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 uufsd *disk = NULL;
138 static struct fs *fs = NULL;
142 * Summary statistics.
144 static uint64_t freefrags;
145 static uint64_t freeblocks;
146 static uint64_t freeinos;
147 static uint64_t freedir;
148 static uint64_t jbytes;
149 static uint64_t jrecs;
151 static jmp_buf jmpbuf;
153 typedef void (*ino_visitor)(ino_t, ufs_lbn_t, ufs2_daddr_t, int);
154 static void err_suj(const char *, ...) __dead2;
155 static void ino_trunc(ino_t, off_t);
156 static void ino_decr(ino_t);
157 static void ino_adjust(struct suj_ino *);
158 static void ino_build(struct suj_ino *);
159 static int blk_isfree(ufs2_daddr_t);
160 static void initsuj(void);
169 err(EX_OSERR, "malloc(%zu)", n);
174 * When hit a fatal error in journalling check, print out
175 * the error and then offer to fallback to normal fsck.
178 err_suj(const char * restrict fmt, ...)
183 (void)fprintf(stdout, "%s: ", cdevname);
186 (void)vfprintf(stdout, fmt, ap);
193 * Open the given provider, load superblock.
196 opendisk(const char *devnam)
200 disk = Malloc(sizeof(*disk));
202 err(EX_OSERR, "malloc(%zu)", sizeof(*disk));
203 if (ufs_disk_fillout(disk, devnam) == -1) {
204 err(EX_OSERR, "ufs_disk_fillout(%s) failed: %s", devnam,
208 if (real_dev_bsize == 0 && ioctl(disk->d_fd, DIOCGSECTORSIZE,
209 &real_dev_bsize) == -1)
210 real_dev_bsize = secsize;
212 printf("dev_bsize %u\n", real_dev_bsize);
216 * Mark file system as clean, write the super-block back, close the disk.
219 closedisk(const char *devnam)
225 * Recompute the fs summary info from correct cs summaries.
227 bzero(&fs->fs_cstotal, sizeof(struct csum_total));
228 for (i = 0; i < fs->fs_ncg; i++) {
229 cgsum = &fs->fs_cs(fs, i);
230 fs->fs_cstotal.cs_nffree += cgsum->cs_nffree;
231 fs->fs_cstotal.cs_nbfree += cgsum->cs_nbfree;
232 fs->fs_cstotal.cs_nifree += cgsum->cs_nifree;
233 fs->fs_cstotal.cs_ndir += cgsum->cs_ndir;
235 fs->fs_pendinginodes = 0;
236 fs->fs_pendingblocks = 0;
238 fs->fs_time = time(NULL);
239 fs->fs_mtime = time(NULL);
240 if (sbwrite(disk, 0) == -1)
241 err(EX_OSERR, "sbwrite(%s)", devnam);
242 if (ufs_disk_close(disk) == -1)
243 err(EX_OSERR, "ufs_disk_close(%s)", devnam);
250 * Lookup a cg by number in the hash so we can keep track of which cgs
251 * need stats rebuilt.
253 static struct suj_cg *
259 if (cgx < 0 || cgx >= fs->fs_ncg)
260 err_suj("Bad cg number %d\n", cgx);
261 if (lastcg && lastcg->sc_cgx == cgx)
263 hd = &cghash[SUJ_HASH(cgx)];
264 LIST_FOREACH(sc, hd, sc_next)
265 if (sc->sc_cgx == cgx) {
269 sc = errmalloc(sizeof(*sc));
270 bzero(sc, sizeof(*sc));
271 sc->sc_cgbuf = errmalloc(fs->fs_bsize);
272 sc->sc_cgp = (struct cg *)sc->sc_cgbuf;
274 LIST_INSERT_HEAD(hd, sc, sc_next);
275 if (bread(disk, fsbtodb(fs, cgtod(fs, sc->sc_cgx)), sc->sc_cgbuf,
277 err_suj("Unable to read cylinder group %d\n", sc->sc_cgx);
283 * Lookup an inode number in the hash and allocate a suj_ino if it does
286 static struct suj_ino *
287 ino_lookup(ino_t ino, int creat)
289 struct suj_ino *sino;
293 sc = cg_lookup(ino_to_cg(fs, ino));
294 if (sc->sc_lastino && sc->sc_lastino->si_ino == ino)
295 return (sc->sc_lastino);
296 hd = &sc->sc_inohash[SUJ_HASH(ino)];
297 LIST_FOREACH(sino, hd, si_next)
298 if (sino->si_ino == ino)
302 sino = errmalloc(sizeof(*sino));
303 bzero(sino, sizeof(*sino));
305 TAILQ_INIT(&sino->si_recs);
306 TAILQ_INIT(&sino->si_newrecs);
307 TAILQ_INIT(&sino->si_movs);
308 LIST_INSERT_HEAD(hd, sino, si_next);
314 * Lookup a block number in the hash and allocate a suj_blk if it does
317 static struct suj_blk *
318 blk_lookup(ufs2_daddr_t blk, int creat)
320 struct suj_blk *sblk;
324 sc = cg_lookup(dtog(fs, blk));
325 if (sc->sc_lastblk && sc->sc_lastblk->sb_blk == blk)
326 return (sc->sc_lastblk);
327 hd = &sc->sc_blkhash[SUJ_HASH(fragstoblks(fs, blk))];
328 LIST_FOREACH(sblk, hd, sb_next)
329 if (sblk->sb_blk == blk)
333 sblk = errmalloc(sizeof(*sblk));
334 bzero(sblk, sizeof(*sblk));
336 TAILQ_INIT(&sblk->sb_recs);
337 LIST_INSERT_HEAD(hd, sblk, sb_next);
342 static struct data_blk *
343 dblk_lookup(ufs2_daddr_t blk)
345 struct data_blk *dblk;
348 hd = &dbhash[SUJ_HASH(fragstoblks(fs, blk))];
349 if (lastblk && lastblk->db_blk == blk)
351 LIST_FOREACH(dblk, hd, db_next)
352 if (dblk->db_blk == blk)
355 * The inode block wasn't located, allocate a new one.
357 dblk = errmalloc(sizeof(*dblk));
358 bzero(dblk, sizeof(*dblk));
359 LIST_INSERT_HEAD(hd, dblk, db_next);
365 dblk_read(ufs2_daddr_t blk, int size)
367 struct data_blk *dblk;
369 dblk = dblk_lookup(blk);
371 * I doubt size mismatches can happen in practice but it is trivial
374 if (size != dblk->db_size) {
377 dblk->db_buf = errmalloc(size);
378 dblk->db_size = size;
379 if (bread(disk, fsbtodb(fs, blk), dblk->db_buf, size) == -1)
380 err_suj("Failed to read data block %jd\n", blk);
382 return (dblk->db_buf);
386 dblk_dirty(ufs2_daddr_t blk)
388 struct data_blk *dblk;
390 dblk = dblk_lookup(blk);
397 struct data_blk *dblk;
400 for (i = 0; i < SUJ_HASHSIZE; i++) {
401 LIST_FOREACH(dblk, &dbhash[i], db_next) {
402 if (dblk->db_dirty == 0 || dblk->db_size == 0)
404 if (bwrite(disk, fsbtodb(fs, dblk->db_blk),
405 dblk->db_buf, dblk->db_size) == -1)
406 err_suj("Unable to write block %jd\n",
412 static union dinode *
415 struct ino_blk *iblk;
421 blk = ino_to_fsba(fs, ino);
422 sc = cg_lookup(ino_to_cg(fs, ino));
423 iblk = sc->sc_lastiblk;
424 if (iblk && iblk->ib_blk == blk)
426 hd = &sc->sc_iblkhash[SUJ_HASH(fragstoblks(fs, blk))];
427 LIST_FOREACH(iblk, hd, ib_next)
428 if (iblk->ib_blk == blk)
431 * The inode block wasn't located, allocate a new one.
433 iblk = errmalloc(sizeof(*iblk));
434 bzero(iblk, sizeof(*iblk));
435 iblk->ib_buf = errmalloc(fs->fs_bsize);
437 LIST_INSERT_HEAD(hd, iblk, ib_next);
438 if (bread(disk, fsbtodb(fs, blk), iblk->ib_buf, fs->fs_bsize) == -1)
439 err_suj("Failed to read inode block %jd\n", blk);
441 sc->sc_lastiblk = iblk;
442 off = ino_to_fsbo(fs, ino);
443 if (fs->fs_magic == FS_UFS1_MAGIC)
444 return (union dinode *)&((struct ufs1_dinode *)iblk->ib_buf)[off];
446 return (union dinode *)&((struct ufs2_dinode *)iblk->ib_buf)[off];
452 struct ino_blk *iblk;
457 blk = ino_to_fsba(fs, ino);
458 sc = cg_lookup(ino_to_cg(fs, ino));
459 iblk = sc->sc_lastiblk;
460 if (iblk && iblk->ib_blk == blk) {
464 hd = &sc->sc_iblkhash[SUJ_HASH(fragstoblks(fs, blk))];
465 LIST_FOREACH(iblk, hd, ib_next) {
466 if (iblk->ib_blk == blk) {
476 iblk_write(struct ino_blk *iblk)
479 if (iblk->ib_dirty == 0)
481 if (bwrite(disk, fsbtodb(fs, iblk->ib_blk), iblk->ib_buf,
483 err_suj("Failed to write inode block %jd\n", iblk->ib_blk);
487 blk_overlaps(struct jblkrec *brec, ufs2_daddr_t start, int frags)
494 bstart = brec->jb_blkno + brec->jb_oldfrags;
495 bend = bstart + brec->jb_frags;
496 if (start < bend && end > bstart)
502 blk_equals(struct jblkrec *brec, ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t start,
506 if (brec->jb_ino != ino || brec->jb_lbn != lbn)
508 if (brec->jb_blkno + brec->jb_oldfrags != start)
510 if (brec->jb_frags < frags)
516 blk_setmask(struct jblkrec *brec, int *mask)
520 for (i = brec->jb_oldfrags; i < brec->jb_oldfrags + brec->jb_frags; i++)
525 * Determine whether a given block has been reallocated to a new location.
526 * Returns a mask of overlapping bits if any frags have been reused or
527 * zero if the block has not been re-used and the contents can be trusted.
529 * This is used to ensure that an orphaned pointer due to truncate is safe
530 * to be freed. The mask value can be used to free partial blocks.
533 blk_freemask(ufs2_daddr_t blk, ino_t ino, ufs_lbn_t lbn, int frags)
535 struct suj_blk *sblk;
536 struct suj_rec *srec;
537 struct jblkrec *brec;
542 * To be certain we're not freeing a reallocated block we lookup
543 * this block in the blk hash and see if there is an allocation
544 * journal record that overlaps with any fragments in the block
545 * we're concerned with. If any fragments have ben reallocated
546 * the block has already been freed and re-used for another purpose.
549 sblk = blk_lookup(blknum(fs, blk), 0);
552 off = blk - sblk->sb_blk;
553 TAILQ_FOREACH(srec, &sblk->sb_recs, sr_next) {
554 brec = (struct jblkrec *)srec->sr_rec;
556 * If the block overlaps but does not match
557 * exactly this record refers to the current
560 if (blk_overlaps(brec, blk, frags) == 0)
562 if (blk_equals(brec, ino, lbn, blk, frags) == 1)
565 blk_setmask(brec, &mask);
568 printf("blk_freemask: blk %jd sblk %jd off %d mask 0x%X\n",
569 blk, sblk->sb_blk, off, mask);
570 return (mask >> off);
574 * Determine whether it is safe to follow an indirect. It is not safe
575 * if any part of the indirect has been reallocated or the last journal
576 * entry was an allocation. Just allocated indirects may not have valid
577 * pointers yet and all of their children will have their own records.
578 * It is also not safe to follow an indirect if the cg bitmap has been
579 * cleared as a new allocation may write to the block prior to the journal
582 * Returns 1 if it's safe to follow the indirect and 0 otherwise.
585 blk_isindir(ufs2_daddr_t blk, ino_t ino, ufs_lbn_t lbn)
587 struct suj_blk *sblk;
588 struct jblkrec *brec;
590 sblk = blk_lookup(blk, 0);
593 if (TAILQ_EMPTY(&sblk->sb_recs))
595 brec = (struct jblkrec *)TAILQ_LAST(&sblk->sb_recs, srechd)->sr_rec;
596 if (blk_equals(brec, ino, lbn, blk, fs->fs_frag))
597 if (brec->jb_op == JOP_FREEBLK)
598 return (!blk_isfree(blk));
603 * Clear an inode from the cg bitmap. If the inode was already clear return
604 * 0 so the caller knows it does not have to check the inode contents.
607 ino_free(ino_t ino, int mode)
614 cg = ino_to_cg(fs, ino);
615 ino = ino % fs->fs_ipg;
618 inosused = cg_inosused(cgp);
620 * The bitmap may never have made it to the disk so we have to
621 * conditionally clear. We can avoid writing the cg in this case.
623 if (isclr(inosused, ino))
626 clrbit(inosused, ino);
627 if (ino < cgp->cg_irotor)
628 cgp->cg_irotor = ino;
629 cgp->cg_cs.cs_nifree++;
630 if ((mode & IFMT) == IFDIR) {
632 cgp->cg_cs.cs_ndir--;
640 * Free 'frags' frags starting at filesystem block 'bno' skipping any frags
644 blk_free(ufs2_daddr_t bno, int mask, int frags)
646 ufs1_daddr_t fragno, cgbno;
653 printf("Freeing %d frags at blk %jd mask 0x%x\n",
658 cgbno = dtogd(fs, bno);
659 blksfree = cg_blksfree(cgp);
662 * If it's not allocated we only wrote the journal entry
663 * and never the bitmaps. Here we unconditionally clear and
664 * resolve the cg summary later.
666 if (frags == fs->fs_frag && mask == 0) {
667 fragno = fragstoblks(fs, cgbno);
668 ffs_setblock(fs, blksfree, fragno);
672 * deallocate the fragment
674 for (i = 0; i < frags; i++)
675 if ((mask & (1 << i)) == 0 && isclr(blksfree, cgbno +i)) {
677 setbit(blksfree, cgbno + i);
684 * Returns 1 if the whole block starting at 'bno' is marked free and 0
688 blk_isfree(ufs2_daddr_t bno)
692 sc = cg_lookup(dtog(fs, bno));
693 return ffs_isblock(fs, cg_blksfree(sc->sc_cgp), dtogd(fs, bno));
697 * Fetch an indirect block to find the block at a given lbn. The lbn
698 * may be negative to fetch a specific indirect block pointer or positive
699 * to fetch a specific block.
702 indir_blkatoff(ufs2_daddr_t blk, ino_t ino, ufs_lbn_t cur, ufs_lbn_t lbn)
713 level = lbn_level(cur);
715 err_suj("Invalid indir lbn %jd\n", lbn);
716 if (level == 0 && lbn < 0)
717 err_suj("Invalid lbn %jd\n", lbn);
718 bap2 = (void *)dblk_read(blk, fs->fs_bsize);
721 base = -(cur + level);
722 for (i = level; i > 0; i--)
723 lbnadd *= NINDIR(fs);
725 i = (lbn - base) / lbnadd;
727 i = (-lbn - base) / lbnadd;
728 if (i < 0 || i >= NINDIR(fs))
729 err_suj("Invalid indirect index %d produced by lbn %jd\n",
732 cur = base + (i * lbnadd);
734 cur = -(base + (i * lbnadd)) - (level - 1);
735 if (fs->fs_magic == FS_UFS1_MAGIC)
742 err_suj("Invalid lbn %jd at level 0\n", lbn);
743 return indir_blkatoff(blk, ino, cur, lbn);
747 * Finds the disk block address at the specified lbn within the inode
748 * specified by ip. This follows the whole tree and honors di_size and
749 * di_extsize so it is a true test of reachability. The lbn may be
750 * negative if an extattr or indirect block is requested.
753 ino_blkatoff(union dinode *ip, ino_t ino, ufs_lbn_t lbn, int *frags)
761 * Handle extattr blocks first.
763 if (lbn < 0 && lbn >= -NXADDR) {
765 if (lbn > lblkno(fs, ip->dp2.di_extsize - 1))
767 *frags = numfrags(fs, sblksize(fs, ip->dp2.di_extsize, lbn));
768 return (ip->dp2.di_extb[lbn]);
771 * Now direct and indirect.
773 if (DIP(ip, di_mode) == IFLNK &&
774 DIP(ip, di_size) < fs->fs_maxsymlinklen)
776 if (lbn >= 0 && lbn < NDADDR) {
777 *frags = numfrags(fs, sblksize(fs, DIP(ip, di_size), lbn));
778 return (DIP(ip, di_db[lbn]));
780 *frags = fs->fs_frag;
782 for (i = 0, tmpval = NINDIR(fs), cur = NDADDR; i < NIADDR; i++,
783 tmpval *= NINDIR(fs), cur = next) {
786 return (DIP(ip, di_ib[i]));
788 * Determine whether the lbn in question is within this tree.
790 if (lbn < 0 && -lbn >= next)
792 if (lbn > 0 && lbn >= next)
794 return indir_blkatoff(DIP(ip, di_ib[i]), ino, -cur - i, lbn);
796 err_suj("lbn %jd not in ino\n", lbn);
801 * Determine whether a block exists at a particular lbn in an inode.
802 * Returns 1 if found, 0 if not. lbn may be negative for indirects
806 blk_isat(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, int *frags)
813 if (DIP(ip, di_nlink) == 0 || DIP(ip, di_mode) == 0)
815 nblk = ino_blkatoff(ip, ino, lbn, frags);
817 return (nblk == blk);
821 * Clear the directory entry at diroff that should point to child. Minimal
822 * checking is done and it is assumed that this path was verified with isat.
825 ino_clrat(ino_t parent, off_t diroff, ino_t child)
837 printf("Clearing inode %ju from parent %ju at offset %jd\n",
838 (uintmax_t)child, (uintmax_t)parent, diroff);
840 lbn = lblkno(fs, diroff);
841 doff = blkoff(fs, diroff);
842 dip = ino_read(parent);
843 blk = ino_blkatoff(dip, parent, lbn, &frags);
844 blksize = sblksize(fs, DIP(dip, di_size), lbn);
845 block = dblk_read(blk, blksize);
846 dp = (struct direct *)&block[doff];
847 if (dp->d_ino != child)
848 errx(1, "Inode %ju does not exist in %ju at %jd",
849 (uintmax_t)child, (uintmax_t)parent, diroff);
853 * The actual .. reference count will already have been removed
854 * from the parent by the .. remref record.
859 * Determines whether a pointer to an inode exists within a directory
860 * at a specified offset. Returns the mode of the found entry.
863 ino_isat(ino_t parent, off_t diroff, ino_t child, int *mode, int *isdot)
876 dip = ino_read(parent);
877 *mode = DIP(dip, di_mode);
878 if ((*mode & IFMT) != IFDIR) {
881 * This can happen if the parent inode
885 printf("Directory %ju has bad mode %o\n",
886 (uintmax_t)parent, *mode);
888 printf("Directory %ju has zero mode\n",
893 lbn = lblkno(fs, diroff);
894 doff = blkoff(fs, diroff);
895 blksize = sblksize(fs, DIP(dip, di_size), lbn);
896 if (diroff + DIRECTSIZ(1) > DIP(dip, di_size) || doff >= blksize) {
898 printf("ino %ju absent from %ju due to offset %jd"
899 " exceeding size %jd\n",
900 (uintmax_t)child, (uintmax_t)parent, diroff,
904 blk = ino_blkatoff(dip, parent, lbn, &frags);
907 printf("Sparse directory %ju", (uintmax_t)parent);
910 block = dblk_read(blk, blksize);
912 * Walk through the records from the start of the block to be
913 * certain we hit a valid record and not some junk in the middle
914 * of a file name. Stop when we reach or pass the expected offset.
916 dpoff = rounddown(doff, DIRBLKSIZ);
918 dp = (struct direct *)&block[dpoff];
921 if (dp->d_reclen == 0)
923 dpoff += dp->d_reclen;
924 } while (dpoff <= doff);
925 if (dpoff > fs->fs_bsize)
926 err_suj("Corrupt directory block in dir ino %ju\n",
931 printf("ino %ju not found in %ju, lbn %jd, dpoff %d\n",
932 (uintmax_t)child, (uintmax_t)parent, lbn, dpoff);
936 * We found the item in question. Record the mode and whether it's
937 * a . or .. link for the caller.
939 if (dp->d_ino == child) {
942 else if (dp->d_namlen == 2 &&
943 dp->d_name[0] == '.' && dp->d_name[1] == '.')
945 *mode = DTTOIF(dp->d_type);
949 printf("ino %ju doesn't match dirent ino %ju in parent %ju\n",
950 (uintmax_t)child, (uintmax_t)dp->d_ino, (uintmax_t)parent);
954 #define VISIT_INDIR 0x0001
955 #define VISIT_EXT 0x0002
956 #define VISIT_ROOT 0x0004 /* Operation came via root & valid pointers. */
959 * Read an indirect level which may or may not be linked into an inode.
962 indir_visit(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, uint64_t *frags,
963 ino_visitor visitor, int flags)
974 * Don't visit indirect blocks with contents we can't trust. This
975 * should only happen when indir_visit() is called to complete a
976 * truncate that never finished and not when a pointer is found via
981 level = lbn_level(lbn);
983 err_suj("Invalid level for lbn %jd\n", lbn);
984 if ((flags & VISIT_ROOT) == 0 && blk_isindir(blk, ino, lbn) == 0) {
986 printf("blk %jd ino %ju lbn %jd(%d) is not indir.\n",
987 blk, (uintmax_t)ino, lbn, level);
991 for (i = level; i > 0; i--)
992 lbnadd *= NINDIR(fs);
993 bap1 = (void *)dblk_read(blk, fs->fs_bsize);
995 for (i = 0; i < NINDIR(fs); i++) {
996 if (fs->fs_magic == FS_UFS1_MAGIC)
1003 nlbn = -lbn + i * lbnadd;
1004 (*frags) += fs->fs_frag;
1005 visitor(ino, nlbn, nblk, fs->fs_frag);
1007 nlbn = (lbn + 1) - (i * lbnadd);
1008 indir_visit(ino, nlbn, nblk, frags, visitor, flags);
1012 if (flags & VISIT_INDIR) {
1013 (*frags) += fs->fs_frag;
1014 visitor(ino, lbn, blk, fs->fs_frag);
1019 * Visit each block in an inode as specified by 'flags' and call a
1020 * callback function. The callback may inspect or free blocks. The
1021 * count of frags found according to the size in the file is returned.
1022 * This is not valid for sparse files but may be used to determine
1023 * the correct di_blocks for a file.
1026 ino_visit(union dinode *ip, ino_t ino, ino_visitor visitor, int flags)
1037 size = DIP(ip, di_size);
1038 mode = DIP(ip, di_mode) & IFMT;
1040 if ((flags & VISIT_EXT) &&
1041 fs->fs_magic == FS_UFS2_MAGIC && ip->dp2.di_extsize) {
1042 for (i = 0; i < NXADDR; i++) {
1043 if (ip->dp2.di_extb[i] == 0)
1045 frags = sblksize(fs, ip->dp2.di_extsize, i);
1046 frags = numfrags(fs, frags);
1048 visitor(ino, -1 - i, ip->dp2.di_extb[i], frags);
1051 /* Skip datablocks for short links and devices. */
1052 if (mode == IFBLK || mode == IFCHR ||
1053 (mode == IFLNK && size < fs->fs_maxsymlinklen))
1055 for (i = 0; i < NDADDR; i++) {
1056 if (DIP(ip, di_db[i]) == 0)
1058 frags = sblksize(fs, size, i);
1059 frags = numfrags(fs, frags);
1061 visitor(ino, i, DIP(ip, di_db[i]), frags);
1064 * We know the following indirects are real as we're following
1065 * real pointers to them.
1067 flags |= VISIT_ROOT;
1068 for (i = 0, tmpval = NINDIR(fs), lbn = NDADDR; i < NIADDR; i++,
1070 nextlbn = lbn + tmpval;
1071 tmpval *= NINDIR(fs);
1072 if (DIP(ip, di_ib[i]) == 0)
1074 indir_visit(ino, -lbn - i, DIP(ip, di_ib[i]), &fragcnt, visitor,
1081 * Null visitor function used when we just want to count blocks and
1086 null_visit(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, int frags)
1093 * Recalculate di_blocks when we discover that a block allocation or
1094 * free was not successfully completed. The kernel does not roll this back
1095 * because it would be too expensive to compute which indirects were
1096 * reachable at the time the inode was written.
1099 ino_adjblks(struct suj_ino *sino)
1110 /* No need to adjust zero'd inodes. */
1111 if (DIP(ip, di_mode) == 0)
1114 * Visit all blocks and count them as well as recording the last
1115 * valid lbn in the file. If the file size doesn't agree with the
1116 * last lbn we need to truncate to fix it. Otherwise just adjust
1120 frags = ino_visit(ip, ino, null_visit, VISIT_INDIR | VISIT_EXT);
1121 blocks = fsbtodb(fs, frags);
1123 * We assume the size and direct block list is kept coherent by
1124 * softdep. For files that have extended into indirects we truncate
1125 * to the size in the inode or the maximum size permitted by
1126 * populated indirects.
1128 if (visitlbn >= NDADDR) {
1129 isize = DIP(ip, di_size);
1130 size = lblktosize(fs, visitlbn + 1);
1133 /* Always truncate to free any unpopulated indirects. */
1134 ino_trunc(sino->si_ino, isize);
1137 if (blocks == DIP(ip, di_blocks))
1140 printf("ino %ju adjusting block count from %jd to %jd\n",
1141 (uintmax_t)ino, DIP(ip, di_blocks), blocks);
1142 DIP_SET(ip, di_blocks, blocks);
1147 blk_free_visit(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, int frags)
1150 blk_free(blk, blk_freemask(blk, ino, lbn, frags), frags);
1154 * Free a block or tree of blocks that was previously rooted in ino at
1155 * the given lbn. If the lbn is an indirect all children are freed
1159 blk_free_lbn(ufs2_daddr_t blk, ino_t ino, ufs_lbn_t lbn, int frags, int follow)
1164 mask = blk_freemask(blk, ino, lbn, frags);
1166 if (lbn <= -NDADDR && follow && mask == 0)
1167 indir_visit(ino, lbn, blk, &resid, blk_free_visit, VISIT_INDIR);
1169 blk_free(blk, mask, frags);
1173 ino_setskip(struct suj_ino *sino, ino_t parent)
1178 if (ino_isat(sino->si_ino, DOTDOT_OFFSET, parent, &mode, &isdot))
1179 sino->si_skipparent = 1;
1183 ino_remref(ino_t parent, ino_t child, uint64_t diroff, int isdotdot)
1185 struct suj_ino *sino;
1186 struct suj_rec *srec;
1187 struct jrefrec *rrec;
1190 * Lookup this inode to see if we have a record for it.
1192 sino = ino_lookup(child, 0);
1194 * Tell any child directories we've already removed their
1195 * parent link cnt. Don't try to adjust our link down again.
1197 if (sino != NULL && isdotdot == 0)
1198 ino_setskip(sino, parent);
1200 * No valid record for this inode. Just drop the on-disk
1203 if (sino == NULL || sino->si_hasrecs == 0) {
1208 * Use ino_adjust() if ino_check() has already processed this
1209 * child. If we lose the last non-dot reference to a
1210 * directory it will be discarded.
1212 if (sino->si_linkadj) {
1215 sino->si_dotlinks--;
1220 * If we haven't yet processed this inode we need to make
1221 * sure we will successfully discover the lost path. If not
1222 * use nlinkadj to remember.
1224 TAILQ_FOREACH(srec, &sino->si_recs, sr_next) {
1225 rrec = (struct jrefrec *)srec->sr_rec;
1226 if (rrec->jr_parent == parent &&
1227 rrec->jr_diroff == diroff)
1230 sino->si_nlinkadj++;
1234 * Free the children of a directory when the directory is discarded.
1237 ino_free_children(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, int frags)
1239 struct suj_ino *sino;
1248 sino = ino_lookup(ino, 0);
1250 skipparent = sino->si_skipparent;
1253 size = lfragtosize(fs, frags);
1254 block = dblk_read(blk, size);
1255 dp = (struct direct *)&block[0];
1256 for (dpoff = 0; dpoff < size && dp->d_reclen; dpoff += dp->d_reclen) {
1257 dp = (struct direct *)&block[dpoff];
1258 if (dp->d_ino == 0 || dp->d_ino == WINO)
1260 if (dp->d_namlen == 1 && dp->d_name[0] == '.')
1262 isdotdot = dp->d_namlen == 2 && dp->d_name[0] == '.' &&
1263 dp->d_name[1] == '.';
1264 if (isdotdot && skipparent == 1)
1267 printf("Directory %ju removing ino %ju name %s\n",
1268 (uintmax_t)ino, (uintmax_t)dp->d_ino, dp->d_name);
1269 diroff = lblktosize(fs, lbn) + dpoff;
1270 ino_remref(ino, dp->d_ino, diroff, isdotdot);
1275 * Reclaim an inode, freeing all blocks and decrementing all children's
1276 * link counts. Free the inode back to the cg.
1279 ino_reclaim(union dinode *ip, ino_t ino, int mode)
1284 err_suj("Attempting to free ROOTINO\n");
1286 printf("Truncating and freeing ino %ju, nlink %d, mode %o\n",
1287 (uintmax_t)ino, DIP(ip, di_nlink), DIP(ip, di_mode));
1289 /* We are freeing an inode or directory. */
1290 if ((DIP(ip, di_mode) & IFMT) == IFDIR)
1291 ino_visit(ip, ino, ino_free_children, 0);
1292 DIP_SET(ip, di_nlink, 0);
1293 ino_visit(ip, ino, blk_free_visit, VISIT_EXT | VISIT_INDIR);
1294 /* Here we have to clear the inode and release any blocks it holds. */
1295 gen = DIP(ip, di_gen);
1296 if (fs->fs_magic == FS_UFS1_MAGIC)
1297 bzero(ip, sizeof(struct ufs1_dinode));
1299 bzero(ip, sizeof(struct ufs2_dinode));
1300 DIP_SET(ip, di_gen, gen);
1302 ino_free(ino, mode);
1307 * Adjust an inode's link count down by one when a directory goes away.
1318 nlink = DIP(ip, di_nlink);
1319 mode = DIP(ip, di_mode);
1321 err_suj("Inode %d link count %d invalid\n", ino, nlink);
1323 err_suj("Inode %d has a link of %d with 0 mode\n", ino, nlink);
1325 if ((mode & IFMT) == IFDIR)
1329 if (nlink < reqlink) {
1331 printf("ino %ju not enough links to live %d < %d\n",
1332 (uintmax_t)ino, nlink, reqlink);
1333 ino_reclaim(ip, ino, mode);
1336 DIP_SET(ip, di_nlink, nlink);
1341 * Adjust the inode link count to 'nlink'. If the count reaches zero
1345 ino_adjust(struct suj_ino *sino)
1347 struct jrefrec *rrec;
1348 struct suj_rec *srec;
1349 struct suj_ino *stmp;
1358 nlink = sino->si_nlink;
1360 mode = sino->si_mode & IFMT;
1362 * If it's a directory with no dot links, it was truncated before
1363 * the name was cleared. We need to clear the dirent that
1366 if (mode == IFDIR && nlink == 1 && sino->si_dotlinks == 0) {
1367 sino->si_nlink = nlink = 0;
1368 TAILQ_FOREACH(srec, &sino->si_recs, sr_next) {
1369 rrec = (struct jrefrec *)srec->sr_rec;
1370 if (ino_isat(rrec->jr_parent, rrec->jr_diroff, ino,
1371 &recmode, &isdot) == 0)
1373 ino_clrat(rrec->jr_parent, rrec->jr_diroff, ino);
1377 errx(1, "Directory %ju name not found", (uintmax_t)ino);
1380 * If it's a directory with no real names pointing to it go ahead
1381 * and truncate it. This will free any children.
1383 if (mode == IFDIR && nlink - sino->si_dotlinks == 0) {
1384 sino->si_nlink = nlink = 0;
1386 * Mark any .. links so they know not to free this inode
1387 * when they are removed.
1389 TAILQ_FOREACH(srec, &sino->si_recs, sr_next) {
1390 rrec = (struct jrefrec *)srec->sr_rec;
1391 if (rrec->jr_diroff == DOTDOT_OFFSET) {
1392 stmp = ino_lookup(rrec->jr_parent, 0);
1394 ino_setskip(stmp, ino);
1399 mode = DIP(ip, di_mode) & IFMT;
1400 if (nlink > LINK_MAX)
1401 err_suj("ino %ju nlink manipulation error, new %ju, old %d\n",
1402 (uintmax_t)ino, (uintmax_t)nlink, DIP(ip, di_nlink));
1404 printf("Adjusting ino %ju, nlink %ju, old link %d lastmode %o\n",
1405 (uintmax_t)ino, (uintmax_t)nlink, DIP(ip, di_nlink),
1409 printf("ino %ju, zero inode freeing bitmap\n",
1411 ino_free(ino, sino->si_mode);
1414 /* XXX Should be an assert? */
1415 if (mode != sino->si_mode && debug)
1416 printf("ino %ju, mode %o != %o\n",
1417 (uintmax_t)ino, mode, sino->si_mode);
1418 if ((mode & IFMT) == IFDIR)
1422 /* If the inode doesn't have enough links to live, free it. */
1423 if (nlink < reqlink) {
1425 printf("ino %ju not enough links to live %ju < %ju\n",
1426 (uintmax_t)ino, (uintmax_t)nlink,
1427 (uintmax_t)reqlink);
1428 ino_reclaim(ip, ino, mode);
1431 /* If required write the updated link count. */
1432 if (DIP(ip, di_nlink) == nlink) {
1434 printf("ino %ju, link matches, skipping.\n",
1438 DIP_SET(ip, di_nlink, nlink);
1443 * Truncate some or all blocks in an indirect, freeing any that are required
1444 * and zeroing the indirect.
1447 indir_trunc(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, ufs_lbn_t lastlbn)
1462 level = lbn_level(lbn);
1464 err_suj("Invalid level for lbn %jd\n", lbn);
1466 for (i = level; i > 0; i--)
1467 lbnadd *= NINDIR(fs);
1468 bap1 = (void *)dblk_read(blk, fs->fs_bsize);
1469 bap2 = (void *)bap1;
1470 for (i = 0; i < NINDIR(fs); i++) {
1471 if (fs->fs_magic == FS_UFS1_MAGIC)
1478 nlbn = (lbn + 1) - (i * lbnadd);
1480 * Calculate the lbn of the next indirect to
1481 * determine if any of this indirect must be
1484 next = -(lbn + level) + ((i+1) * lbnadd);
1485 if (next <= lastlbn)
1487 indir_trunc(ino, nlbn, nblk, lastlbn);
1488 /* If all of this indirect was reclaimed, free it. */
1489 nlbn = next - lbnadd;
1493 nlbn = -lbn + i * lbnadd;
1498 blk_free(nblk, 0, fs->fs_frag);
1499 if (fs->fs_magic == FS_UFS1_MAGIC)
1509 * Truncate an inode to the minimum of the given size or the last populated
1510 * block after any over size have been discarded. The kernel would allocate
1511 * the last block in the file but fsck does not and neither do we. This
1512 * code never extends files, only shrinks them.
1515 ino_trunc(ino_t ino, off_t size)
1519 uint64_t totalfrags;
1531 mode = DIP(ip, di_mode) & IFMT;
1532 cursize = DIP(ip, di_size);
1534 printf("Truncating ino %ju, mode %o to size %jd from size %jd\n",
1535 (uintmax_t)ino, mode, size, cursize);
1537 /* Skip datablocks for short links and devices. */
1538 if (mode == 0 || mode == IFBLK || mode == IFCHR ||
1539 (mode == IFLNK && cursize < fs->fs_maxsymlinklen))
1544 lastlbn = lblkno(fs, blkroundup(fs, size));
1545 for (i = lastlbn; i < NDADDR; i++) {
1546 if (DIP(ip, di_db[i]) == 0)
1548 frags = sblksize(fs, cursize, i);
1549 frags = numfrags(fs, frags);
1550 blk_free(DIP(ip, di_db[i]), 0, frags);
1551 DIP_SET(ip, di_db[i], 0);
1554 * Follow indirect blocks, freeing anything required.
1556 for (i = 0, tmpval = NINDIR(fs), lbn = NDADDR; i < NIADDR; i++,
1558 nextlbn = lbn + tmpval;
1559 tmpval *= NINDIR(fs);
1560 /* If we're not freeing any in this indirect range skip it. */
1561 if (lastlbn >= nextlbn)
1563 if (DIP(ip, di_ib[i]) == 0)
1565 indir_trunc(ino, -lbn - i, DIP(ip, di_ib[i]), lastlbn);
1566 /* If we freed everything in this indirect free the indir. */
1569 blk_free(DIP(ip, di_ib[i]), 0, frags);
1570 DIP_SET(ip, di_ib[i], 0);
1574 * Now that we've freed any whole blocks that exceed the desired
1575 * truncation size, figure out how many blocks remain and what the
1576 * last populated lbn is. We will set the size to this last lbn
1577 * rather than worrying about allocating the final lbn as the kernel
1578 * would've done. This is consistent with normal fsck behavior.
1581 totalfrags = ino_visit(ip, ino, null_visit, VISIT_INDIR | VISIT_EXT);
1582 if (size > lblktosize(fs, visitlbn + 1))
1583 size = lblktosize(fs, visitlbn + 1);
1585 * If we're truncating direct blocks we have to adjust frags
1588 if (visitlbn < NDADDR && totalfrags) {
1589 long oldspace, newspace;
1591 bn = DIP(ip, di_db[visitlbn]);
1593 err_suj("Bad blk at ino %ju lbn %jd\n",
1594 (uintmax_t)ino, visitlbn);
1595 oldspace = sblksize(fs, cursize, visitlbn);
1596 newspace = sblksize(fs, size, visitlbn);
1597 if (oldspace != newspace) {
1598 bn += numfrags(fs, newspace);
1599 frags = numfrags(fs, oldspace - newspace);
1600 blk_free(bn, 0, frags);
1601 totalfrags -= frags;
1604 DIP_SET(ip, di_blocks, fsbtodb(fs, totalfrags));
1605 DIP_SET(ip, di_size, size);
1607 * If we've truncated into the middle of a block or frag we have
1608 * to zero it here. Otherwise the file could extend into
1609 * uninitialized space later.
1611 off = blkoff(fs, size);
1612 if (off && DIP(ip, di_mode) != IFDIR) {
1616 bn = ino_blkatoff(ip, ino, visitlbn, &frags);
1618 err_suj("Block missing from ino %ju at lbn %jd\n",
1619 (uintmax_t)ino, visitlbn);
1620 clrsize = frags * fs->fs_fsize;
1621 buf = dblk_read(bn, clrsize);
1624 bzero(buf, clrsize);
1631 * Process records available for one inode and determine whether the
1632 * link count is correct or needs adjusting.
1635 ino_check(struct suj_ino *sino)
1637 struct suj_rec *srec;
1638 struct jrefrec *rrec;
1648 if (sino->si_hasrecs == 0)
1651 rrec = (struct jrefrec *)TAILQ_FIRST(&sino->si_recs)->sr_rec;
1652 nlink = rrec->jr_nlink;
1655 removes = sino->si_nlinkadj;
1656 TAILQ_FOREACH(srec, &sino->si_recs, sr_next) {
1657 rrec = (struct jrefrec *)srec->sr_rec;
1658 isat = ino_isat(rrec->jr_parent, rrec->jr_diroff,
1659 rrec->jr_ino, &mode, &isdot);
1660 if (isat && (mode & IFMT) != (rrec->jr_mode & IFMT))
1661 err_suj("Inode mode/directory type mismatch %o != %o\n",
1662 mode, rrec->jr_mode);
1664 printf("jrefrec: op %d ino %ju, nlink %ju, parent %ju, "
1665 "diroff %jd, mode %o, isat %d, isdot %d\n",
1666 rrec->jr_op, (uintmax_t)rrec->jr_ino,
1667 (uintmax_t)rrec->jr_nlink,
1668 (uintmax_t)rrec->jr_parent,
1669 (uintmax_t)rrec->jr_diroff,
1670 rrec->jr_mode, isat, isdot);
1671 mode = rrec->jr_mode & IFMT;
1672 if (rrec->jr_op == JOP_REMREF)
1679 * The number of links that remain are the starting link count
1680 * subtracted by the total number of removes with the total
1681 * links discovered back in. An incomplete remove thus
1682 * makes no change to the link count but an add increases
1687 "ino %ju nlink %ju newlinks %ju removes %ju dotlinks %ju\n",
1688 (uintmax_t)ino, (uintmax_t)nlink, (uintmax_t)newlinks,
1689 (uintmax_t)removes, (uintmax_t)dotlinks);
1692 sino->si_linkadj = 1;
1693 sino->si_nlink = nlink;
1694 sino->si_dotlinks = dotlinks;
1695 sino->si_mode = mode;
1700 * Process records available for one block and determine whether it is
1701 * still allocated and whether the owning inode needs to be updated or
1705 blk_check(struct suj_blk *sblk)
1707 struct suj_rec *srec;
1708 struct jblkrec *brec;
1709 struct suj_ino *sino;
1716 * Each suj_blk actually contains records for any fragments in that
1717 * block. As a result we must evaluate each record individually.
1720 TAILQ_FOREACH(srec, &sblk->sb_recs, sr_next) {
1721 brec = (struct jblkrec *)srec->sr_rec;
1722 frags = brec->jb_frags;
1723 blk = brec->jb_blkno + brec->jb_oldfrags;
1724 isat = blk_isat(brec->jb_ino, brec->jb_lbn, blk, &frags);
1725 if (sino == NULL || sino->si_ino != brec->jb_ino) {
1726 sino = ino_lookup(brec->jb_ino, 1);
1727 sino->si_blkadj = 1;
1730 printf("op %d blk %jd ino %ju lbn %jd frags %d isat %d (%d)\n",
1731 brec->jb_op, blk, (uintmax_t)brec->jb_ino,
1732 brec->jb_lbn, brec->jb_frags, isat, frags);
1734 * If we found the block at this address we still have to
1735 * determine if we need to free the tail end that was
1736 * added by adding contiguous fragments from the same block.
1739 if (frags == brec->jb_frags)
1741 mask = blk_freemask(blk, brec->jb_ino, brec->jb_lbn,
1745 frags = brec->jb_frags - frags;
1746 blk_free(blk, mask, frags);
1750 * The block wasn't found, attempt to free it. It won't be
1751 * freed if it was actually reallocated. If this was an
1752 * allocation we don't want to follow indirects as they
1753 * may not be written yet. Any children of the indirect will
1754 * have their own records. If it's a free we need to
1755 * recursively free children.
1757 blk_free_lbn(blk, brec->jb_ino, brec->jb_lbn, brec->jb_frags,
1758 brec->jb_op == JOP_FREEBLK);
1763 * Walk the list of inode records for this cg and resolve moved and duplicate
1764 * inode references now that we have a complete picture.
1767 cg_build(struct suj_cg *sc)
1769 struct suj_ino *sino;
1772 for (i = 0; i < SUJ_HASHSIZE; i++)
1773 LIST_FOREACH(sino, &sc->sc_inohash[i], si_next)
1778 * Handle inodes requiring truncation. This must be done prior to
1779 * looking up any inodes in directories.
1782 cg_trunc(struct suj_cg *sc)
1784 struct suj_ino *sino;
1787 for (i = 0; i < SUJ_HASHSIZE; i++) {
1788 LIST_FOREACH(sino, &sc->sc_inohash[i], si_next) {
1789 if (sino->si_trunc) {
1790 ino_trunc(sino->si_ino,
1791 sino->si_trunc->jt_size);
1792 sino->si_blkadj = 0;
1793 sino->si_trunc = NULL;
1795 if (sino->si_blkadj)
1802 cg_adj_blk(struct suj_cg *sc)
1804 struct suj_ino *sino;
1807 for (i = 0; i < SUJ_HASHSIZE; i++) {
1808 LIST_FOREACH(sino, &sc->sc_inohash[i], si_next) {
1809 if (sino->si_blkadj)
1816 * Free any partially allocated blocks and then resolve inode block
1820 cg_check_blk(struct suj_cg *sc)
1822 struct suj_blk *sblk;
1826 for (i = 0; i < SUJ_HASHSIZE; i++)
1827 LIST_FOREACH(sblk, &sc->sc_blkhash[i], sb_next)
1832 * Walk the list of inode records for this cg, recovering any
1833 * changes which were not complete at the time of crash.
1836 cg_check_ino(struct suj_cg *sc)
1838 struct suj_ino *sino;
1841 for (i = 0; i < SUJ_HASHSIZE; i++)
1842 LIST_FOREACH(sino, &sc->sc_inohash[i], si_next)
1847 * Write a potentially dirty cg. Recalculate the summary information and
1848 * update the superblock summary.
1851 cg_write(struct suj_cg *sc)
1853 ufs1_daddr_t fragno, cgbno, maxbno;
1859 if (sc->sc_dirty == 0)
1862 * Fix the frag and cluster summary.
1865 cgp->cg_cs.cs_nbfree = 0;
1866 cgp->cg_cs.cs_nffree = 0;
1867 bzero(&cgp->cg_frsum, sizeof(cgp->cg_frsum));
1868 maxbno = fragstoblks(fs, fs->fs_fpg);
1869 if (fs->fs_contigsumsize > 0) {
1870 for (i = 1; i <= fs->fs_contigsumsize; i++)
1871 cg_clustersum(cgp)[i] = 0;
1872 bzero(cg_clustersfree(cgp), howmany(maxbno, CHAR_BIT));
1874 blksfree = cg_blksfree(cgp);
1875 for (cgbno = 0; cgbno < maxbno; cgbno++) {
1876 if (ffs_isfreeblock(fs, blksfree, cgbno))
1878 if (ffs_isblock(fs, blksfree, cgbno)) {
1879 ffs_clusteracct(fs, cgp, cgbno, 1);
1880 cgp->cg_cs.cs_nbfree++;
1883 fragno = blkstofrags(fs, cgbno);
1884 blk = blkmap(fs, blksfree, fragno);
1885 ffs_fragacct(fs, blk, cgp->cg_frsum, 1);
1886 for (i = 0; i < fs->fs_frag; i++)
1887 if (isset(blksfree, fragno + i))
1888 cgp->cg_cs.cs_nffree++;
1891 * Update the superblock cg summary from our now correct values
1892 * before writing the block.
1894 fs->fs_cs(fs, sc->sc_cgx) = cgp->cg_cs;
1895 if (bwrite(disk, fsbtodb(fs, cgtod(fs, sc->sc_cgx)), sc->sc_cgbuf,
1896 fs->fs_bsize) == -1)
1897 err_suj("Unable to write cylinder group %d\n", sc->sc_cgx);
1901 * Write out any modified inodes.
1904 cg_write_inos(struct suj_cg *sc)
1906 struct ino_blk *iblk;
1909 for (i = 0; i < SUJ_HASHSIZE; i++)
1910 LIST_FOREACH(iblk, &sc->sc_iblkhash[i], ib_next)
1916 cg_apply(void (*apply)(struct suj_cg *))
1921 for (i = 0; i < SUJ_HASHSIZE; i++)
1922 LIST_FOREACH(scg, &cghash[i], sc_next)
1927 * Process the unlinked but referenced file list. Freeing all inodes.
1937 ino = fs->fs_sujfree;
1941 mode = DIP(ip, di_mode) & IFMT;
1942 inon = DIP(ip, di_freelink);
1943 DIP_SET(ip, di_freelink, 0);
1945 * XXX Should this be an errx?
1947 if (DIP(ip, di_nlink) == 0) {
1949 printf("Freeing unlinked ino %ju mode %o\n",
1950 (uintmax_t)ino, mode);
1951 ino_reclaim(ip, ino, mode);
1953 printf("Skipping ino %ju mode %o with link %d\n",
1954 (uintmax_t)ino, mode, DIP(ip, di_nlink));
1960 * Append a new record to the list of records requiring processing.
1963 ino_append(union jrec *rec)
1965 struct jrefrec *refrec;
1966 struct jmvrec *mvrec;
1967 struct suj_ino *sino;
1968 struct suj_rec *srec;
1970 mvrec = &rec->rec_jmvrec;
1971 refrec = &rec->rec_jrefrec;
1972 if (debug && mvrec->jm_op == JOP_MVREF)
1973 printf("ino move: ino %ju, parent %ju, "
1974 "diroff %jd, oldoff %jd\n",
1975 (uintmax_t)mvrec->jm_ino, (uintmax_t)mvrec->jm_parent,
1976 (uintmax_t)mvrec->jm_newoff, (uintmax_t)mvrec->jm_oldoff);
1978 (refrec->jr_op == JOP_ADDREF || refrec->jr_op == JOP_REMREF))
1979 printf("ino ref: op %d, ino %ju, nlink %ju, "
1980 "parent %ju, diroff %jd\n",
1981 refrec->jr_op, (uintmax_t)refrec->jr_ino,
1982 (uintmax_t)refrec->jr_nlink,
1983 (uintmax_t)refrec->jr_parent, (uintmax_t)refrec->jr_diroff);
1984 sino = ino_lookup(((struct jrefrec *)rec)->jr_ino, 1);
1985 sino->si_hasrecs = 1;
1986 srec = errmalloc(sizeof(*srec));
1988 TAILQ_INSERT_TAIL(&sino->si_newrecs, srec, sr_next);
1992 * Add a reference adjustment to the sino list and eliminate dups. The
1993 * primary loop in ino_build_ref() checks for dups but new ones may be
1994 * created as a result of offset adjustments.
1997 ino_add_ref(struct suj_ino *sino, struct suj_rec *srec)
1999 struct jrefrec *refrec;
2000 struct suj_rec *srn;
2001 struct jrefrec *rrn;
2003 refrec = (struct jrefrec *)srec->sr_rec;
2005 * We walk backwards so that the oldest link count is preserved. If
2006 * an add record conflicts with a remove keep the remove. Redundant
2007 * removes are eliminated in ino_build_ref. Otherwise we keep the
2008 * oldest record at a given location.
2010 for (srn = TAILQ_LAST(&sino->si_recs, srechd); srn;
2011 srn = TAILQ_PREV(srn, srechd, sr_next)) {
2012 rrn = (struct jrefrec *)srn->sr_rec;
2013 if (rrn->jr_parent != refrec->jr_parent ||
2014 rrn->jr_diroff != refrec->jr_diroff)
2016 if (rrn->jr_op == JOP_REMREF || refrec->jr_op == JOP_ADDREF) {
2017 rrn->jr_mode = refrec->jr_mode;
2023 * Replace the record in place with the old nlink in case
2024 * we replace the head of the list. Abandon srec as a dup.
2026 refrec->jr_nlink = rrn->jr_nlink;
2027 srn->sr_rec = srec->sr_rec;
2030 TAILQ_INSERT_TAIL(&sino->si_recs, srec, sr_next);
2034 * Create a duplicate of a reference at a previous location.
2037 ino_dup_ref(struct suj_ino *sino, struct jrefrec *refrec, off_t diroff)
2039 struct jrefrec *rrn;
2040 struct suj_rec *srn;
2042 rrn = errmalloc(sizeof(*refrec));
2044 rrn->jr_op = JOP_ADDREF;
2045 rrn->jr_diroff = diroff;
2046 srn = errmalloc(sizeof(*srn));
2047 srn->sr_rec = (union jrec *)rrn;
2048 ino_add_ref(sino, srn);
2052 * Add a reference to the list at all known locations. We follow the offset
2053 * changes for a single instance and create duplicate add refs at each so
2054 * that we can tolerate any version of the directory block. Eliminate
2055 * removes which collide with adds that are seen in the journal. They should
2056 * not adjust the link count down.
2059 ino_build_ref(struct suj_ino *sino, struct suj_rec *srec)
2061 struct jrefrec *refrec;
2062 struct jmvrec *mvrec;
2063 struct suj_rec *srp;
2064 struct suj_rec *srn;
2065 struct jrefrec *rrn;
2068 refrec = (struct jrefrec *)srec->sr_rec;
2070 * Search for a mvrec that matches this offset. Whether it's an add
2071 * or a remove we can delete the mvref after creating a dup record in
2074 if (!TAILQ_EMPTY(&sino->si_movs)) {
2075 diroff = refrec->jr_diroff;
2076 for (srn = TAILQ_LAST(&sino->si_movs, srechd); srn; srn = srp) {
2077 srp = TAILQ_PREV(srn, srechd, sr_next);
2078 mvrec = (struct jmvrec *)srn->sr_rec;
2079 if (mvrec->jm_parent != refrec->jr_parent ||
2080 mvrec->jm_newoff != diroff)
2082 diroff = mvrec->jm_oldoff;
2083 TAILQ_REMOVE(&sino->si_movs, srn, sr_next);
2085 ino_dup_ref(sino, refrec, diroff);
2089 * If a remove wasn't eliminated by an earlier add just append it to
2092 if (refrec->jr_op == JOP_REMREF) {
2093 ino_add_ref(sino, srec);
2097 * Walk the list of records waiting to be added to the list. We
2098 * must check for moves that apply to our current offset and remove
2099 * them from the list. Remove any duplicates to eliminate removes
2100 * with corresponding adds.
2102 TAILQ_FOREACH_SAFE(srn, &sino->si_newrecs, sr_next, srp) {
2103 switch (srn->sr_rec->rec_jrefrec.jr_op) {
2106 * This should actually be an error we should
2107 * have a remove for every add journaled.
2109 rrn = (struct jrefrec *)srn->sr_rec;
2110 if (rrn->jr_parent != refrec->jr_parent ||
2111 rrn->jr_diroff != refrec->jr_diroff)
2113 TAILQ_REMOVE(&sino->si_newrecs, srn, sr_next);
2117 * Once we remove the current iteration of the
2118 * record at this address we're done.
2120 rrn = (struct jrefrec *)srn->sr_rec;
2121 if (rrn->jr_parent != refrec->jr_parent ||
2122 rrn->jr_diroff != refrec->jr_diroff)
2124 TAILQ_REMOVE(&sino->si_newrecs, srn, sr_next);
2125 ino_add_ref(sino, srec);
2129 * Update our diroff based on any moves that match
2130 * and remove the move.
2132 mvrec = (struct jmvrec *)srn->sr_rec;
2133 if (mvrec->jm_parent != refrec->jr_parent ||
2134 mvrec->jm_oldoff != refrec->jr_diroff)
2136 ino_dup_ref(sino, refrec, mvrec->jm_oldoff);
2137 refrec->jr_diroff = mvrec->jm_newoff;
2138 TAILQ_REMOVE(&sino->si_newrecs, srn, sr_next);
2141 err_suj("ino_build_ref: Unknown op %d\n",
2142 srn->sr_rec->rec_jrefrec.jr_op);
2145 ino_add_ref(sino, srec);
2149 * Walk the list of new records and add them in-order resolving any
2150 * dups and adjusted offsets.
2153 ino_build(struct suj_ino *sino)
2155 struct suj_rec *srec;
2157 while ((srec = TAILQ_FIRST(&sino->si_newrecs)) != NULL) {
2158 TAILQ_REMOVE(&sino->si_newrecs, srec, sr_next);
2159 switch (srec->sr_rec->rec_jrefrec.jr_op) {
2162 ino_build_ref(sino, srec);
2166 * Add this mvrec to the queue of pending mvs.
2168 TAILQ_INSERT_TAIL(&sino->si_movs, srec, sr_next);
2171 err_suj("ino_build: Unknown op %d\n",
2172 srec->sr_rec->rec_jrefrec.jr_op);
2175 if (TAILQ_EMPTY(&sino->si_recs))
2176 sino->si_hasrecs = 0;
2180 * Modify journal records so they refer to the base block number
2181 * and a start and end frag range. This is to facilitate the discovery
2182 * of overlapping fragment allocations.
2185 blk_build(struct jblkrec *blkrec)
2187 struct suj_rec *srec;
2188 struct suj_blk *sblk;
2189 struct jblkrec *blkrn;
2194 printf("blk_build: op %d blkno %jd frags %d oldfrags %d "
2195 "ino %ju lbn %jd\n",
2196 blkrec->jb_op, (uintmax_t)blkrec->jb_blkno,
2197 blkrec->jb_frags, blkrec->jb_oldfrags,
2198 (uintmax_t)blkrec->jb_ino, (uintmax_t)blkrec->jb_lbn);
2200 blk = blknum(fs, blkrec->jb_blkno);
2201 frag = fragnum(fs, blkrec->jb_blkno);
2202 sblk = blk_lookup(blk, 1);
2204 * Rewrite the record using oldfrags to indicate the offset into
2205 * the block. Leave jb_frags as the actual allocated count.
2207 blkrec->jb_blkno -= frag;
2208 blkrec->jb_oldfrags = frag;
2209 if (blkrec->jb_oldfrags + blkrec->jb_frags > fs->fs_frag)
2210 err_suj("Invalid fragment count %d oldfrags %d\n",
2211 blkrec->jb_frags, frag);
2213 * Detect dups. If we detect a dup we always discard the oldest
2214 * record as it is superseded by the new record. This speeds up
2215 * later stages but also eliminates free records which are used
2216 * to indicate that the contents of indirects can be trusted.
2218 TAILQ_FOREACH(srec, &sblk->sb_recs, sr_next) {
2219 blkrn = (struct jblkrec *)srec->sr_rec;
2220 if (blkrn->jb_ino != blkrec->jb_ino ||
2221 blkrn->jb_lbn != blkrec->jb_lbn ||
2222 blkrn->jb_blkno != blkrec->jb_blkno ||
2223 blkrn->jb_frags != blkrec->jb_frags ||
2224 blkrn->jb_oldfrags != blkrec->jb_oldfrags)
2227 printf("Removed dup.\n");
2228 /* Discard the free which is a dup with an alloc. */
2229 if (blkrec->jb_op == JOP_FREEBLK)
2231 TAILQ_REMOVE(&sblk->sb_recs, srec, sr_next);
2235 srec = errmalloc(sizeof(*srec));
2236 srec->sr_rec = (union jrec *)blkrec;
2237 TAILQ_INSERT_TAIL(&sblk->sb_recs, srec, sr_next);
2241 ino_build_trunc(struct jtrncrec *rec)
2243 struct suj_ino *sino;
2246 printf("ino_build_trunc: op %d ino %ju, size %jd\n",
2247 rec->jt_op, (uintmax_t)rec->jt_ino,
2248 (uintmax_t)rec->jt_size);
2249 sino = ino_lookup(rec->jt_ino, 1);
2250 if (rec->jt_op == JOP_SYNC) {
2251 sino->si_trunc = NULL;
2254 if (sino->si_trunc == NULL || sino->si_trunc->jt_size > rec->jt_size)
2255 sino->si_trunc = rec;
2259 * Build up tables of the operations we need to recover.
2264 struct suj_seg *seg;
2269 TAILQ_FOREACH(seg, &allsegs, ss_next) {
2271 printf("seg %jd has %d records, oldseq %jd.\n",
2272 seg->ss_rec.jsr_seq, seg->ss_rec.jsr_cnt,
2273 seg->ss_rec.jsr_oldest);
2275 rec = (union jrec *)seg->ss_blk;
2276 for (i = 0; i < seg->ss_rec.jsr_cnt; off += JREC_SIZE, rec++) {
2277 /* skip the segrec. */
2278 if ((off % real_dev_bsize) == 0)
2280 switch (rec->rec_jrefrec.jr_op) {
2288 blk_build((struct jblkrec *)rec);
2292 ino_build_trunc((struct jtrncrec *)rec);
2295 err_suj("Unknown journal operation %d (%d)\n",
2296 rec->rec_jrefrec.jr_op, off);
2304 * Prune the journal segments to those we care about based on the
2305 * oldest sequence in the newest segment. Order the segment list
2306 * based on sequence number.
2311 struct suj_seg *seg;
2312 struct suj_seg *segn;
2317 printf("Pruning up to %jd\n", oldseq);
2318 /* First free the expired segments. */
2319 TAILQ_FOREACH_SAFE(seg, &allsegs, ss_next, segn) {
2320 if (seg->ss_rec.jsr_seq >= oldseq)
2322 TAILQ_REMOVE(&allsegs, seg, ss_next);
2326 /* Next ensure that segments are ordered properly. */
2327 seg = TAILQ_FIRST(&allsegs);
2330 printf("Empty journal\n");
2333 newseq = seg->ss_rec.jsr_seq;
2335 seg = TAILQ_LAST(&allsegs, seghd);
2336 if (seg->ss_rec.jsr_seq >= newseq)
2338 TAILQ_REMOVE(&allsegs, seg, ss_next);
2339 TAILQ_INSERT_HEAD(&allsegs, seg, ss_next);
2340 newseq = seg->ss_rec.jsr_seq;
2343 if (newseq != oldseq) {
2344 TAILQ_FOREACH(seg, &allsegs, ss_next) {
2345 printf("%jd, ", seg->ss_rec.jsr_seq);
2348 err_suj("Journal file sequence mismatch %jd != %jd\n",
2352 * The kernel may asynchronously write segments which can create
2353 * gaps in the sequence space. Throw away any segments after the
2354 * gap as the kernel guarantees only those that are contiguously
2355 * reachable are marked as completed.
2358 TAILQ_FOREACH_SAFE(seg, &allsegs, ss_next, segn) {
2359 if (!discard && newseq++ == seg->ss_rec.jsr_seq) {
2360 jrecs += seg->ss_rec.jsr_cnt;
2361 jbytes += seg->ss_rec.jsr_blocks * real_dev_bsize;
2366 printf("Journal order mismatch %jd != %jd pruning\n",
2367 newseq-1, seg->ss_rec.jsr_seq);
2368 TAILQ_REMOVE(&allsegs, seg, ss_next);
2373 printf("Processing journal segments from %jd to %jd\n",
2378 * Verify the journal inode before attempting to read records.
2381 suj_verifyino(union dinode *ip)
2384 if (DIP(ip, di_nlink) != 1) {
2385 printf("Invalid link count %d for journal inode %ju\n",
2386 DIP(ip, di_nlink), (uintmax_t)sujino);
2390 if ((DIP(ip, di_flags) & (SF_IMMUTABLE | SF_NOUNLINK)) !=
2391 (SF_IMMUTABLE | SF_NOUNLINK)) {
2392 printf("Invalid flags 0x%X for journal inode %ju\n",
2393 DIP(ip, di_flags), (uintmax_t)sujino);
2397 if (DIP(ip, di_mode) != (IFREG | IREAD)) {
2398 printf("Invalid mode %o for journal inode %ju\n",
2399 DIP(ip, di_mode), (uintmax_t)sujino);
2403 if (DIP(ip, di_size) < SUJ_MIN) {
2404 printf("Invalid size %jd for journal inode %ju\n",
2405 DIP(ip, di_size), (uintmax_t)sujino);
2409 if (DIP(ip, di_modrev) != fs->fs_mtime) {
2410 printf("Journal timestamp does not match fs mount time\n");
2418 struct jextent *jb_extent; /* Extent array. */
2419 int jb_avail; /* Available extents. */
2420 int jb_used; /* Last used extent. */
2421 int jb_head; /* Allocator head. */
2422 int jb_off; /* Allocator extent offset. */
2425 ufs2_daddr_t je_daddr; /* Disk block address. */
2426 int je_blocks; /* Disk block count. */
2429 static struct jblocks *suj_jblocks;
2431 static struct jblocks *
2432 jblocks_create(void)
2434 struct jblocks *jblocks;
2437 jblocks = errmalloc(sizeof(*jblocks));
2438 jblocks->jb_avail = 10;
2439 jblocks->jb_used = 0;
2440 jblocks->jb_head = 0;
2441 jblocks->jb_off = 0;
2442 size = sizeof(struct jextent) * jblocks->jb_avail;
2443 jblocks->jb_extent = errmalloc(size);
2444 bzero(jblocks->jb_extent, size);
2450 * Return the next available disk block and the amount of contiguous
2451 * free space it contains.
2454 jblocks_next(struct jblocks *jblocks, int bytes, int *actual)
2456 struct jextent *jext;
2461 blocks = bytes / disk->d_bsize;
2462 jext = &jblocks->jb_extent[jblocks->jb_head];
2463 freecnt = jext->je_blocks - jblocks->jb_off;
2465 jblocks->jb_off = 0;
2466 if (++jblocks->jb_head > jblocks->jb_used)
2468 jext = &jblocks->jb_extent[jblocks->jb_head];
2469 freecnt = jext->je_blocks;
2471 if (freecnt > blocks)
2473 *actual = freecnt * disk->d_bsize;
2474 daddr = jext->je_daddr + jblocks->jb_off;
2480 * Advance the allocation head by a specified number of bytes, consuming
2481 * one journal segment.
2484 jblocks_advance(struct jblocks *jblocks, int bytes)
2487 jblocks->jb_off += bytes / disk->d_bsize;
2491 jblocks_destroy(struct jblocks *jblocks)
2494 free(jblocks->jb_extent);
2499 jblocks_add(struct jblocks *jblocks, ufs2_daddr_t daddr, int blocks)
2501 struct jextent *jext;
2504 jext = &jblocks->jb_extent[jblocks->jb_used];
2505 /* Adding the first block. */
2506 if (jext->je_daddr == 0) {
2507 jext->je_daddr = daddr;
2508 jext->je_blocks = blocks;
2511 /* Extending the last extent. */
2512 if (jext->je_daddr + jext->je_blocks == daddr) {
2513 jext->je_blocks += blocks;
2516 /* Adding a new extent. */
2517 if (++jblocks->jb_used == jblocks->jb_avail) {
2518 jblocks->jb_avail *= 2;
2519 size = sizeof(struct jextent) * jblocks->jb_avail;
2520 jext = errmalloc(size);
2522 bcopy(jblocks->jb_extent, jext,
2523 sizeof(struct jextent) * jblocks->jb_used);
2524 free(jblocks->jb_extent);
2525 jblocks->jb_extent = jext;
2527 jext = &jblocks->jb_extent[jblocks->jb_used];
2528 jext->je_daddr = daddr;
2529 jext->je_blocks = blocks;
2535 * Add a file block from the journal to the extent map. We can't read
2536 * each file block individually because the kernel treats it as a circular
2537 * buffer and segments may span mutliple contiguous blocks.
2540 suj_add_block(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, int frags)
2543 jblocks_add(suj_jblocks, fsbtodb(fs, blk), fsbtodb(fs, frags));
2549 uint8_t block[1 * 1024 * 1024];
2550 struct suj_seg *seg;
2551 struct jsegrec *recn;
2552 struct jsegrec *rec;
2561 * Read records until we exhaust the journal space. If we find
2562 * an invalid record we start searching for a valid segment header
2563 * at the next block. This is because we don't have a head/tail
2564 * pointer and must recover the information indirectly. At the gap
2565 * between the head and tail we won't necessarily have a valid
2570 size = sizeof(block);
2571 blk = jblocks_next(suj_jblocks, size, &readsize);
2576 * Read 1MB at a time and scan for records within this block.
2578 if (bread(disk, blk, &block, size) == -1) {
2579 err_suj("Error reading journal block %jd\n",
2582 for (rec = (void *)block; size; size -= recsize,
2583 rec = (struct jsegrec *)((uintptr_t)rec + recsize)) {
2584 recsize = real_dev_bsize;
2585 if (rec->jsr_time != fs->fs_mtime) {
2587 printf("Rec time %jd != fs mtime %jd\n",
2588 rec->jsr_time, fs->fs_mtime);
2589 jblocks_advance(suj_jblocks, recsize);
2592 if (rec->jsr_cnt == 0) {
2594 printf("Found illegal count %d\n",
2596 jblocks_advance(suj_jblocks, recsize);
2599 blocks = rec->jsr_blocks;
2600 recsize = blocks * real_dev_bsize;
2601 if (recsize > size) {
2603 * We may just have run out of buffer, restart
2604 * the loop to re-read from this spot.
2606 if (size < fs->fs_bsize &&
2608 recsize <= fs->fs_bsize)
2611 printf("Found invalid segsize %d > %d\n",
2613 recsize = real_dev_bsize;
2614 jblocks_advance(suj_jblocks, recsize);
2618 * Verify that all blocks in the segment are present.
2620 for (i = 1; i < blocks; i++) {
2621 recn = (void *)((uintptr_t)rec) + i *
2623 if (recn->jsr_seq == rec->jsr_seq &&
2624 recn->jsr_time == rec->jsr_time)
2627 printf("Incomplete record %jd (%d)\n",
2629 recsize = i * real_dev_bsize;
2630 jblocks_advance(suj_jblocks, recsize);
2633 seg = errmalloc(sizeof(*seg));
2634 seg->ss_blk = errmalloc(recsize);
2636 bcopy((void *)rec, seg->ss_blk, recsize);
2637 if (rec->jsr_oldest > oldseq)
2638 oldseq = rec->jsr_oldest;
2639 TAILQ_INSERT_TAIL(&allsegs, seg, ss_next);
2640 jblocks_advance(suj_jblocks, recsize);
2646 * Search a directory block for the SUJ_FILE.
2649 suj_find(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, int frags)
2651 char block[MAXBSIZE];
2658 bytes = lfragtosize(fs, frags);
2659 if (bread(disk, fsbtodb(fs, blk), block, bytes) <= 0)
2660 err_suj("Failed to read ROOTINO directory block %jd\n", blk);
2661 for (off = 0; off < bytes; off += dp->d_reclen) {
2662 dp = (struct direct *)&block[off];
2663 if (dp->d_reclen == 0)
2667 if (dp->d_namlen != strlen(SUJ_FILE))
2669 if (bcmp(dp->d_name, SUJ_FILE, dp->d_namlen) != 0)
2677 * Orchestrate the verification of a filesystem via the softupdates journal.
2680 suj_check(const char *filesys)
2686 struct suj_seg *seg;
2687 struct suj_seg *segn;
2693 * Set an exit point when SUJ check failed
2695 retval = setjmp(jmpbuf);
2697 pwarn("UNEXPECTED SU+J INCONSISTENCY\n");
2698 TAILQ_FOREACH_SAFE(seg, &allsegs, ss_next, segn) {
2699 TAILQ_REMOVE(&allsegs, seg, ss_next);
2703 if (reply("FALLBACK TO FULL FSCK") == 0) {
2711 * Find the journal inode.
2713 ip = ino_read(ROOTINO);
2715 ino_visit(ip, ROOTINO, suj_find, 0);
2717 printf("Journal inode removed. Use tunefs to re-create.\n");
2718 sblock.fs_flags &= ~FS_SUJ;
2719 sblock.fs_sujfree = 0;
2723 * Fetch the journal inode and verify it.
2725 jip = ino_read(sujino);
2726 printf("** SU+J Recovering %s\n", filesys);
2727 if (suj_verifyino(jip) != 0)
2730 * Build a list of journal blocks in jblocks before parsing the
2731 * available journal blocks in with suj_read().
2733 printf("** Reading %jd byte journal from inode %ju.\n",
2734 DIP(jip, di_size), (uintmax_t)sujino);
2735 suj_jblocks = jblocks_create();
2736 blocks = ino_visit(jip, sujino, suj_add_block, 0);
2737 if (blocks != numfrags(fs, DIP(jip, di_size))) {
2738 printf("Sparse journal inode %ju.\n", (uintmax_t)sujino);
2742 jblocks_destroy(suj_jblocks);
2744 if (preen || reply("RECOVER")) {
2745 printf("** Building recovery table.\n");
2749 printf("** Resolving unreferenced inode list.\n");
2751 printf("** Processing journal entries.\n");
2753 cg_apply(cg_check_blk);
2754 cg_apply(cg_adj_blk);
2755 cg_apply(cg_check_ino);
2757 if (preen == 0 && (jrecs > 0 || jbytes > 0) && reply("WRITE CHANGES") == 0)
2760 * To remain idempotent with partial truncations the free bitmaps
2761 * must be written followed by indirect blocks and lastly inode
2762 * blocks. This preserves access to the modified pointers until
2767 cg_apply(cg_write_inos);
2768 /* Write back superblock. */
2770 if (jrecs > 0 || jbytes > 0) {
2771 printf("** %jd journal records in %jd bytes for %.2f%% utilization\n",
2772 jrecs, jbytes, ((float)jrecs / (float)(jbytes / JREC_SIZE)) * 100);
2773 printf("** Freed %jd inodes (%jd dirs) %jd blocks, and %jd frags.\n",
2774 freeinos, freedir, freeblocks, freefrags);
2785 for (i = 0; i < SUJ_HASHSIZE; i++) {
2786 LIST_INIT(&cghash[i]);
2787 LIST_INIT(&dbhash[i]);
2791 TAILQ_INIT(&allsegs);