xref: /freebsd/sbin/fsck_ffs/suj.c (revision f61e92ca5a23450bc28169bbdd71d7674df98c19)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright 2009, 2010 Jeffrey W. Roberson <jeff@FreeBSD.org>
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
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.
15  *
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
26  * SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include <sys/param.h>
33 #include <sys/disk.h>
34 #include <sys/disklabel.h>
35 #include <sys/mount.h>
36 #include <sys/stat.h>
37 
38 #include <ufs/ufs/ufsmount.h>
39 #include <ufs/ufs/dinode.h>
40 #include <ufs/ufs/dir.h>
41 #include <ufs/ffs/fs.h>
42 
43 #include <assert.h>
44 #include <err.h>
45 #include <setjmp.h>
46 #include <stdarg.h>
47 #include <stdio.h>
48 #include <stdlib.h>
49 #include <stdint.h>
50 #include <libufs.h>
51 #include <string.h>
52 #include <strings.h>
53 #include <sysexits.h>
54 #include <time.h>
55 
56 #include "fsck.h"
57 
58 #define	DOTDOT_OFFSET	DIRECTSIZ(1)
59 
60 struct suj_seg {
61 	TAILQ_ENTRY(suj_seg) ss_next;
62 	struct jsegrec	ss_rec;
63 	uint8_t		*ss_blk;
64 };
65 
66 struct suj_rec {
67 	TAILQ_ENTRY(suj_rec) sr_next;
68 	union jrec	*sr_rec;
69 };
70 TAILQ_HEAD(srechd, suj_rec);
71 
72 struct suj_ino {
73 	LIST_ENTRY(suj_ino)	si_next;
74 	struct srechd		si_recs;
75 	struct srechd		si_newrecs;
76 	struct srechd		si_movs;
77 	struct jtrncrec		*si_trunc;
78 	ino_t			si_ino;
79 	char			si_skipparent;
80 	char			si_hasrecs;
81 	char			si_blkadj;
82 	char			si_linkadj;
83 	int			si_mode;
84 	nlink_t			si_nlinkadj;
85 	nlink_t			si_nlink;
86 	nlink_t			si_dotlinks;
87 };
88 LIST_HEAD(inohd, suj_ino);
89 
90 struct suj_blk {
91 	LIST_ENTRY(suj_blk)	sb_next;
92 	struct srechd		sb_recs;
93 	ufs2_daddr_t		sb_blk;
94 };
95 LIST_HEAD(blkhd, suj_blk);
96 
97 struct suj_cg {
98 	LIST_ENTRY(suj_cg)	sc_next;
99 	struct blkhd		sc_blkhash[HASHSIZE];
100 	struct inohd		sc_inohash[HASHSIZE];
101 	struct ino_blk		*sc_lastiblk;
102 	struct suj_ino		*sc_lastino;
103 	struct suj_blk		*sc_lastblk;
104 	struct bufarea		*sc_cgbp;
105 	struct cg		*sc_cgp;
106 	int			sc_cgx;
107 };
108 
109 static LIST_HEAD(cghd, suj_cg) cghash[HASHSIZE];
110 static struct suj_cg *lastcg;
111 
112 static TAILQ_HEAD(seghd, suj_seg) allsegs;
113 static uint64_t oldseq;
114 static struct fs *fs = NULL;
115 static ino_t sujino;
116 
117 /*
118  * Summary statistics.
119  */
120 static uint64_t freefrags;
121 static uint64_t freeblocks;
122 static uint64_t freeinos;
123 static uint64_t freedir;
124 static uint64_t jbytes;
125 static uint64_t jrecs;
126 
127 static jmp_buf	jmpbuf;
128 
129 typedef void (*ino_visitor)(ino_t, ufs_lbn_t, ufs2_daddr_t, int);
130 static void err_suj(const char *, ...) __dead2;
131 static void ino_trunc(ino_t, off_t);
132 static void ino_decr(ino_t);
133 static void ino_adjust(struct suj_ino *);
134 static void ino_build(struct suj_ino *);
135 static int blk_isfree(ufs2_daddr_t);
136 static void initsuj(void);
137 
138 static void *
139 errmalloc(size_t n)
140 {
141 	void *a;
142 
143 	a = Malloc(n);
144 	if (a == NULL)
145 		err(EX_OSERR, "malloc(%zu)", n);
146 	return (a);
147 }
148 
149 /*
150  * When hit a fatal error in journalling check, print out
151  * the error and then offer to fallback to normal fsck.
152  */
153 static void
154 err_suj(const char * restrict fmt, ...)
155 {
156 	va_list ap;
157 
158 	if (preen)
159 		(void)fprintf(stdout, "%s: ", cdevname);
160 
161 	va_start(ap, fmt);
162 	(void)vfprintf(stdout, fmt, ap);
163 	va_end(ap);
164 
165 	longjmp(jmpbuf, -1);
166 }
167 
168 /*
169  * Lookup a cg by number in the hash so we can keep track of which cgs
170  * need stats rebuilt.
171  */
172 static struct suj_cg *
173 cg_lookup(int cgx)
174 {
175 	struct cghd *hd;
176 	struct suj_cg *sc;
177 	struct bufarea *cgbp;
178 
179 	if (cgx < 0 || cgx >= fs->fs_ncg)
180 		err_suj("Bad cg number %d\n", cgx);
181 	if (lastcg && lastcg->sc_cgx == cgx)
182 		return (lastcg);
183 	cgbp = cglookup(cgx);
184 	if (!check_cgmagic(cgx, cgbp, 0))
185 		err_suj("UNABLE TO REBUILD CYLINDER GROUP %d", cgx);
186 	hd = &cghash[HASH(cgx)];
187 	LIST_FOREACH(sc, hd, sc_next)
188 		if (sc->sc_cgx == cgx) {
189 			sc->sc_cgbp = cgbp;
190 			sc->sc_cgp = sc->sc_cgbp->b_un.b_cg;
191 			lastcg = sc;
192 			return (sc);
193 		}
194 	sc = errmalloc(sizeof(*sc));
195 	bzero(sc, sizeof(*sc));
196 	sc->sc_cgbp = cgbp;
197 	sc->sc_cgp = sc->sc_cgbp->b_un.b_cg;
198 	sc->sc_cgx = cgx;
199 	LIST_INSERT_HEAD(hd, sc, sc_next);
200 	return (sc);
201 }
202 
203 /*
204  * Lookup an inode number in the hash and allocate a suj_ino if it does
205  * not exist.
206  */
207 static struct suj_ino *
208 ino_lookup(ino_t ino, int creat)
209 {
210 	struct suj_ino *sino;
211 	struct inohd *hd;
212 	struct suj_cg *sc;
213 
214 	sc = cg_lookup(ino_to_cg(fs, ino));
215 	if (sc->sc_lastino && sc->sc_lastino->si_ino == ino)
216 		return (sc->sc_lastino);
217 	hd = &sc->sc_inohash[HASH(ino)];
218 	LIST_FOREACH(sino, hd, si_next)
219 		if (sino->si_ino == ino)
220 			return (sino);
221 	if (creat == 0)
222 		return (NULL);
223 	sino = errmalloc(sizeof(*sino));
224 	bzero(sino, sizeof(*sino));
225 	sino->si_ino = ino;
226 	TAILQ_INIT(&sino->si_recs);
227 	TAILQ_INIT(&sino->si_newrecs);
228 	TAILQ_INIT(&sino->si_movs);
229 	LIST_INSERT_HEAD(hd, sino, si_next);
230 
231 	return (sino);
232 }
233 
234 /*
235  * Lookup a block number in the hash and allocate a suj_blk if it does
236  * not exist.
237  */
238 static struct suj_blk *
239 blk_lookup(ufs2_daddr_t blk, int creat)
240 {
241 	struct suj_blk *sblk;
242 	struct suj_cg *sc;
243 	struct blkhd *hd;
244 
245 	sc = cg_lookup(dtog(fs, blk));
246 	if (sc->sc_lastblk && sc->sc_lastblk->sb_blk == blk)
247 		return (sc->sc_lastblk);
248 	hd = &sc->sc_blkhash[HASH(fragstoblks(fs, blk))];
249 	LIST_FOREACH(sblk, hd, sb_next)
250 		if (sblk->sb_blk == blk)
251 			return (sblk);
252 	if (creat == 0)
253 		return (NULL);
254 	sblk = errmalloc(sizeof(*sblk));
255 	bzero(sblk, sizeof(*sblk));
256 	sblk->sb_blk = blk;
257 	TAILQ_INIT(&sblk->sb_recs);
258 	LIST_INSERT_HEAD(hd, sblk, sb_next);
259 
260 	return (sblk);
261 }
262 
263 static int
264 blk_overlaps(struct jblkrec *brec, ufs2_daddr_t start, int frags)
265 {
266 	ufs2_daddr_t bstart;
267 	ufs2_daddr_t bend;
268 	ufs2_daddr_t end;
269 
270 	end = start + frags;
271 	bstart = brec->jb_blkno + brec->jb_oldfrags;
272 	bend = bstart + brec->jb_frags;
273 	if (start < bend && end > bstart)
274 		return (1);
275 	return (0);
276 }
277 
278 static int
279 blk_equals(struct jblkrec *brec, ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t start,
280     int frags)
281 {
282 
283 	if (brec->jb_ino != ino || brec->jb_lbn != lbn)
284 		return (0);
285 	if (brec->jb_blkno + brec->jb_oldfrags != start)
286 		return (0);
287 	if (brec->jb_frags < frags)
288 		return (0);
289 	return (1);
290 }
291 
292 static void
293 blk_setmask(struct jblkrec *brec, int *mask)
294 {
295 	int i;
296 
297 	for (i = brec->jb_oldfrags; i < brec->jb_oldfrags + brec->jb_frags; i++)
298 		*mask |= 1 << i;
299 }
300 
301 /*
302  * Determine whether a given block has been reallocated to a new location.
303  * Returns a mask of overlapping bits if any frags have been reused or
304  * zero if the block has not been re-used and the contents can be trusted.
305  *
306  * This is used to ensure that an orphaned pointer due to truncate is safe
307  * to be freed.  The mask value can be used to free partial blocks.
308  */
309 static int
310 blk_freemask(ufs2_daddr_t blk, ino_t ino, ufs_lbn_t lbn, int frags)
311 {
312 	struct suj_blk *sblk;
313 	struct suj_rec *srec;
314 	struct jblkrec *brec;
315 	int mask;
316 	int off;
317 
318 	/*
319 	 * To be certain we're not freeing a reallocated block we lookup
320 	 * this block in the blk hash and see if there is an allocation
321 	 * journal record that overlaps with any fragments in the block
322 	 * we're concerned with.  If any fragments have ben reallocated
323 	 * the block has already been freed and re-used for another purpose.
324 	 */
325 	mask = 0;
326 	sblk = blk_lookup(blknum(fs, blk), 0);
327 	if (sblk == NULL)
328 		return (0);
329 	off = blk - sblk->sb_blk;
330 	TAILQ_FOREACH(srec, &sblk->sb_recs, sr_next) {
331 		brec = (struct jblkrec *)srec->sr_rec;
332 		/*
333 		 * If the block overlaps but does not match
334 		 * exactly this record refers to the current
335 		 * location.
336 		 */
337 		if (blk_overlaps(brec, blk, frags) == 0)
338 			continue;
339 		if (blk_equals(brec, ino, lbn, blk, frags) == 1)
340 			mask = 0;
341 		else
342 			blk_setmask(brec, &mask);
343 	}
344 	if (debug)
345 		printf("blk_freemask: blk %jd sblk %jd off %d mask 0x%X\n",
346 		    blk, sblk->sb_blk, off, mask);
347 	return (mask >> off);
348 }
349 
350 /*
351  * Determine whether it is safe to follow an indirect.  It is not safe
352  * if any part of the indirect has been reallocated or the last journal
353  * entry was an allocation.  Just allocated indirects may not have valid
354  * pointers yet and all of their children will have their own records.
355  * It is also not safe to follow an indirect if the cg bitmap has been
356  * cleared as a new allocation may write to the block prior to the journal
357  * being written.
358  *
359  * Returns 1 if it's safe to follow the indirect and 0 otherwise.
360  */
361 static int
362 blk_isindir(ufs2_daddr_t blk, ino_t ino, ufs_lbn_t lbn)
363 {
364 	struct suj_blk *sblk;
365 	struct jblkrec *brec;
366 
367 	sblk = blk_lookup(blk, 0);
368 	if (sblk == NULL)
369 		return (1);
370 	if (TAILQ_EMPTY(&sblk->sb_recs))
371 		return (1);
372 	brec = (struct jblkrec *)TAILQ_LAST(&sblk->sb_recs, srechd)->sr_rec;
373 	if (blk_equals(brec, ino, lbn, blk, fs->fs_frag))
374 		if (brec->jb_op == JOP_FREEBLK)
375 			return (!blk_isfree(blk));
376 	return (0);
377 }
378 
379 /*
380  * Clear an inode from the cg bitmap.  If the inode was already clear return
381  * 0 so the caller knows it does not have to check the inode contents.
382  */
383 static int
384 ino_free(ino_t ino, int mode)
385 {
386 	struct suj_cg *sc;
387 	uint8_t *inosused;
388 	struct cg *cgp;
389 	int cg;
390 
391 	cg = ino_to_cg(fs, ino);
392 	ino = ino % fs->fs_ipg;
393 	sc = cg_lookup(cg);
394 	cgp = sc->sc_cgp;
395 	inosused = cg_inosused(cgp);
396 	/*
397 	 * The bitmap may never have made it to the disk so we have to
398 	 * conditionally clear.  We can avoid writing the cg in this case.
399 	 */
400 	if (isclr(inosused, ino))
401 		return (0);
402 	freeinos++;
403 	clrbit(inosused, ino);
404 	if (ino < cgp->cg_irotor)
405 		cgp->cg_irotor = ino;
406 	cgp->cg_cs.cs_nifree++;
407 	if ((mode & IFMT) == IFDIR) {
408 		freedir++;
409 		cgp->cg_cs.cs_ndir--;
410 	}
411 	cgdirty(sc->sc_cgbp);
412 
413 	return (1);
414 }
415 
416 /*
417  * Free 'frags' frags starting at filesystem block 'bno' skipping any frags
418  * set in the mask.
419  */
420 static void
421 blk_free(ufs2_daddr_t bno, int mask, int frags)
422 {
423 	ufs1_daddr_t fragno, cgbno;
424 	struct suj_cg *sc;
425 	struct cg *cgp;
426 	int i, cg;
427 	uint8_t *blksfree;
428 
429 	if (debug)
430 		printf("Freeing %d frags at blk %jd mask 0x%x\n",
431 		    frags, bno, mask);
432 	cg = dtog(fs, bno);
433 	sc = cg_lookup(cg);
434 	cgp = sc->sc_cgp;
435 	cgbno = dtogd(fs, bno);
436 	blksfree = cg_blksfree(cgp);
437 
438 	/*
439 	 * If it's not allocated we only wrote the journal entry
440 	 * and never the bitmaps.  Here we unconditionally clear and
441 	 * resolve the cg summary later.
442 	 */
443 	if (frags == fs->fs_frag && mask == 0) {
444 		fragno = fragstoblks(fs, cgbno);
445 		ffs_setblock(fs, blksfree, fragno);
446 		freeblocks++;
447 	} else {
448 		/*
449 		 * deallocate the fragment
450 		 */
451 		for (i = 0; i < frags; i++)
452 			if ((mask & (1 << i)) == 0 && isclr(blksfree, cgbno +i)) {
453 				freefrags++;
454 				setbit(blksfree, cgbno + i);
455 			}
456 	}
457 	cgdirty(sc->sc_cgbp);
458 }
459 
460 /*
461  * Returns 1 if the whole block starting at 'bno' is marked free and 0
462  * otherwise.
463  */
464 static int
465 blk_isfree(ufs2_daddr_t bno)
466 {
467 	struct suj_cg *sc;
468 
469 	sc = cg_lookup(dtog(fs, bno));
470 	return ffs_isblock(fs, cg_blksfree(sc->sc_cgp), dtogd(fs, bno));
471 }
472 
473 /*
474  * Determine whether a block exists at a particular lbn in an inode.
475  * Returns 1 if found, 0 if not.  lbn may be negative for indirects
476  * or ext blocks.
477  */
478 static int
479 blk_isat(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, int *frags)
480 {
481 	struct inode ip;
482 	union dinode *dp;
483 	ufs2_daddr_t nblk;
484 
485 	ginode(ino, &ip);
486 	dp = ip.i_dp;
487 	if (DIP(dp, di_nlink) == 0 || DIP(dp, di_mode) == 0) {
488 		irelse(&ip);
489 		return (0);
490 	}
491 	nblk = ino_blkatoff(dp, ino, lbn, frags, NULL);
492 	irelse(&ip);
493 	return (nblk == blk);
494 }
495 
496 /*
497  * Clear the directory entry at diroff that should point to child.  Minimal
498  * checking is done and it is assumed that this path was verified with isat.
499  */
500 static void
501 ino_clrat(ino_t parent, off_t diroff, ino_t child)
502 {
503 	union dinode *dip;
504 	struct direct *dp;
505 	struct inode ip;
506 	ufs2_daddr_t blk;
507 	struct bufarea *bp;
508 	ufs_lbn_t lbn;
509 	int blksize;
510 	int frags;
511 	int doff;
512 
513 	if (debug)
514 		printf("Clearing inode %ju from parent %ju at offset %jd\n",
515 		    (uintmax_t)child, (uintmax_t)parent, diroff);
516 
517 	lbn = lblkno(fs, diroff);
518 	doff = blkoff(fs, diroff);
519 	ginode(parent, &ip);
520 	dip = ip.i_dp;
521 	blk = ino_blkatoff(dip, parent, lbn, &frags, NULL);
522 	blksize = sblksize(fs, DIP(dip, di_size), lbn);
523 	irelse(&ip);
524 	bp = getdatablk(blk, blksize, BT_DIRDATA);
525 	if (bp->b_errs != 0)
526 		err_suj("ino_clrat: UNRECOVERABLE I/O ERROR");
527 	dp = (struct direct *)&bp->b_un.b_buf[doff];
528 	if (dp->d_ino != child)
529 		errx(1, "Inode %ju does not exist in %ju at %jd",
530 		    (uintmax_t)child, (uintmax_t)parent, diroff);
531 	dp->d_ino = 0;
532 	dirty(bp);
533 	brelse(bp);
534 	/*
535 	 * The actual .. reference count will already have been removed
536 	 * from the parent by the .. remref record.
537 	 */
538 }
539 
540 /*
541  * Determines whether a pointer to an inode exists within a directory
542  * at a specified offset.  Returns the mode of the found entry.
543  */
544 static int
545 ino_isat(ino_t parent, off_t diroff, ino_t child, int *mode, int *isdot)
546 {
547 	struct inode ip;
548 	union dinode *dip;
549 	struct bufarea *bp;
550 	struct direct *dp;
551 	ufs2_daddr_t blk;
552 	ufs_lbn_t lbn;
553 	int blksize;
554 	int frags;
555 	int dpoff;
556 	int doff;
557 
558 	*isdot = 0;
559 	ginode(parent, &ip);
560 	dip = ip.i_dp;
561 	*mode = DIP(dip, di_mode);
562 	if ((*mode & IFMT) != IFDIR) {
563 		if (debug) {
564 			/*
565 			 * This can happen if the parent inode
566 			 * was reallocated.
567 			 */
568 			if (*mode != 0)
569 				printf("Directory %ju has bad mode %o\n",
570 				    (uintmax_t)parent, *mode);
571 			else
572 				printf("Directory %ju has zero mode\n",
573 				    (uintmax_t)parent);
574 		}
575 		irelse(&ip);
576 		return (0);
577 	}
578 	lbn = lblkno(fs, diroff);
579 	doff = blkoff(fs, diroff);
580 	blksize = sblksize(fs, DIP(dip, di_size), lbn);
581 	if (diroff + DIRECTSIZ(1) > DIP(dip, di_size) || doff >= blksize) {
582 		if (debug)
583 			printf("ino %ju absent from %ju due to offset %jd"
584 			    " exceeding size %jd\n",
585 			    (uintmax_t)child, (uintmax_t)parent, diroff,
586 			    DIP(dip, di_size));
587 		irelse(&ip);
588 		return (0);
589 	}
590 	blk = ino_blkatoff(dip, parent, lbn, &frags, NULL);
591 	irelse(&ip);
592 	if (blk <= 0) {
593 		if (debug)
594 			printf("Sparse directory %ju", (uintmax_t)parent);
595 		return (0);
596 	}
597 	bp = getdatablk(blk, blksize, BT_DIRDATA);
598 	if (bp->b_errs != 0)
599 		err_suj("ino_isat: UNRECOVERABLE I/O ERROR");
600 	/*
601 	 * Walk through the records from the start of the block to be
602 	 * certain we hit a valid record and not some junk in the middle
603 	 * of a file name.  Stop when we reach or pass the expected offset.
604 	 */
605 	dpoff = rounddown(doff, DIRBLKSIZ);
606 	do {
607 		dp = (struct direct *)&bp->b_un.b_buf[dpoff];
608 		if (dpoff == doff)
609 			break;
610 		if (dp->d_reclen == 0)
611 			break;
612 		dpoff += dp->d_reclen;
613 	} while (dpoff <= doff);
614 	if (dpoff > fs->fs_bsize)
615 		err_suj("Corrupt directory block in dir ino %ju\n",
616 		    (uintmax_t)parent);
617 	/* Not found. */
618 	if (dpoff != doff) {
619 		if (debug)
620 			printf("ino %ju not found in %ju, lbn %jd, dpoff %d\n",
621 			    (uintmax_t)child, (uintmax_t)parent, lbn, dpoff);
622 		brelse(bp);
623 		return (0);
624 	}
625 	/*
626 	 * We found the item in question.  Record the mode and whether it's
627 	 * a . or .. link for the caller.
628 	 */
629 	if (dp->d_ino == child) {
630 		if (child == parent)
631 			*isdot = 1;
632 		else if (dp->d_namlen == 2 &&
633 		    dp->d_name[0] == '.' && dp->d_name[1] == '.')
634 			*isdot = 1;
635 		*mode = DTTOIF(dp->d_type);
636 		brelse(bp);
637 		return (1);
638 	}
639 	if (debug)
640 		printf("ino %ju doesn't match dirent ino %ju in parent %ju\n",
641 		    (uintmax_t)child, (uintmax_t)dp->d_ino, (uintmax_t)parent);
642 	brelse(bp);
643 	return (0);
644 }
645 
646 #define	VISIT_INDIR	0x0001
647 #define	VISIT_EXT	0x0002
648 #define	VISIT_ROOT	0x0004	/* Operation came via root & valid pointers. */
649 
650 /*
651  * Read an indirect level which may or may not be linked into an inode.
652  */
653 static void
654 indir_visit(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, uint64_t *frags,
655     ino_visitor visitor, int flags)
656 {
657 	struct bufarea *bp;
658 	ufs_lbn_t lbnadd;
659 	ufs2_daddr_t nblk;
660 	ufs_lbn_t nlbn;
661 	int level;
662 	int i;
663 
664 	/*
665 	 * Don't visit indirect blocks with contents we can't trust.  This
666 	 * should only happen when indir_visit() is called to complete a
667 	 * truncate that never finished and not when a pointer is found via
668 	 * an inode.
669 	 */
670 	if (blk == 0)
671 		return;
672 	level = lbn_level(lbn);
673 	if (level == -1)
674 		err_suj("Invalid level for lbn %jd\n", lbn);
675 	if ((flags & VISIT_ROOT) == 0 && blk_isindir(blk, ino, lbn) == 0) {
676 		if (debug)
677 			printf("blk %jd ino %ju lbn %jd(%d) is not indir.\n",
678 			    blk, (uintmax_t)ino, lbn, level);
679 		goto out;
680 	}
681 	lbnadd = 1;
682 	for (i = level; i > 0; i--)
683 		lbnadd *= NINDIR(fs);
684 	bp = getdatablk(blk, fs->fs_bsize, BT_LEVEL1 + level);
685 	if (bp->b_errs != 0)
686 		err_suj("indir_visit: UNRECOVERABLE I/O ERROR");
687 	for (i = 0; i < NINDIR(fs); i++) {
688 		if ((nblk = IBLK(bp, i)) == 0)
689 			continue;
690 		if (level == 0) {
691 			nlbn = -lbn + i * lbnadd;
692 			(*frags) += fs->fs_frag;
693 			visitor(ino, nlbn, nblk, fs->fs_frag);
694 		} else {
695 			nlbn = (lbn + 1) - (i * lbnadd);
696 			indir_visit(ino, nlbn, nblk, frags, visitor, flags);
697 		}
698 	}
699 	brelse(bp);
700 out:
701 	if (flags & VISIT_INDIR) {
702 		(*frags) += fs->fs_frag;
703 		visitor(ino, lbn, blk, fs->fs_frag);
704 	}
705 }
706 
707 /*
708  * Visit each block in an inode as specified by 'flags' and call a
709  * callback function.  The callback may inspect or free blocks.  The
710  * count of frags found according to the size in the file is returned.
711  * This is not valid for sparse files but may be used to determine
712  * the correct di_blocks for a file.
713  */
714 static uint64_t
715 ino_visit(union dinode *dp, ino_t ino, ino_visitor visitor, int flags)
716 {
717 	ufs_lbn_t nextlbn;
718 	ufs_lbn_t tmpval;
719 	ufs_lbn_t lbn;
720 	uint64_t size;
721 	uint64_t fragcnt;
722 	int mode;
723 	int frags;
724 	int i;
725 
726 	size = DIP(dp, di_size);
727 	mode = DIP(dp, di_mode) & IFMT;
728 	fragcnt = 0;
729 	if ((flags & VISIT_EXT) &&
730 	    fs->fs_magic == FS_UFS2_MAGIC && dp->dp2.di_extsize) {
731 		for (i = 0; i < UFS_NXADDR; i++) {
732 			if (dp->dp2.di_extb[i] == 0)
733 				continue;
734 			frags = sblksize(fs, dp->dp2.di_extsize, i);
735 			frags = numfrags(fs, frags);
736 			fragcnt += frags;
737 			visitor(ino, -1 - i, dp->dp2.di_extb[i], frags);
738 		}
739 	}
740 	/* Skip datablocks for short links and devices. */
741 	if (mode == IFBLK || mode == IFCHR ||
742 	    (mode == IFLNK && size < fs->fs_maxsymlinklen))
743 		return (fragcnt);
744 	for (i = 0; i < UFS_NDADDR; i++) {
745 		if (DIP(dp, di_db[i]) == 0)
746 			continue;
747 		frags = sblksize(fs, size, i);
748 		frags = numfrags(fs, frags);
749 		fragcnt += frags;
750 		visitor(ino, i, DIP(dp, di_db[i]), frags);
751 	}
752 	/*
753 	 * We know the following indirects are real as we're following
754 	 * real pointers to them.
755 	 */
756 	flags |= VISIT_ROOT;
757 	for (i = 0, tmpval = NINDIR(fs), lbn = UFS_NDADDR; i < UFS_NIADDR; i++,
758 	    lbn = nextlbn) {
759 		nextlbn = lbn + tmpval;
760 		tmpval *= NINDIR(fs);
761 		if (DIP(dp, di_ib[i]) == 0)
762 			continue;
763 		indir_visit(ino, -lbn - i, DIP(dp, di_ib[i]), &fragcnt, visitor,
764 		    flags);
765 	}
766 	return (fragcnt);
767 }
768 
769 /*
770  * Null visitor function used when we just want to count blocks and
771  * record the lbn.
772  */
773 ufs_lbn_t visitlbn;
774 static void
775 null_visit(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, int frags)
776 {
777 	if (lbn > 0)
778 		visitlbn = lbn;
779 }
780 
781 /*
782  * Recalculate di_blocks when we discover that a block allocation or
783  * free was not successfully completed.  The kernel does not roll this back
784  * because it would be too expensive to compute which indirects were
785  * reachable at the time the inode was written.
786  */
787 static void
788 ino_adjblks(struct suj_ino *sino)
789 {
790 	struct inode ip;
791 	union dinode *dp;
792 	uint64_t blocks;
793 	uint64_t frags;
794 	off_t isize;
795 	off_t size;
796 	ino_t ino;
797 
798 	ino = sino->si_ino;
799 	ginode(ino, &ip);
800 	dp = ip.i_dp;
801 	/* No need to adjust zero'd inodes. */
802 	if (DIP(dp, di_mode) == 0) {
803 		irelse(&ip);
804 		return;
805 	}
806 	/*
807 	 * Visit all blocks and count them as well as recording the last
808 	 * valid lbn in the file.  If the file size doesn't agree with the
809 	 * last lbn we need to truncate to fix it.  Otherwise just adjust
810 	 * the blocks count.
811 	 */
812 	visitlbn = 0;
813 	frags = ino_visit(dp, ino, null_visit, VISIT_INDIR | VISIT_EXT);
814 	blocks = fsbtodb(fs, frags);
815 	/*
816 	 * We assume the size and direct block list is kept coherent by
817 	 * softdep.  For files that have extended into indirects we truncate
818 	 * to the size in the inode or the maximum size permitted by
819 	 * populated indirects.
820 	 */
821 	if (visitlbn >= UFS_NDADDR) {
822 		isize = DIP(dp, di_size);
823 		size = lblktosize(fs, visitlbn + 1);
824 		if (isize > size)
825 			isize = size;
826 		/* Always truncate to free any unpopulated indirects. */
827 		ino_trunc(ino, isize);
828 		irelse(&ip);
829 		return;
830 	}
831 	if (blocks == DIP(dp, di_blocks)) {
832 		irelse(&ip);
833 		return;
834 	}
835 	if (debug)
836 		printf("ino %ju adjusting block count from %jd to %jd\n",
837 		    (uintmax_t)ino, DIP(dp, di_blocks), blocks);
838 	DIP_SET(dp, di_blocks, blocks);
839 	inodirty(&ip);
840 	irelse(&ip);
841 }
842 
843 static void
844 blk_free_visit(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, int frags)
845 {
846 
847 	blk_free(blk, blk_freemask(blk, ino, lbn, frags), frags);
848 }
849 
850 /*
851  * Free a block or tree of blocks that was previously rooted in ino at
852  * the given lbn.  If the lbn is an indirect all children are freed
853  * recursively.
854  */
855 static void
856 blk_free_lbn(ufs2_daddr_t blk, ino_t ino, ufs_lbn_t lbn, int frags, int follow)
857 {
858 	uint64_t resid;
859 	int mask;
860 
861 	mask = blk_freemask(blk, ino, lbn, frags);
862 	resid = 0;
863 	if (lbn <= -UFS_NDADDR && follow && mask == 0)
864 		indir_visit(ino, lbn, blk, &resid, blk_free_visit, VISIT_INDIR);
865 	else
866 		blk_free(blk, mask, frags);
867 }
868 
869 static void
870 ino_setskip(struct suj_ino *sino, ino_t parent)
871 {
872 	int isdot;
873 	int mode;
874 
875 	if (ino_isat(sino->si_ino, DOTDOT_OFFSET, parent, &mode, &isdot))
876 		sino->si_skipparent = 1;
877 }
878 
879 static void
880 ino_remref(ino_t parent, ino_t child, uint64_t diroff, int isdotdot)
881 {
882 	struct suj_ino *sino;
883 	struct suj_rec *srec;
884 	struct jrefrec *rrec;
885 
886 	/*
887 	 * Lookup this inode to see if we have a record for it.
888 	 */
889 	sino = ino_lookup(child, 0);
890 	/*
891 	 * Tell any child directories we've already removed their
892 	 * parent link cnt.  Don't try to adjust our link down again.
893 	 */
894 	if (sino != NULL && isdotdot == 0)
895 		ino_setskip(sino, parent);
896 	/*
897 	 * No valid record for this inode.  Just drop the on-disk
898 	 * link by one.
899 	 */
900 	if (sino == NULL || sino->si_hasrecs == 0) {
901 		ino_decr(child);
902 		return;
903 	}
904 	/*
905 	 * Use ino_adjust() if ino_check() has already processed this
906 	 * child.  If we lose the last non-dot reference to a
907 	 * directory it will be discarded.
908 	 */
909 	if (sino->si_linkadj) {
910 		if (sino->si_nlink == 0)
911 			err_suj("ino_remref: ino %ld mode 0%o about to go "
912 			    "negative\n", sino->si_ino, sino->si_mode);
913 		sino->si_nlink--;
914 		if (isdotdot)
915 			sino->si_dotlinks--;
916 		ino_adjust(sino);
917 		return;
918 	}
919 	/*
920 	 * If we haven't yet processed this inode we need to make
921 	 * sure we will successfully discover the lost path.  If not
922 	 * use nlinkadj to remember.
923 	 */
924 	TAILQ_FOREACH(srec, &sino->si_recs, sr_next) {
925 		rrec = (struct jrefrec *)srec->sr_rec;
926 		if (rrec->jr_parent == parent &&
927 		    rrec->jr_diroff == diroff)
928 			return;
929 	}
930 	sino->si_nlinkadj++;
931 }
932 
933 /*
934  * Free the children of a directory when the directory is discarded.
935  */
936 static void
937 ino_free_children(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, int frags)
938 {
939 	struct suj_ino *sino;
940 	struct bufarea *bp;
941 	struct direct *dp;
942 	off_t diroff;
943 	int skipparent;
944 	int isdotdot;
945 	int dpoff;
946 	int size;
947 
948 	sino = ino_lookup(ino, 0);
949 	if (sino)
950 		skipparent = sino->si_skipparent;
951 	else
952 		skipparent = 0;
953 	size = lfragtosize(fs, frags);
954 	bp = getdatablk(blk, size, BT_DIRDATA);
955 	if (bp->b_errs != 0)
956 		err_suj("ino_free_children: UNRECOVERABLE I/O ERROR");
957 	dp = (struct direct *)&bp->b_un.b_buf[0];
958 	for (dpoff = 0; dpoff < size && dp->d_reclen; dpoff += dp->d_reclen) {
959 		dp = (struct direct *)&bp->b_un.b_buf[dpoff];
960 		if (dp->d_ino == 0 || dp->d_ino == UFS_WINO)
961 			continue;
962 		if (dp->d_namlen == 1 && dp->d_name[0] == '.')
963 			continue;
964 		isdotdot = dp->d_namlen == 2 && dp->d_name[0] == '.' &&
965 		    dp->d_name[1] == '.';
966 		if (isdotdot && skipparent == 1)
967 			continue;
968 		if (debug)
969 			printf("Directory %ju removing ino %ju name %s\n",
970 			    (uintmax_t)ino, (uintmax_t)dp->d_ino, dp->d_name);
971 		diroff = lblktosize(fs, lbn) + dpoff;
972 		ino_remref(ino, dp->d_ino, diroff, isdotdot);
973 	}
974 	brelse(bp);
975 }
976 
977 /*
978  * Reclaim an inode, freeing all blocks and decrementing all children's
979  * link counts.  Free the inode back to the cg.
980  */
981 static void
982 ino_reclaim(struct inode *ip, ino_t ino, int mode)
983 {
984 	union dinode *dp;
985 	uint32_t gen;
986 
987 	dp = ip->i_dp;
988 	if (ino == UFS_ROOTINO)
989 		err_suj("Attempting to free UFS_ROOTINO\n");
990 	if (debug)
991 		printf("Truncating and freeing ino %ju, nlink %d, mode %o\n",
992 		    (uintmax_t)ino, DIP(dp, di_nlink), DIP(dp, di_mode));
993 
994 	/* We are freeing an inode or directory. */
995 	if ((DIP(dp, di_mode) & IFMT) == IFDIR)
996 		ino_visit(dp, ino, ino_free_children, 0);
997 	DIP_SET(dp, di_nlink, 0);
998 	ino_visit(dp, ino, blk_free_visit, VISIT_EXT | VISIT_INDIR);
999 	/* Here we have to clear the inode and release any blocks it holds. */
1000 	gen = DIP(dp, di_gen);
1001 	if (fs->fs_magic == FS_UFS1_MAGIC)
1002 		bzero(dp, sizeof(struct ufs1_dinode));
1003 	else
1004 		bzero(dp, sizeof(struct ufs2_dinode));
1005 	DIP_SET(dp, di_gen, gen);
1006 	inodirty(ip);
1007 	ino_free(ino, mode);
1008 	return;
1009 }
1010 
1011 /*
1012  * Adjust an inode's link count down by one when a directory goes away.
1013  */
1014 static void
1015 ino_decr(ino_t ino)
1016 {
1017 	struct inode ip;
1018 	union dinode *dp;
1019 	int reqlink;
1020 	int nlink;
1021 	int mode;
1022 
1023 	ginode(ino, &ip);
1024 	dp = ip.i_dp;
1025 	nlink = DIP(dp, di_nlink);
1026 	mode = DIP(dp, di_mode);
1027 	if (nlink < 1)
1028 		err_suj("Inode %d link count %d invalid\n", ino, nlink);
1029 	if (mode == 0)
1030 		err_suj("Inode %d has a link of %d with 0 mode\n", ino, nlink);
1031 	nlink--;
1032 	if ((mode & IFMT) == IFDIR)
1033 		reqlink = 2;
1034 	else
1035 		reqlink = 1;
1036 	if (nlink < reqlink) {
1037 		if (debug)
1038 			printf("ino %ju not enough links to live %d < %d\n",
1039 			    (uintmax_t)ino, nlink, reqlink);
1040 		ino_reclaim(&ip, ino, mode);
1041 		irelse(&ip);
1042 		return;
1043 	}
1044 	DIP_SET(dp, di_nlink, nlink);
1045 	inodirty(&ip);
1046 	irelse(&ip);
1047 }
1048 
1049 /*
1050  * Adjust the inode link count to 'nlink'.  If the count reaches zero
1051  * free it.
1052  */
1053 static void
1054 ino_adjust(struct suj_ino *sino)
1055 {
1056 	struct jrefrec *rrec;
1057 	struct suj_rec *srec;
1058 	struct suj_ino *stmp;
1059 	union dinode *dp;
1060 	struct inode ip;
1061 	nlink_t nlink;
1062 	nlink_t reqlink;
1063 	int recmode;
1064 	int isdot;
1065 	int mode;
1066 	ino_t ino;
1067 
1068 	nlink = sino->si_nlink;
1069 	ino = sino->si_ino;
1070 	mode = sino->si_mode & IFMT;
1071 	/*
1072 	 * If it's a directory with no dot links, it was truncated before
1073 	 * the name was cleared.  We need to clear the dirent that
1074 	 * points at it.
1075 	 */
1076 	if (mode == IFDIR && nlink == 1 && sino->si_dotlinks == 0) {
1077 		sino->si_nlink = nlink = 0;
1078 		TAILQ_FOREACH(srec, &sino->si_recs, sr_next) {
1079 			rrec = (struct jrefrec *)srec->sr_rec;
1080 			if (ino_isat(rrec->jr_parent, rrec->jr_diroff, ino,
1081 			    &recmode, &isdot) == 0)
1082 				continue;
1083 			ino_clrat(rrec->jr_parent, rrec->jr_diroff, ino);
1084 			break;
1085 		}
1086 		if (srec == NULL)
1087 			errx(1, "Directory %ju name not found", (uintmax_t)ino);
1088 	}
1089 	/*
1090 	 * If it's a directory with no real names pointing to it go ahead
1091 	 * and truncate it.  This will free any children.
1092 	 */
1093 	if (mode == IFDIR && nlink - sino->si_dotlinks == 0) {
1094 		sino->si_nlink = nlink = 0;
1095 		/*
1096 		 * Mark any .. links so they know not to free this inode
1097 		 * when they are removed.
1098 		 */
1099 		TAILQ_FOREACH(srec, &sino->si_recs, sr_next) {
1100 			rrec = (struct jrefrec *)srec->sr_rec;
1101 			if (rrec->jr_diroff == DOTDOT_OFFSET) {
1102 				stmp = ino_lookup(rrec->jr_parent, 0);
1103 				if (stmp)
1104 					ino_setskip(stmp, ino);
1105 			}
1106 		}
1107 	}
1108 	ginode(ino, &ip);
1109 	dp = ip.i_dp;
1110 	mode = DIP(dp, di_mode) & IFMT;
1111 	if (nlink > UFS_LINK_MAX)
1112 		err_suj("ino %ju nlink manipulation error, new %ju, old %d\n",
1113 		    (uintmax_t)ino, (uintmax_t)nlink, DIP(dp, di_nlink));
1114 	if (debug)
1115 	       printf("Adjusting ino %ju, nlink %ju, old link %d lastmode %o\n",
1116 		    (uintmax_t)ino, (uintmax_t)nlink, DIP(dp, di_nlink),
1117 		    sino->si_mode);
1118 	if (mode == 0) {
1119 		if (debug)
1120 			printf("ino %ju, zero inode freeing bitmap\n",
1121 			    (uintmax_t)ino);
1122 		ino_free(ino, sino->si_mode);
1123 		irelse(&ip);
1124 		return;
1125 	}
1126 	/* XXX Should be an assert? */
1127 	if (mode != sino->si_mode && debug)
1128 		printf("ino %ju, mode %o != %o\n",
1129 		    (uintmax_t)ino, mode, sino->si_mode);
1130 	if ((mode & IFMT) == IFDIR)
1131 		reqlink = 2;
1132 	else
1133 		reqlink = 1;
1134 	/* If the inode doesn't have enough links to live, free it. */
1135 	if (nlink < reqlink) {
1136 		if (debug)
1137 			printf("ino %ju not enough links to live %ju < %ju\n",
1138 			    (uintmax_t)ino, (uintmax_t)nlink,
1139 			    (uintmax_t)reqlink);
1140 		ino_reclaim(&ip, ino, mode);
1141 		irelse(&ip);
1142 		return;
1143 	}
1144 	/* If required write the updated link count. */
1145 	if (DIP(dp, di_nlink) == nlink) {
1146 		if (debug)
1147 			printf("ino %ju, link matches, skipping.\n",
1148 			    (uintmax_t)ino);
1149 		irelse(&ip);
1150 		return;
1151 	}
1152 	DIP_SET(dp, di_nlink, nlink);
1153 	inodirty(&ip);
1154 	irelse(&ip);
1155 }
1156 
1157 /*
1158  * Truncate some or all blocks in an indirect, freeing any that are required
1159  * and zeroing the indirect.
1160  */
1161 static void
1162 indir_trunc(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, ufs_lbn_t lastlbn,
1163 	union dinode *dp)
1164 {
1165 	struct bufarea *bp;
1166 	ufs_lbn_t lbnadd;
1167 	ufs2_daddr_t nblk;
1168 	ufs_lbn_t next;
1169 	ufs_lbn_t nlbn;
1170 	int isdirty;
1171 	int level;
1172 	int i;
1173 
1174 	if (blk == 0)
1175 		return;
1176 	isdirty = 0;
1177 	level = lbn_level(lbn);
1178 	if (level == -1)
1179 		err_suj("Invalid level for lbn %jd\n", lbn);
1180 	lbnadd = 1;
1181 	for (i = level; i > 0; i--)
1182 		lbnadd *= NINDIR(fs);
1183 	bp = getdatablk(blk, fs->fs_bsize, BT_LEVEL1 + level);
1184 	if (bp->b_errs != 0)
1185 		err_suj("indir_trunc: UNRECOVERABLE I/O ERROR");
1186 	for (i = 0; i < NINDIR(fs); i++) {
1187 		if ((nblk = IBLK(bp, i)) == 0)
1188 			continue;
1189 		if (level != 0) {
1190 			nlbn = (lbn + 1) - (i * lbnadd);
1191 			/*
1192 			 * Calculate the lbn of the next indirect to
1193 			 * determine if any of this indirect must be
1194 			 * reclaimed.
1195 			 */
1196 			next = -(lbn + level) + ((i+1) * lbnadd);
1197 			if (next <= lastlbn)
1198 				continue;
1199 			indir_trunc(ino, nlbn, nblk, lastlbn, dp);
1200 			/* If all of this indirect was reclaimed, free it. */
1201 			nlbn = next - lbnadd;
1202 			if (nlbn < lastlbn)
1203 				continue;
1204 		} else {
1205 			nlbn = -lbn + i * lbnadd;
1206 			if (nlbn < lastlbn)
1207 				continue;
1208 		}
1209 		isdirty = 1;
1210 		blk_free(nblk, 0, fs->fs_frag);
1211 		IBLK_SET(bp, i, 0);
1212 	}
1213 	if (isdirty)
1214 		dirty(bp);
1215 	brelse(bp);
1216 }
1217 
1218 /*
1219  * Truncate an inode to the minimum of the given size or the last populated
1220  * block after any over size have been discarded.  The kernel would allocate
1221  * the last block in the file but fsck does not and neither do we.  This
1222  * code never extends files, only shrinks them.
1223  */
1224 static void
1225 ino_trunc(ino_t ino, off_t size)
1226 {
1227 	struct inode ip;
1228 	union dinode *dp;
1229 	struct bufarea *bp;
1230 	ufs2_daddr_t bn;
1231 	uint64_t totalfrags;
1232 	ufs_lbn_t nextlbn;
1233 	ufs_lbn_t lastlbn;
1234 	ufs_lbn_t tmpval;
1235 	ufs_lbn_t lbn;
1236 	ufs_lbn_t i;
1237 	int blksize, frags;
1238 	off_t cursize;
1239 	off_t off;
1240 	int mode;
1241 
1242 	ginode(ino, &ip);
1243 	dp = ip.i_dp;
1244 	mode = DIP(dp, di_mode) & IFMT;
1245 	cursize = DIP(dp, di_size);
1246 	if (debug)
1247 		printf("Truncating ino %ju, mode %o to size %jd from size %jd\n",
1248 		    (uintmax_t)ino, mode, size, cursize);
1249 
1250 	/* Skip datablocks for short links and devices. */
1251 	if (mode == 0 || mode == IFBLK || mode == IFCHR ||
1252 	    (mode == IFLNK && cursize < fs->fs_maxsymlinklen)) {
1253 		irelse(&ip);
1254 		return;
1255 	}
1256 	/* Don't extend. */
1257 	if (size > cursize) {
1258 		irelse(&ip);
1259 		return;
1260 	}
1261 	if ((DIP(dp, di_flags) & SF_SNAPSHOT) != 0) {
1262 		if (size > 0)
1263 			err_suj("Partial truncation of ino %ju snapshot file\n",
1264 			    (uintmax_t)ino);
1265 	}
1266 	lastlbn = lblkno(fs, blkroundup(fs, size));
1267 	for (i = lastlbn; i < UFS_NDADDR; i++) {
1268 		if ((bn = DIP(dp, di_db[i])) == 0)
1269 			continue;
1270 		blksize = sblksize(fs, cursize, i);
1271 		blk_free(bn, 0, numfrags(fs, blksize));
1272 		DIP_SET(dp, di_db[i], 0);
1273 	}
1274 	/*
1275 	 * Follow indirect blocks, freeing anything required.
1276 	 */
1277 	for (i = 0, tmpval = NINDIR(fs), lbn = UFS_NDADDR; i < UFS_NIADDR; i++,
1278 	    lbn = nextlbn) {
1279 		nextlbn = lbn + tmpval;
1280 		tmpval *= NINDIR(fs);
1281 		/* If we're not freeing any in this indirect range skip it. */
1282 		if (lastlbn >= nextlbn)
1283 			continue;
1284 		if (DIP(dp, di_ib[i]) == 0)
1285 			continue;
1286 		indir_trunc(ino, -lbn - i, DIP(dp, di_ib[i]), lastlbn, dp);
1287 		/* If we freed everything in this indirect free the indir. */
1288 		if (lastlbn > lbn)
1289 			continue;
1290 		blk_free(DIP(dp, di_ib[i]), 0, fs->fs_frag);
1291 		DIP_SET(dp, di_ib[i], 0);
1292 	}
1293 	/*
1294 	 * Now that we've freed any whole blocks that exceed the desired
1295 	 * truncation size, figure out how many blocks remain and what the
1296 	 * last populated lbn is.  We will set the size to this last lbn
1297 	 * rather than worrying about allocating the final lbn as the kernel
1298 	 * would've done.  This is consistent with normal fsck behavior.
1299 	 */
1300 	visitlbn = 0;
1301 	totalfrags = ino_visit(dp, ino, null_visit, VISIT_INDIR | VISIT_EXT);
1302 	if (size > lblktosize(fs, visitlbn + 1))
1303 		size = lblktosize(fs, visitlbn + 1);
1304 	/*
1305 	 * If we're truncating direct blocks we have to adjust frags
1306 	 * accordingly.
1307 	 */
1308 	if (visitlbn < UFS_NDADDR && totalfrags) {
1309 		long oldspace, newspace;
1310 
1311 		bn = DIP(dp, di_db[visitlbn]);
1312 		if (bn == 0)
1313 			err_suj("Bad blk at ino %ju lbn %jd\n",
1314 			    (uintmax_t)ino, visitlbn);
1315 		oldspace = sblksize(fs, cursize, visitlbn);
1316 		newspace = sblksize(fs, size, visitlbn);
1317 		if (oldspace != newspace) {
1318 			bn += numfrags(fs, newspace);
1319 			frags = numfrags(fs, oldspace - newspace);
1320 			blk_free(bn, 0, frags);
1321 			totalfrags -= frags;
1322 		}
1323 	}
1324 	DIP_SET(dp, di_blocks, fsbtodb(fs, totalfrags));
1325 	DIP_SET(dp, di_size, size);
1326 	inodirty(&ip);
1327 	/*
1328 	 * If we've truncated into the middle of a block or frag we have
1329 	 * to zero it here.  Otherwise the file could extend into
1330 	 * uninitialized space later.
1331 	 */
1332 	off = blkoff(fs, size);
1333 	if (off && DIP(dp, di_mode) != IFDIR) {
1334 		long clrsize;
1335 
1336 		bn = ino_blkatoff(dp, ino, visitlbn, &frags, NULL);
1337 		if (bn == 0)
1338 			err_suj("Block missing from ino %ju at lbn %jd\n",
1339 			    (uintmax_t)ino, visitlbn);
1340 		clrsize = frags * fs->fs_fsize;
1341 		bp = getdatablk(bn, clrsize, BT_DATA);
1342 		if (bp->b_errs != 0)
1343 			err_suj("ino_trunc: UNRECOVERABLE I/O ERROR");
1344 		clrsize -= off;
1345 		bzero(&bp->b_un.b_buf[off], clrsize);
1346 		dirty(bp);
1347 		brelse(bp);
1348 	}
1349 	irelse(&ip);
1350 	return;
1351 }
1352 
1353 /*
1354  * Process records available for one inode and determine whether the
1355  * link count is correct or needs adjusting.
1356  */
1357 static void
1358 ino_check(struct suj_ino *sino)
1359 {
1360 	struct suj_rec *srec;
1361 	struct jrefrec *rrec;
1362 	nlink_t dotlinks;
1363 	nlink_t newlinks;
1364 	nlink_t removes;
1365 	nlink_t nlink;
1366 	ino_t ino;
1367 	int isdot;
1368 	int isat;
1369 	int mode;
1370 
1371 	if (sino->si_hasrecs == 0)
1372 		return;
1373 	ino = sino->si_ino;
1374 	rrec = (struct jrefrec *)TAILQ_FIRST(&sino->si_recs)->sr_rec;
1375 	nlink = rrec->jr_nlink;
1376 	newlinks = 0;
1377 	dotlinks = 0;
1378 	removes = sino->si_nlinkadj;
1379 	TAILQ_FOREACH(srec, &sino->si_recs, sr_next) {
1380 		rrec = (struct jrefrec *)srec->sr_rec;
1381 		isat = ino_isat(rrec->jr_parent, rrec->jr_diroff,
1382 		    rrec->jr_ino, &mode, &isdot);
1383 		if (isat && (mode & IFMT) != (rrec->jr_mode & IFMT))
1384 			err_suj("Inode mode/directory type mismatch %o != %o\n",
1385 			    mode, rrec->jr_mode);
1386 		if (debug)
1387 			printf("jrefrec: op %d ino %ju, nlink %ju, parent %ju, "
1388 			    "diroff %jd, mode %o, isat %d, isdot %d\n",
1389 			    rrec->jr_op, (uintmax_t)rrec->jr_ino,
1390 			    (uintmax_t)rrec->jr_nlink,
1391 			    (uintmax_t)rrec->jr_parent,
1392 			    (uintmax_t)rrec->jr_diroff,
1393 			    rrec->jr_mode, isat, isdot);
1394 		mode = rrec->jr_mode & IFMT;
1395 		if (rrec->jr_op == JOP_REMREF)
1396 			removes++;
1397 		newlinks += isat;
1398 		if (isdot)
1399 			dotlinks += isat;
1400 	}
1401 	/*
1402 	 * The number of links that remain are the starting link count
1403 	 * subtracted by the total number of removes with the total
1404 	 * links discovered back in.  An incomplete remove thus
1405 	 * makes no change to the link count but an add increases
1406 	 * by one.
1407 	 */
1408 	if (debug)
1409 		printf(
1410 		    "ino %ju nlink %ju newlinks %ju removes %ju dotlinks %ju\n",
1411 		    (uintmax_t)ino, (uintmax_t)nlink, (uintmax_t)newlinks,
1412 		    (uintmax_t)removes, (uintmax_t)dotlinks);
1413 	nlink += newlinks;
1414 	nlink -= removes;
1415 	sino->si_linkadj = 1;
1416 	sino->si_nlink = nlink;
1417 	sino->si_dotlinks = dotlinks;
1418 	sino->si_mode = mode;
1419 	ino_adjust(sino);
1420 }
1421 
1422 /*
1423  * Process records available for one block and determine whether it is
1424  * still allocated and whether the owning inode needs to be updated or
1425  * a free completed.
1426  */
1427 static void
1428 blk_check(struct suj_blk *sblk)
1429 {
1430 	struct suj_rec *srec;
1431 	struct jblkrec *brec;
1432 	struct suj_ino *sino;
1433 	ufs2_daddr_t blk;
1434 	int mask;
1435 	int frags;
1436 	int isat;
1437 
1438 	/*
1439 	 * Each suj_blk actually contains records for any fragments in that
1440 	 * block.  As a result we must evaluate each record individually.
1441 	 */
1442 	sino = NULL;
1443 	TAILQ_FOREACH(srec, &sblk->sb_recs, sr_next) {
1444 		brec = (struct jblkrec *)srec->sr_rec;
1445 		frags = brec->jb_frags;
1446 		blk = brec->jb_blkno + brec->jb_oldfrags;
1447 		isat = blk_isat(brec->jb_ino, brec->jb_lbn, blk, &frags);
1448 		if (sino == NULL || sino->si_ino != brec->jb_ino) {
1449 			sino = ino_lookup(brec->jb_ino, 1);
1450 			sino->si_blkadj = 1;
1451 		}
1452 		if (debug)
1453 			printf("op %d blk %jd ino %ju lbn %jd frags %d isat %d (%d)\n",
1454 			    brec->jb_op, blk, (uintmax_t)brec->jb_ino,
1455 			    brec->jb_lbn, brec->jb_frags, isat, frags);
1456 		/*
1457 		 * If we found the block at this address we still have to
1458 		 * determine if we need to free the tail end that was
1459 		 * added by adding contiguous fragments from the same block.
1460 		 */
1461 		if (isat == 1) {
1462 			if (frags == brec->jb_frags)
1463 				continue;
1464 			mask = blk_freemask(blk, brec->jb_ino, brec->jb_lbn,
1465 			    brec->jb_frags);
1466 			mask >>= frags;
1467 			blk += frags;
1468 			frags = brec->jb_frags - frags;
1469 			blk_free(blk, mask, frags);
1470 			continue;
1471 		}
1472 		/*
1473 	 	 * The block wasn't found, attempt to free it.  It won't be
1474 		 * freed if it was actually reallocated.  If this was an
1475 		 * allocation we don't want to follow indirects as they
1476 		 * may not be written yet.  Any children of the indirect will
1477 		 * have their own records.  If it's a free we need to
1478 		 * recursively free children.
1479 		 */
1480 		blk_free_lbn(blk, brec->jb_ino, brec->jb_lbn, brec->jb_frags,
1481 		    brec->jb_op == JOP_FREEBLK);
1482 	}
1483 }
1484 
1485 /*
1486  * Walk the list of inode records for this cg and resolve moved and duplicate
1487  * inode references now that we have a complete picture.
1488  */
1489 static void
1490 cg_build(struct suj_cg *sc)
1491 {
1492 	struct suj_ino *sino;
1493 	int i;
1494 
1495 	for (i = 0; i < HASHSIZE; i++)
1496 		LIST_FOREACH(sino, &sc->sc_inohash[i], si_next)
1497 			ino_build(sino);
1498 }
1499 
1500 /*
1501  * Handle inodes requiring truncation.  This must be done prior to
1502  * looking up any inodes in directories.
1503  */
1504 static void
1505 cg_trunc(struct suj_cg *sc)
1506 {
1507 	struct suj_ino *sino;
1508 	int i;
1509 
1510 	for (i = 0; i < HASHSIZE; i++) {
1511 		LIST_FOREACH(sino, &sc->sc_inohash[i], si_next) {
1512 			if (sino->si_trunc) {
1513 				ino_trunc(sino->si_ino,
1514 				    sino->si_trunc->jt_size);
1515 				sino->si_blkadj = 0;
1516 				sino->si_trunc = NULL;
1517 			}
1518 			if (sino->si_blkadj)
1519 				ino_adjblks(sino);
1520 		}
1521 	}
1522 }
1523 
1524 static void
1525 cg_adj_blk(struct suj_cg *sc)
1526 {
1527 	struct suj_ino *sino;
1528 	int i;
1529 
1530 	for (i = 0; i < HASHSIZE; i++) {
1531 		LIST_FOREACH(sino, &sc->sc_inohash[i], si_next) {
1532 			if (sino->si_blkadj)
1533 				ino_adjblks(sino);
1534 		}
1535 	}
1536 }
1537 
1538 /*
1539  * Free any partially allocated blocks and then resolve inode block
1540  * counts.
1541  */
1542 static void
1543 cg_check_blk(struct suj_cg *sc)
1544 {
1545 	struct suj_blk *sblk;
1546 	int i;
1547 
1548 
1549 	for (i = 0; i < HASHSIZE; i++)
1550 		LIST_FOREACH(sblk, &sc->sc_blkhash[i], sb_next)
1551 			blk_check(sblk);
1552 }
1553 
1554 /*
1555  * Walk the list of inode records for this cg, recovering any
1556  * changes which were not complete at the time of crash.
1557  */
1558 static void
1559 cg_check_ino(struct suj_cg *sc)
1560 {
1561 	struct suj_ino *sino;
1562 	int i;
1563 
1564 	for (i = 0; i < HASHSIZE; i++)
1565 		LIST_FOREACH(sino, &sc->sc_inohash[i], si_next)
1566 			ino_check(sino);
1567 }
1568 
1569 static void
1570 cg_apply(void (*apply)(struct suj_cg *))
1571 {
1572 	struct suj_cg *scg;
1573 	int i;
1574 
1575 	for (i = 0; i < HASHSIZE; i++)
1576 		LIST_FOREACH(scg, &cghash[i], sc_next)
1577 			apply(scg);
1578 }
1579 
1580 /*
1581  * Process the unlinked but referenced file list.  Freeing all inodes.
1582  */
1583 static void
1584 ino_unlinked(void)
1585 {
1586 	struct inode ip;
1587 	union dinode *dp;
1588 	uint16_t mode;
1589 	ino_t inon;
1590 	ino_t ino;
1591 
1592 	ino = fs->fs_sujfree;
1593 	fs->fs_sujfree = 0;
1594 	while (ino != 0) {
1595 		ginode(ino, &ip);
1596 		dp = ip.i_dp;
1597 		mode = DIP(dp, di_mode) & IFMT;
1598 		inon = DIP(dp, di_freelink);
1599 		DIP_SET(dp, di_freelink, 0);
1600 		inodirty(&ip);
1601 		/*
1602 		 * XXX Should this be an errx?
1603 		 */
1604 		if (DIP(dp, di_nlink) == 0) {
1605 			if (debug)
1606 				printf("Freeing unlinked ino %ju mode %o\n",
1607 				    (uintmax_t)ino, mode);
1608 			ino_reclaim(&ip, ino, mode);
1609 		} else if (debug)
1610 			printf("Skipping ino %ju mode %o with link %d\n",
1611 			    (uintmax_t)ino, mode, DIP(dp, di_nlink));
1612 		ino = inon;
1613 		irelse(&ip);
1614 	}
1615 }
1616 
1617 /*
1618  * Append a new record to the list of records requiring processing.
1619  */
1620 static void
1621 ino_append(union jrec *rec)
1622 {
1623 	struct jrefrec *refrec;
1624 	struct jmvrec *mvrec;
1625 	struct suj_ino *sino;
1626 	struct suj_rec *srec;
1627 
1628 	mvrec = &rec->rec_jmvrec;
1629 	refrec = &rec->rec_jrefrec;
1630 	if (debug && mvrec->jm_op == JOP_MVREF)
1631 		printf("ino move: ino %ju, parent %ju, "
1632 		    "diroff %jd, oldoff %jd\n",
1633 		    (uintmax_t)mvrec->jm_ino, (uintmax_t)mvrec->jm_parent,
1634 		    (uintmax_t)mvrec->jm_newoff, (uintmax_t)mvrec->jm_oldoff);
1635 	else if (debug &&
1636 	    (refrec->jr_op == JOP_ADDREF || refrec->jr_op == JOP_REMREF))
1637 		printf("ino ref: op %d, ino %ju, nlink %ju, "
1638 		    "parent %ju, diroff %jd\n",
1639 		    refrec->jr_op, (uintmax_t)refrec->jr_ino,
1640 		    (uintmax_t)refrec->jr_nlink,
1641 		    (uintmax_t)refrec->jr_parent, (uintmax_t)refrec->jr_diroff);
1642 	sino = ino_lookup(((struct jrefrec *)rec)->jr_ino, 1);
1643 	sino->si_hasrecs = 1;
1644 	srec = errmalloc(sizeof(*srec));
1645 	srec->sr_rec = rec;
1646 	TAILQ_INSERT_TAIL(&sino->si_newrecs, srec, sr_next);
1647 }
1648 
1649 /*
1650  * Add a reference adjustment to the sino list and eliminate dups.  The
1651  * primary loop in ino_build_ref() checks for dups but new ones may be
1652  * created as a result of offset adjustments.
1653  */
1654 static void
1655 ino_add_ref(struct suj_ino *sino, struct suj_rec *srec)
1656 {
1657 	struct jrefrec *refrec;
1658 	struct suj_rec *srn;
1659 	struct jrefrec *rrn;
1660 
1661 	refrec = (struct jrefrec *)srec->sr_rec;
1662 	/*
1663 	 * We walk backwards so that the oldest link count is preserved.  If
1664 	 * an add record conflicts with a remove keep the remove.  Redundant
1665 	 * removes are eliminated in ino_build_ref.  Otherwise we keep the
1666 	 * oldest record at a given location.
1667 	 */
1668 	for (srn = TAILQ_LAST(&sino->si_recs, srechd); srn;
1669 	    srn = TAILQ_PREV(srn, srechd, sr_next)) {
1670 		rrn = (struct jrefrec *)srn->sr_rec;
1671 		if (rrn->jr_parent != refrec->jr_parent ||
1672 		    rrn->jr_diroff != refrec->jr_diroff)
1673 			continue;
1674 		if (rrn->jr_op == JOP_REMREF || refrec->jr_op == JOP_ADDREF) {
1675 			rrn->jr_mode = refrec->jr_mode;
1676 			return;
1677 		}
1678 		/*
1679 		 * Adding a remove.
1680 		 *
1681 		 * Replace the record in place with the old nlink in case
1682 		 * we replace the head of the list.  Abandon srec as a dup.
1683 		 */
1684 		refrec->jr_nlink = rrn->jr_nlink;
1685 		srn->sr_rec = srec->sr_rec;
1686 		return;
1687 	}
1688 	TAILQ_INSERT_TAIL(&sino->si_recs, srec, sr_next);
1689 }
1690 
1691 /*
1692  * Create a duplicate of a reference at a previous location.
1693  */
1694 static void
1695 ino_dup_ref(struct suj_ino *sino, struct jrefrec *refrec, off_t diroff)
1696 {
1697 	struct jrefrec *rrn;
1698 	struct suj_rec *srn;
1699 
1700 	rrn = errmalloc(sizeof(*refrec));
1701 	*rrn = *refrec;
1702 	rrn->jr_op = JOP_ADDREF;
1703 	rrn->jr_diroff = diroff;
1704 	srn = errmalloc(sizeof(*srn));
1705 	srn->sr_rec = (union jrec *)rrn;
1706 	ino_add_ref(sino, srn);
1707 }
1708 
1709 /*
1710  * Add a reference to the list at all known locations.  We follow the offset
1711  * changes for a single instance and create duplicate add refs at each so
1712  * that we can tolerate any version of the directory block.  Eliminate
1713  * removes which collide with adds that are seen in the journal.  They should
1714  * not adjust the link count down.
1715  */
1716 static void
1717 ino_build_ref(struct suj_ino *sino, struct suj_rec *srec)
1718 {
1719 	struct jrefrec *refrec;
1720 	struct jmvrec *mvrec;
1721 	struct suj_rec *srp;
1722 	struct suj_rec *srn;
1723 	struct jrefrec *rrn;
1724 	off_t diroff;
1725 
1726 	refrec = (struct jrefrec *)srec->sr_rec;
1727 	/*
1728 	 * Search for a mvrec that matches this offset.  Whether it's an add
1729 	 * or a remove we can delete the mvref after creating a dup record in
1730 	 * the old location.
1731 	 */
1732 	if (!TAILQ_EMPTY(&sino->si_movs)) {
1733 		diroff = refrec->jr_diroff;
1734 		for (srn = TAILQ_LAST(&sino->si_movs, srechd); srn; srn = srp) {
1735 			srp = TAILQ_PREV(srn, srechd, sr_next);
1736 			mvrec = (struct jmvrec *)srn->sr_rec;
1737 			if (mvrec->jm_parent != refrec->jr_parent ||
1738 			    mvrec->jm_newoff != diroff)
1739 				continue;
1740 			diroff = mvrec->jm_oldoff;
1741 			TAILQ_REMOVE(&sino->si_movs, srn, sr_next);
1742 			free(srn);
1743 			ino_dup_ref(sino, refrec, diroff);
1744 		}
1745 	}
1746 	/*
1747 	 * If a remove wasn't eliminated by an earlier add just append it to
1748 	 * the list.
1749 	 */
1750 	if (refrec->jr_op == JOP_REMREF) {
1751 		ino_add_ref(sino, srec);
1752 		return;
1753 	}
1754 	/*
1755 	 * Walk the list of records waiting to be added to the list.  We
1756 	 * must check for moves that apply to our current offset and remove
1757 	 * them from the list.  Remove any duplicates to eliminate removes
1758 	 * with corresponding adds.
1759 	 */
1760 	TAILQ_FOREACH_SAFE(srn, &sino->si_newrecs, sr_next, srp) {
1761 		switch (srn->sr_rec->rec_jrefrec.jr_op) {
1762 		case JOP_ADDREF:
1763 			/*
1764 			 * This should actually be an error we should
1765 			 * have a remove for every add journaled.
1766 			 */
1767 			rrn = (struct jrefrec *)srn->sr_rec;
1768 			if (rrn->jr_parent != refrec->jr_parent ||
1769 			    rrn->jr_diroff != refrec->jr_diroff)
1770 				break;
1771 			TAILQ_REMOVE(&sino->si_newrecs, srn, sr_next);
1772 			break;
1773 		case JOP_REMREF:
1774 			/*
1775 			 * Once we remove the current iteration of the
1776 			 * record at this address we're done.
1777 			 */
1778 			rrn = (struct jrefrec *)srn->sr_rec;
1779 			if (rrn->jr_parent != refrec->jr_parent ||
1780 			    rrn->jr_diroff != refrec->jr_diroff)
1781 				break;
1782 			TAILQ_REMOVE(&sino->si_newrecs, srn, sr_next);
1783 			ino_add_ref(sino, srec);
1784 			return;
1785 		case JOP_MVREF:
1786 			/*
1787 			 * Update our diroff based on any moves that match
1788 			 * and remove the move.
1789 			 */
1790 			mvrec = (struct jmvrec *)srn->sr_rec;
1791 			if (mvrec->jm_parent != refrec->jr_parent ||
1792 			    mvrec->jm_oldoff != refrec->jr_diroff)
1793 				break;
1794 			ino_dup_ref(sino, refrec, mvrec->jm_oldoff);
1795 			refrec->jr_diroff = mvrec->jm_newoff;
1796 			TAILQ_REMOVE(&sino->si_newrecs, srn, sr_next);
1797 			break;
1798 		default:
1799 			err_suj("ino_build_ref: Unknown op %d\n",
1800 			    srn->sr_rec->rec_jrefrec.jr_op);
1801 		}
1802 	}
1803 	ino_add_ref(sino, srec);
1804 }
1805 
1806 /*
1807  * Walk the list of new records and add them in-order resolving any
1808  * dups and adjusted offsets.
1809  */
1810 static void
1811 ino_build(struct suj_ino *sino)
1812 {
1813 	struct suj_rec *srec;
1814 
1815 	while ((srec = TAILQ_FIRST(&sino->si_newrecs)) != NULL) {
1816 		TAILQ_REMOVE(&sino->si_newrecs, srec, sr_next);
1817 		switch (srec->sr_rec->rec_jrefrec.jr_op) {
1818 		case JOP_ADDREF:
1819 		case JOP_REMREF:
1820 			ino_build_ref(sino, srec);
1821 			break;
1822 		case JOP_MVREF:
1823 			/*
1824 			 * Add this mvrec to the queue of pending mvs.
1825 			 */
1826 			TAILQ_INSERT_TAIL(&sino->si_movs, srec, sr_next);
1827 			break;
1828 		default:
1829 			err_suj("ino_build: Unknown op %d\n",
1830 			    srec->sr_rec->rec_jrefrec.jr_op);
1831 		}
1832 	}
1833 	if (TAILQ_EMPTY(&sino->si_recs))
1834 		sino->si_hasrecs = 0;
1835 }
1836 
1837 /*
1838  * Modify journal records so they refer to the base block number
1839  * and a start and end frag range.  This is to facilitate the discovery
1840  * of overlapping fragment allocations.
1841  */
1842 static void
1843 blk_build(struct jblkrec *blkrec)
1844 {
1845 	struct suj_rec *srec;
1846 	struct suj_blk *sblk;
1847 	struct jblkrec *blkrn;
1848 	ufs2_daddr_t blk;
1849 	int frag;
1850 
1851 	if (debug)
1852 		printf("blk_build: op %d blkno %jd frags %d oldfrags %d "
1853 		    "ino %ju lbn %jd\n",
1854 		    blkrec->jb_op, (uintmax_t)blkrec->jb_blkno,
1855 		    blkrec->jb_frags, blkrec->jb_oldfrags,
1856 		    (uintmax_t)blkrec->jb_ino, (uintmax_t)blkrec->jb_lbn);
1857 
1858 	blk = blknum(fs, blkrec->jb_blkno);
1859 	frag = fragnum(fs, blkrec->jb_blkno);
1860 	sblk = blk_lookup(blk, 1);
1861 	/*
1862 	 * Rewrite the record using oldfrags to indicate the offset into
1863 	 * the block.  Leave jb_frags as the actual allocated count.
1864 	 */
1865 	blkrec->jb_blkno -= frag;
1866 	blkrec->jb_oldfrags = frag;
1867 	if (blkrec->jb_oldfrags + blkrec->jb_frags > fs->fs_frag)
1868 		err_suj("Invalid fragment count %d oldfrags %d\n",
1869 		    blkrec->jb_frags, frag);
1870 	/*
1871 	 * Detect dups.  If we detect a dup we always discard the oldest
1872 	 * record as it is superseded by the new record.  This speeds up
1873 	 * later stages but also eliminates free records which are used
1874 	 * to indicate that the contents of indirects can be trusted.
1875 	 */
1876 	TAILQ_FOREACH(srec, &sblk->sb_recs, sr_next) {
1877 		blkrn = (struct jblkrec *)srec->sr_rec;
1878 		if (blkrn->jb_ino != blkrec->jb_ino ||
1879 		    blkrn->jb_lbn != blkrec->jb_lbn ||
1880 		    blkrn->jb_blkno != blkrec->jb_blkno ||
1881 		    blkrn->jb_frags != blkrec->jb_frags ||
1882 		    blkrn->jb_oldfrags != blkrec->jb_oldfrags)
1883 			continue;
1884 		if (debug)
1885 			printf("Removed dup.\n");
1886 		/* Discard the free which is a dup with an alloc. */
1887 		if (blkrec->jb_op == JOP_FREEBLK)
1888 			return;
1889 		TAILQ_REMOVE(&sblk->sb_recs, srec, sr_next);
1890 		free(srec);
1891 		break;
1892 	}
1893 	srec = errmalloc(sizeof(*srec));
1894 	srec->sr_rec = (union jrec *)blkrec;
1895 	TAILQ_INSERT_TAIL(&sblk->sb_recs, srec, sr_next);
1896 }
1897 
1898 static void
1899 ino_build_trunc(struct jtrncrec *rec)
1900 {
1901 	struct suj_ino *sino;
1902 
1903 	if (debug)
1904 		printf("ino_build_trunc: op %d ino %ju, size %jd\n",
1905 		    rec->jt_op, (uintmax_t)rec->jt_ino,
1906 		    (uintmax_t)rec->jt_size);
1907 	sino = ino_lookup(rec->jt_ino, 1);
1908 	if (rec->jt_op == JOP_SYNC) {
1909 		sino->si_trunc = NULL;
1910 		return;
1911 	}
1912 	if (sino->si_trunc == NULL || sino->si_trunc->jt_size > rec->jt_size)
1913 		sino->si_trunc = rec;
1914 }
1915 
1916 /*
1917  * Build up tables of the operations we need to recover.
1918  */
1919 static void
1920 suj_build(void)
1921 {
1922 	struct suj_seg *seg;
1923 	union jrec *rec;
1924 	int off;
1925 	int i;
1926 
1927 	TAILQ_FOREACH(seg, &allsegs, ss_next) {
1928 		if (debug)
1929 			printf("seg %jd has %d records, oldseq %jd.\n",
1930 			    seg->ss_rec.jsr_seq, seg->ss_rec.jsr_cnt,
1931 			    seg->ss_rec.jsr_oldest);
1932 		off = 0;
1933 		rec = (union jrec *)seg->ss_blk;
1934 		for (i = 0; i < seg->ss_rec.jsr_cnt; off += JREC_SIZE, rec++) {
1935 			/* skip the segrec. */
1936 			if ((off % real_dev_bsize) == 0)
1937 				continue;
1938 			switch (rec->rec_jrefrec.jr_op) {
1939 			case JOP_ADDREF:
1940 			case JOP_REMREF:
1941 			case JOP_MVREF:
1942 				ino_append(rec);
1943 				break;
1944 			case JOP_NEWBLK:
1945 			case JOP_FREEBLK:
1946 				blk_build((struct jblkrec *)rec);
1947 				break;
1948 			case JOP_TRUNC:
1949 			case JOP_SYNC:
1950 				ino_build_trunc((struct jtrncrec *)rec);
1951 				break;
1952 			default:
1953 				err_suj("Unknown journal operation %d (%d)\n",
1954 				    rec->rec_jrefrec.jr_op, off);
1955 			}
1956 			i++;
1957 		}
1958 	}
1959 }
1960 
1961 /*
1962  * Prune the journal segments to those we care about based on the
1963  * oldest sequence in the newest segment.  Order the segment list
1964  * based on sequence number.
1965  */
1966 static void
1967 suj_prune(void)
1968 {
1969 	struct suj_seg *seg;
1970 	struct suj_seg *segn;
1971 	uint64_t newseq;
1972 	int discard;
1973 
1974 	if (debug)
1975 		printf("Pruning up to %jd\n", oldseq);
1976 	/* First free the expired segments. */
1977 	TAILQ_FOREACH_SAFE(seg, &allsegs, ss_next, segn) {
1978 		if (seg->ss_rec.jsr_seq >= oldseq)
1979 			continue;
1980 		TAILQ_REMOVE(&allsegs, seg, ss_next);
1981 		free(seg->ss_blk);
1982 		free(seg);
1983 	}
1984 	/* Next ensure that segments are ordered properly. */
1985 	seg = TAILQ_FIRST(&allsegs);
1986 	if (seg == NULL) {
1987 		if (debug)
1988 			printf("Empty journal\n");
1989 		return;
1990 	}
1991 	newseq = seg->ss_rec.jsr_seq;
1992 	for (;;) {
1993 		seg = TAILQ_LAST(&allsegs, seghd);
1994 		if (seg->ss_rec.jsr_seq >= newseq)
1995 			break;
1996 		TAILQ_REMOVE(&allsegs, seg, ss_next);
1997 		TAILQ_INSERT_HEAD(&allsegs, seg, ss_next);
1998 		newseq = seg->ss_rec.jsr_seq;
1999 
2000 	}
2001 	if (newseq != oldseq) {
2002 		TAILQ_FOREACH(seg, &allsegs, ss_next) {
2003 			printf("%jd, ", seg->ss_rec.jsr_seq);
2004 		}
2005 		printf("\n");
2006 		err_suj("Journal file sequence mismatch %jd != %jd\n",
2007 		    newseq, oldseq);
2008 	}
2009 	/*
2010 	 * The kernel may asynchronously write segments which can create
2011 	 * gaps in the sequence space.  Throw away any segments after the
2012 	 * gap as the kernel guarantees only those that are contiguously
2013 	 * reachable are marked as completed.
2014 	 */
2015 	discard = 0;
2016 	TAILQ_FOREACH_SAFE(seg, &allsegs, ss_next, segn) {
2017 		if (!discard && newseq++ == seg->ss_rec.jsr_seq) {
2018 			jrecs += seg->ss_rec.jsr_cnt;
2019 			jbytes += seg->ss_rec.jsr_blocks * real_dev_bsize;
2020 			continue;
2021 		}
2022 		discard = 1;
2023 		if (debug)
2024 			printf("Journal order mismatch %jd != %jd pruning\n",
2025 			    newseq-1, seg->ss_rec.jsr_seq);
2026 		TAILQ_REMOVE(&allsegs, seg, ss_next);
2027 		free(seg->ss_blk);
2028 		free(seg);
2029 	}
2030 	if (debug)
2031 		printf("Processing journal segments from %jd to %jd\n",
2032 		    oldseq, newseq-1);
2033 }
2034 
2035 /*
2036  * Verify the journal inode before attempting to read records.
2037  */
2038 static int
2039 suj_verifyino(union dinode *dp)
2040 {
2041 
2042 	if (DIP(dp, di_nlink) != 1) {
2043 		printf("Invalid link count %d for journal inode %ju\n",
2044 		    DIP(dp, di_nlink), (uintmax_t)sujino);
2045 		return (-1);
2046 	}
2047 
2048 	if ((DIP(dp, di_flags) & (SF_IMMUTABLE | SF_NOUNLINK)) !=
2049 	    (SF_IMMUTABLE | SF_NOUNLINK)) {
2050 		printf("Invalid flags 0x%X for journal inode %ju\n",
2051 		    DIP(dp, di_flags), (uintmax_t)sujino);
2052 		return (-1);
2053 	}
2054 
2055 	if (DIP(dp, di_mode) != (IFREG | IREAD)) {
2056 		printf("Invalid mode %o for journal inode %ju\n",
2057 		    DIP(dp, di_mode), (uintmax_t)sujino);
2058 		return (-1);
2059 	}
2060 
2061 	if (DIP(dp, di_size) < SUJ_MIN) {
2062 		printf("Invalid size %jd for journal inode %ju\n",
2063 		    DIP(dp, di_size), (uintmax_t)sujino);
2064 		return (-1);
2065 	}
2066 
2067 	if (DIP(dp, di_modrev) != fs->fs_mtime) {
2068 		printf("Journal timestamp does not match fs mount time\n");
2069 		return (-1);
2070 	}
2071 
2072 	return (0);
2073 }
2074 
2075 struct jblocks {
2076 	struct jextent *jb_extent;	/* Extent array. */
2077 	int		jb_avail;	/* Available extents. */
2078 	int		jb_used;	/* Last used extent. */
2079 	int		jb_head;	/* Allocator head. */
2080 	int		jb_off;		/* Allocator extent offset. */
2081 };
2082 struct jextent {
2083 	ufs2_daddr_t	je_daddr;	/* Disk block address. */
2084 	int		je_blocks;	/* Disk block count. */
2085 };
2086 
2087 static struct jblocks *suj_jblocks;
2088 
2089 static struct jblocks *
2090 jblocks_create(void)
2091 {
2092 	struct jblocks *jblocks;
2093 	int size;
2094 
2095 	jblocks = errmalloc(sizeof(*jblocks));
2096 	jblocks->jb_avail = 10;
2097 	jblocks->jb_used = 0;
2098 	jblocks->jb_head = 0;
2099 	jblocks->jb_off = 0;
2100 	size = sizeof(struct jextent) * jblocks->jb_avail;
2101 	jblocks->jb_extent = errmalloc(size);
2102 	bzero(jblocks->jb_extent, size);
2103 
2104 	return (jblocks);
2105 }
2106 
2107 /*
2108  * Return the next available disk block and the amount of contiguous
2109  * free space it contains.
2110  */
2111 static ufs2_daddr_t
2112 jblocks_next(struct jblocks *jblocks, int bytes, int *actual)
2113 {
2114 	struct jextent *jext;
2115 	ufs2_daddr_t daddr;
2116 	int freecnt;
2117 	int blocks;
2118 
2119 	blocks = btodb(bytes);
2120 	jext = &jblocks->jb_extent[jblocks->jb_head];
2121 	freecnt = jext->je_blocks - jblocks->jb_off;
2122 	if (freecnt == 0) {
2123 		jblocks->jb_off = 0;
2124 		if (++jblocks->jb_head > jblocks->jb_used)
2125 			return (0);
2126 		jext = &jblocks->jb_extent[jblocks->jb_head];
2127 		freecnt = jext->je_blocks;
2128 	}
2129 	if (freecnt > blocks)
2130 		freecnt = blocks;
2131 	*actual = dbtob(freecnt);
2132 	daddr = jext->je_daddr + jblocks->jb_off;
2133 
2134 	return (daddr);
2135 }
2136 
2137 /*
2138  * Advance the allocation head by a specified number of bytes, consuming
2139  * one journal segment.
2140  */
2141 static void
2142 jblocks_advance(struct jblocks *jblocks, int bytes)
2143 {
2144 
2145 	jblocks->jb_off += btodb(bytes);
2146 }
2147 
2148 static void
2149 jblocks_destroy(struct jblocks *jblocks)
2150 {
2151 
2152 	free(jblocks->jb_extent);
2153 	free(jblocks);
2154 }
2155 
2156 static void
2157 jblocks_add(struct jblocks *jblocks, ufs2_daddr_t daddr, int blocks)
2158 {
2159 	struct jextent *jext;
2160 	int size;
2161 
2162 	jext = &jblocks->jb_extent[jblocks->jb_used];
2163 	/* Adding the first block. */
2164 	if (jext->je_daddr == 0) {
2165 		jext->je_daddr = daddr;
2166 		jext->je_blocks = blocks;
2167 		return;
2168 	}
2169 	/* Extending the last extent. */
2170 	if (jext->je_daddr + jext->je_blocks == daddr) {
2171 		jext->je_blocks += blocks;
2172 		return;
2173 	}
2174 	/* Adding a new extent. */
2175 	if (++jblocks->jb_used == jblocks->jb_avail) {
2176 		jblocks->jb_avail *= 2;
2177 		size = sizeof(struct jextent) * jblocks->jb_avail;
2178 		jext = errmalloc(size);
2179 		bzero(jext, size);
2180 		bcopy(jblocks->jb_extent, jext,
2181 		    sizeof(struct jextent) * jblocks->jb_used);
2182 		free(jblocks->jb_extent);
2183 		jblocks->jb_extent = jext;
2184 	}
2185 	jext = &jblocks->jb_extent[jblocks->jb_used];
2186 	jext->je_daddr = daddr;
2187 	jext->je_blocks = blocks;
2188 
2189 	return;
2190 }
2191 
2192 /*
2193  * Add a file block from the journal to the extent map.  We can't read
2194  * each file block individually because the kernel treats it as a circular
2195  * buffer and segments may span mutliple contiguous blocks.
2196  */
2197 static void
2198 suj_add_block(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, int frags)
2199 {
2200 
2201 	jblocks_add(suj_jblocks, fsbtodb(fs, blk), fsbtodb(fs, frags));
2202 }
2203 
2204 static void
2205 suj_read(void)
2206 {
2207 	uint8_t block[1 * 1024 * 1024];
2208 	struct suj_seg *seg;
2209 	struct jsegrec *recn;
2210 	struct jsegrec *rec;
2211 	ufs2_daddr_t blk;
2212 	int readsize;
2213 	int blocks;
2214 	int recsize;
2215 	int size;
2216 	int i;
2217 
2218 	/*
2219 	 * Read records until we exhaust the journal space.  If we find
2220 	 * an invalid record we start searching for a valid segment header
2221 	 * at the next block.  This is because we don't have a head/tail
2222 	 * pointer and must recover the information indirectly.  At the gap
2223 	 * between the head and tail we won't necessarily have a valid
2224 	 * segment.
2225 	 */
2226 restart:
2227 	for (;;) {
2228 		size = sizeof(block);
2229 		blk = jblocks_next(suj_jblocks, size, &readsize);
2230 		if (blk == 0)
2231 			return;
2232 		size = readsize;
2233 		/*
2234 		 * Read 1MB at a time and scan for records within this block.
2235 		 */
2236 		if (pread(fsreadfd, &block, size, dbtob(blk)) != size) {
2237 			err_suj("Error reading journal block %jd\n",
2238 			    (intmax_t)blk);
2239 		}
2240 		for (rec = (void *)block; size; size -= recsize,
2241 		    rec = (struct jsegrec *)((uintptr_t)rec + recsize)) {
2242 			recsize = real_dev_bsize;
2243 			if (rec->jsr_time != fs->fs_mtime) {
2244 #ifdef notdef
2245 				if (debug)
2246 					printf("Rec time %jd != fs mtime %jd\n",
2247 					    rec->jsr_time, fs->fs_mtime);
2248 #endif
2249 				jblocks_advance(suj_jblocks, recsize);
2250 				continue;
2251 			}
2252 			if (rec->jsr_cnt == 0) {
2253 				if (debug)
2254 					printf("Found illegal count %d\n",
2255 					    rec->jsr_cnt);
2256 				jblocks_advance(suj_jblocks, recsize);
2257 				continue;
2258 			}
2259 			blocks = rec->jsr_blocks;
2260 			recsize = blocks * real_dev_bsize;
2261 			if (recsize > size) {
2262 				/*
2263 				 * We may just have run out of buffer, restart
2264 				 * the loop to re-read from this spot.
2265 				 */
2266 				if (size < fs->fs_bsize &&
2267 				    size != readsize &&
2268 				    recsize <= fs->fs_bsize)
2269 					goto restart;
2270 				if (debug)
2271 					printf("Found invalid segsize %d > %d\n",
2272 					    recsize, size);
2273 				recsize = real_dev_bsize;
2274 				jblocks_advance(suj_jblocks, recsize);
2275 				continue;
2276 			}
2277 			/*
2278 			 * Verify that all blocks in the segment are present.
2279 			 */
2280 			for (i = 1; i < blocks; i++) {
2281 				recn = (void *)((uintptr_t)rec) + i *
2282 				    real_dev_bsize;
2283 				if (recn->jsr_seq == rec->jsr_seq &&
2284 				    recn->jsr_time == rec->jsr_time)
2285 					continue;
2286 				if (debug)
2287 					printf("Incomplete record %jd (%d)\n",
2288 					    rec->jsr_seq, i);
2289 				recsize = i * real_dev_bsize;
2290 				jblocks_advance(suj_jblocks, recsize);
2291 				goto restart;
2292 			}
2293 			seg = errmalloc(sizeof(*seg));
2294 			seg->ss_blk = errmalloc(recsize);
2295 			seg->ss_rec = *rec;
2296 			bcopy((void *)rec, seg->ss_blk, recsize);
2297 			if (rec->jsr_oldest > oldseq)
2298 				oldseq = rec->jsr_oldest;
2299 			TAILQ_INSERT_TAIL(&allsegs, seg, ss_next);
2300 			jblocks_advance(suj_jblocks, recsize);
2301 		}
2302 	}
2303 }
2304 
2305 /*
2306  * Orchestrate the verification of a filesystem via the softupdates journal.
2307  */
2308 int
2309 suj_check(const char *filesys)
2310 {
2311 	struct inodesc idesc;
2312 	struct csum *cgsum;
2313 	union dinode *jip;
2314 	struct inode ip;
2315 	uint64_t blocks;
2316 	int i, retval;
2317 	struct suj_seg *seg;
2318 	struct suj_seg *segn;
2319 
2320 	initsuj();
2321 	fs = &sblock;
2322 	if (real_dev_bsize == 0 && ioctl(fsreadfd, DIOCGSECTORSIZE,
2323 	    &real_dev_bsize) == -1)
2324 		real_dev_bsize = secsize;
2325 	if (debug)
2326 		printf("dev_bsize %u\n", real_dev_bsize);
2327 
2328 	/*
2329 	 * Set an exit point when SUJ check failed
2330 	 */
2331 	retval = setjmp(jmpbuf);
2332 	if (retval != 0) {
2333 		pwarn("UNEXPECTED SU+J INCONSISTENCY\n");
2334 		TAILQ_FOREACH_SAFE(seg, &allsegs, ss_next, segn) {
2335 			TAILQ_REMOVE(&allsegs, seg, ss_next);
2336 				free(seg->ss_blk);
2337 				free(seg);
2338 		}
2339 		if (reply("FALLBACK TO FULL FSCK") == 0) {
2340 			ckfini(0);
2341 			exit(EEXIT);
2342 		} else
2343 			return (-1);
2344 	}
2345 
2346 	/*
2347 	 * Search the root directory for the SUJ_FILE.
2348 	 */
2349 	idesc.id_type = DATA;
2350 	idesc.id_fix = IGNORE;
2351 	idesc.id_number = UFS_ROOTINO;
2352 	idesc.id_func = findino;
2353 	idesc.id_name = SUJ_FILE;
2354 	ginode(UFS_ROOTINO, &ip);
2355 	if ((ckinode(ip.i_dp, &idesc) & FOUND) == FOUND) {
2356 		sujino = idesc.id_parent;
2357 		irelse(&ip);
2358 	} else {
2359 		printf("Journal inode removed.  Use tunefs to re-create.\n");
2360 		sblock.fs_flags &= ~FS_SUJ;
2361 		sblock.fs_sujfree = 0;
2362 		irelse(&ip);
2363 		return (-1);
2364 	}
2365 	/*
2366 	 * Fetch the journal inode and verify it.
2367 	 */
2368 	ginode(sujino, &ip);
2369 	jip = ip.i_dp;
2370 	printf("** SU+J Recovering %s\n", filesys);
2371 	if (suj_verifyino(jip) != 0 || (!preen && !reply("USE JOURNAL"))) {
2372 		irelse(&ip);
2373 		return (-1);
2374 	}
2375 	/*
2376 	 * Build a list of journal blocks in jblocks before parsing the
2377 	 * available journal blocks in with suj_read().
2378 	 */
2379 	printf("** Reading %jd byte journal from inode %ju.\n",
2380 	    DIP(jip, di_size), (uintmax_t)sujino);
2381 	suj_jblocks = jblocks_create();
2382 	blocks = ino_visit(jip, sujino, suj_add_block, 0);
2383 	if (blocks != numfrags(fs, DIP(jip, di_size))) {
2384 		printf("Sparse journal inode %ju.\n", (uintmax_t)sujino);
2385 		irelse(&ip);
2386 		return (-1);
2387 	}
2388 	irelse(&ip);
2389 	suj_read();
2390 	jblocks_destroy(suj_jblocks);
2391 	suj_jblocks = NULL;
2392 	if (preen || reply("RECOVER")) {
2393 		printf("** Building recovery table.\n");
2394 		suj_prune();
2395 		suj_build();
2396 		cg_apply(cg_build);
2397 		printf("** Resolving unreferenced inode list.\n");
2398 		ino_unlinked();
2399 		printf("** Processing journal entries.\n");
2400 		cg_apply(cg_trunc);
2401 		cg_apply(cg_check_blk);
2402 		cg_apply(cg_adj_blk);
2403 		cg_apply(cg_check_ino);
2404 	}
2405 	if (preen == 0 && (jrecs > 0 || jbytes > 0) && reply("WRITE CHANGES") == 0)
2406 		return (0);
2407 	/*
2408 	 * Recompute the fs summary info from correct cs summaries.
2409 	 */
2410 	bzero(&fs->fs_cstotal, sizeof(struct csum_total));
2411 	for (i = 0; i < fs->fs_ncg; i++) {
2412 		cgsum = &fs->fs_cs(fs, i);
2413 		fs->fs_cstotal.cs_nffree += cgsum->cs_nffree;
2414 		fs->fs_cstotal.cs_nbfree += cgsum->cs_nbfree;
2415 		fs->fs_cstotal.cs_nifree += cgsum->cs_nifree;
2416 		fs->fs_cstotal.cs_ndir += cgsum->cs_ndir;
2417 	}
2418 	fs->fs_pendinginodes = 0;
2419 	fs->fs_pendingblocks = 0;
2420 	fs->fs_clean = 1;
2421 	fs->fs_time = time(NULL);
2422 	fs->fs_mtime = time(NULL);
2423 	sbdirty();
2424 	ckfini(1);
2425 	if (jrecs > 0 || jbytes > 0) {
2426 		printf("** %jd journal records in %jd bytes for %.2f%% utilization\n",
2427 		    jrecs, jbytes, ((float)jrecs / (float)(jbytes / JREC_SIZE)) * 100);
2428 		printf("** Freed %jd inodes (%jd dirs) %jd blocks, and %jd frags.\n",
2429 		    freeinos, freedir, freeblocks, freefrags);
2430 	}
2431 
2432 	return (0);
2433 }
2434 
2435 static void
2436 initsuj(void)
2437 {
2438 	int i;
2439 
2440 	for (i = 0; i < HASHSIZE; i++)
2441 		LIST_INIT(&cghash[i]);
2442 	lastcg = NULL;
2443 	TAILQ_INIT(&allsegs);
2444 	oldseq = 0;
2445 	fs = NULL;
2446 	sujino = 0;
2447 	freefrags = 0;
2448 	freeblocks = 0;
2449 	freeinos = 0;
2450 	freedir = 0;
2451 	jbytes = 0;
2452 	jrecs = 0;
2453 	suj_jblocks = NULL;
2454 }
2455