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;
436 blk = ino_to_fsba(fs, ino);
437 sc = cg_lookup(ino_to_cg(fs, ino));
438 iblk = sc->sc_lastiblk;
439 if (iblk && iblk->ib_blk == blk) {
440 if (fs->fs_magic == FS_UFS2_MAGIC) {
441 off = ino_to_fsbo(fs, ino);
442 ffs_update_dinode_ckhash(fs,
443 &((struct ufs2_dinode *)iblk->ib_buf)[off]);
448 hd = &sc->sc_iblkhash[SUJ_HASH(fragstoblks(fs, blk))];
449 LIST_FOREACH(iblk, hd, ib_next) {
450 if (iblk->ib_blk == blk) {
451 if (fs->fs_magic == FS_UFS2_MAGIC) {
452 off = ino_to_fsbo(fs, ino);
453 ffs_update_dinode_ckhash(fs,
454 &((struct ufs2_dinode *)iblk->ib_buf)[off]);
465 iblk_write(struct ino_blk *iblk)
468 if (iblk->ib_dirty == 0)
470 if (bwrite(&disk, fsbtodb(fs, iblk->ib_blk), iblk->ib_buf,
472 err_suj("Failed to write inode block %jd\n", iblk->ib_blk);
476 blk_overlaps(struct jblkrec *brec, ufs2_daddr_t start, int frags)
483 bstart = brec->jb_blkno + brec->jb_oldfrags;
484 bend = bstart + brec->jb_frags;
485 if (start < bend && end > bstart)
491 blk_equals(struct jblkrec *brec, ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t start,
495 if (brec->jb_ino != ino || brec->jb_lbn != lbn)
497 if (brec->jb_blkno + brec->jb_oldfrags != start)
499 if (brec->jb_frags < frags)
505 blk_setmask(struct jblkrec *brec, int *mask)
509 for (i = brec->jb_oldfrags; i < brec->jb_oldfrags + brec->jb_frags; i++)
514 * Determine whether a given block has been reallocated to a new location.
515 * Returns a mask of overlapping bits if any frags have been reused or
516 * zero if the block has not been re-used and the contents can be trusted.
518 * This is used to ensure that an orphaned pointer due to truncate is safe
519 * to be freed. The mask value can be used to free partial blocks.
522 blk_freemask(ufs2_daddr_t blk, ino_t ino, ufs_lbn_t lbn, int frags)
524 struct suj_blk *sblk;
525 struct suj_rec *srec;
526 struct jblkrec *brec;
531 * To be certain we're not freeing a reallocated block we lookup
532 * this block in the blk hash and see if there is an allocation
533 * journal record that overlaps with any fragments in the block
534 * we're concerned with. If any fragments have ben reallocated
535 * the block has already been freed and re-used for another purpose.
538 sblk = blk_lookup(blknum(fs, blk), 0);
541 off = blk - sblk->sb_blk;
542 TAILQ_FOREACH(srec, &sblk->sb_recs, sr_next) {
543 brec = (struct jblkrec *)srec->sr_rec;
545 * If the block overlaps but does not match
546 * exactly this record refers to the current
549 if (blk_overlaps(brec, blk, frags) == 0)
551 if (blk_equals(brec, ino, lbn, blk, frags) == 1)
554 blk_setmask(brec, &mask);
557 printf("blk_freemask: blk %jd sblk %jd off %d mask 0x%X\n",
558 blk, sblk->sb_blk, off, mask);
559 return (mask >> off);
563 * Determine whether it is safe to follow an indirect. It is not safe
564 * if any part of the indirect has been reallocated or the last journal
565 * entry was an allocation. Just allocated indirects may not have valid
566 * pointers yet and all of their children will have their own records.
567 * It is also not safe to follow an indirect if the cg bitmap has been
568 * cleared as a new allocation may write to the block prior to the journal
571 * Returns 1 if it's safe to follow the indirect and 0 otherwise.
574 blk_isindir(ufs2_daddr_t blk, ino_t ino, ufs_lbn_t lbn)
576 struct suj_blk *sblk;
577 struct jblkrec *brec;
579 sblk = blk_lookup(blk, 0);
582 if (TAILQ_EMPTY(&sblk->sb_recs))
584 brec = (struct jblkrec *)TAILQ_LAST(&sblk->sb_recs, srechd)->sr_rec;
585 if (blk_equals(brec, ino, lbn, blk, fs->fs_frag))
586 if (brec->jb_op == JOP_FREEBLK)
587 return (!blk_isfree(blk));
592 * Clear an inode from the cg bitmap. If the inode was already clear return
593 * 0 so the caller knows it does not have to check the inode contents.
596 ino_free(ino_t ino, int mode)
603 cg = ino_to_cg(fs, ino);
604 ino = ino % fs->fs_ipg;
607 inosused = cg_inosused(cgp);
609 * The bitmap may never have made it to the disk so we have to
610 * conditionally clear. We can avoid writing the cg in this case.
612 if (isclr(inosused, ino))
615 clrbit(inosused, ino);
616 if (ino < cgp->cg_irotor)
617 cgp->cg_irotor = ino;
618 cgp->cg_cs.cs_nifree++;
619 if ((mode & IFMT) == IFDIR) {
621 cgp->cg_cs.cs_ndir--;
629 * Free 'frags' frags starting at filesystem block 'bno' skipping any frags
633 blk_free(ufs2_daddr_t bno, int mask, int frags)
635 ufs1_daddr_t fragno, cgbno;
642 printf("Freeing %d frags at blk %jd mask 0x%x\n",
647 cgbno = dtogd(fs, bno);
648 blksfree = cg_blksfree(cgp);
651 * If it's not allocated we only wrote the journal entry
652 * and never the bitmaps. Here we unconditionally clear and
653 * resolve the cg summary later.
655 if (frags == fs->fs_frag && mask == 0) {
656 fragno = fragstoblks(fs, cgbno);
657 ffs_setblock(fs, blksfree, fragno);
661 * deallocate the fragment
663 for (i = 0; i < frags; i++)
664 if ((mask & (1 << i)) == 0 && isclr(blksfree, cgbno +i)) {
666 setbit(blksfree, cgbno + i);
673 * Returns 1 if the whole block starting at 'bno' is marked free and 0
677 blk_isfree(ufs2_daddr_t bno)
681 sc = cg_lookup(dtog(fs, bno));
682 return ffs_isblock(fs, cg_blksfree(sc->sc_cgp), dtogd(fs, bno));
686 * Fetch an indirect block to find the block at a given lbn. The lbn
687 * may be negative to fetch a specific indirect block pointer or positive
688 * to fetch a specific block.
691 indir_blkatoff(ufs2_daddr_t blk, ino_t ino, ufs_lbn_t cur, ufs_lbn_t lbn)
702 level = lbn_level(cur);
704 err_suj("Invalid indir lbn %jd\n", lbn);
705 if (level == 0 && lbn < 0)
706 err_suj("Invalid lbn %jd\n", lbn);
707 bap2 = (void *)dblk_read(blk, fs->fs_bsize);
710 base = -(cur + level);
711 for (i = level; i > 0; i--)
712 lbnadd *= NINDIR(fs);
714 i = (lbn - base) / lbnadd;
716 i = (-lbn - base) / lbnadd;
717 if (i < 0 || i >= NINDIR(fs))
718 err_suj("Invalid indirect index %d produced by lbn %jd\n",
721 cur = base + (i * lbnadd);
723 cur = -(base + (i * lbnadd)) - (level - 1);
724 if (fs->fs_magic == FS_UFS1_MAGIC)
731 err_suj("Invalid lbn %jd at level 0\n", lbn);
732 return indir_blkatoff(blk, ino, cur, lbn);
736 * Finds the disk block address at the specified lbn within the inode
737 * specified by ip. This follows the whole tree and honors di_size and
738 * di_extsize so it is a true test of reachability. The lbn may be
739 * negative if an extattr or indirect block is requested.
742 ino_blkatoff(union dinode *ip, ino_t ino, ufs_lbn_t lbn, int *frags)
750 * Handle extattr blocks first.
752 if (lbn < 0 && lbn >= -UFS_NXADDR) {
754 if (lbn > lblkno(fs, ip->dp2.di_extsize - 1))
756 *frags = numfrags(fs, sblksize(fs, ip->dp2.di_extsize, lbn));
757 return (ip->dp2.di_extb[lbn]);
760 * Now direct and indirect.
762 if (DIP(ip, di_mode) == IFLNK &&
763 DIP(ip, di_size) < fs->fs_maxsymlinklen)
765 if (lbn >= 0 && lbn < UFS_NDADDR) {
766 *frags = numfrags(fs, sblksize(fs, DIP(ip, di_size), lbn));
767 return (DIP(ip, di_db[lbn]));
769 *frags = fs->fs_frag;
771 for (i = 0, tmpval = NINDIR(fs), cur = UFS_NDADDR; i < UFS_NIADDR; i++,
772 tmpval *= NINDIR(fs), cur = next) {
775 return (DIP(ip, di_ib[i]));
777 * Determine whether the lbn in question is within this tree.
779 if (lbn < 0 && -lbn >= next)
781 if (lbn > 0 && lbn >= next)
783 return indir_blkatoff(DIP(ip, di_ib[i]), ino, -cur - i, lbn);
785 err_suj("lbn %jd not in ino\n", lbn);
790 * Determine whether a block exists at a particular lbn in an inode.
791 * Returns 1 if found, 0 if not. lbn may be negative for indirects
795 blk_isat(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, int *frags)
802 if (DIP(ip, di_nlink) == 0 || DIP(ip, di_mode) == 0)
804 nblk = ino_blkatoff(ip, ino, lbn, frags);
806 return (nblk == blk);
810 * Clear the directory entry at diroff that should point to child. Minimal
811 * checking is done and it is assumed that this path was verified with isat.
814 ino_clrat(ino_t parent, off_t diroff, ino_t child)
826 printf("Clearing inode %ju from parent %ju at offset %jd\n",
827 (uintmax_t)child, (uintmax_t)parent, diroff);
829 lbn = lblkno(fs, diroff);
830 doff = blkoff(fs, diroff);
831 dip = ino_read(parent);
832 blk = ino_blkatoff(dip, parent, lbn, &frags);
833 blksize = sblksize(fs, DIP(dip, di_size), lbn);
834 block = dblk_read(blk, blksize);
835 dp = (struct direct *)&block[doff];
836 if (dp->d_ino != child)
837 errx(1, "Inode %ju does not exist in %ju at %jd",
838 (uintmax_t)child, (uintmax_t)parent, diroff);
842 * The actual .. reference count will already have been removed
843 * from the parent by the .. remref record.
848 * Determines whether a pointer to an inode exists within a directory
849 * at a specified offset. Returns the mode of the found entry.
852 ino_isat(ino_t parent, off_t diroff, ino_t child, int *mode, int *isdot)
865 dip = ino_read(parent);
866 *mode = DIP(dip, di_mode);
867 if ((*mode & IFMT) != IFDIR) {
870 * This can happen if the parent inode
874 printf("Directory %ju has bad mode %o\n",
875 (uintmax_t)parent, *mode);
877 printf("Directory %ju has zero mode\n",
882 lbn = lblkno(fs, diroff);
883 doff = blkoff(fs, diroff);
884 blksize = sblksize(fs, DIP(dip, di_size), lbn);
885 if (diroff + DIRECTSIZ(1) > DIP(dip, di_size) || doff >= blksize) {
887 printf("ino %ju absent from %ju due to offset %jd"
888 " exceeding size %jd\n",
889 (uintmax_t)child, (uintmax_t)parent, diroff,
893 blk = ino_blkatoff(dip, parent, lbn, &frags);
896 printf("Sparse directory %ju", (uintmax_t)parent);
899 block = dblk_read(blk, blksize);
901 * Walk through the records from the start of the block to be
902 * certain we hit a valid record and not some junk in the middle
903 * of a file name. Stop when we reach or pass the expected offset.
905 dpoff = rounddown(doff, DIRBLKSIZ);
907 dp = (struct direct *)&block[dpoff];
910 if (dp->d_reclen == 0)
912 dpoff += dp->d_reclen;
913 } while (dpoff <= doff);
914 if (dpoff > fs->fs_bsize)
915 err_suj("Corrupt directory block in dir ino %ju\n",
920 printf("ino %ju not found in %ju, lbn %jd, dpoff %d\n",
921 (uintmax_t)child, (uintmax_t)parent, lbn, dpoff);
925 * We found the item in question. Record the mode and whether it's
926 * a . or .. link for the caller.
928 if (dp->d_ino == child) {
931 else if (dp->d_namlen == 2 &&
932 dp->d_name[0] == '.' && dp->d_name[1] == '.')
934 *mode = DTTOIF(dp->d_type);
938 printf("ino %ju doesn't match dirent ino %ju in parent %ju\n",
939 (uintmax_t)child, (uintmax_t)dp->d_ino, (uintmax_t)parent);
943 #define VISIT_INDIR 0x0001
944 #define VISIT_EXT 0x0002
945 #define VISIT_ROOT 0x0004 /* Operation came via root & valid pointers. */
948 * Read an indirect level which may or may not be linked into an inode.
951 indir_visit(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, uint64_t *frags,
952 ino_visitor visitor, int flags)
963 * Don't visit indirect blocks with contents we can't trust. This
964 * should only happen when indir_visit() is called to complete a
965 * truncate that never finished and not when a pointer is found via
970 level = lbn_level(lbn);
972 err_suj("Invalid level for lbn %jd\n", lbn);
973 if ((flags & VISIT_ROOT) == 0 && blk_isindir(blk, ino, lbn) == 0) {
975 printf("blk %jd ino %ju lbn %jd(%d) is not indir.\n",
976 blk, (uintmax_t)ino, lbn, level);
980 for (i = level; i > 0; i--)
981 lbnadd *= NINDIR(fs);
982 bap1 = (void *)dblk_read(blk, fs->fs_bsize);
984 for (i = 0; i < NINDIR(fs); i++) {
985 if (fs->fs_magic == FS_UFS1_MAGIC)
992 nlbn = -lbn + i * lbnadd;
993 (*frags) += fs->fs_frag;
994 visitor(ino, nlbn, nblk, fs->fs_frag);
996 nlbn = (lbn + 1) - (i * lbnadd);
997 indir_visit(ino, nlbn, nblk, frags, visitor, flags);
1001 if (flags & VISIT_INDIR) {
1002 (*frags) += fs->fs_frag;
1003 visitor(ino, lbn, blk, fs->fs_frag);
1008 * Visit each block in an inode as specified by 'flags' and call a
1009 * callback function. The callback may inspect or free blocks. The
1010 * count of frags found according to the size in the file is returned.
1011 * This is not valid for sparse files but may be used to determine
1012 * the correct di_blocks for a file.
1015 ino_visit(union dinode *ip, ino_t ino, ino_visitor visitor, int flags)
1026 size = DIP(ip, di_size);
1027 mode = DIP(ip, di_mode) & IFMT;
1029 if ((flags & VISIT_EXT) &&
1030 fs->fs_magic == FS_UFS2_MAGIC && ip->dp2.di_extsize) {
1031 for (i = 0; i < UFS_NXADDR; i++) {
1032 if (ip->dp2.di_extb[i] == 0)
1034 frags = sblksize(fs, ip->dp2.di_extsize, i);
1035 frags = numfrags(fs, frags);
1037 visitor(ino, -1 - i, ip->dp2.di_extb[i], frags);
1040 /* Skip datablocks for short links and devices. */
1041 if (mode == IFBLK || mode == IFCHR ||
1042 (mode == IFLNK && size < fs->fs_maxsymlinklen))
1044 for (i = 0; i < UFS_NDADDR; i++) {
1045 if (DIP(ip, di_db[i]) == 0)
1047 frags = sblksize(fs, size, i);
1048 frags = numfrags(fs, frags);
1050 visitor(ino, i, DIP(ip, di_db[i]), frags);
1053 * We know the following indirects are real as we're following
1054 * real pointers to them.
1056 flags |= VISIT_ROOT;
1057 for (i = 0, tmpval = NINDIR(fs), lbn = UFS_NDADDR; i < UFS_NIADDR; i++,
1059 nextlbn = lbn + tmpval;
1060 tmpval *= NINDIR(fs);
1061 if (DIP(ip, di_ib[i]) == 0)
1063 indir_visit(ino, -lbn - i, DIP(ip, di_ib[i]), &fragcnt, visitor,
1070 * Null visitor function used when we just want to count blocks and
1075 null_visit(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, int frags)
1082 * Recalculate di_blocks when we discover that a block allocation or
1083 * free was not successfully completed. The kernel does not roll this back
1084 * because it would be too expensive to compute which indirects were
1085 * reachable at the time the inode was written.
1088 ino_adjblks(struct suj_ino *sino)
1099 /* No need to adjust zero'd inodes. */
1100 if (DIP(ip, di_mode) == 0)
1103 * Visit all blocks and count them as well as recording the last
1104 * valid lbn in the file. If the file size doesn't agree with the
1105 * last lbn we need to truncate to fix it. Otherwise just adjust
1109 frags = ino_visit(ip, ino, null_visit, VISIT_INDIR | VISIT_EXT);
1110 blocks = fsbtodb(fs, frags);
1112 * We assume the size and direct block list is kept coherent by
1113 * softdep. For files that have extended into indirects we truncate
1114 * to the size in the inode or the maximum size permitted by
1115 * populated indirects.
1117 if (visitlbn >= UFS_NDADDR) {
1118 isize = DIP(ip, di_size);
1119 size = lblktosize(fs, visitlbn + 1);
1122 /* Always truncate to free any unpopulated indirects. */
1123 ino_trunc(sino->si_ino, isize);
1126 if (blocks == DIP(ip, di_blocks))
1129 printf("ino %ju adjusting block count from %jd to %jd\n",
1130 (uintmax_t)ino, DIP(ip, di_blocks), blocks);
1131 DIP_SET(ip, di_blocks, blocks);
1136 blk_free_visit(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, int frags)
1139 blk_free(blk, blk_freemask(blk, ino, lbn, frags), frags);
1143 * Free a block or tree of blocks that was previously rooted in ino at
1144 * the given lbn. If the lbn is an indirect all children are freed
1148 blk_free_lbn(ufs2_daddr_t blk, ino_t ino, ufs_lbn_t lbn, int frags, int follow)
1153 mask = blk_freemask(blk, ino, lbn, frags);
1155 if (lbn <= -UFS_NDADDR && follow && mask == 0)
1156 indir_visit(ino, lbn, blk, &resid, blk_free_visit, VISIT_INDIR);
1158 blk_free(blk, mask, frags);
1162 ino_setskip(struct suj_ino *sino, ino_t parent)
1167 if (ino_isat(sino->si_ino, DOTDOT_OFFSET, parent, &mode, &isdot))
1168 sino->si_skipparent = 1;
1172 ino_remref(ino_t parent, ino_t child, uint64_t diroff, int isdotdot)
1174 struct suj_ino *sino;
1175 struct suj_rec *srec;
1176 struct jrefrec *rrec;
1179 * Lookup this inode to see if we have a record for it.
1181 sino = ino_lookup(child, 0);
1183 * Tell any child directories we've already removed their
1184 * parent link cnt. Don't try to adjust our link down again.
1186 if (sino != NULL && isdotdot == 0)
1187 ino_setskip(sino, parent);
1189 * No valid record for this inode. Just drop the on-disk
1192 if (sino == NULL || sino->si_hasrecs == 0) {
1197 * Use ino_adjust() if ino_check() has already processed this
1198 * child. If we lose the last non-dot reference to a
1199 * directory it will be discarded.
1201 if (sino->si_linkadj) {
1204 sino->si_dotlinks--;
1209 * If we haven't yet processed this inode we need to make
1210 * sure we will successfully discover the lost path. If not
1211 * use nlinkadj to remember.
1213 TAILQ_FOREACH(srec, &sino->si_recs, sr_next) {
1214 rrec = (struct jrefrec *)srec->sr_rec;
1215 if (rrec->jr_parent == parent &&
1216 rrec->jr_diroff == diroff)
1219 sino->si_nlinkadj++;
1223 * Free the children of a directory when the directory is discarded.
1226 ino_free_children(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, int frags)
1228 struct suj_ino *sino;
1237 sino = ino_lookup(ino, 0);
1239 skipparent = sino->si_skipparent;
1242 size = lfragtosize(fs, frags);
1243 block = dblk_read(blk, size);
1244 dp = (struct direct *)&block[0];
1245 for (dpoff = 0; dpoff < size && dp->d_reclen; dpoff += dp->d_reclen) {
1246 dp = (struct direct *)&block[dpoff];
1247 if (dp->d_ino == 0 || dp->d_ino == UFS_WINO)
1249 if (dp->d_namlen == 1 && dp->d_name[0] == '.')
1251 isdotdot = dp->d_namlen == 2 && dp->d_name[0] == '.' &&
1252 dp->d_name[1] == '.';
1253 if (isdotdot && skipparent == 1)
1256 printf("Directory %ju removing ino %ju name %s\n",
1257 (uintmax_t)ino, (uintmax_t)dp->d_ino, dp->d_name);
1258 diroff = lblktosize(fs, lbn) + dpoff;
1259 ino_remref(ino, dp->d_ino, diroff, isdotdot);
1264 * Reclaim an inode, freeing all blocks and decrementing all children's
1265 * link counts. Free the inode back to the cg.
1268 ino_reclaim(union dinode *ip, ino_t ino, int mode)
1272 if (ino == UFS_ROOTINO)
1273 err_suj("Attempting to free UFS_ROOTINO\n");
1275 printf("Truncating and freeing ino %ju, nlink %d, mode %o\n",
1276 (uintmax_t)ino, DIP(ip, di_nlink), DIP(ip, di_mode));
1278 /* We are freeing an inode or directory. */
1279 if ((DIP(ip, di_mode) & IFMT) == IFDIR)
1280 ino_visit(ip, ino, ino_free_children, 0);
1281 DIP_SET(ip, di_nlink, 0);
1282 ino_visit(ip, ino, blk_free_visit, VISIT_EXT | VISIT_INDIR);
1283 /* Here we have to clear the inode and release any blocks it holds. */
1284 gen = DIP(ip, di_gen);
1285 if (fs->fs_magic == FS_UFS1_MAGIC)
1286 bzero(ip, sizeof(struct ufs1_dinode));
1288 bzero(ip, sizeof(struct ufs2_dinode));
1289 DIP_SET(ip, di_gen, gen);
1291 ino_free(ino, mode);
1296 * Adjust an inode's link count down by one when a directory goes away.
1307 nlink = DIP(ip, di_nlink);
1308 mode = DIP(ip, di_mode);
1310 err_suj("Inode %d link count %d invalid\n", ino, nlink);
1312 err_suj("Inode %d has a link of %d with 0 mode\n", ino, nlink);
1314 if ((mode & IFMT) == IFDIR)
1318 if (nlink < reqlink) {
1320 printf("ino %ju not enough links to live %d < %d\n",
1321 (uintmax_t)ino, nlink, reqlink);
1322 ino_reclaim(ip, ino, mode);
1325 DIP_SET(ip, di_nlink, nlink);
1330 * Adjust the inode link count to 'nlink'. If the count reaches zero
1334 ino_adjust(struct suj_ino *sino)
1336 struct jrefrec *rrec;
1337 struct suj_rec *srec;
1338 struct suj_ino *stmp;
1347 nlink = sino->si_nlink;
1349 mode = sino->si_mode & IFMT;
1351 * If it's a directory with no dot links, it was truncated before
1352 * the name was cleared. We need to clear the dirent that
1355 if (mode == IFDIR && nlink == 1 && sino->si_dotlinks == 0) {
1356 sino->si_nlink = nlink = 0;
1357 TAILQ_FOREACH(srec, &sino->si_recs, sr_next) {
1358 rrec = (struct jrefrec *)srec->sr_rec;
1359 if (ino_isat(rrec->jr_parent, rrec->jr_diroff, ino,
1360 &recmode, &isdot) == 0)
1362 ino_clrat(rrec->jr_parent, rrec->jr_diroff, ino);
1366 errx(1, "Directory %ju name not found", (uintmax_t)ino);
1369 * If it's a directory with no real names pointing to it go ahead
1370 * and truncate it. This will free any children.
1372 if (mode == IFDIR && nlink - sino->si_dotlinks == 0) {
1373 sino->si_nlink = nlink = 0;
1375 * Mark any .. links so they know not to free this inode
1376 * when they are removed.
1378 TAILQ_FOREACH(srec, &sino->si_recs, sr_next) {
1379 rrec = (struct jrefrec *)srec->sr_rec;
1380 if (rrec->jr_diroff == DOTDOT_OFFSET) {
1381 stmp = ino_lookup(rrec->jr_parent, 0);
1383 ino_setskip(stmp, ino);
1388 mode = DIP(ip, di_mode) & IFMT;
1389 if (nlink > UFS_LINK_MAX)
1390 err_suj("ino %ju nlink manipulation error, new %ju, old %d\n",
1391 (uintmax_t)ino, (uintmax_t)nlink, DIP(ip, di_nlink));
1393 printf("Adjusting ino %ju, nlink %ju, old link %d lastmode %o\n",
1394 (uintmax_t)ino, (uintmax_t)nlink, DIP(ip, di_nlink),
1398 printf("ino %ju, zero inode freeing bitmap\n",
1400 ino_free(ino, sino->si_mode);
1403 /* XXX Should be an assert? */
1404 if (mode != sino->si_mode && debug)
1405 printf("ino %ju, mode %o != %o\n",
1406 (uintmax_t)ino, mode, sino->si_mode);
1407 if ((mode & IFMT) == IFDIR)
1411 /* If the inode doesn't have enough links to live, free it. */
1412 if (nlink < reqlink) {
1414 printf("ino %ju not enough links to live %ju < %ju\n",
1415 (uintmax_t)ino, (uintmax_t)nlink,
1416 (uintmax_t)reqlink);
1417 ino_reclaim(ip, ino, mode);
1420 /* If required write the updated link count. */
1421 if (DIP(ip, di_nlink) == nlink) {
1423 printf("ino %ju, link matches, skipping.\n",
1427 DIP_SET(ip, di_nlink, nlink);
1432 * Truncate some or all blocks in an indirect, freeing any that are required
1433 * and zeroing the indirect.
1436 indir_trunc(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, ufs_lbn_t lastlbn)
1451 level = lbn_level(lbn);
1453 err_suj("Invalid level for lbn %jd\n", lbn);
1455 for (i = level; i > 0; i--)
1456 lbnadd *= NINDIR(fs);
1457 bap1 = (void *)dblk_read(blk, fs->fs_bsize);
1458 bap2 = (void *)bap1;
1459 for (i = 0; i < NINDIR(fs); i++) {
1460 if (fs->fs_magic == FS_UFS1_MAGIC)
1467 nlbn = (lbn + 1) - (i * lbnadd);
1469 * Calculate the lbn of the next indirect to
1470 * determine if any of this indirect must be
1473 next = -(lbn + level) + ((i+1) * lbnadd);
1474 if (next <= lastlbn)
1476 indir_trunc(ino, nlbn, nblk, lastlbn);
1477 /* If all of this indirect was reclaimed, free it. */
1478 nlbn = next - lbnadd;
1482 nlbn = -lbn + i * lbnadd;
1487 blk_free(nblk, 0, fs->fs_frag);
1488 if (fs->fs_magic == FS_UFS1_MAGIC)
1498 * Truncate an inode to the minimum of the given size or the last populated
1499 * block after any over size have been discarded. The kernel would allocate
1500 * the last block in the file but fsck does not and neither do we. This
1501 * code never extends files, only shrinks them.
1504 ino_trunc(ino_t ino, off_t size)
1508 uint64_t totalfrags;
1520 mode = DIP(ip, di_mode) & IFMT;
1521 cursize = DIP(ip, di_size);
1523 printf("Truncating ino %ju, mode %o to size %jd from size %jd\n",
1524 (uintmax_t)ino, mode, size, cursize);
1526 /* Skip datablocks for short links and devices. */
1527 if (mode == 0 || mode == IFBLK || mode == IFCHR ||
1528 (mode == IFLNK && cursize < fs->fs_maxsymlinklen))
1533 lastlbn = lblkno(fs, blkroundup(fs, size));
1534 for (i = lastlbn; i < UFS_NDADDR; i++) {
1535 if (DIP(ip, di_db[i]) == 0)
1537 frags = sblksize(fs, cursize, i);
1538 frags = numfrags(fs, frags);
1539 blk_free(DIP(ip, di_db[i]), 0, frags);
1540 DIP_SET(ip, di_db[i], 0);
1543 * Follow indirect blocks, freeing anything required.
1545 for (i = 0, tmpval = NINDIR(fs), lbn = UFS_NDADDR; i < UFS_NIADDR; i++,
1547 nextlbn = lbn + tmpval;
1548 tmpval *= NINDIR(fs);
1549 /* If we're not freeing any in this indirect range skip it. */
1550 if (lastlbn >= nextlbn)
1552 if (DIP(ip, di_ib[i]) == 0)
1554 indir_trunc(ino, -lbn - i, DIP(ip, di_ib[i]), lastlbn);
1555 /* If we freed everything in this indirect free the indir. */
1558 blk_free(DIP(ip, di_ib[i]), 0, fs->fs_frag);
1559 DIP_SET(ip, di_ib[i], 0);
1563 * Now that we've freed any whole blocks that exceed the desired
1564 * truncation size, figure out how many blocks remain and what the
1565 * last populated lbn is. We will set the size to this last lbn
1566 * rather than worrying about allocating the final lbn as the kernel
1567 * would've done. This is consistent with normal fsck behavior.
1570 totalfrags = ino_visit(ip, ino, null_visit, VISIT_INDIR | VISIT_EXT);
1571 if (size > lblktosize(fs, visitlbn + 1))
1572 size = lblktosize(fs, visitlbn + 1);
1574 * If we're truncating direct blocks we have to adjust frags
1577 if (visitlbn < UFS_NDADDR && totalfrags) {
1578 long oldspace, newspace;
1580 bn = DIP(ip, di_db[visitlbn]);
1582 err_suj("Bad blk at ino %ju lbn %jd\n",
1583 (uintmax_t)ino, visitlbn);
1584 oldspace = sblksize(fs, cursize, visitlbn);
1585 newspace = sblksize(fs, size, visitlbn);
1586 if (oldspace != newspace) {
1587 bn += numfrags(fs, newspace);
1588 frags = numfrags(fs, oldspace - newspace);
1589 blk_free(bn, 0, frags);
1590 totalfrags -= frags;
1593 DIP_SET(ip, di_blocks, fsbtodb(fs, totalfrags));
1594 DIP_SET(ip, di_size, size);
1596 * If we've truncated into the middle of a block or frag we have
1597 * to zero it here. Otherwise the file could extend into
1598 * uninitialized space later.
1600 off = blkoff(fs, size);
1601 if (off && DIP(ip, di_mode) != IFDIR) {
1605 bn = ino_blkatoff(ip, ino, visitlbn, &frags);
1607 err_suj("Block missing from ino %ju at lbn %jd\n",
1608 (uintmax_t)ino, visitlbn);
1609 clrsize = frags * fs->fs_fsize;
1610 buf = dblk_read(bn, clrsize);
1613 bzero(buf, clrsize);
1620 * Process records available for one inode and determine whether the
1621 * link count is correct or needs adjusting.
1624 ino_check(struct suj_ino *sino)
1626 struct suj_rec *srec;
1627 struct jrefrec *rrec;
1637 if (sino->si_hasrecs == 0)
1640 rrec = (struct jrefrec *)TAILQ_FIRST(&sino->si_recs)->sr_rec;
1641 nlink = rrec->jr_nlink;
1644 removes = sino->si_nlinkadj;
1645 TAILQ_FOREACH(srec, &sino->si_recs, sr_next) {
1646 rrec = (struct jrefrec *)srec->sr_rec;
1647 isat = ino_isat(rrec->jr_parent, rrec->jr_diroff,
1648 rrec->jr_ino, &mode, &isdot);
1649 if (isat && (mode & IFMT) != (rrec->jr_mode & IFMT))
1650 err_suj("Inode mode/directory type mismatch %o != %o\n",
1651 mode, rrec->jr_mode);
1653 printf("jrefrec: op %d ino %ju, nlink %ju, parent %ju, "
1654 "diroff %jd, mode %o, isat %d, isdot %d\n",
1655 rrec->jr_op, (uintmax_t)rrec->jr_ino,
1656 (uintmax_t)rrec->jr_nlink,
1657 (uintmax_t)rrec->jr_parent,
1658 (uintmax_t)rrec->jr_diroff,
1659 rrec->jr_mode, isat, isdot);
1660 mode = rrec->jr_mode & IFMT;
1661 if (rrec->jr_op == JOP_REMREF)
1668 * The number of links that remain are the starting link count
1669 * subtracted by the total number of removes with the total
1670 * links discovered back in. An incomplete remove thus
1671 * makes no change to the link count but an add increases
1676 "ino %ju nlink %ju newlinks %ju removes %ju dotlinks %ju\n",
1677 (uintmax_t)ino, (uintmax_t)nlink, (uintmax_t)newlinks,
1678 (uintmax_t)removes, (uintmax_t)dotlinks);
1681 sino->si_linkadj = 1;
1682 sino->si_nlink = nlink;
1683 sino->si_dotlinks = dotlinks;
1684 sino->si_mode = mode;
1689 * Process records available for one block and determine whether it is
1690 * still allocated and whether the owning inode needs to be updated or
1694 blk_check(struct suj_blk *sblk)
1696 struct suj_rec *srec;
1697 struct jblkrec *brec;
1698 struct suj_ino *sino;
1705 * Each suj_blk actually contains records for any fragments in that
1706 * block. As a result we must evaluate each record individually.
1709 TAILQ_FOREACH(srec, &sblk->sb_recs, sr_next) {
1710 brec = (struct jblkrec *)srec->sr_rec;
1711 frags = brec->jb_frags;
1712 blk = brec->jb_blkno + brec->jb_oldfrags;
1713 isat = blk_isat(brec->jb_ino, brec->jb_lbn, blk, &frags);
1714 if (sino == NULL || sino->si_ino != brec->jb_ino) {
1715 sino = ino_lookup(brec->jb_ino, 1);
1716 sino->si_blkadj = 1;
1719 printf("op %d blk %jd ino %ju lbn %jd frags %d isat %d (%d)\n",
1720 brec->jb_op, blk, (uintmax_t)brec->jb_ino,
1721 brec->jb_lbn, brec->jb_frags, isat, frags);
1723 * If we found the block at this address we still have to
1724 * determine if we need to free the tail end that was
1725 * added by adding contiguous fragments from the same block.
1728 if (frags == brec->jb_frags)
1730 mask = blk_freemask(blk, brec->jb_ino, brec->jb_lbn,
1734 frags = brec->jb_frags - frags;
1735 blk_free(blk, mask, frags);
1739 * The block wasn't found, attempt to free it. It won't be
1740 * freed if it was actually reallocated. If this was an
1741 * allocation we don't want to follow indirects as they
1742 * may not be written yet. Any children of the indirect will
1743 * have their own records. If it's a free we need to
1744 * recursively free children.
1746 blk_free_lbn(blk, brec->jb_ino, brec->jb_lbn, brec->jb_frags,
1747 brec->jb_op == JOP_FREEBLK);
1752 * Walk the list of inode records for this cg and resolve moved and duplicate
1753 * inode references now that we have a complete picture.
1756 cg_build(struct suj_cg *sc)
1758 struct suj_ino *sino;
1761 for (i = 0; i < SUJ_HASHSIZE; i++)
1762 LIST_FOREACH(sino, &sc->sc_inohash[i], si_next)
1767 * Handle inodes requiring truncation. This must be done prior to
1768 * looking up any inodes in directories.
1771 cg_trunc(struct suj_cg *sc)
1773 struct suj_ino *sino;
1776 for (i = 0; i < SUJ_HASHSIZE; i++) {
1777 LIST_FOREACH(sino, &sc->sc_inohash[i], si_next) {
1778 if (sino->si_trunc) {
1779 ino_trunc(sino->si_ino,
1780 sino->si_trunc->jt_size);
1781 sino->si_blkadj = 0;
1782 sino->si_trunc = NULL;
1784 if (sino->si_blkadj)
1791 cg_adj_blk(struct suj_cg *sc)
1793 struct suj_ino *sino;
1796 for (i = 0; i < SUJ_HASHSIZE; i++) {
1797 LIST_FOREACH(sino, &sc->sc_inohash[i], si_next) {
1798 if (sino->si_blkadj)
1805 * Free any partially allocated blocks and then resolve inode block
1809 cg_check_blk(struct suj_cg *sc)
1811 struct suj_blk *sblk;
1815 for (i = 0; i < SUJ_HASHSIZE; i++)
1816 LIST_FOREACH(sblk, &sc->sc_blkhash[i], sb_next)
1821 * Walk the list of inode records for this cg, recovering any
1822 * changes which were not complete at the time of crash.
1825 cg_check_ino(struct suj_cg *sc)
1827 struct suj_ino *sino;
1830 for (i = 0; i < SUJ_HASHSIZE; i++)
1831 LIST_FOREACH(sino, &sc->sc_inohash[i], si_next)
1836 * Write a potentially dirty cg. Recalculate the summary information and
1837 * update the superblock summary.
1840 cg_write(struct suj_cg *sc)
1842 ufs1_daddr_t fragno, cgbno, maxbno;
1848 if (sc->sc_dirty == 0)
1851 * Fix the frag and cluster summary.
1854 cgp->cg_cs.cs_nbfree = 0;
1855 cgp->cg_cs.cs_nffree = 0;
1856 bzero(&cgp->cg_frsum, sizeof(cgp->cg_frsum));
1857 maxbno = fragstoblks(fs, fs->fs_fpg);
1858 if (fs->fs_contigsumsize > 0) {
1859 for (i = 1; i <= fs->fs_contigsumsize; i++)
1860 cg_clustersum(cgp)[i] = 0;
1861 bzero(cg_clustersfree(cgp), howmany(maxbno, CHAR_BIT));
1863 blksfree = cg_blksfree(cgp);
1864 for (cgbno = 0; cgbno < maxbno; cgbno++) {
1865 if (ffs_isfreeblock(fs, blksfree, cgbno))
1867 if (ffs_isblock(fs, blksfree, cgbno)) {
1868 ffs_clusteracct(fs, cgp, cgbno, 1);
1869 cgp->cg_cs.cs_nbfree++;
1872 fragno = blkstofrags(fs, cgbno);
1873 blk = blkmap(fs, blksfree, fragno);
1874 ffs_fragacct(fs, blk, cgp->cg_frsum, 1);
1875 for (i = 0; i < fs->fs_frag; i++)
1876 if (isset(blksfree, fragno + i))
1877 cgp->cg_cs.cs_nffree++;
1880 * Update the superblock cg summary from our now correct values
1881 * before writing the block.
1883 fs->fs_cs(fs, sc->sc_cgx) = cgp->cg_cs;
1884 if (cgput(&disk, cgp) == -1)
1885 err_suj("Unable to write cylinder group %d\n", sc->sc_cgx);
1889 * Write out any modified inodes.
1892 cg_write_inos(struct suj_cg *sc)
1894 struct ino_blk *iblk;
1897 for (i = 0; i < SUJ_HASHSIZE; i++)
1898 LIST_FOREACH(iblk, &sc->sc_iblkhash[i], ib_next)
1904 cg_apply(void (*apply)(struct suj_cg *))
1909 for (i = 0; i < SUJ_HASHSIZE; i++)
1910 LIST_FOREACH(scg, &cghash[i], sc_next)
1915 * Process the unlinked but referenced file list. Freeing all inodes.
1925 ino = fs->fs_sujfree;
1929 mode = DIP(ip, di_mode) & IFMT;
1930 inon = DIP(ip, di_freelink);
1931 DIP_SET(ip, di_freelink, 0);
1933 * XXX Should this be an errx?
1935 if (DIP(ip, di_nlink) == 0) {
1937 printf("Freeing unlinked ino %ju mode %o\n",
1938 (uintmax_t)ino, mode);
1939 ino_reclaim(ip, ino, mode);
1941 printf("Skipping ino %ju mode %o with link %d\n",
1942 (uintmax_t)ino, mode, DIP(ip, di_nlink));
1948 * Append a new record to the list of records requiring processing.
1951 ino_append(union jrec *rec)
1953 struct jrefrec *refrec;
1954 struct jmvrec *mvrec;
1955 struct suj_ino *sino;
1956 struct suj_rec *srec;
1958 mvrec = &rec->rec_jmvrec;
1959 refrec = &rec->rec_jrefrec;
1960 if (debug && mvrec->jm_op == JOP_MVREF)
1961 printf("ino move: ino %ju, parent %ju, "
1962 "diroff %jd, oldoff %jd\n",
1963 (uintmax_t)mvrec->jm_ino, (uintmax_t)mvrec->jm_parent,
1964 (uintmax_t)mvrec->jm_newoff, (uintmax_t)mvrec->jm_oldoff);
1966 (refrec->jr_op == JOP_ADDREF || refrec->jr_op == JOP_REMREF))
1967 printf("ino ref: op %d, ino %ju, nlink %ju, "
1968 "parent %ju, diroff %jd\n",
1969 refrec->jr_op, (uintmax_t)refrec->jr_ino,
1970 (uintmax_t)refrec->jr_nlink,
1971 (uintmax_t)refrec->jr_parent, (uintmax_t)refrec->jr_diroff);
1972 sino = ino_lookup(((struct jrefrec *)rec)->jr_ino, 1);
1973 sino->si_hasrecs = 1;
1974 srec = errmalloc(sizeof(*srec));
1976 TAILQ_INSERT_TAIL(&sino->si_newrecs, srec, sr_next);
1980 * Add a reference adjustment to the sino list and eliminate dups. The
1981 * primary loop in ino_build_ref() checks for dups but new ones may be
1982 * created as a result of offset adjustments.
1985 ino_add_ref(struct suj_ino *sino, struct suj_rec *srec)
1987 struct jrefrec *refrec;
1988 struct suj_rec *srn;
1989 struct jrefrec *rrn;
1991 refrec = (struct jrefrec *)srec->sr_rec;
1993 * We walk backwards so that the oldest link count is preserved. If
1994 * an add record conflicts with a remove keep the remove. Redundant
1995 * removes are eliminated in ino_build_ref. Otherwise we keep the
1996 * oldest record at a given location.
1998 for (srn = TAILQ_LAST(&sino->si_recs, srechd); srn;
1999 srn = TAILQ_PREV(srn, srechd, sr_next)) {
2000 rrn = (struct jrefrec *)srn->sr_rec;
2001 if (rrn->jr_parent != refrec->jr_parent ||
2002 rrn->jr_diroff != refrec->jr_diroff)
2004 if (rrn->jr_op == JOP_REMREF || refrec->jr_op == JOP_ADDREF) {
2005 rrn->jr_mode = refrec->jr_mode;
2011 * Replace the record in place with the old nlink in case
2012 * we replace the head of the list. Abandon srec as a dup.
2014 refrec->jr_nlink = rrn->jr_nlink;
2015 srn->sr_rec = srec->sr_rec;
2018 TAILQ_INSERT_TAIL(&sino->si_recs, srec, sr_next);
2022 * Create a duplicate of a reference at a previous location.
2025 ino_dup_ref(struct suj_ino *sino, struct jrefrec *refrec, off_t diroff)
2027 struct jrefrec *rrn;
2028 struct suj_rec *srn;
2030 rrn = errmalloc(sizeof(*refrec));
2032 rrn->jr_op = JOP_ADDREF;
2033 rrn->jr_diroff = diroff;
2034 srn = errmalloc(sizeof(*srn));
2035 srn->sr_rec = (union jrec *)rrn;
2036 ino_add_ref(sino, srn);
2040 * Add a reference to the list at all known locations. We follow the offset
2041 * changes for a single instance and create duplicate add refs at each so
2042 * that we can tolerate any version of the directory block. Eliminate
2043 * removes which collide with adds that are seen in the journal. They should
2044 * not adjust the link count down.
2047 ino_build_ref(struct suj_ino *sino, struct suj_rec *srec)
2049 struct jrefrec *refrec;
2050 struct jmvrec *mvrec;
2051 struct suj_rec *srp;
2052 struct suj_rec *srn;
2053 struct jrefrec *rrn;
2056 refrec = (struct jrefrec *)srec->sr_rec;
2058 * Search for a mvrec that matches this offset. Whether it's an add
2059 * or a remove we can delete the mvref after creating a dup record in
2062 if (!TAILQ_EMPTY(&sino->si_movs)) {
2063 diroff = refrec->jr_diroff;
2064 for (srn = TAILQ_LAST(&sino->si_movs, srechd); srn; srn = srp) {
2065 srp = TAILQ_PREV(srn, srechd, sr_next);
2066 mvrec = (struct jmvrec *)srn->sr_rec;
2067 if (mvrec->jm_parent != refrec->jr_parent ||
2068 mvrec->jm_newoff != diroff)
2070 diroff = mvrec->jm_oldoff;
2071 TAILQ_REMOVE(&sino->si_movs, srn, sr_next);
2073 ino_dup_ref(sino, refrec, diroff);
2077 * If a remove wasn't eliminated by an earlier add just append it to
2080 if (refrec->jr_op == JOP_REMREF) {
2081 ino_add_ref(sino, srec);
2085 * Walk the list of records waiting to be added to the list. We
2086 * must check for moves that apply to our current offset and remove
2087 * them from the list. Remove any duplicates to eliminate removes
2088 * with corresponding adds.
2090 TAILQ_FOREACH_SAFE(srn, &sino->si_newrecs, sr_next, srp) {
2091 switch (srn->sr_rec->rec_jrefrec.jr_op) {
2094 * This should actually be an error we should
2095 * have a remove for every add journaled.
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);
2105 * Once we remove the current iteration of the
2106 * record at this address we're done.
2108 rrn = (struct jrefrec *)srn->sr_rec;
2109 if (rrn->jr_parent != refrec->jr_parent ||
2110 rrn->jr_diroff != refrec->jr_diroff)
2112 TAILQ_REMOVE(&sino->si_newrecs, srn, sr_next);
2113 ino_add_ref(sino, srec);
2117 * Update our diroff based on any moves that match
2118 * and remove the move.
2120 mvrec = (struct jmvrec *)srn->sr_rec;
2121 if (mvrec->jm_parent != refrec->jr_parent ||
2122 mvrec->jm_oldoff != refrec->jr_diroff)
2124 ino_dup_ref(sino, refrec, mvrec->jm_oldoff);
2125 refrec->jr_diroff = mvrec->jm_newoff;
2126 TAILQ_REMOVE(&sino->si_newrecs, srn, sr_next);
2129 err_suj("ino_build_ref: Unknown op %d\n",
2130 srn->sr_rec->rec_jrefrec.jr_op);
2133 ino_add_ref(sino, srec);
2137 * Walk the list of new records and add them in-order resolving any
2138 * dups and adjusted offsets.
2141 ino_build(struct suj_ino *sino)
2143 struct suj_rec *srec;
2145 while ((srec = TAILQ_FIRST(&sino->si_newrecs)) != NULL) {
2146 TAILQ_REMOVE(&sino->si_newrecs, srec, sr_next);
2147 switch (srec->sr_rec->rec_jrefrec.jr_op) {
2150 ino_build_ref(sino, srec);
2154 * Add this mvrec to the queue of pending mvs.
2156 TAILQ_INSERT_TAIL(&sino->si_movs, srec, sr_next);
2159 err_suj("ino_build: Unknown op %d\n",
2160 srec->sr_rec->rec_jrefrec.jr_op);
2163 if (TAILQ_EMPTY(&sino->si_recs))
2164 sino->si_hasrecs = 0;
2168 * Modify journal records so they refer to the base block number
2169 * and a start and end frag range. This is to facilitate the discovery
2170 * of overlapping fragment allocations.
2173 blk_build(struct jblkrec *blkrec)
2175 struct suj_rec *srec;
2176 struct suj_blk *sblk;
2177 struct jblkrec *blkrn;
2182 printf("blk_build: op %d blkno %jd frags %d oldfrags %d "
2183 "ino %ju lbn %jd\n",
2184 blkrec->jb_op, (uintmax_t)blkrec->jb_blkno,
2185 blkrec->jb_frags, blkrec->jb_oldfrags,
2186 (uintmax_t)blkrec->jb_ino, (uintmax_t)blkrec->jb_lbn);
2188 blk = blknum(fs, blkrec->jb_blkno);
2189 frag = fragnum(fs, blkrec->jb_blkno);
2190 sblk = blk_lookup(blk, 1);
2192 * Rewrite the record using oldfrags to indicate the offset into
2193 * the block. Leave jb_frags as the actual allocated count.
2195 blkrec->jb_blkno -= frag;
2196 blkrec->jb_oldfrags = frag;
2197 if (blkrec->jb_oldfrags + blkrec->jb_frags > fs->fs_frag)
2198 err_suj("Invalid fragment count %d oldfrags %d\n",
2199 blkrec->jb_frags, frag);
2201 * Detect dups. If we detect a dup we always discard the oldest
2202 * record as it is superseded by the new record. This speeds up
2203 * later stages but also eliminates free records which are used
2204 * to indicate that the contents of indirects can be trusted.
2206 TAILQ_FOREACH(srec, &sblk->sb_recs, sr_next) {
2207 blkrn = (struct jblkrec *)srec->sr_rec;
2208 if (blkrn->jb_ino != blkrec->jb_ino ||
2209 blkrn->jb_lbn != blkrec->jb_lbn ||
2210 blkrn->jb_blkno != blkrec->jb_blkno ||
2211 blkrn->jb_frags != blkrec->jb_frags ||
2212 blkrn->jb_oldfrags != blkrec->jb_oldfrags)
2215 printf("Removed dup.\n");
2216 /* Discard the free which is a dup with an alloc. */
2217 if (blkrec->jb_op == JOP_FREEBLK)
2219 TAILQ_REMOVE(&sblk->sb_recs, srec, sr_next);
2223 srec = errmalloc(sizeof(*srec));
2224 srec->sr_rec = (union jrec *)blkrec;
2225 TAILQ_INSERT_TAIL(&sblk->sb_recs, srec, sr_next);
2229 ino_build_trunc(struct jtrncrec *rec)
2231 struct suj_ino *sino;
2234 printf("ino_build_trunc: op %d ino %ju, size %jd\n",
2235 rec->jt_op, (uintmax_t)rec->jt_ino,
2236 (uintmax_t)rec->jt_size);
2237 sino = ino_lookup(rec->jt_ino, 1);
2238 if (rec->jt_op == JOP_SYNC) {
2239 sino->si_trunc = NULL;
2242 if (sino->si_trunc == NULL || sino->si_trunc->jt_size > rec->jt_size)
2243 sino->si_trunc = rec;
2247 * Build up tables of the operations we need to recover.
2252 struct suj_seg *seg;
2257 TAILQ_FOREACH(seg, &allsegs, ss_next) {
2259 printf("seg %jd has %d records, oldseq %jd.\n",
2260 seg->ss_rec.jsr_seq, seg->ss_rec.jsr_cnt,
2261 seg->ss_rec.jsr_oldest);
2263 rec = (union jrec *)seg->ss_blk;
2264 for (i = 0; i < seg->ss_rec.jsr_cnt; off += JREC_SIZE, rec++) {
2265 /* skip the segrec. */
2266 if ((off % real_dev_bsize) == 0)
2268 switch (rec->rec_jrefrec.jr_op) {
2276 blk_build((struct jblkrec *)rec);
2280 ino_build_trunc((struct jtrncrec *)rec);
2283 err_suj("Unknown journal operation %d (%d)\n",
2284 rec->rec_jrefrec.jr_op, off);
2292 * Prune the journal segments to those we care about based on the
2293 * oldest sequence in the newest segment. Order the segment list
2294 * based on sequence number.
2299 struct suj_seg *seg;
2300 struct suj_seg *segn;
2305 printf("Pruning up to %jd\n", oldseq);
2306 /* First free the expired segments. */
2307 TAILQ_FOREACH_SAFE(seg, &allsegs, ss_next, segn) {
2308 if (seg->ss_rec.jsr_seq >= oldseq)
2310 TAILQ_REMOVE(&allsegs, seg, ss_next);
2314 /* Next ensure that segments are ordered properly. */
2315 seg = TAILQ_FIRST(&allsegs);
2318 printf("Empty journal\n");
2321 newseq = seg->ss_rec.jsr_seq;
2323 seg = TAILQ_LAST(&allsegs, seghd);
2324 if (seg->ss_rec.jsr_seq >= newseq)
2326 TAILQ_REMOVE(&allsegs, seg, ss_next);
2327 TAILQ_INSERT_HEAD(&allsegs, seg, ss_next);
2328 newseq = seg->ss_rec.jsr_seq;
2331 if (newseq != oldseq) {
2332 TAILQ_FOREACH(seg, &allsegs, ss_next) {
2333 printf("%jd, ", seg->ss_rec.jsr_seq);
2336 err_suj("Journal file sequence mismatch %jd != %jd\n",
2340 * The kernel may asynchronously write segments which can create
2341 * gaps in the sequence space. Throw away any segments after the
2342 * gap as the kernel guarantees only those that are contiguously
2343 * reachable are marked as completed.
2346 TAILQ_FOREACH_SAFE(seg, &allsegs, ss_next, segn) {
2347 if (!discard && newseq++ == seg->ss_rec.jsr_seq) {
2348 jrecs += seg->ss_rec.jsr_cnt;
2349 jbytes += seg->ss_rec.jsr_blocks * real_dev_bsize;
2354 printf("Journal order mismatch %jd != %jd pruning\n",
2355 newseq-1, seg->ss_rec.jsr_seq);
2356 TAILQ_REMOVE(&allsegs, seg, ss_next);
2361 printf("Processing journal segments from %jd to %jd\n",
2366 * Verify the journal inode before attempting to read records.
2369 suj_verifyino(union dinode *ip)
2372 if (DIP(ip, di_nlink) != 1) {
2373 printf("Invalid link count %d for journal inode %ju\n",
2374 DIP(ip, di_nlink), (uintmax_t)sujino);
2378 if ((DIP(ip, di_flags) & (SF_IMMUTABLE | SF_NOUNLINK)) !=
2379 (SF_IMMUTABLE | SF_NOUNLINK)) {
2380 printf("Invalid flags 0x%X for journal inode %ju\n",
2381 DIP(ip, di_flags), (uintmax_t)sujino);
2385 if (DIP(ip, di_mode) != (IFREG | IREAD)) {
2386 printf("Invalid mode %o for journal inode %ju\n",
2387 DIP(ip, di_mode), (uintmax_t)sujino);
2391 if (DIP(ip, di_size) < SUJ_MIN) {
2392 printf("Invalid size %jd for journal inode %ju\n",
2393 DIP(ip, di_size), (uintmax_t)sujino);
2397 if (DIP(ip, di_modrev) != fs->fs_mtime) {
2398 printf("Journal timestamp does not match fs mount time\n");
2406 struct jextent *jb_extent; /* Extent array. */
2407 int jb_avail; /* Available extents. */
2408 int jb_used; /* Last used extent. */
2409 int jb_head; /* Allocator head. */
2410 int jb_off; /* Allocator extent offset. */
2413 ufs2_daddr_t je_daddr; /* Disk block address. */
2414 int je_blocks; /* Disk block count. */
2417 static struct jblocks *suj_jblocks;
2419 static struct jblocks *
2420 jblocks_create(void)
2422 struct jblocks *jblocks;
2425 jblocks = errmalloc(sizeof(*jblocks));
2426 jblocks->jb_avail = 10;
2427 jblocks->jb_used = 0;
2428 jblocks->jb_head = 0;
2429 jblocks->jb_off = 0;
2430 size = sizeof(struct jextent) * jblocks->jb_avail;
2431 jblocks->jb_extent = errmalloc(size);
2432 bzero(jblocks->jb_extent, size);
2438 * Return the next available disk block and the amount of contiguous
2439 * free space it contains.
2442 jblocks_next(struct jblocks *jblocks, int bytes, int *actual)
2444 struct jextent *jext;
2449 blocks = bytes / disk.d_bsize;
2450 jext = &jblocks->jb_extent[jblocks->jb_head];
2451 freecnt = jext->je_blocks - jblocks->jb_off;
2453 jblocks->jb_off = 0;
2454 if (++jblocks->jb_head > jblocks->jb_used)
2456 jext = &jblocks->jb_extent[jblocks->jb_head];
2457 freecnt = jext->je_blocks;
2459 if (freecnt > blocks)
2461 *actual = freecnt * disk.d_bsize;
2462 daddr = jext->je_daddr + jblocks->jb_off;
2468 * Advance the allocation head by a specified number of bytes, consuming
2469 * one journal segment.
2472 jblocks_advance(struct jblocks *jblocks, int bytes)
2475 jblocks->jb_off += bytes / disk.d_bsize;
2479 jblocks_destroy(struct jblocks *jblocks)
2482 free(jblocks->jb_extent);
2487 jblocks_add(struct jblocks *jblocks, ufs2_daddr_t daddr, int blocks)
2489 struct jextent *jext;
2492 jext = &jblocks->jb_extent[jblocks->jb_used];
2493 /* Adding the first block. */
2494 if (jext->je_daddr == 0) {
2495 jext->je_daddr = daddr;
2496 jext->je_blocks = blocks;
2499 /* Extending the last extent. */
2500 if (jext->je_daddr + jext->je_blocks == daddr) {
2501 jext->je_blocks += blocks;
2504 /* Adding a new extent. */
2505 if (++jblocks->jb_used == jblocks->jb_avail) {
2506 jblocks->jb_avail *= 2;
2507 size = sizeof(struct jextent) * jblocks->jb_avail;
2508 jext = errmalloc(size);
2510 bcopy(jblocks->jb_extent, jext,
2511 sizeof(struct jextent) * jblocks->jb_used);
2512 free(jblocks->jb_extent);
2513 jblocks->jb_extent = jext;
2515 jext = &jblocks->jb_extent[jblocks->jb_used];
2516 jext->je_daddr = daddr;
2517 jext->je_blocks = blocks;
2523 * Add a file block from the journal to the extent map. We can't read
2524 * each file block individually because the kernel treats it as a circular
2525 * buffer and segments may span mutliple contiguous blocks.
2528 suj_add_block(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, int frags)
2531 jblocks_add(suj_jblocks, fsbtodb(fs, blk), fsbtodb(fs, frags));
2537 uint8_t block[1 * 1024 * 1024];
2538 struct suj_seg *seg;
2539 struct jsegrec *recn;
2540 struct jsegrec *rec;
2549 * Read records until we exhaust the journal space. If we find
2550 * an invalid record we start searching for a valid segment header
2551 * at the next block. This is because we don't have a head/tail
2552 * pointer and must recover the information indirectly. At the gap
2553 * between the head and tail we won't necessarily have a valid
2558 size = sizeof(block);
2559 blk = jblocks_next(suj_jblocks, size, &readsize);
2564 * Read 1MB at a time and scan for records within this block.
2566 if (bread(&disk, blk, &block, size) == -1) {
2567 err_suj("Error reading journal block %jd\n",
2570 for (rec = (void *)block; size; size -= recsize,
2571 rec = (struct jsegrec *)((uintptr_t)rec + recsize)) {
2572 recsize = real_dev_bsize;
2573 if (rec->jsr_time != fs->fs_mtime) {
2575 printf("Rec time %jd != fs mtime %jd\n",
2576 rec->jsr_time, fs->fs_mtime);
2577 jblocks_advance(suj_jblocks, recsize);
2580 if (rec->jsr_cnt == 0) {
2582 printf("Found illegal count %d\n",
2584 jblocks_advance(suj_jblocks, recsize);
2587 blocks = rec->jsr_blocks;
2588 recsize = blocks * real_dev_bsize;
2589 if (recsize > size) {
2591 * We may just have run out of buffer, restart
2592 * the loop to re-read from this spot.
2594 if (size < fs->fs_bsize &&
2596 recsize <= fs->fs_bsize)
2599 printf("Found invalid segsize %d > %d\n",
2601 recsize = real_dev_bsize;
2602 jblocks_advance(suj_jblocks, recsize);
2606 * Verify that all blocks in the segment are present.
2608 for (i = 1; i < blocks; i++) {
2609 recn = (void *)((uintptr_t)rec) + i *
2611 if (recn->jsr_seq == rec->jsr_seq &&
2612 recn->jsr_time == rec->jsr_time)
2615 printf("Incomplete record %jd (%d)\n",
2617 recsize = i * real_dev_bsize;
2618 jblocks_advance(suj_jblocks, recsize);
2621 seg = errmalloc(sizeof(*seg));
2622 seg->ss_blk = errmalloc(recsize);
2624 bcopy((void *)rec, seg->ss_blk, recsize);
2625 if (rec->jsr_oldest > oldseq)
2626 oldseq = rec->jsr_oldest;
2627 TAILQ_INSERT_TAIL(&allsegs, seg, ss_next);
2628 jblocks_advance(suj_jblocks, recsize);
2634 * Search a directory block for the SUJ_FILE.
2637 suj_find(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, int frags)
2639 char block[MAXBSIZE];
2646 bytes = lfragtosize(fs, frags);
2647 if (bread(&disk, fsbtodb(fs, blk), block, bytes) <= 0)
2648 err_suj("Failed to read UFS_ROOTINO directory block %jd\n",
2650 for (off = 0; off < bytes; off += dp->d_reclen) {
2651 dp = (struct direct *)&block[off];
2652 if (dp->d_reclen == 0)
2656 if (dp->d_namlen != strlen(SUJ_FILE))
2658 if (bcmp(dp->d_name, SUJ_FILE, dp->d_namlen) != 0)
2666 * Orchestrate the verification of a filesystem via the softupdates journal.
2669 suj_check(const char *filesys)
2675 struct suj_seg *seg;
2676 struct suj_seg *segn;
2680 if (real_dev_bsize == 0 && ioctl(disk.d_fd, DIOCGSECTORSIZE,
2681 &real_dev_bsize) == -1)
2682 real_dev_bsize = secsize;
2684 printf("dev_bsize %u\n", real_dev_bsize);
2687 * Set an exit point when SUJ check failed
2689 retval = setjmp(jmpbuf);
2691 pwarn("UNEXPECTED SU+J INCONSISTENCY\n");
2692 TAILQ_FOREACH_SAFE(seg, &allsegs, ss_next, segn) {
2693 TAILQ_REMOVE(&allsegs, seg, ss_next);
2697 if (reply("FALLBACK TO FULL FSCK") == 0) {
2705 * Find the journal inode.
2707 ip = ino_read(UFS_ROOTINO);
2709 ino_visit(ip, UFS_ROOTINO, suj_find, 0);
2711 printf("Journal inode removed. Use tunefs to re-create.\n");
2712 sblock.fs_flags &= ~FS_SUJ;
2713 sblock.fs_sujfree = 0;
2717 * Fetch the journal inode and verify it.
2719 jip = ino_read(sujino);
2720 printf("** SU+J Recovering %s\n", filesys);
2721 if (suj_verifyino(jip) != 0)
2723 if (!preen && !reply("USE JOURNAL"))
2726 * Build a list of journal blocks in jblocks before parsing the
2727 * available journal blocks in with suj_read().
2729 printf("** Reading %jd byte journal from inode %ju.\n",
2730 DIP(jip, di_size), (uintmax_t)sujino);
2731 suj_jblocks = jblocks_create();
2732 blocks = ino_visit(jip, sujino, suj_add_block, 0);
2733 if (blocks != numfrags(fs, DIP(jip, di_size))) {
2734 printf("Sparse journal inode %ju.\n", (uintmax_t)sujino);
2738 jblocks_destroy(suj_jblocks);
2740 if (preen || reply("RECOVER")) {
2741 printf("** Building recovery table.\n");
2745 printf("** Resolving unreferenced inode list.\n");
2747 printf("** Processing journal entries.\n");
2749 cg_apply(cg_check_blk);
2750 cg_apply(cg_adj_blk);
2751 cg_apply(cg_check_ino);
2753 if (preen == 0 && (jrecs > 0 || jbytes > 0) && reply("WRITE CHANGES") == 0)
2756 * To remain idempotent with partial truncations the free bitmaps
2757 * must be written followed by indirect blocks and lastly inode
2758 * blocks. This preserves access to the modified pointers until
2763 cg_apply(cg_write_inos);
2764 /* Write back superblock. */
2766 if (jrecs > 0 || jbytes > 0) {
2767 printf("** %jd journal records in %jd bytes for %.2f%% utilization\n",
2768 jrecs, jbytes, ((float)jrecs / (float)(jbytes / JREC_SIZE)) * 100);
2769 printf("** Freed %jd inodes (%jd dirs) %jd blocks, and %jd frags.\n",
2770 freeinos, freedir, freeblocks, freefrags);
2781 for (i = 0; i < SUJ_HASHSIZE; i++) {
2782 LIST_INIT(&cghash[i]);
2783 LIST_INIT(&dbhash[i]);
2787 TAILQ_INIT(&allsegs);