xref: /freebsd/sbin/fsck_ffs/suj.c (revision f7c32ed617858bcd22f8d1b03199099d50125721)
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/extattr.h>
39 #include <ufs/ufs/quota.h>
40 #include <ufs/ufs/ufsmount.h>
41 #include <ufs/ufs/dinode.h>
42 #include <ufs/ufs/dir.h>
43 #include <ufs/ffs/fs.h>
44 
45 #include <assert.h>
46 #include <err.h>
47 #include <setjmp.h>
48 #include <stdarg.h>
49 #include <stdio.h>
50 #include <stdlib.h>
51 #include <stdint.h>
52 #include <libufs.h>
53 #include <string.h>
54 #include <strings.h>
55 #include <sysexits.h>
56 #include <time.h>
57 
58 #include "fsck.h"
59 
60 #define	DOTDOT_OFFSET	DIRECTSIZ(1)
61 
62 struct suj_seg {
63 	TAILQ_ENTRY(suj_seg) ss_next;
64 	struct jsegrec	ss_rec;
65 	uint8_t		*ss_blk;
66 };
67 
68 struct suj_rec {
69 	TAILQ_ENTRY(suj_rec) sr_next;
70 	union jrec	*sr_rec;
71 };
72 TAILQ_HEAD(srechd, suj_rec);
73 
74 struct suj_ino {
75 	LIST_ENTRY(suj_ino)	si_next;
76 	struct srechd		si_recs;
77 	struct srechd		si_newrecs;
78 	struct srechd		si_movs;
79 	struct jtrncrec		*si_trunc;
80 	ino_t			si_ino;
81 	char			si_skipparent;
82 	char			si_hasrecs;
83 	char			si_blkadj;
84 	char			si_linkadj;
85 	int			si_mode;
86 	nlink_t			si_nlinkadj;
87 	nlink_t			si_nlink;
88 	nlink_t			si_dotlinks;
89 };
90 LIST_HEAD(inohd, suj_ino);
91 
92 struct suj_blk {
93 	LIST_ENTRY(suj_blk)	sb_next;
94 	struct srechd		sb_recs;
95 	ufs2_daddr_t		sb_blk;
96 };
97 LIST_HEAD(blkhd, suj_blk);
98 
99 struct suj_cg {
100 	LIST_ENTRY(suj_cg)	sc_next;
101 	struct blkhd		sc_blkhash[HASHSIZE];
102 	struct inohd		sc_inohash[HASHSIZE];
103 	struct ino_blk		*sc_lastiblk;
104 	struct suj_ino		*sc_lastino;
105 	struct suj_blk		*sc_lastblk;
106 	struct bufarea		*sc_cgbp;
107 	struct cg		*sc_cgp;
108 	int			sc_cgx;
109 };
110 
111 static LIST_HEAD(cghd, suj_cg) cghash[HASHSIZE];
112 static struct suj_cg *lastcg;
113 
114 static TAILQ_HEAD(seghd, suj_seg) allsegs;
115 static uint64_t oldseq;
116 static struct fs *fs = NULL;
117 static ino_t sujino;
118 
119 /*
120  * Summary statistics.
121  */
122 static uint64_t freefrags;
123 static uint64_t freeblocks;
124 static uint64_t freeinos;
125 static uint64_t freedir;
126 static uint64_t jbytes;
127 static uint64_t jrecs;
128 
129 static jmp_buf	jmpbuf;
130 
131 typedef void (*ino_visitor)(ino_t, ufs_lbn_t, ufs2_daddr_t, int);
132 static void err_suj(const char *, ...) __dead2;
133 static void ino_trunc(ino_t, off_t);
134 static void ino_decr(ino_t);
135 static void ino_adjust(struct suj_ino *);
136 static void ino_build(struct suj_ino *);
137 static int blk_isfree(ufs2_daddr_t);
138 static void initsuj(void);
139 
140 static void *
141 errmalloc(size_t n)
142 {
143 	void *a;
144 
145 	a = Malloc(n);
146 	if (a == NULL)
147 		err(EX_OSERR, "malloc(%zu)", n);
148 	return (a);
149 }
150 
151 /*
152  * When hit a fatal error in journalling check, print out
153  * the error and then offer to fallback to normal fsck.
154  */
155 static void
156 err_suj(const char * restrict fmt, ...)
157 {
158 	va_list ap;
159 
160 	if (preen)
161 		(void)fprintf(stdout, "%s: ", cdevname);
162 
163 	va_start(ap, fmt);
164 	(void)vfprintf(stdout, fmt, ap);
165 	va_end(ap);
166 
167 	longjmp(jmpbuf, -1);
168 }
169 
170 /*
171  * Lookup a cg by number in the hash so we can keep track of which cgs
172  * need stats rebuilt.
173  */
174 static struct suj_cg *
175 cg_lookup(int cgx)
176 {
177 	struct cghd *hd;
178 	struct suj_cg *sc;
179 	struct bufarea *cgbp;
180 
181 	if (cgx < 0 || cgx >= fs->fs_ncg)
182 		err_suj("Bad cg number %d\n", cgx);
183 	if (lastcg && lastcg->sc_cgx == cgx)
184 		return (lastcg);
185 	cgbp = cglookup(cgx);
186 	if (!check_cgmagic(cgx, cgbp, 0))
187 		err_suj("UNABLE TO REBUILD CYLINDER GROUP %d", cgx);
188 	hd = &cghash[HASH(cgx)];
189 	LIST_FOREACH(sc, hd, sc_next)
190 		if (sc->sc_cgx == cgx) {
191 			sc->sc_cgbp = cgbp;
192 			sc->sc_cgp = sc->sc_cgbp->b_un.b_cg;
193 			lastcg = sc;
194 			return (sc);
195 		}
196 	sc = errmalloc(sizeof(*sc));
197 	bzero(sc, sizeof(*sc));
198 	sc->sc_cgbp = cgbp;
199 	sc->sc_cgp = sc->sc_cgbp->b_un.b_cg;
200 	sc->sc_cgx = cgx;
201 	LIST_INSERT_HEAD(hd, sc, sc_next);
202 	return (sc);
203 }
204 
205 /*
206  * Lookup an inode number in the hash and allocate a suj_ino if it does
207  * not exist.
208  */
209 static struct suj_ino *
210 ino_lookup(ino_t ino, int creat)
211 {
212 	struct suj_ino *sino;
213 	struct inohd *hd;
214 	struct suj_cg *sc;
215 
216 	sc = cg_lookup(ino_to_cg(fs, ino));
217 	if (sc->sc_lastino && sc->sc_lastino->si_ino == ino)
218 		return (sc->sc_lastino);
219 	hd = &sc->sc_inohash[HASH(ino)];
220 	LIST_FOREACH(sino, hd, si_next)
221 		if (sino->si_ino == ino)
222 			return (sino);
223 	if (creat == 0)
224 		return (NULL);
225 	sino = errmalloc(sizeof(*sino));
226 	bzero(sino, sizeof(*sino));
227 	sino->si_ino = ino;
228 	TAILQ_INIT(&sino->si_recs);
229 	TAILQ_INIT(&sino->si_newrecs);
230 	TAILQ_INIT(&sino->si_movs);
231 	LIST_INSERT_HEAD(hd, sino, si_next);
232 
233 	return (sino);
234 }
235 
236 /*
237  * Lookup a block number in the hash and allocate a suj_blk if it does
238  * not exist.
239  */
240 static struct suj_blk *
241 blk_lookup(ufs2_daddr_t blk, int creat)
242 {
243 	struct suj_blk *sblk;
244 	struct suj_cg *sc;
245 	struct blkhd *hd;
246 
247 	sc = cg_lookup(dtog(fs, blk));
248 	if (sc->sc_lastblk && sc->sc_lastblk->sb_blk == blk)
249 		return (sc->sc_lastblk);
250 	hd = &sc->sc_blkhash[HASH(fragstoblks(fs, blk))];
251 	LIST_FOREACH(sblk, hd, sb_next)
252 		if (sblk->sb_blk == blk)
253 			return (sblk);
254 	if (creat == 0)
255 		return (NULL);
256 	sblk = errmalloc(sizeof(*sblk));
257 	bzero(sblk, sizeof(*sblk));
258 	sblk->sb_blk = blk;
259 	TAILQ_INIT(&sblk->sb_recs);
260 	LIST_INSERT_HEAD(hd, sblk, sb_next);
261 
262 	return (sblk);
263 }
264 
265 static int
266 blk_overlaps(struct jblkrec *brec, ufs2_daddr_t start, int frags)
267 {
268 	ufs2_daddr_t bstart;
269 	ufs2_daddr_t bend;
270 	ufs2_daddr_t end;
271 
272 	end = start + frags;
273 	bstart = brec->jb_blkno + brec->jb_oldfrags;
274 	bend = bstart + brec->jb_frags;
275 	if (start < bend && end > bstart)
276 		return (1);
277 	return (0);
278 }
279 
280 static int
281 blk_equals(struct jblkrec *brec, ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t start,
282     int frags)
283 {
284 
285 	if (brec->jb_ino != ino || brec->jb_lbn != lbn)
286 		return (0);
287 	if (brec->jb_blkno + brec->jb_oldfrags != start)
288 		return (0);
289 	if (brec->jb_frags < frags)
290 		return (0);
291 	return (1);
292 }
293 
294 static void
295 blk_setmask(struct jblkrec *brec, int *mask)
296 {
297 	int i;
298 
299 	for (i = brec->jb_oldfrags; i < brec->jb_oldfrags + brec->jb_frags; i++)
300 		*mask |= 1 << i;
301 }
302 
303 /*
304  * Determine whether a given block has been reallocated to a new location.
305  * Returns a mask of overlapping bits if any frags have been reused or
306  * zero if the block has not been re-used and the contents can be trusted.
307  *
308  * This is used to ensure that an orphaned pointer due to truncate is safe
309  * to be freed.  The mask value can be used to free partial blocks.
310  */
311 static int
312 blk_freemask(ufs2_daddr_t blk, ino_t ino, ufs_lbn_t lbn, int frags)
313 {
314 	struct suj_blk *sblk;
315 	struct suj_rec *srec;
316 	struct jblkrec *brec;
317 	int mask;
318 	int off;
319 
320 	/*
321 	 * To be certain we're not freeing a reallocated block we lookup
322 	 * this block in the blk hash and see if there is an allocation
323 	 * journal record that overlaps with any fragments in the block
324 	 * we're concerned with.  If any fragments have ben reallocated
325 	 * the block has already been freed and re-used for another purpose.
326 	 */
327 	mask = 0;
328 	sblk = blk_lookup(blknum(fs, blk), 0);
329 	if (sblk == NULL)
330 		return (0);
331 	off = blk - sblk->sb_blk;
332 	TAILQ_FOREACH(srec, &sblk->sb_recs, sr_next) {
333 		brec = (struct jblkrec *)srec->sr_rec;
334 		/*
335 		 * If the block overlaps but does not match
336 		 * exactly this record refers to the current
337 		 * location.
338 		 */
339 		if (blk_overlaps(brec, blk, frags) == 0)
340 			continue;
341 		if (blk_equals(brec, ino, lbn, blk, frags) == 1)
342 			mask = 0;
343 		else
344 			blk_setmask(brec, &mask);
345 	}
346 	if (debug)
347 		printf("blk_freemask: blk %jd sblk %jd off %d mask 0x%X\n",
348 		    blk, sblk->sb_blk, off, mask);
349 	return (mask >> off);
350 }
351 
352 /*
353  * Determine whether it is safe to follow an indirect.  It is not safe
354  * if any part of the indirect has been reallocated or the last journal
355  * entry was an allocation.  Just allocated indirects may not have valid
356  * pointers yet and all of their children will have their own records.
357  * It is also not safe to follow an indirect if the cg bitmap has been
358  * cleared as a new allocation may write to the block prior to the journal
359  * being written.
360  *
361  * Returns 1 if it's safe to follow the indirect and 0 otherwise.
362  */
363 static int
364 blk_isindir(ufs2_daddr_t blk, ino_t ino, ufs_lbn_t lbn)
365 {
366 	struct suj_blk *sblk;
367 	struct jblkrec *brec;
368 
369 	sblk = blk_lookup(blk, 0);
370 	if (sblk == NULL)
371 		return (1);
372 	if (TAILQ_EMPTY(&sblk->sb_recs))
373 		return (1);
374 	brec = (struct jblkrec *)TAILQ_LAST(&sblk->sb_recs, srechd)->sr_rec;
375 	if (blk_equals(brec, ino, lbn, blk, fs->fs_frag))
376 		if (brec->jb_op == JOP_FREEBLK)
377 			return (!blk_isfree(blk));
378 	return (0);
379 }
380 
381 /*
382  * Clear an inode from the cg bitmap.  If the inode was already clear return
383  * 0 so the caller knows it does not have to check the inode contents.
384  */
385 static int
386 ino_free(ino_t ino, int mode)
387 {
388 	struct suj_cg *sc;
389 	uint8_t *inosused;
390 	struct cg *cgp;
391 	int cg;
392 
393 	cg = ino_to_cg(fs, ino);
394 	ino = ino % fs->fs_ipg;
395 	sc = cg_lookup(cg);
396 	cgp = sc->sc_cgp;
397 	inosused = cg_inosused(cgp);
398 	/*
399 	 * The bitmap may never have made it to the disk so we have to
400 	 * conditionally clear.  We can avoid writing the cg in this case.
401 	 */
402 	if (isclr(inosused, ino))
403 		return (0);
404 	freeinos++;
405 	clrbit(inosused, ino);
406 	if (ino < cgp->cg_irotor)
407 		cgp->cg_irotor = ino;
408 	cgp->cg_cs.cs_nifree++;
409 	if ((mode & IFMT) == IFDIR) {
410 		freedir++;
411 		cgp->cg_cs.cs_ndir--;
412 	}
413 	cgdirty(sc->sc_cgbp);
414 
415 	return (1);
416 }
417 
418 /*
419  * Free 'frags' frags starting at filesystem block 'bno' skipping any frags
420  * set in the mask.
421  */
422 static void
423 blk_free(ufs2_daddr_t bno, int mask, int frags)
424 {
425 	ufs1_daddr_t fragno, cgbno;
426 	struct suj_cg *sc;
427 	struct cg *cgp;
428 	int i, cg;
429 	uint8_t *blksfree;
430 
431 	if (debug)
432 		printf("Freeing %d frags at blk %jd mask 0x%x\n",
433 		    frags, bno, mask);
434 	cg = dtog(fs, bno);
435 	sc = cg_lookup(cg);
436 	cgp = sc->sc_cgp;
437 	cgbno = dtogd(fs, bno);
438 	blksfree = cg_blksfree(cgp);
439 
440 	/*
441 	 * If it's not allocated we only wrote the journal entry
442 	 * and never the bitmaps.  Here we unconditionally clear and
443 	 * resolve the cg summary later.
444 	 */
445 	if (frags == fs->fs_frag && mask == 0) {
446 		fragno = fragstoblks(fs, cgbno);
447 		ffs_setblock(fs, blksfree, fragno);
448 		freeblocks++;
449 	} else {
450 		/*
451 		 * deallocate the fragment
452 		 */
453 		for (i = 0; i < frags; i++)
454 			if ((mask & (1 << i)) == 0 && isclr(blksfree, cgbno +i)) {
455 				freefrags++;
456 				setbit(blksfree, cgbno + i);
457 			}
458 	}
459 	cgdirty(sc->sc_cgbp);
460 }
461 
462 /*
463  * Returns 1 if the whole block starting at 'bno' is marked free and 0
464  * otherwise.
465  */
466 static int
467 blk_isfree(ufs2_daddr_t bno)
468 {
469 	struct suj_cg *sc;
470 
471 	sc = cg_lookup(dtog(fs, bno));
472 	return ffs_isblock(fs, cg_blksfree(sc->sc_cgp), dtogd(fs, bno));
473 }
474 
475 /*
476  * Determine whether a block exists at a particular lbn in an inode.
477  * Returns 1 if found, 0 if not.  lbn may be negative for indirects
478  * or ext blocks.
479  */
480 static int
481 blk_isat(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, int *frags)
482 {
483 	struct inode ip;
484 	union dinode *dp;
485 	ufs2_daddr_t nblk;
486 
487 	ginode(ino, &ip);
488 	dp = ip.i_dp;
489 	if (DIP(dp, di_nlink) == 0 || DIP(dp, di_mode) == 0) {
490 		irelse(&ip);
491 		return (0);
492 	}
493 	nblk = ino_blkatoff(dp, ino, lbn, frags, NULL);
494 	irelse(&ip);
495 	return (nblk == blk);
496 }
497 
498 /*
499  * Clear the directory entry at diroff that should point to child.  Minimal
500  * checking is done and it is assumed that this path was verified with isat.
501  */
502 static void
503 ino_clrat(ino_t parent, off_t diroff, ino_t child)
504 {
505 	union dinode *dip;
506 	struct direct *dp;
507 	struct inode ip;
508 	ufs2_daddr_t blk;
509 	struct bufarea *bp;
510 	ufs_lbn_t lbn;
511 	int blksize;
512 	int frags;
513 	int doff;
514 
515 	if (debug)
516 		printf("Clearing inode %ju from parent %ju at offset %jd\n",
517 		    (uintmax_t)child, (uintmax_t)parent, diroff);
518 
519 	lbn = lblkno(fs, diroff);
520 	doff = blkoff(fs, diroff);
521 	ginode(parent, &ip);
522 	dip = ip.i_dp;
523 	blk = ino_blkatoff(dip, parent, lbn, &frags, NULL);
524 	blksize = sblksize(fs, DIP(dip, di_size), lbn);
525 	irelse(&ip);
526 	bp = getdatablk(blk, blksize, BT_DIRDATA);
527 	if (bp->b_errs != 0)
528 		err_suj("ino_clrat: UNRECOVERABLE I/O ERROR");
529 	dp = (struct direct *)&bp->b_un.b_buf[doff];
530 	if (dp->d_ino != child)
531 		errx(1, "Inode %ju does not exist in %ju at %jd",
532 		    (uintmax_t)child, (uintmax_t)parent, diroff);
533 	dp->d_ino = 0;
534 	dirty(bp);
535 	brelse(bp);
536 	/*
537 	 * The actual .. reference count will already have been removed
538 	 * from the parent by the .. remref record.
539 	 */
540 }
541 
542 /*
543  * Determines whether a pointer to an inode exists within a directory
544  * at a specified offset.  Returns the mode of the found entry.
545  */
546 static int
547 ino_isat(ino_t parent, off_t diroff, ino_t child, int *mode, int *isdot)
548 {
549 	struct inode ip;
550 	union dinode *dip;
551 	struct bufarea *bp;
552 	struct direct *dp;
553 	ufs2_daddr_t blk;
554 	ufs_lbn_t lbn;
555 	int blksize;
556 	int frags;
557 	int dpoff;
558 	int doff;
559 
560 	*isdot = 0;
561 	ginode(parent, &ip);
562 	dip = ip.i_dp;
563 	*mode = DIP(dip, di_mode);
564 	if ((*mode & IFMT) != IFDIR) {
565 		if (debug) {
566 			/*
567 			 * This can happen if the parent inode
568 			 * was reallocated.
569 			 */
570 			if (*mode != 0)
571 				printf("Directory %ju has bad mode %o\n",
572 				    (uintmax_t)parent, *mode);
573 			else
574 				printf("Directory %ju has zero mode\n",
575 				    (uintmax_t)parent);
576 		}
577 		irelse(&ip);
578 		return (0);
579 	}
580 	lbn = lblkno(fs, diroff);
581 	doff = blkoff(fs, diroff);
582 	blksize = sblksize(fs, DIP(dip, di_size), lbn);
583 	if (diroff + DIRECTSIZ(1) > DIP(dip, di_size) || doff >= blksize) {
584 		if (debug)
585 			printf("ino %ju absent from %ju due to offset %jd"
586 			    " exceeding size %jd\n",
587 			    (uintmax_t)child, (uintmax_t)parent, diroff,
588 			    DIP(dip, di_size));
589 		irelse(&ip);
590 		return (0);
591 	}
592 	blk = ino_blkatoff(dip, parent, lbn, &frags, NULL);
593 	irelse(&ip);
594 	if (blk <= 0) {
595 		if (debug)
596 			printf("Sparse directory %ju", (uintmax_t)parent);
597 		return (0);
598 	}
599 	bp = getdatablk(blk, blksize, BT_DIRDATA);
600 	if (bp->b_errs != 0)
601 		err_suj("ino_isat: UNRECOVERABLE I/O ERROR");
602 	/*
603 	 * Walk through the records from the start of the block to be
604 	 * certain we hit a valid record and not some junk in the middle
605 	 * of a file name.  Stop when we reach or pass the expected offset.
606 	 */
607 	dpoff = rounddown(doff, DIRBLKSIZ);
608 	do {
609 		dp = (struct direct *)&bp->b_un.b_buf[dpoff];
610 		if (dpoff == doff)
611 			break;
612 		if (dp->d_reclen == 0)
613 			break;
614 		dpoff += dp->d_reclen;
615 	} while (dpoff <= doff);
616 	if (dpoff > fs->fs_bsize)
617 		err_suj("Corrupt directory block in dir ino %ju\n",
618 		    (uintmax_t)parent);
619 	/* Not found. */
620 	if (dpoff != doff) {
621 		if (debug)
622 			printf("ino %ju not found in %ju, lbn %jd, dpoff %d\n",
623 			    (uintmax_t)child, (uintmax_t)parent, lbn, dpoff);
624 		brelse(bp);
625 		return (0);
626 	}
627 	/*
628 	 * We found the item in question.  Record the mode and whether it's
629 	 * a . or .. link for the caller.
630 	 */
631 	if (dp->d_ino == child) {
632 		if (child == parent)
633 			*isdot = 1;
634 		else if (dp->d_namlen == 2 &&
635 		    dp->d_name[0] == '.' && dp->d_name[1] == '.')
636 			*isdot = 1;
637 		*mode = DTTOIF(dp->d_type);
638 		brelse(bp);
639 		return (1);
640 	}
641 	if (debug)
642 		printf("ino %ju doesn't match dirent ino %ju in parent %ju\n",
643 		    (uintmax_t)child, (uintmax_t)dp->d_ino, (uintmax_t)parent);
644 	brelse(bp);
645 	return (0);
646 }
647 
648 #define	VISIT_INDIR	0x0001
649 #define	VISIT_EXT	0x0002
650 #define	VISIT_ROOT	0x0004	/* Operation came via root & valid pointers. */
651 
652 /*
653  * Read an indirect level which may or may not be linked into an inode.
654  */
655 static void
656 indir_visit(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, uint64_t *frags,
657     ino_visitor visitor, int flags)
658 {
659 	struct bufarea *bp;
660 	ufs_lbn_t lbnadd;
661 	ufs2_daddr_t nblk;
662 	ufs_lbn_t nlbn;
663 	int level;
664 	int i;
665 
666 	/*
667 	 * Don't visit indirect blocks with contents we can't trust.  This
668 	 * should only happen when indir_visit() is called to complete a
669 	 * truncate that never finished and not when a pointer is found via
670 	 * an inode.
671 	 */
672 	if (blk == 0)
673 		return;
674 	level = lbn_level(lbn);
675 	if (level == -1)
676 		err_suj("Invalid level for lbn %jd\n", lbn);
677 	if ((flags & VISIT_ROOT) == 0 && blk_isindir(blk, ino, lbn) == 0) {
678 		if (debug)
679 			printf("blk %jd ino %ju lbn %jd(%d) is not indir.\n",
680 			    blk, (uintmax_t)ino, lbn, level);
681 		goto out;
682 	}
683 	lbnadd = 1;
684 	for (i = level; i > 0; i--)
685 		lbnadd *= NINDIR(fs);
686 	bp = getdatablk(blk, fs->fs_bsize, BT_LEVEL1 + level);
687 	if (bp->b_errs != 0)
688 		err_suj("indir_visit: UNRECOVERABLE I/O ERROR");
689 	for (i = 0; i < NINDIR(fs); i++) {
690 		if ((nblk = IBLK(bp, i)) == 0)
691 			continue;
692 		if (level == 0) {
693 			nlbn = -lbn + i * lbnadd;
694 			(*frags) += fs->fs_frag;
695 			visitor(ino, nlbn, nblk, fs->fs_frag);
696 		} else {
697 			nlbn = (lbn + 1) - (i * lbnadd);
698 			indir_visit(ino, nlbn, nblk, frags, visitor, flags);
699 		}
700 	}
701 	brelse(bp);
702 out:
703 	if (flags & VISIT_INDIR) {
704 		(*frags) += fs->fs_frag;
705 		visitor(ino, lbn, blk, fs->fs_frag);
706 	}
707 }
708 
709 /*
710  * Visit each block in an inode as specified by 'flags' and call a
711  * callback function.  The callback may inspect or free blocks.  The
712  * count of frags found according to the size in the file is returned.
713  * This is not valid for sparse files but may be used to determine
714  * the correct di_blocks for a file.
715  */
716 static uint64_t
717 ino_visit(union dinode *dp, ino_t ino, ino_visitor visitor, int flags)
718 {
719 	ufs_lbn_t nextlbn;
720 	ufs_lbn_t tmpval;
721 	ufs_lbn_t lbn;
722 	uint64_t size;
723 	uint64_t fragcnt;
724 	int mode;
725 	int frags;
726 	int i;
727 
728 	size = DIP(dp, di_size);
729 	mode = DIP(dp, di_mode) & IFMT;
730 	fragcnt = 0;
731 	if ((flags & VISIT_EXT) &&
732 	    fs->fs_magic == FS_UFS2_MAGIC && dp->dp2.di_extsize) {
733 		for (i = 0; i < UFS_NXADDR; i++) {
734 			if (dp->dp2.di_extb[i] == 0)
735 				continue;
736 			frags = sblksize(fs, dp->dp2.di_extsize, i);
737 			frags = numfrags(fs, frags);
738 			fragcnt += frags;
739 			visitor(ino, -1 - i, dp->dp2.di_extb[i], frags);
740 		}
741 	}
742 	/* Skip datablocks for short links and devices. */
743 	if (mode == IFBLK || mode == IFCHR ||
744 	    (mode == IFLNK && size < fs->fs_maxsymlinklen))
745 		return (fragcnt);
746 	for (i = 0; i < UFS_NDADDR; i++) {
747 		if (DIP(dp, di_db[i]) == 0)
748 			continue;
749 		frags = sblksize(fs, size, i);
750 		frags = numfrags(fs, frags);
751 		fragcnt += frags;
752 		visitor(ino, i, DIP(dp, di_db[i]), frags);
753 	}
754 	/*
755 	 * We know the following indirects are real as we're following
756 	 * real pointers to them.
757 	 */
758 	flags |= VISIT_ROOT;
759 	for (i = 0, tmpval = NINDIR(fs), lbn = UFS_NDADDR; i < UFS_NIADDR; i++,
760 	    lbn = nextlbn) {
761 		nextlbn = lbn + tmpval;
762 		tmpval *= NINDIR(fs);
763 		if (DIP(dp, di_ib[i]) == 0)
764 			continue;
765 		indir_visit(ino, -lbn - i, DIP(dp, di_ib[i]), &fragcnt, visitor,
766 		    flags);
767 	}
768 	return (fragcnt);
769 }
770 
771 /*
772  * Null visitor function used when we just want to count blocks and
773  * record the lbn.
774  */
775 ufs_lbn_t visitlbn;
776 static void
777 null_visit(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, int frags)
778 {
779 	if (lbn > 0)
780 		visitlbn = lbn;
781 }
782 
783 /*
784  * Recalculate di_blocks when we discover that a block allocation or
785  * free was not successfully completed.  The kernel does not roll this back
786  * because it would be too expensive to compute which indirects were
787  * reachable at the time the inode was written.
788  */
789 static void
790 ino_adjblks(struct suj_ino *sino)
791 {
792 	struct inode ip;
793 	union dinode *dp;
794 	uint64_t blocks;
795 	uint64_t frags;
796 	off_t isize;
797 	off_t size;
798 	ino_t ino;
799 
800 	ino = sino->si_ino;
801 	ginode(ino, &ip);
802 	dp = ip.i_dp;
803 	/* No need to adjust zero'd inodes. */
804 	if (DIP(dp, di_mode) == 0) {
805 		irelse(&ip);
806 		return;
807 	}
808 	/*
809 	 * Visit all blocks and count them as well as recording the last
810 	 * valid lbn in the file.  If the file size doesn't agree with the
811 	 * last lbn we need to truncate to fix it.  Otherwise just adjust
812 	 * the blocks count.
813 	 */
814 	visitlbn = 0;
815 	frags = ino_visit(dp, ino, null_visit, VISIT_INDIR | VISIT_EXT);
816 	blocks = fsbtodb(fs, frags);
817 	/*
818 	 * We assume the size and direct block list is kept coherent by
819 	 * softdep.  For files that have extended into indirects we truncate
820 	 * to the size in the inode or the maximum size permitted by
821 	 * populated indirects.
822 	 */
823 	if (visitlbn >= UFS_NDADDR) {
824 		isize = DIP(dp, di_size);
825 		size = lblktosize(fs, visitlbn + 1);
826 		if (isize > size)
827 			isize = size;
828 		/* Always truncate to free any unpopulated indirects. */
829 		ino_trunc(ino, isize);
830 		irelse(&ip);
831 		return;
832 	}
833 	if (blocks == DIP(dp, di_blocks)) {
834 		irelse(&ip);
835 		return;
836 	}
837 	if (debug)
838 		printf("ino %ju adjusting block count from %jd to %jd\n",
839 		    (uintmax_t)ino, DIP(dp, di_blocks), blocks);
840 	DIP_SET(dp, di_blocks, blocks);
841 	inodirty(&ip);
842 	irelse(&ip);
843 }
844 
845 static void
846 blk_free_visit(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, int frags)
847 {
848 
849 	blk_free(blk, blk_freemask(blk, ino, lbn, frags), frags);
850 }
851 
852 /*
853  * Free a block or tree of blocks that was previously rooted in ino at
854  * the given lbn.  If the lbn is an indirect all children are freed
855  * recursively.
856  */
857 static void
858 blk_free_lbn(ufs2_daddr_t blk, ino_t ino, ufs_lbn_t lbn, int frags, int follow)
859 {
860 	uint64_t resid;
861 	int mask;
862 
863 	mask = blk_freemask(blk, ino, lbn, frags);
864 	resid = 0;
865 	if (lbn <= -UFS_NDADDR && follow && mask == 0)
866 		indir_visit(ino, lbn, blk, &resid, blk_free_visit, VISIT_INDIR);
867 	else
868 		blk_free(blk, mask, frags);
869 }
870 
871 static void
872 ino_setskip(struct suj_ino *sino, ino_t parent)
873 {
874 	int isdot;
875 	int mode;
876 
877 	if (ino_isat(sino->si_ino, DOTDOT_OFFSET, parent, &mode, &isdot))
878 		sino->si_skipparent = 1;
879 }
880 
881 static void
882 ino_remref(ino_t parent, ino_t child, uint64_t diroff, int isdotdot)
883 {
884 	struct suj_ino *sino;
885 	struct suj_rec *srec;
886 	struct jrefrec *rrec;
887 
888 	/*
889 	 * Lookup this inode to see if we have a record for it.
890 	 */
891 	sino = ino_lookup(child, 0);
892 	/*
893 	 * Tell any child directories we've already removed their
894 	 * parent link cnt.  Don't try to adjust our link down again.
895 	 */
896 	if (sino != NULL && isdotdot == 0)
897 		ino_setskip(sino, parent);
898 	/*
899 	 * No valid record for this inode.  Just drop the on-disk
900 	 * link by one.
901 	 */
902 	if (sino == NULL || sino->si_hasrecs == 0) {
903 		ino_decr(child);
904 		return;
905 	}
906 	/*
907 	 * Use ino_adjust() if ino_check() has already processed this
908 	 * child.  If we lose the last non-dot reference to a
909 	 * directory it will be discarded.
910 	 */
911 	if (sino->si_linkadj) {
912 		if (sino->si_nlink == 0)
913 			err_suj("ino_remref: ino %ld mode 0%o about to go "
914 			    "negative\n", sino->si_ino, sino->si_mode);
915 		sino->si_nlink--;
916 		if (isdotdot)
917 			sino->si_dotlinks--;
918 		ino_adjust(sino);
919 		return;
920 	}
921 	/*
922 	 * If we haven't yet processed this inode we need to make
923 	 * sure we will successfully discover the lost path.  If not
924 	 * use nlinkadj to remember.
925 	 */
926 	TAILQ_FOREACH(srec, &sino->si_recs, sr_next) {
927 		rrec = (struct jrefrec *)srec->sr_rec;
928 		if (rrec->jr_parent == parent &&
929 		    rrec->jr_diroff == diroff)
930 			return;
931 	}
932 	sino->si_nlinkadj++;
933 }
934 
935 /*
936  * Free the children of a directory when the directory is discarded.
937  */
938 static void
939 ino_free_children(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, int frags)
940 {
941 	struct suj_ino *sino;
942 	struct bufarea *bp;
943 	struct direct *dp;
944 	off_t diroff;
945 	int skipparent;
946 	int isdotdot;
947 	int dpoff;
948 	int size;
949 
950 	sino = ino_lookup(ino, 0);
951 	if (sino)
952 		skipparent = sino->si_skipparent;
953 	else
954 		skipparent = 0;
955 	size = lfragtosize(fs, frags);
956 	bp = getdatablk(blk, size, BT_DIRDATA);
957 	if (bp->b_errs != 0)
958 		err_suj("ino_free_children: UNRECOVERABLE I/O ERROR");
959 	dp = (struct direct *)&bp->b_un.b_buf[0];
960 	for (dpoff = 0; dpoff < size && dp->d_reclen; dpoff += dp->d_reclen) {
961 		dp = (struct direct *)&bp->b_un.b_buf[dpoff];
962 		if (dp->d_ino == 0 || dp->d_ino == UFS_WINO)
963 			continue;
964 		if (dp->d_namlen == 1 && dp->d_name[0] == '.')
965 			continue;
966 		isdotdot = dp->d_namlen == 2 && dp->d_name[0] == '.' &&
967 		    dp->d_name[1] == '.';
968 		if (isdotdot && skipparent == 1)
969 			continue;
970 		if (debug)
971 			printf("Directory %ju removing ino %ju name %s\n",
972 			    (uintmax_t)ino, (uintmax_t)dp->d_ino, dp->d_name);
973 		diroff = lblktosize(fs, lbn) + dpoff;
974 		ino_remref(ino, dp->d_ino, diroff, isdotdot);
975 	}
976 	brelse(bp);
977 }
978 
979 /*
980  * Reclaim an inode, freeing all blocks and decrementing all children's
981  * link counts.  Free the inode back to the cg.
982  */
983 static void
984 ino_reclaim(struct inode *ip, ino_t ino, int mode)
985 {
986 	union dinode *dp;
987 	uint32_t gen;
988 
989 	dp = ip->i_dp;
990 	if (ino == UFS_ROOTINO)
991 		err_suj("Attempting to free UFS_ROOTINO\n");
992 	if (debug)
993 		printf("Truncating and freeing ino %ju, nlink %d, mode %o\n",
994 		    (uintmax_t)ino, DIP(dp, di_nlink), DIP(dp, di_mode));
995 
996 	/* We are freeing an inode or directory. */
997 	if ((DIP(dp, di_mode) & IFMT) == IFDIR)
998 		ino_visit(dp, ino, ino_free_children, 0);
999 	DIP_SET(dp, di_nlink, 0);
1000 	ino_visit(dp, ino, blk_free_visit, VISIT_EXT | VISIT_INDIR);
1001 	/* Here we have to clear the inode and release any blocks it holds. */
1002 	gen = DIP(dp, di_gen);
1003 	if (fs->fs_magic == FS_UFS1_MAGIC)
1004 		bzero(dp, sizeof(struct ufs1_dinode));
1005 	else
1006 		bzero(dp, sizeof(struct ufs2_dinode));
1007 	DIP_SET(dp, di_gen, gen);
1008 	inodirty(ip);
1009 	ino_free(ino, mode);
1010 	return;
1011 }
1012 
1013 /*
1014  * Adjust an inode's link count down by one when a directory goes away.
1015  */
1016 static void
1017 ino_decr(ino_t ino)
1018 {
1019 	struct inode ip;
1020 	union dinode *dp;
1021 	int reqlink;
1022 	int nlink;
1023 	int mode;
1024 
1025 	ginode(ino, &ip);
1026 	dp = ip.i_dp;
1027 	nlink = DIP(dp, di_nlink);
1028 	mode = DIP(dp, di_mode);
1029 	if (nlink < 1)
1030 		err_suj("Inode %d link count %d invalid\n", ino, nlink);
1031 	if (mode == 0)
1032 		err_suj("Inode %d has a link of %d with 0 mode\n", ino, nlink);
1033 	nlink--;
1034 	if ((mode & IFMT) == IFDIR)
1035 		reqlink = 2;
1036 	else
1037 		reqlink = 1;
1038 	if (nlink < reqlink) {
1039 		if (debug)
1040 			printf("ino %ju not enough links to live %d < %d\n",
1041 			    (uintmax_t)ino, nlink, reqlink);
1042 		ino_reclaim(&ip, ino, mode);
1043 		irelse(&ip);
1044 		return;
1045 	}
1046 	DIP_SET(dp, di_nlink, nlink);
1047 	inodirty(&ip);
1048 	irelse(&ip);
1049 }
1050 
1051 /*
1052  * Adjust the inode link count to 'nlink'.  If the count reaches zero
1053  * free it.
1054  */
1055 static void
1056 ino_adjust(struct suj_ino *sino)
1057 {
1058 	struct jrefrec *rrec;
1059 	struct suj_rec *srec;
1060 	struct suj_ino *stmp;
1061 	union dinode *dp;
1062 	struct inode ip;
1063 	nlink_t nlink;
1064 	nlink_t reqlink;
1065 	int recmode;
1066 	int isdot;
1067 	int mode;
1068 	ino_t ino;
1069 
1070 	nlink = sino->si_nlink;
1071 	ino = sino->si_ino;
1072 	mode = sino->si_mode & IFMT;
1073 	/*
1074 	 * If it's a directory with no dot links, it was truncated before
1075 	 * the name was cleared.  We need to clear the dirent that
1076 	 * points at it.
1077 	 */
1078 	if (mode == IFDIR && nlink == 1 && sino->si_dotlinks == 0) {
1079 		sino->si_nlink = nlink = 0;
1080 		TAILQ_FOREACH(srec, &sino->si_recs, sr_next) {
1081 			rrec = (struct jrefrec *)srec->sr_rec;
1082 			if (ino_isat(rrec->jr_parent, rrec->jr_diroff, ino,
1083 			    &recmode, &isdot) == 0)
1084 				continue;
1085 			ino_clrat(rrec->jr_parent, rrec->jr_diroff, ino);
1086 			break;
1087 		}
1088 		if (srec == NULL)
1089 			errx(1, "Directory %ju name not found", (uintmax_t)ino);
1090 	}
1091 	/*
1092 	 * If it's a directory with no real names pointing to it go ahead
1093 	 * and truncate it.  This will free any children.
1094 	 */
1095 	if (mode == IFDIR && nlink - sino->si_dotlinks == 0) {
1096 		sino->si_nlink = nlink = 0;
1097 		/*
1098 		 * Mark any .. links so they know not to free this inode
1099 		 * when they are removed.
1100 		 */
1101 		TAILQ_FOREACH(srec, &sino->si_recs, sr_next) {
1102 			rrec = (struct jrefrec *)srec->sr_rec;
1103 			if (rrec->jr_diroff == DOTDOT_OFFSET) {
1104 				stmp = ino_lookup(rrec->jr_parent, 0);
1105 				if (stmp)
1106 					ino_setskip(stmp, ino);
1107 			}
1108 		}
1109 	}
1110 	ginode(ino, &ip);
1111 	dp = ip.i_dp;
1112 	mode = DIP(dp, di_mode) & IFMT;
1113 	if (nlink > UFS_LINK_MAX)
1114 		err_suj("ino %ju nlink manipulation error, new %ju, old %d\n",
1115 		    (uintmax_t)ino, (uintmax_t)nlink, DIP(dp, di_nlink));
1116 	if (debug)
1117 	       printf("Adjusting ino %ju, nlink %ju, old link %d lastmode %o\n",
1118 		    (uintmax_t)ino, (uintmax_t)nlink, DIP(dp, di_nlink),
1119 		    sino->si_mode);
1120 	if (mode == 0) {
1121 		if (debug)
1122 			printf("ino %ju, zero inode freeing bitmap\n",
1123 			    (uintmax_t)ino);
1124 		ino_free(ino, sino->si_mode);
1125 		irelse(&ip);
1126 		return;
1127 	}
1128 	/* XXX Should be an assert? */
1129 	if (mode != sino->si_mode && debug)
1130 		printf("ino %ju, mode %o != %o\n",
1131 		    (uintmax_t)ino, mode, sino->si_mode);
1132 	if ((mode & IFMT) == IFDIR)
1133 		reqlink = 2;
1134 	else
1135 		reqlink = 1;
1136 	/* If the inode doesn't have enough links to live, free it. */
1137 	if (nlink < reqlink) {
1138 		if (debug)
1139 			printf("ino %ju not enough links to live %ju < %ju\n",
1140 			    (uintmax_t)ino, (uintmax_t)nlink,
1141 			    (uintmax_t)reqlink);
1142 		ino_reclaim(&ip, ino, mode);
1143 		irelse(&ip);
1144 		return;
1145 	}
1146 	/* If required write the updated link count. */
1147 	if (DIP(dp, di_nlink) == nlink) {
1148 		if (debug)
1149 			printf("ino %ju, link matches, skipping.\n",
1150 			    (uintmax_t)ino);
1151 		irelse(&ip);
1152 		return;
1153 	}
1154 	DIP_SET(dp, di_nlink, nlink);
1155 	inodirty(&ip);
1156 	irelse(&ip);
1157 }
1158 
1159 /*
1160  * Truncate some or all blocks in an indirect, freeing any that are required
1161  * and zeroing the indirect.
1162  */
1163 static void
1164 indir_trunc(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, ufs_lbn_t lastlbn,
1165 	union dinode *dp)
1166 {
1167 	struct bufarea *bp;
1168 	ufs_lbn_t lbnadd;
1169 	ufs2_daddr_t nblk;
1170 	ufs_lbn_t next;
1171 	ufs_lbn_t nlbn;
1172 	int isdirty;
1173 	int level;
1174 	int i;
1175 
1176 	if (blk == 0)
1177 		return;
1178 	isdirty = 0;
1179 	level = lbn_level(lbn);
1180 	if (level == -1)
1181 		err_suj("Invalid level for lbn %jd\n", lbn);
1182 	lbnadd = 1;
1183 	for (i = level; i > 0; i--)
1184 		lbnadd *= NINDIR(fs);
1185 	bp = getdatablk(blk, fs->fs_bsize, BT_LEVEL1 + level);
1186 	if (bp->b_errs != 0)
1187 		err_suj("indir_trunc: UNRECOVERABLE I/O ERROR");
1188 	for (i = 0; i < NINDIR(fs); i++) {
1189 		if ((nblk = IBLK(bp, i)) == 0)
1190 			continue;
1191 		if (level != 0) {
1192 			nlbn = (lbn + 1) - (i * lbnadd);
1193 			/*
1194 			 * Calculate the lbn of the next indirect to
1195 			 * determine if any of this indirect must be
1196 			 * reclaimed.
1197 			 */
1198 			next = -(lbn + level) + ((i+1) * lbnadd);
1199 			if (next <= lastlbn)
1200 				continue;
1201 			indir_trunc(ino, nlbn, nblk, lastlbn, dp);
1202 			/* If all of this indirect was reclaimed, free it. */
1203 			nlbn = next - lbnadd;
1204 			if (nlbn < lastlbn)
1205 				continue;
1206 		} else {
1207 			nlbn = -lbn + i * lbnadd;
1208 			if (nlbn < lastlbn)
1209 				continue;
1210 		}
1211 		isdirty = 1;
1212 		blk_free(nblk, 0, fs->fs_frag);
1213 		IBLK_SET(bp, i, 0);
1214 	}
1215 	if (isdirty)
1216 		dirty(bp);
1217 	brelse(bp);
1218 }
1219 
1220 /*
1221  * Truncate an inode to the minimum of the given size or the last populated
1222  * block after any over size have been discarded.  The kernel would allocate
1223  * the last block in the file but fsck does not and neither do we.  This
1224  * code never extends files, only shrinks them.
1225  */
1226 static void
1227 ino_trunc(ino_t ino, off_t size)
1228 {
1229 	struct inode ip;
1230 	union dinode *dp;
1231 	struct bufarea *bp;
1232 	ufs2_daddr_t bn;
1233 	uint64_t totalfrags;
1234 	ufs_lbn_t nextlbn;
1235 	ufs_lbn_t lastlbn;
1236 	ufs_lbn_t tmpval;
1237 	ufs_lbn_t lbn;
1238 	ufs_lbn_t i;
1239 	int blksize, frags;
1240 	off_t cursize;
1241 	off_t off;
1242 	int mode;
1243 
1244 	ginode(ino, &ip);
1245 	dp = ip.i_dp;
1246 	mode = DIP(dp, di_mode) & IFMT;
1247 	cursize = DIP(dp, di_size);
1248 	if (debug)
1249 		printf("Truncating ino %ju, mode %o to size %jd from size %jd\n",
1250 		    (uintmax_t)ino, mode, size, cursize);
1251 
1252 	/* Skip datablocks for short links and devices. */
1253 	if (mode == 0 || mode == IFBLK || mode == IFCHR ||
1254 	    (mode == IFLNK && cursize < fs->fs_maxsymlinklen)) {
1255 		irelse(&ip);
1256 		return;
1257 	}
1258 	/* Don't extend. */
1259 	if (size > cursize) {
1260 		irelse(&ip);
1261 		return;
1262 	}
1263 	if ((DIP(dp, di_flags) & SF_SNAPSHOT) != 0) {
1264 		if (size > 0)
1265 			err_suj("Partial truncation of ino %ju snapshot file\n",
1266 			    (uintmax_t)ino);
1267 	}
1268 	lastlbn = lblkno(fs, blkroundup(fs, size));
1269 	for (i = lastlbn; i < UFS_NDADDR; i++) {
1270 		if ((bn = DIP(dp, di_db[i])) == 0)
1271 			continue;
1272 		blksize = sblksize(fs, cursize, i);
1273 		blk_free(bn, 0, numfrags(fs, blksize));
1274 		DIP_SET(dp, di_db[i], 0);
1275 	}
1276 	/*
1277 	 * Follow indirect blocks, freeing anything required.
1278 	 */
1279 	for (i = 0, tmpval = NINDIR(fs), lbn = UFS_NDADDR; i < UFS_NIADDR; i++,
1280 	    lbn = nextlbn) {
1281 		nextlbn = lbn + tmpval;
1282 		tmpval *= NINDIR(fs);
1283 		/* If we're not freeing any in this indirect range skip it. */
1284 		if (lastlbn >= nextlbn)
1285 			continue;
1286 		if (DIP(dp, di_ib[i]) == 0)
1287 			continue;
1288 		indir_trunc(ino, -lbn - i, DIP(dp, di_ib[i]), lastlbn, dp);
1289 		/* If we freed everything in this indirect free the indir. */
1290 		if (lastlbn > lbn)
1291 			continue;
1292 		blk_free(DIP(dp, di_ib[i]), 0, fs->fs_frag);
1293 		DIP_SET(dp, di_ib[i], 0);
1294 	}
1295 	/*
1296 	 * Now that we've freed any whole blocks that exceed the desired
1297 	 * truncation size, figure out how many blocks remain and what the
1298 	 * last populated lbn is.  We will set the size to this last lbn
1299 	 * rather than worrying about allocating the final lbn as the kernel
1300 	 * would've done.  This is consistent with normal fsck behavior.
1301 	 */
1302 	visitlbn = 0;
1303 	totalfrags = ino_visit(dp, ino, null_visit, VISIT_INDIR | VISIT_EXT);
1304 	if (size > lblktosize(fs, visitlbn + 1))
1305 		size = lblktosize(fs, visitlbn + 1);
1306 	/*
1307 	 * If we're truncating direct blocks we have to adjust frags
1308 	 * accordingly.
1309 	 */
1310 	if (visitlbn < UFS_NDADDR && totalfrags) {
1311 		long oldspace, newspace;
1312 
1313 		bn = DIP(dp, di_db[visitlbn]);
1314 		if (bn == 0)
1315 			err_suj("Bad blk at ino %ju lbn %jd\n",
1316 			    (uintmax_t)ino, visitlbn);
1317 		oldspace = sblksize(fs, cursize, visitlbn);
1318 		newspace = sblksize(fs, size, visitlbn);
1319 		if (oldspace != newspace) {
1320 			bn += numfrags(fs, newspace);
1321 			frags = numfrags(fs, oldspace - newspace);
1322 			blk_free(bn, 0, frags);
1323 			totalfrags -= frags;
1324 		}
1325 	}
1326 	DIP_SET(dp, di_blocks, fsbtodb(fs, totalfrags));
1327 	DIP_SET(dp, di_size, size);
1328 	inodirty(&ip);
1329 	/*
1330 	 * If we've truncated into the middle of a block or frag we have
1331 	 * to zero it here.  Otherwise the file could extend into
1332 	 * uninitialized space later.
1333 	 */
1334 	off = blkoff(fs, size);
1335 	if (off && DIP(dp, di_mode) != IFDIR) {
1336 		long clrsize;
1337 
1338 		bn = ino_blkatoff(dp, ino, visitlbn, &frags, NULL);
1339 		if (bn == 0)
1340 			err_suj("Block missing from ino %ju at lbn %jd\n",
1341 			    (uintmax_t)ino, visitlbn);
1342 		clrsize = frags * fs->fs_fsize;
1343 		bp = getdatablk(bn, clrsize, BT_DATA);
1344 		if (bp->b_errs != 0)
1345 			err_suj("ino_trunc: UNRECOVERABLE I/O ERROR");
1346 		clrsize -= off;
1347 		bzero(&bp->b_un.b_buf[off], clrsize);
1348 		dirty(bp);
1349 		brelse(bp);
1350 	}
1351 	irelse(&ip);
1352 	return;
1353 }
1354 
1355 /*
1356  * Process records available for one inode and determine whether the
1357  * link count is correct or needs adjusting.
1358  */
1359 static void
1360 ino_check(struct suj_ino *sino)
1361 {
1362 	struct suj_rec *srec;
1363 	struct jrefrec *rrec;
1364 	nlink_t dotlinks;
1365 	nlink_t newlinks;
1366 	nlink_t removes;
1367 	nlink_t nlink;
1368 	ino_t ino;
1369 	int isdot;
1370 	int isat;
1371 	int mode;
1372 
1373 	if (sino->si_hasrecs == 0)
1374 		return;
1375 	ino = sino->si_ino;
1376 	rrec = (struct jrefrec *)TAILQ_FIRST(&sino->si_recs)->sr_rec;
1377 	nlink = rrec->jr_nlink;
1378 	newlinks = 0;
1379 	dotlinks = 0;
1380 	removes = sino->si_nlinkadj;
1381 	TAILQ_FOREACH(srec, &sino->si_recs, sr_next) {
1382 		rrec = (struct jrefrec *)srec->sr_rec;
1383 		isat = ino_isat(rrec->jr_parent, rrec->jr_diroff,
1384 		    rrec->jr_ino, &mode, &isdot);
1385 		if (isat && (mode & IFMT) != (rrec->jr_mode & IFMT))
1386 			err_suj("Inode mode/directory type mismatch %o != %o\n",
1387 			    mode, rrec->jr_mode);
1388 		if (debug)
1389 			printf("jrefrec: op %d ino %ju, nlink %ju, parent %ju, "
1390 			    "diroff %jd, mode %o, isat %d, isdot %d\n",
1391 			    rrec->jr_op, (uintmax_t)rrec->jr_ino,
1392 			    (uintmax_t)rrec->jr_nlink,
1393 			    (uintmax_t)rrec->jr_parent,
1394 			    (uintmax_t)rrec->jr_diroff,
1395 			    rrec->jr_mode, isat, isdot);
1396 		mode = rrec->jr_mode & IFMT;
1397 		if (rrec->jr_op == JOP_REMREF)
1398 			removes++;
1399 		newlinks += isat;
1400 		if (isdot)
1401 			dotlinks += isat;
1402 	}
1403 	/*
1404 	 * The number of links that remain are the starting link count
1405 	 * subtracted by the total number of removes with the total
1406 	 * links discovered back in.  An incomplete remove thus
1407 	 * makes no change to the link count but an add increases
1408 	 * by one.
1409 	 */
1410 	if (debug)
1411 		printf(
1412 		    "ino %ju nlink %ju newlinks %ju removes %ju dotlinks %ju\n",
1413 		    (uintmax_t)ino, (uintmax_t)nlink, (uintmax_t)newlinks,
1414 		    (uintmax_t)removes, (uintmax_t)dotlinks);
1415 	nlink += newlinks;
1416 	nlink -= removes;
1417 	sino->si_linkadj = 1;
1418 	sino->si_nlink = nlink;
1419 	sino->si_dotlinks = dotlinks;
1420 	sino->si_mode = mode;
1421 	ino_adjust(sino);
1422 }
1423 
1424 /*
1425  * Process records available for one block and determine whether it is
1426  * still allocated and whether the owning inode needs to be updated or
1427  * a free completed.
1428  */
1429 static void
1430 blk_check(struct suj_blk *sblk)
1431 {
1432 	struct suj_rec *srec;
1433 	struct jblkrec *brec;
1434 	struct suj_ino *sino;
1435 	ufs2_daddr_t blk;
1436 	int mask;
1437 	int frags;
1438 	int isat;
1439 
1440 	/*
1441 	 * Each suj_blk actually contains records for any fragments in that
1442 	 * block.  As a result we must evaluate each record individually.
1443 	 */
1444 	sino = NULL;
1445 	TAILQ_FOREACH(srec, &sblk->sb_recs, sr_next) {
1446 		brec = (struct jblkrec *)srec->sr_rec;
1447 		frags = brec->jb_frags;
1448 		blk = brec->jb_blkno + brec->jb_oldfrags;
1449 		isat = blk_isat(brec->jb_ino, brec->jb_lbn, blk, &frags);
1450 		if (sino == NULL || sino->si_ino != brec->jb_ino) {
1451 			sino = ino_lookup(brec->jb_ino, 1);
1452 			sino->si_blkadj = 1;
1453 		}
1454 		if (debug)
1455 			printf("op %d blk %jd ino %ju lbn %jd frags %d isat %d (%d)\n",
1456 			    brec->jb_op, blk, (uintmax_t)brec->jb_ino,
1457 			    brec->jb_lbn, brec->jb_frags, isat, frags);
1458 		/*
1459 		 * If we found the block at this address we still have to
1460 		 * determine if we need to free the tail end that was
1461 		 * added by adding contiguous fragments from the same block.
1462 		 */
1463 		if (isat == 1) {
1464 			if (frags == brec->jb_frags)
1465 				continue;
1466 			mask = blk_freemask(blk, brec->jb_ino, brec->jb_lbn,
1467 			    brec->jb_frags);
1468 			mask >>= frags;
1469 			blk += frags;
1470 			frags = brec->jb_frags - frags;
1471 			blk_free(blk, mask, frags);
1472 			continue;
1473 		}
1474 		/*
1475 	 	 * The block wasn't found, attempt to free it.  It won't be
1476 		 * freed if it was actually reallocated.  If this was an
1477 		 * allocation we don't want to follow indirects as they
1478 		 * may not be written yet.  Any children of the indirect will
1479 		 * have their own records.  If it's a free we need to
1480 		 * recursively free children.
1481 		 */
1482 		blk_free_lbn(blk, brec->jb_ino, brec->jb_lbn, brec->jb_frags,
1483 		    brec->jb_op == JOP_FREEBLK);
1484 	}
1485 }
1486 
1487 /*
1488  * Walk the list of inode records for this cg and resolve moved and duplicate
1489  * inode references now that we have a complete picture.
1490  */
1491 static void
1492 cg_build(struct suj_cg *sc)
1493 {
1494 	struct suj_ino *sino;
1495 	int i;
1496 
1497 	for (i = 0; i < HASHSIZE; i++)
1498 		LIST_FOREACH(sino, &sc->sc_inohash[i], si_next)
1499 			ino_build(sino);
1500 }
1501 
1502 /*
1503  * Handle inodes requiring truncation.  This must be done prior to
1504  * looking up any inodes in directories.
1505  */
1506 static void
1507 cg_trunc(struct suj_cg *sc)
1508 {
1509 	struct suj_ino *sino;
1510 	int i;
1511 
1512 	for (i = 0; i < HASHSIZE; i++) {
1513 		LIST_FOREACH(sino, &sc->sc_inohash[i], si_next) {
1514 			if (sino->si_trunc) {
1515 				ino_trunc(sino->si_ino,
1516 				    sino->si_trunc->jt_size);
1517 				sino->si_blkadj = 0;
1518 				sino->si_trunc = NULL;
1519 			}
1520 			if (sino->si_blkadj)
1521 				ino_adjblks(sino);
1522 		}
1523 	}
1524 }
1525 
1526 static void
1527 cg_adj_blk(struct suj_cg *sc)
1528 {
1529 	struct suj_ino *sino;
1530 	int i;
1531 
1532 	for (i = 0; i < HASHSIZE; i++) {
1533 		LIST_FOREACH(sino, &sc->sc_inohash[i], si_next) {
1534 			if (sino->si_blkadj)
1535 				ino_adjblks(sino);
1536 		}
1537 	}
1538 }
1539 
1540 /*
1541  * Free any partially allocated blocks and then resolve inode block
1542  * counts.
1543  */
1544 static void
1545 cg_check_blk(struct suj_cg *sc)
1546 {
1547 	struct suj_blk *sblk;
1548 	int i;
1549 
1550 
1551 	for (i = 0; i < HASHSIZE; i++)
1552 		LIST_FOREACH(sblk, &sc->sc_blkhash[i], sb_next)
1553 			blk_check(sblk);
1554 }
1555 
1556 /*
1557  * Walk the list of inode records for this cg, recovering any
1558  * changes which were not complete at the time of crash.
1559  */
1560 static void
1561 cg_check_ino(struct suj_cg *sc)
1562 {
1563 	struct suj_ino *sino;
1564 	int i;
1565 
1566 	for (i = 0; i < HASHSIZE; i++)
1567 		LIST_FOREACH(sino, &sc->sc_inohash[i], si_next)
1568 			ino_check(sino);
1569 }
1570 
1571 static void
1572 cg_apply(void (*apply)(struct suj_cg *))
1573 {
1574 	struct suj_cg *scg;
1575 	int i;
1576 
1577 	for (i = 0; i < HASHSIZE; i++)
1578 		LIST_FOREACH(scg, &cghash[i], sc_next)
1579 			apply(scg);
1580 }
1581 
1582 /*
1583  * Process the unlinked but referenced file list.  Freeing all inodes.
1584  */
1585 static void
1586 ino_unlinked(void)
1587 {
1588 	struct inode ip;
1589 	union dinode *dp;
1590 	uint16_t mode;
1591 	ino_t inon;
1592 	ino_t ino;
1593 
1594 	ino = fs->fs_sujfree;
1595 	fs->fs_sujfree = 0;
1596 	while (ino != 0) {
1597 		ginode(ino, &ip);
1598 		dp = ip.i_dp;
1599 		mode = DIP(dp, di_mode) & IFMT;
1600 		inon = DIP(dp, di_freelink);
1601 		DIP_SET(dp, di_freelink, 0);
1602 		inodirty(&ip);
1603 		/*
1604 		 * XXX Should this be an errx?
1605 		 */
1606 		if (DIP(dp, di_nlink) == 0) {
1607 			if (debug)
1608 				printf("Freeing unlinked ino %ju mode %o\n",
1609 				    (uintmax_t)ino, mode);
1610 			ino_reclaim(&ip, ino, mode);
1611 		} else if (debug)
1612 			printf("Skipping ino %ju mode %o with link %d\n",
1613 			    (uintmax_t)ino, mode, DIP(dp, di_nlink));
1614 		ino = inon;
1615 		irelse(&ip);
1616 	}
1617 }
1618 
1619 /*
1620  * Append a new record to the list of records requiring processing.
1621  */
1622 static void
1623 ino_append(union jrec *rec)
1624 {
1625 	struct jrefrec *refrec;
1626 	struct jmvrec *mvrec;
1627 	struct suj_ino *sino;
1628 	struct suj_rec *srec;
1629 
1630 	mvrec = &rec->rec_jmvrec;
1631 	refrec = &rec->rec_jrefrec;
1632 	if (debug && mvrec->jm_op == JOP_MVREF)
1633 		printf("ino move: ino %ju, parent %ju, "
1634 		    "diroff %jd, oldoff %jd\n",
1635 		    (uintmax_t)mvrec->jm_ino, (uintmax_t)mvrec->jm_parent,
1636 		    (uintmax_t)mvrec->jm_newoff, (uintmax_t)mvrec->jm_oldoff);
1637 	else if (debug &&
1638 	    (refrec->jr_op == JOP_ADDREF || refrec->jr_op == JOP_REMREF))
1639 		printf("ino ref: op %d, ino %ju, nlink %ju, "
1640 		    "parent %ju, diroff %jd\n",
1641 		    refrec->jr_op, (uintmax_t)refrec->jr_ino,
1642 		    (uintmax_t)refrec->jr_nlink,
1643 		    (uintmax_t)refrec->jr_parent, (uintmax_t)refrec->jr_diroff);
1644 	sino = ino_lookup(((struct jrefrec *)rec)->jr_ino, 1);
1645 	sino->si_hasrecs = 1;
1646 	srec = errmalloc(sizeof(*srec));
1647 	srec->sr_rec = rec;
1648 	TAILQ_INSERT_TAIL(&sino->si_newrecs, srec, sr_next);
1649 }
1650 
1651 /*
1652  * Add a reference adjustment to the sino list and eliminate dups.  The
1653  * primary loop in ino_build_ref() checks for dups but new ones may be
1654  * created as a result of offset adjustments.
1655  */
1656 static void
1657 ino_add_ref(struct suj_ino *sino, struct suj_rec *srec)
1658 {
1659 	struct jrefrec *refrec;
1660 	struct suj_rec *srn;
1661 	struct jrefrec *rrn;
1662 
1663 	refrec = (struct jrefrec *)srec->sr_rec;
1664 	/*
1665 	 * We walk backwards so that the oldest link count is preserved.  If
1666 	 * an add record conflicts with a remove keep the remove.  Redundant
1667 	 * removes are eliminated in ino_build_ref.  Otherwise we keep the
1668 	 * oldest record at a given location.
1669 	 */
1670 	for (srn = TAILQ_LAST(&sino->si_recs, srechd); srn;
1671 	    srn = TAILQ_PREV(srn, srechd, sr_next)) {
1672 		rrn = (struct jrefrec *)srn->sr_rec;
1673 		if (rrn->jr_parent != refrec->jr_parent ||
1674 		    rrn->jr_diroff != refrec->jr_diroff)
1675 			continue;
1676 		if (rrn->jr_op == JOP_REMREF || refrec->jr_op == JOP_ADDREF) {
1677 			rrn->jr_mode = refrec->jr_mode;
1678 			return;
1679 		}
1680 		/*
1681 		 * Adding a remove.
1682 		 *
1683 		 * Replace the record in place with the old nlink in case
1684 		 * we replace the head of the list.  Abandon srec as a dup.
1685 		 */
1686 		refrec->jr_nlink = rrn->jr_nlink;
1687 		srn->sr_rec = srec->sr_rec;
1688 		return;
1689 	}
1690 	TAILQ_INSERT_TAIL(&sino->si_recs, srec, sr_next);
1691 }
1692 
1693 /*
1694  * Create a duplicate of a reference at a previous location.
1695  */
1696 static void
1697 ino_dup_ref(struct suj_ino *sino, struct jrefrec *refrec, off_t diroff)
1698 {
1699 	struct jrefrec *rrn;
1700 	struct suj_rec *srn;
1701 
1702 	rrn = errmalloc(sizeof(*refrec));
1703 	*rrn = *refrec;
1704 	rrn->jr_op = JOP_ADDREF;
1705 	rrn->jr_diroff = diroff;
1706 	srn = errmalloc(sizeof(*srn));
1707 	srn->sr_rec = (union jrec *)rrn;
1708 	ino_add_ref(sino, srn);
1709 }
1710 
1711 /*
1712  * Add a reference to the list at all known locations.  We follow the offset
1713  * changes for a single instance and create duplicate add refs at each so
1714  * that we can tolerate any version of the directory block.  Eliminate
1715  * removes which collide with adds that are seen in the journal.  They should
1716  * not adjust the link count down.
1717  */
1718 static void
1719 ino_build_ref(struct suj_ino *sino, struct suj_rec *srec)
1720 {
1721 	struct jrefrec *refrec;
1722 	struct jmvrec *mvrec;
1723 	struct suj_rec *srp;
1724 	struct suj_rec *srn;
1725 	struct jrefrec *rrn;
1726 	off_t diroff;
1727 
1728 	refrec = (struct jrefrec *)srec->sr_rec;
1729 	/*
1730 	 * Search for a mvrec that matches this offset.  Whether it's an add
1731 	 * or a remove we can delete the mvref after creating a dup record in
1732 	 * the old location.
1733 	 */
1734 	if (!TAILQ_EMPTY(&sino->si_movs)) {
1735 		diroff = refrec->jr_diroff;
1736 		for (srn = TAILQ_LAST(&sino->si_movs, srechd); srn; srn = srp) {
1737 			srp = TAILQ_PREV(srn, srechd, sr_next);
1738 			mvrec = (struct jmvrec *)srn->sr_rec;
1739 			if (mvrec->jm_parent != refrec->jr_parent ||
1740 			    mvrec->jm_newoff != diroff)
1741 				continue;
1742 			diroff = mvrec->jm_oldoff;
1743 			TAILQ_REMOVE(&sino->si_movs, srn, sr_next);
1744 			free(srn);
1745 			ino_dup_ref(sino, refrec, diroff);
1746 		}
1747 	}
1748 	/*
1749 	 * If a remove wasn't eliminated by an earlier add just append it to
1750 	 * the list.
1751 	 */
1752 	if (refrec->jr_op == JOP_REMREF) {
1753 		ino_add_ref(sino, srec);
1754 		return;
1755 	}
1756 	/*
1757 	 * Walk the list of records waiting to be added to the list.  We
1758 	 * must check for moves that apply to our current offset and remove
1759 	 * them from the list.  Remove any duplicates to eliminate removes
1760 	 * with corresponding adds.
1761 	 */
1762 	TAILQ_FOREACH_SAFE(srn, &sino->si_newrecs, sr_next, srp) {
1763 		switch (srn->sr_rec->rec_jrefrec.jr_op) {
1764 		case JOP_ADDREF:
1765 			/*
1766 			 * This should actually be an error we should
1767 			 * have a remove for every add journaled.
1768 			 */
1769 			rrn = (struct jrefrec *)srn->sr_rec;
1770 			if (rrn->jr_parent != refrec->jr_parent ||
1771 			    rrn->jr_diroff != refrec->jr_diroff)
1772 				break;
1773 			TAILQ_REMOVE(&sino->si_newrecs, srn, sr_next);
1774 			break;
1775 		case JOP_REMREF:
1776 			/*
1777 			 * Once we remove the current iteration of the
1778 			 * record at this address we're done.
1779 			 */
1780 			rrn = (struct jrefrec *)srn->sr_rec;
1781 			if (rrn->jr_parent != refrec->jr_parent ||
1782 			    rrn->jr_diroff != refrec->jr_diroff)
1783 				break;
1784 			TAILQ_REMOVE(&sino->si_newrecs, srn, sr_next);
1785 			ino_add_ref(sino, srec);
1786 			return;
1787 		case JOP_MVREF:
1788 			/*
1789 			 * Update our diroff based on any moves that match
1790 			 * and remove the move.
1791 			 */
1792 			mvrec = (struct jmvrec *)srn->sr_rec;
1793 			if (mvrec->jm_parent != refrec->jr_parent ||
1794 			    mvrec->jm_oldoff != refrec->jr_diroff)
1795 				break;
1796 			ino_dup_ref(sino, refrec, mvrec->jm_oldoff);
1797 			refrec->jr_diroff = mvrec->jm_newoff;
1798 			TAILQ_REMOVE(&sino->si_newrecs, srn, sr_next);
1799 			break;
1800 		default:
1801 			err_suj("ino_build_ref: Unknown op %d\n",
1802 			    srn->sr_rec->rec_jrefrec.jr_op);
1803 		}
1804 	}
1805 	ino_add_ref(sino, srec);
1806 }
1807 
1808 /*
1809  * Walk the list of new records and add them in-order resolving any
1810  * dups and adjusted offsets.
1811  */
1812 static void
1813 ino_build(struct suj_ino *sino)
1814 {
1815 	struct suj_rec *srec;
1816 
1817 	while ((srec = TAILQ_FIRST(&sino->si_newrecs)) != NULL) {
1818 		TAILQ_REMOVE(&sino->si_newrecs, srec, sr_next);
1819 		switch (srec->sr_rec->rec_jrefrec.jr_op) {
1820 		case JOP_ADDREF:
1821 		case JOP_REMREF:
1822 			ino_build_ref(sino, srec);
1823 			break;
1824 		case JOP_MVREF:
1825 			/*
1826 			 * Add this mvrec to the queue of pending mvs.
1827 			 */
1828 			TAILQ_INSERT_TAIL(&sino->si_movs, srec, sr_next);
1829 			break;
1830 		default:
1831 			err_suj("ino_build: Unknown op %d\n",
1832 			    srec->sr_rec->rec_jrefrec.jr_op);
1833 		}
1834 	}
1835 	if (TAILQ_EMPTY(&sino->si_recs))
1836 		sino->si_hasrecs = 0;
1837 }
1838 
1839 /*
1840  * Modify journal records so they refer to the base block number
1841  * and a start and end frag range.  This is to facilitate the discovery
1842  * of overlapping fragment allocations.
1843  */
1844 static void
1845 blk_build(struct jblkrec *blkrec)
1846 {
1847 	struct suj_rec *srec;
1848 	struct suj_blk *sblk;
1849 	struct jblkrec *blkrn;
1850 	ufs2_daddr_t blk;
1851 	int frag;
1852 
1853 	if (debug)
1854 		printf("blk_build: op %d blkno %jd frags %d oldfrags %d "
1855 		    "ino %ju lbn %jd\n",
1856 		    blkrec->jb_op, (uintmax_t)blkrec->jb_blkno,
1857 		    blkrec->jb_frags, blkrec->jb_oldfrags,
1858 		    (uintmax_t)blkrec->jb_ino, (uintmax_t)blkrec->jb_lbn);
1859 
1860 	blk = blknum(fs, blkrec->jb_blkno);
1861 	frag = fragnum(fs, blkrec->jb_blkno);
1862 	sblk = blk_lookup(blk, 1);
1863 	/*
1864 	 * Rewrite the record using oldfrags to indicate the offset into
1865 	 * the block.  Leave jb_frags as the actual allocated count.
1866 	 */
1867 	blkrec->jb_blkno -= frag;
1868 	blkrec->jb_oldfrags = frag;
1869 	if (blkrec->jb_oldfrags + blkrec->jb_frags > fs->fs_frag)
1870 		err_suj("Invalid fragment count %d oldfrags %d\n",
1871 		    blkrec->jb_frags, frag);
1872 	/*
1873 	 * Detect dups.  If we detect a dup we always discard the oldest
1874 	 * record as it is superseded by the new record.  This speeds up
1875 	 * later stages but also eliminates free records which are used
1876 	 * to indicate that the contents of indirects can be trusted.
1877 	 */
1878 	TAILQ_FOREACH(srec, &sblk->sb_recs, sr_next) {
1879 		blkrn = (struct jblkrec *)srec->sr_rec;
1880 		if (blkrn->jb_ino != blkrec->jb_ino ||
1881 		    blkrn->jb_lbn != blkrec->jb_lbn ||
1882 		    blkrn->jb_blkno != blkrec->jb_blkno ||
1883 		    blkrn->jb_frags != blkrec->jb_frags ||
1884 		    blkrn->jb_oldfrags != blkrec->jb_oldfrags)
1885 			continue;
1886 		if (debug)
1887 			printf("Removed dup.\n");
1888 		/* Discard the free which is a dup with an alloc. */
1889 		if (blkrec->jb_op == JOP_FREEBLK)
1890 			return;
1891 		TAILQ_REMOVE(&sblk->sb_recs, srec, sr_next);
1892 		free(srec);
1893 		break;
1894 	}
1895 	srec = errmalloc(sizeof(*srec));
1896 	srec->sr_rec = (union jrec *)blkrec;
1897 	TAILQ_INSERT_TAIL(&sblk->sb_recs, srec, sr_next);
1898 }
1899 
1900 static void
1901 ino_build_trunc(struct jtrncrec *rec)
1902 {
1903 	struct suj_ino *sino;
1904 
1905 	if (debug)
1906 		printf("ino_build_trunc: op %d ino %ju, size %jd\n",
1907 		    rec->jt_op, (uintmax_t)rec->jt_ino,
1908 		    (uintmax_t)rec->jt_size);
1909 	sino = ino_lookup(rec->jt_ino, 1);
1910 	if (rec->jt_op == JOP_SYNC) {
1911 		sino->si_trunc = NULL;
1912 		return;
1913 	}
1914 	if (sino->si_trunc == NULL || sino->si_trunc->jt_size > rec->jt_size)
1915 		sino->si_trunc = rec;
1916 }
1917 
1918 /*
1919  * Build up tables of the operations we need to recover.
1920  */
1921 static void
1922 suj_build(void)
1923 {
1924 	struct suj_seg *seg;
1925 	union jrec *rec;
1926 	int off;
1927 	int i;
1928 
1929 	TAILQ_FOREACH(seg, &allsegs, ss_next) {
1930 		if (debug)
1931 			printf("seg %jd has %d records, oldseq %jd.\n",
1932 			    seg->ss_rec.jsr_seq, seg->ss_rec.jsr_cnt,
1933 			    seg->ss_rec.jsr_oldest);
1934 		off = 0;
1935 		rec = (union jrec *)seg->ss_blk;
1936 		for (i = 0; i < seg->ss_rec.jsr_cnt; off += JREC_SIZE, rec++) {
1937 			/* skip the segrec. */
1938 			if ((off % real_dev_bsize) == 0)
1939 				continue;
1940 			switch (rec->rec_jrefrec.jr_op) {
1941 			case JOP_ADDREF:
1942 			case JOP_REMREF:
1943 			case JOP_MVREF:
1944 				ino_append(rec);
1945 				break;
1946 			case JOP_NEWBLK:
1947 			case JOP_FREEBLK:
1948 				blk_build((struct jblkrec *)rec);
1949 				break;
1950 			case JOP_TRUNC:
1951 			case JOP_SYNC:
1952 				ino_build_trunc((struct jtrncrec *)rec);
1953 				break;
1954 			default:
1955 				err_suj("Unknown journal operation %d (%d)\n",
1956 				    rec->rec_jrefrec.jr_op, off);
1957 			}
1958 			i++;
1959 		}
1960 	}
1961 }
1962 
1963 /*
1964  * Prune the journal segments to those we care about based on the
1965  * oldest sequence in the newest segment.  Order the segment list
1966  * based on sequence number.
1967  */
1968 static void
1969 suj_prune(void)
1970 {
1971 	struct suj_seg *seg;
1972 	struct suj_seg *segn;
1973 	uint64_t newseq;
1974 	int discard;
1975 
1976 	if (debug)
1977 		printf("Pruning up to %jd\n", oldseq);
1978 	/* First free the expired segments. */
1979 	TAILQ_FOREACH_SAFE(seg, &allsegs, ss_next, segn) {
1980 		if (seg->ss_rec.jsr_seq >= oldseq)
1981 			continue;
1982 		TAILQ_REMOVE(&allsegs, seg, ss_next);
1983 		free(seg->ss_blk);
1984 		free(seg);
1985 	}
1986 	/* Next ensure that segments are ordered properly. */
1987 	seg = TAILQ_FIRST(&allsegs);
1988 	if (seg == NULL) {
1989 		if (debug)
1990 			printf("Empty journal\n");
1991 		return;
1992 	}
1993 	newseq = seg->ss_rec.jsr_seq;
1994 	for (;;) {
1995 		seg = TAILQ_LAST(&allsegs, seghd);
1996 		if (seg->ss_rec.jsr_seq >= newseq)
1997 			break;
1998 		TAILQ_REMOVE(&allsegs, seg, ss_next);
1999 		TAILQ_INSERT_HEAD(&allsegs, seg, ss_next);
2000 		newseq = seg->ss_rec.jsr_seq;
2001 
2002 	}
2003 	if (newseq != oldseq) {
2004 		TAILQ_FOREACH(seg, &allsegs, ss_next) {
2005 			printf("%jd, ", seg->ss_rec.jsr_seq);
2006 		}
2007 		printf("\n");
2008 		err_suj("Journal file sequence mismatch %jd != %jd\n",
2009 		    newseq, oldseq);
2010 	}
2011 	/*
2012 	 * The kernel may asynchronously write segments which can create
2013 	 * gaps in the sequence space.  Throw away any segments after the
2014 	 * gap as the kernel guarantees only those that are contiguously
2015 	 * reachable are marked as completed.
2016 	 */
2017 	discard = 0;
2018 	TAILQ_FOREACH_SAFE(seg, &allsegs, ss_next, segn) {
2019 		if (!discard && newseq++ == seg->ss_rec.jsr_seq) {
2020 			jrecs += seg->ss_rec.jsr_cnt;
2021 			jbytes += seg->ss_rec.jsr_blocks * real_dev_bsize;
2022 			continue;
2023 		}
2024 		discard = 1;
2025 		if (debug)
2026 			printf("Journal order mismatch %jd != %jd pruning\n",
2027 			    newseq-1, seg->ss_rec.jsr_seq);
2028 		TAILQ_REMOVE(&allsegs, seg, ss_next);
2029 		free(seg->ss_blk);
2030 		free(seg);
2031 	}
2032 	if (debug)
2033 		printf("Processing journal segments from %jd to %jd\n",
2034 		    oldseq, newseq-1);
2035 }
2036 
2037 /*
2038  * Verify the journal inode before attempting to read records.
2039  */
2040 static int
2041 suj_verifyino(union dinode *dp)
2042 {
2043 
2044 	if (DIP(dp, di_nlink) != 1) {
2045 		printf("Invalid link count %d for journal inode %ju\n",
2046 		    DIP(dp, di_nlink), (uintmax_t)sujino);
2047 		return (-1);
2048 	}
2049 
2050 	if ((DIP(dp, di_flags) & (SF_IMMUTABLE | SF_NOUNLINK)) !=
2051 	    (SF_IMMUTABLE | SF_NOUNLINK)) {
2052 		printf("Invalid flags 0x%X for journal inode %ju\n",
2053 		    DIP(dp, di_flags), (uintmax_t)sujino);
2054 		return (-1);
2055 	}
2056 
2057 	if (DIP(dp, di_mode) != (IFREG | IREAD)) {
2058 		printf("Invalid mode %o for journal inode %ju\n",
2059 		    DIP(dp, di_mode), (uintmax_t)sujino);
2060 		return (-1);
2061 	}
2062 
2063 	if (DIP(dp, di_size) < SUJ_MIN) {
2064 		printf("Invalid size %jd for journal inode %ju\n",
2065 		    DIP(dp, di_size), (uintmax_t)sujino);
2066 		return (-1);
2067 	}
2068 
2069 	if (DIP(dp, di_modrev) != fs->fs_mtime) {
2070 		printf("Journal timestamp does not match fs mount time\n");
2071 		return (-1);
2072 	}
2073 
2074 	return (0);
2075 }
2076 
2077 struct jblocks {
2078 	struct jextent *jb_extent;	/* Extent array. */
2079 	int		jb_avail;	/* Available extents. */
2080 	int		jb_used;	/* Last used extent. */
2081 	int		jb_head;	/* Allocator head. */
2082 	int		jb_off;		/* Allocator extent offset. */
2083 };
2084 struct jextent {
2085 	ufs2_daddr_t	je_daddr;	/* Disk block address. */
2086 	int		je_blocks;	/* Disk block count. */
2087 };
2088 
2089 static struct jblocks *suj_jblocks;
2090 
2091 static struct jblocks *
2092 jblocks_create(void)
2093 {
2094 	struct jblocks *jblocks;
2095 	int size;
2096 
2097 	jblocks = errmalloc(sizeof(*jblocks));
2098 	jblocks->jb_avail = 10;
2099 	jblocks->jb_used = 0;
2100 	jblocks->jb_head = 0;
2101 	jblocks->jb_off = 0;
2102 	size = sizeof(struct jextent) * jblocks->jb_avail;
2103 	jblocks->jb_extent = errmalloc(size);
2104 	bzero(jblocks->jb_extent, size);
2105 
2106 	return (jblocks);
2107 }
2108 
2109 /*
2110  * Return the next available disk block and the amount of contiguous
2111  * free space it contains.
2112  */
2113 static ufs2_daddr_t
2114 jblocks_next(struct jblocks *jblocks, int bytes, int *actual)
2115 {
2116 	struct jextent *jext;
2117 	ufs2_daddr_t daddr;
2118 	int freecnt;
2119 	int blocks;
2120 
2121 	blocks = btodb(bytes);
2122 	jext = &jblocks->jb_extent[jblocks->jb_head];
2123 	freecnt = jext->je_blocks - jblocks->jb_off;
2124 	if (freecnt == 0) {
2125 		jblocks->jb_off = 0;
2126 		if (++jblocks->jb_head > jblocks->jb_used)
2127 			return (0);
2128 		jext = &jblocks->jb_extent[jblocks->jb_head];
2129 		freecnt = jext->je_blocks;
2130 	}
2131 	if (freecnt > blocks)
2132 		freecnt = blocks;
2133 	*actual = dbtob(freecnt);
2134 	daddr = jext->je_daddr + jblocks->jb_off;
2135 
2136 	return (daddr);
2137 }
2138 
2139 /*
2140  * Advance the allocation head by a specified number of bytes, consuming
2141  * one journal segment.
2142  */
2143 static void
2144 jblocks_advance(struct jblocks *jblocks, int bytes)
2145 {
2146 
2147 	jblocks->jb_off += btodb(bytes);
2148 }
2149 
2150 static void
2151 jblocks_destroy(struct jblocks *jblocks)
2152 {
2153 
2154 	free(jblocks->jb_extent);
2155 	free(jblocks);
2156 }
2157 
2158 static void
2159 jblocks_add(struct jblocks *jblocks, ufs2_daddr_t daddr, int blocks)
2160 {
2161 	struct jextent *jext;
2162 	int size;
2163 
2164 	jext = &jblocks->jb_extent[jblocks->jb_used];
2165 	/* Adding the first block. */
2166 	if (jext->je_daddr == 0) {
2167 		jext->je_daddr = daddr;
2168 		jext->je_blocks = blocks;
2169 		return;
2170 	}
2171 	/* Extending the last extent. */
2172 	if (jext->je_daddr + jext->je_blocks == daddr) {
2173 		jext->je_blocks += blocks;
2174 		return;
2175 	}
2176 	/* Adding a new extent. */
2177 	if (++jblocks->jb_used == jblocks->jb_avail) {
2178 		jblocks->jb_avail *= 2;
2179 		size = sizeof(struct jextent) * jblocks->jb_avail;
2180 		jext = errmalloc(size);
2181 		bzero(jext, size);
2182 		bcopy(jblocks->jb_extent, jext,
2183 		    sizeof(struct jextent) * jblocks->jb_used);
2184 		free(jblocks->jb_extent);
2185 		jblocks->jb_extent = jext;
2186 	}
2187 	jext = &jblocks->jb_extent[jblocks->jb_used];
2188 	jext->je_daddr = daddr;
2189 	jext->je_blocks = blocks;
2190 
2191 	return;
2192 }
2193 
2194 /*
2195  * Add a file block from the journal to the extent map.  We can't read
2196  * each file block individually because the kernel treats it as a circular
2197  * buffer and segments may span mutliple contiguous blocks.
2198  */
2199 static void
2200 suj_add_block(ino_t ino, ufs_lbn_t lbn, ufs2_daddr_t blk, int frags)
2201 {
2202 
2203 	jblocks_add(suj_jblocks, fsbtodb(fs, blk), fsbtodb(fs, frags));
2204 }
2205 
2206 static void
2207 suj_read(void)
2208 {
2209 	uint8_t block[1 * 1024 * 1024];
2210 	struct suj_seg *seg;
2211 	struct jsegrec *recn;
2212 	struct jsegrec *rec;
2213 	ufs2_daddr_t blk;
2214 	int readsize;
2215 	int blocks;
2216 	int recsize;
2217 	int size;
2218 	int i;
2219 
2220 	/*
2221 	 * Read records until we exhaust the journal space.  If we find
2222 	 * an invalid record we start searching for a valid segment header
2223 	 * at the next block.  This is because we don't have a head/tail
2224 	 * pointer and must recover the information indirectly.  At the gap
2225 	 * between the head and tail we won't necessarily have a valid
2226 	 * segment.
2227 	 */
2228 restart:
2229 	for (;;) {
2230 		size = sizeof(block);
2231 		blk = jblocks_next(suj_jblocks, size, &readsize);
2232 		if (blk == 0)
2233 			return;
2234 		size = readsize;
2235 		/*
2236 		 * Read 1MB at a time and scan for records within this block.
2237 		 */
2238 		if (pread(fsreadfd, &block, size, dbtob(blk)) != size) {
2239 			err_suj("Error reading journal block %jd\n",
2240 			    (intmax_t)blk);
2241 		}
2242 		for (rec = (void *)block; size; size -= recsize,
2243 		    rec = (struct jsegrec *)((uintptr_t)rec + recsize)) {
2244 			recsize = real_dev_bsize;
2245 			if (rec->jsr_time != fs->fs_mtime) {
2246 #ifdef notdef
2247 				if (debug)
2248 					printf("Rec time %jd != fs mtime %jd\n",
2249 					    rec->jsr_time, fs->fs_mtime);
2250 #endif
2251 				jblocks_advance(suj_jblocks, recsize);
2252 				continue;
2253 			}
2254 			if (rec->jsr_cnt == 0) {
2255 				if (debug)
2256 					printf("Found illegal count %d\n",
2257 					    rec->jsr_cnt);
2258 				jblocks_advance(suj_jblocks, recsize);
2259 				continue;
2260 			}
2261 			blocks = rec->jsr_blocks;
2262 			recsize = blocks * real_dev_bsize;
2263 			if (recsize > size) {
2264 				/*
2265 				 * We may just have run out of buffer, restart
2266 				 * the loop to re-read from this spot.
2267 				 */
2268 				if (size < fs->fs_bsize &&
2269 				    size != readsize &&
2270 				    recsize <= fs->fs_bsize)
2271 					goto restart;
2272 				if (debug)
2273 					printf("Found invalid segsize %d > %d\n",
2274 					    recsize, size);
2275 				recsize = real_dev_bsize;
2276 				jblocks_advance(suj_jblocks, recsize);
2277 				continue;
2278 			}
2279 			/*
2280 			 * Verify that all blocks in the segment are present.
2281 			 */
2282 			for (i = 1; i < blocks; i++) {
2283 				recn = (void *)((uintptr_t)rec) + i *
2284 				    real_dev_bsize;
2285 				if (recn->jsr_seq == rec->jsr_seq &&
2286 				    recn->jsr_time == rec->jsr_time)
2287 					continue;
2288 				if (debug)
2289 					printf("Incomplete record %jd (%d)\n",
2290 					    rec->jsr_seq, i);
2291 				recsize = i * real_dev_bsize;
2292 				jblocks_advance(suj_jblocks, recsize);
2293 				goto restart;
2294 			}
2295 			seg = errmalloc(sizeof(*seg));
2296 			seg->ss_blk = errmalloc(recsize);
2297 			seg->ss_rec = *rec;
2298 			bcopy((void *)rec, seg->ss_blk, recsize);
2299 			if (rec->jsr_oldest > oldseq)
2300 				oldseq = rec->jsr_oldest;
2301 			TAILQ_INSERT_TAIL(&allsegs, seg, ss_next);
2302 			jblocks_advance(suj_jblocks, recsize);
2303 		}
2304 	}
2305 }
2306 
2307 /*
2308  * Orchestrate the verification of a filesystem via the softupdates journal.
2309  */
2310 int
2311 suj_check(const char *filesys)
2312 {
2313 	struct inodesc idesc;
2314 	struct csum *cgsum;
2315 	union dinode *jip;
2316 	struct inode ip;
2317 	uint64_t blocks;
2318 	int i, retval;
2319 	struct suj_seg *seg;
2320 	struct suj_seg *segn;
2321 
2322 	initsuj();
2323 	fs = &sblock;
2324 	if (real_dev_bsize == 0 && ioctl(fsreadfd, DIOCGSECTORSIZE,
2325 	    &real_dev_bsize) == -1)
2326 		real_dev_bsize = secsize;
2327 	if (debug)
2328 		printf("dev_bsize %u\n", real_dev_bsize);
2329 
2330 	/*
2331 	 * Set an exit point when SUJ check failed
2332 	 */
2333 	retval = setjmp(jmpbuf);
2334 	if (retval != 0) {
2335 		pwarn("UNEXPECTED SU+J INCONSISTENCY\n");
2336 		TAILQ_FOREACH_SAFE(seg, &allsegs, ss_next, segn) {
2337 			TAILQ_REMOVE(&allsegs, seg, ss_next);
2338 				free(seg->ss_blk);
2339 				free(seg);
2340 		}
2341 		if (reply("FALLBACK TO FULL FSCK") == 0) {
2342 			ckfini(0);
2343 			exit(EEXIT);
2344 		} else
2345 			return (-1);
2346 	}
2347 
2348 	/*
2349 	 * Search the root directory for the SUJ_FILE.
2350 	 */
2351 	idesc.id_type = DATA;
2352 	idesc.id_fix = IGNORE;
2353 	idesc.id_number = UFS_ROOTINO;
2354 	idesc.id_func = findino;
2355 	idesc.id_name = SUJ_FILE;
2356 	ginode(UFS_ROOTINO, &ip);
2357 	if ((ckinode(ip.i_dp, &idesc) & FOUND) == FOUND) {
2358 		sujino = idesc.id_parent;
2359 		irelse(&ip);
2360 	} else {
2361 		printf("Journal inode removed.  Use tunefs to re-create.\n");
2362 		sblock.fs_flags &= ~FS_SUJ;
2363 		sblock.fs_sujfree = 0;
2364 		irelse(&ip);
2365 		return (-1);
2366 	}
2367 	/*
2368 	 * Fetch the journal inode and verify it.
2369 	 */
2370 	ginode(sujino, &ip);
2371 	jip = ip.i_dp;
2372 	printf("** SU+J Recovering %s\n", filesys);
2373 	if (suj_verifyino(jip) != 0 || (!preen && !reply("USE JOURNAL"))) {
2374 		irelse(&ip);
2375 		return (-1);
2376 	}
2377 	/*
2378 	 * Build a list of journal blocks in jblocks before parsing the
2379 	 * available journal blocks in with suj_read().
2380 	 */
2381 	printf("** Reading %jd byte journal from inode %ju.\n",
2382 	    DIP(jip, di_size), (uintmax_t)sujino);
2383 	suj_jblocks = jblocks_create();
2384 	blocks = ino_visit(jip, sujino, suj_add_block, 0);
2385 	if (blocks != numfrags(fs, DIP(jip, di_size))) {
2386 		printf("Sparse journal inode %ju.\n", (uintmax_t)sujino);
2387 		irelse(&ip);
2388 		return (-1);
2389 	}
2390 	irelse(&ip);
2391 	suj_read();
2392 	jblocks_destroy(suj_jblocks);
2393 	suj_jblocks = NULL;
2394 	if (preen || reply("RECOVER")) {
2395 		printf("** Building recovery table.\n");
2396 		suj_prune();
2397 		suj_build();
2398 		cg_apply(cg_build);
2399 		printf("** Resolving unreferenced inode list.\n");
2400 		ino_unlinked();
2401 		printf("** Processing journal entries.\n");
2402 		cg_apply(cg_trunc);
2403 		cg_apply(cg_check_blk);
2404 		cg_apply(cg_adj_blk);
2405 		cg_apply(cg_check_ino);
2406 	}
2407 	if (preen == 0 && (jrecs > 0 || jbytes > 0) && reply("WRITE CHANGES") == 0)
2408 		return (0);
2409 	/*
2410 	 * Recompute the fs summary info from correct cs summaries.
2411 	 */
2412 	bzero(&fs->fs_cstotal, sizeof(struct csum_total));
2413 	for (i = 0; i < fs->fs_ncg; i++) {
2414 		cgsum = &fs->fs_cs(fs, i);
2415 		fs->fs_cstotal.cs_nffree += cgsum->cs_nffree;
2416 		fs->fs_cstotal.cs_nbfree += cgsum->cs_nbfree;
2417 		fs->fs_cstotal.cs_nifree += cgsum->cs_nifree;
2418 		fs->fs_cstotal.cs_ndir += cgsum->cs_ndir;
2419 	}
2420 	fs->fs_pendinginodes = 0;
2421 	fs->fs_pendingblocks = 0;
2422 	fs->fs_clean = 1;
2423 	fs->fs_time = time(NULL);
2424 	fs->fs_mtime = time(NULL);
2425 	sbdirty();
2426 	ckfini(1);
2427 	if (jrecs > 0 || jbytes > 0) {
2428 		printf("** %jd journal records in %jd bytes for %.2f%% utilization\n",
2429 		    jrecs, jbytes, ((float)jrecs / (float)(jbytes / JREC_SIZE)) * 100);
2430 		printf("** Freed %jd inodes (%jd dirs) %jd blocks, and %jd frags.\n",
2431 		    freeinos, freedir, freeblocks, freefrags);
2432 	}
2433 
2434 	return (0);
2435 }
2436 
2437 static void
2438 initsuj(void)
2439 {
2440 	int i;
2441 
2442 	for (i = 0; i < HASHSIZE; i++)
2443 		LIST_INIT(&cghash[i]);
2444 	lastcg = NULL;
2445 	TAILQ_INIT(&allsegs);
2446 	oldseq = 0;
2447 	fs = NULL;
2448 	sujino = 0;
2449 	freefrags = 0;
2450 	freeblocks = 0;
2451 	freeinos = 0;
2452 	freedir = 0;
2453 	jbytes = 0;
2454 	jrecs = 0;
2455 	suj_jblocks = NULL;
2456 }
2457