, seghd);
2313 if (seg->ss_rec.jsr_seq >= newseq)
2315 TAILQ_REMOVE(&allsegs, seg, ss_next);
2316 TAILQ_INSERT_HEAD(&allsegs, seg, ss_next);
2317 newseq = seg->ss_rec.jsr_seq;
2320 if (newseq != oldseq) {
2321 TAILQ_FOREACH(seg, &allsegs, ss_next) {
2322 printf("%jd, ", seg->ss_rec.jsr_seq);
2325 err_suj("Journal file sequence mismatch %jd != %jd\n",
2329 * The kernel may asynchronously write segments which can create
2330 * gaps in the sequence space. Throw away any segments after the
2331 * gap as the kernel guarantees only those that are contiguously
2332 * reachable are marked as completed.
2335 TAILQ_FOREACH_SAFE(seg, &allsegs, ss_next, segn) {
2336 if (!discard && newseq++ == seg->ss_rec.jsr_seq) {
2337 jrecs += seg->ss_rec.jsr_cnt;
2338 jbytes += seg->ss_rec.jsr_blocks * real_dev_bsize;
2343 printf("Journal order mismatch %jd != %jd pruning\n",
2344 newseq-1, seg->ss_rec.jsr_seq);
2345 TAILQ_REMOVE(&allsegs, seg, ss_next);
2350 printf("Processing journal segments from %jd to %jd\n",
2355 * Verify the journal inode before attempting to read records.
2358 suj_verifyino(union dinode *ip)
2361 if (DIP(ip, di_nlink) != 1) {
2362 printf("Invalid link count %d for journal inode %ju\n",
2363 DIP(ip, di_nlink), (uintmax_t)sujino);
2367 if ((DIP(ip, di_flags) & (SF_IMMUTABLE | SF_NOUNLINK)) !=
2368 (SF_IMMUTABLE | SF_NOUNLINK)) {
2369 printf("Invalid flags 0x%X for journal inode %ju\n",
2370 DIP(ip, di_flags), (uintmax_t)sujino);
2374 if (DIP(ip, di_mode) != (IFREG | IREAD)) {
2375 printf("Invalid mode %o for journal inode %ju\n",
2376 DIP(ip, di_mode), (uintmax_t)sujino);
2380 if (DIP(ip, di_size) < SUJ_MIN) {
2381 printf("Invalid size %jd for journal inode %ju\n",
2382 DIP(ip, di_size), (uintmax_t)sujino);
2386 if (DIP(ip, di_modrev) != fs->fs_mtime) {
2387 printf("Journal timestamp does not match fs mount time\n");
2395 struct jextent *jb_extent; /* Extent array. */
2396 int jb_avail; /* Available extents. */
2397 int jb_used; /* Last used extent. */
2398 int jb_head; /* Allocator head. */
2399 int jb_off; /* Allocator extent offset. */
2402 ufs2_daddr_t je_daddr; /* Disk block address. */
2403 int je_blocks; /* Disk block count. */
2406 static struct jblocks *suj_jblocks;
2408 static struct jblocks *
2409 jblocks_create(void)
2411 struct jblocks *jblocks;
2414 jblocks = errmalloc(sizeof(*jblocks));
2415 jblocks->jb_avail = 10;
2416 jblocks->jb_used = 0;
2417 jblocks->jb_head = 0;
2418 jblocks->jb_off = 0;
2419 size = sizeof(struct jextent) * jblocks->jb_avail;
2420 jblocks->jb_extent = errmalloc(size);
2421 bzero(jblocks->jb_extent, size);
2427 * Return the next available disk block and the amount of contiguous
2428 * free space it contains.
2431 jblocks_next(struct jblocks *jblocks, int bytes, int *actual)
2433 struct jextent *jext;
2438 blocks = bytes / disk.d_bsize;
2439 jext = &jblocks->jb_extent[jblocks->jb_head];
2440 freecnt = jext->je_blocks - jblocks->jb_off;
2442 jblocks->jb_off = 0;
2443 if (++jblocks->jb_head > jblocks->jb_used)
2445 jext = &jblocks->jb_extent[jblocks->jb_head];
2446 freecnt = jext->je_blocks;
2448 if (freecnt > blocks)
2450 *actual = freecnt * disk.d_bsize;
2451 daddr = jext->je_daddr + jblocks->jb_off;
2457 * Advance the allocation head by a specified number of bytes, consuming
2458 * one journal segment.
2461 jblocks_advance(struct jblocks *jblocks, int bytes)
2464 jblocks->jb_off += bytes / disk.d_bsize;
2468 jblocks_destroy(struct jblocks *jblocks)
2471 free(jblocks->jb_extent);
2476 jblocks_add(struct jblocks *jblocks, ufs2_daddr_t daddr, int blocks)
2478 struct jextent *jext;
2481 jext = &jblocks->jb_extent[jblocks->jb_used];
2482 /* Adding the first block. */
2483 if (jext->je_daddr == 0) {
2484 jext->je_daddr = daddr;
2485 jext->je_blocks = blocks;
2488 /* Extending the last extent. */
2489 if (jext->je_daddr + jext->je_blocks == daddr) {
2490 jext->je_blocks += blocks;
2493 /* Adding a new extent. */
2494 if (++jblocks->jb_used == jblocks->jb_avail) {
2495 jblocks->jb_avail *= 2;
2496 size = sizeof(struct jextent) * jblocks->jb_avail;
2497 jext = errmalloc(size);
2499 bcopy(jblocks->jb_extent, jext,
2500 sizeof(struct jextent) * jblocks->jb_used);
2501 free(jblocks->jb_extent);
2502 jblocks->jb_extent = jext;
2504 jext = &jblocks->jb_extent[jblocks->jb_used];
2505 jext->je_daddr = daddr;
2506 jext->je_blocks = blocks;
2512 * Add a file block from the journal to the extent map. We can't read
2513 * each file block individually because the kernel treats it as a circular
2514 * buffer and segments may span mutliple contiguous blocks.
2517 suj_add_block(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, int frags)
2520 jblocks_add(suj_jblocks, fsbtodb(fs, blk), fsbtodb(fs, frags));
2526 uint8_t block[1 * 1024 * 1024];
2527 struct suj_seg *seg;
2528 struct jsegrec *recn;
2529 struct jsegrec *rec;
2538 * Read records until we exhaust the journal space. If we find
2539 * an invalid record we start searching for a valid segment header
2540 * at the next block. This is because we don't have a head/tail
2541 * pointer and must recover the information indirectly. At the gap
2542 * between the head and tail we won't necessarily have a valid
2547 size = sizeof(block);
2548 blk = jblocks_next(suj_jblocks, size, &readsize);
2553 * Read 1MB at a time and scan for records within this block.
2555 if (bread(&disk, blk, &block, size) == -1) {
2556 err_suj("Error reading journal block %jd\n",
2559 for (rec = (void *)block; size; size -= recsize,
2560 rec = (struct jsegrec *)((uintptr_t)rec + recsize)) {
2561 recsize = real_dev_bsize;
2562 if (rec->jsr_time != fs->fs_mtime) {
2564 printf("Rec time %jd != fs mtime %jd\n",
2565 rec->jsr_time, fs->fs_mtime);
2566 jblocks_advance(suj_jblocks, recsize);
2569 if (rec->jsr_cnt == 0) {
2571 printf("Found illegal count %d\n",
2573 jblocks_advance(suj_jblocks, recsize);
2576 blocks = rec->jsr_blocks;
2577 recsize = blocks * real_dev_bsize;
2578 if (recsize > size) {
2580 * We may just have run out of buffer, restart
2581 * the loop to re-read from this spot.
2583 if (size < fs->fs_bsize &&
2585 recsize <= fs->fs_bsize)
2588 printf("Found invalid segsize %d > %d\n",
2590 recsize = real_dev_bsize;
2591 jblocks_advance(suj_jblocks, recsize);
2595 * Verify that all blocks in the segment are present.
2597 for (i = 1; i < blocks; i++) {
2598 recn = (void *)((uintptr_t)rec) + i *
2600 if (recn->jsr_seq == rec->jsr_seq &&
2601 recn->jsr_time == rec->jsr_time)
2604 printf("Incomplete record %jd (%d)\n",
2606 recsize = i * real_dev_bsize;
2607 jblocks_advance(suj_jblocks, recsize);
2610 seg = errmalloc(sizeof(*seg));
2611 seg->ss_blk = errmalloc(recsize);
2613 bcopy((void *)rec, seg->ss_blk, recsize);
2614 if (rec->jsr_oldest > oldseq)
2615 oldseq = rec->jsr_oldest;
2616 TAILQ_INSERT_TAIL(&allsegs, seg, ss_next);
2617 jblocks_advance(suj_jblocks, recsize);
2623 * Search a directory block for the SUJ_FILE.
2626 suj_find(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, int frags)
2628 char block[MAXBSIZE];
2635 bytes = lfragtosize(fs, frags);
2636 if (bread(&disk, fsbtodb(fs, blk), block, bytes) <= 0)
2637 err_suj("Failed to read UFS_ROOTINO directory block %jd\n",
2639 for (off = 0; off < bytes; off += dp->d_reclen) {
2640 dp = (struct direct *)&block[off];
2641 if (dp->d_reclen == 0)
2645 if (dp->d_namlen != strlen(SUJ_FILE))
2647 if (bcmp(dp->d_name, SUJ_FILE, dp->d_namlen) != 0)
2655 * Orchestrate the verification of a filesystem via the softupdates journal.
2658 suj_check(const char *filesys)
2664 struct suj_seg *seg;
2665 struct suj_seg *segn;
2669 if (real_dev_bsize == 0 && ioctl(disk.d_fd, DIOCGSECTORSIZE,
2670 &real_dev_bsize) == -1)
2671 real_dev_bsize = secsize;
2673 printf("dev_bsize %u\n", real_dev_bsize);
2676 * Set an exit point when SUJ check failed
2678 retval = setjmp(jmpbuf);
2680 pwarn("UNEXPECTED SU+J INCONSISTENCY\n");
2681 TAILQ_FOREACH_SAFE(seg, &allsegs, ss_next, segn) {
2682 TAILQ_REMOVE(&allsegs, seg, ss_next);
2686 if (reply("FALLBACK TO FULL FSCK") == 0) {
2694 * Find the journal inode.
2696 ip = ino_read(UFS_ROOTINO);
2698 ino_visit(ip, UFS_ROOTINO, suj_find, 0);
2700 printf("Journal inode removed. Use tunefs to re-create.\n");
2701 sblock.fs_flags &= ~FS_SUJ;
2702 sblock.fs_sujfree = 0;
2706 * Fetch the journal inode and verify it.
2708 jip = ino_read(sujino);
2709 printf("** SU+J Recovering %s\n", filesys);
2710 if (suj_verifyino(jip) != 0)
2713 * Build a list of journal blocks in jblocks before parsing the
2714 * available journal blocks in with suj_read().
2716 printf("** Reading %jd byte journal from inode %ju.\n",
2717 DIP(jip, di_size), (uintmax_t)sujino);
2718 suj_jblocks = jblocks_create();
2719 blocks = ino_visit(jip, sujino, suj_add_block, 0);
2720 if (blocks != numfrags(fs, DIP(jip, di_size))) {
2721 printf("Sparse journal inode %ju.\n", (uintmax_t)sujino);
2725 jblocks_destroy(suj_jblocks);
2727 if (preen || reply("RECOVER")) {
2728 printf("** Building recovery table.\n");
2732 printf("** Resolving unreferenced inode list.\n");
2734 printf("** Processing journal entries.\n");
2736 cg_apply(cg_check_blk);
2737 cg_apply(cg_adj_blk);
2738 cg_apply(cg_check_ino);
2740 if (preen == 0 && (jrecs > 0 || jbytes > 0) && reply("WRITE CHANGES") == 0)
2743 * To remain idempotent with partial truncations the free bitmaps
2744 * must be written followed by indirect blocks and lastly inode
2745 * blocks. This preserves access to the modified pointers until
2750 cg_apply(cg_write_inos);
2751 /* Write back superblock. */
2753 if (jrecs > 0 || jbytes > 0) {
2754 printf("** %jd journal records in %jd bytes for %.2f%% utilization\n",
2755 jrecs, jbytes, ((float)jrecs / (float)(jbytes / JREC_SIZE)) * 100);
2756 printf("** Freed %jd inodes (%jd dirs) %jd blocks, and %jd frags.\n",
2757 freeinos, freedir, freeblocks, freefrags);
2768 for (i = 0; i < SUJ_HASHSIZE; i++) {
2769 LIST_INIT(&cghash[i]);
2770 LIST_INIT(&dbhash[i]);
2774 TAILQ_INIT(&allsegs);