xref: /netbsd-src/sbin/fsck_lfs/pass6.c (revision 8b0f9554ff8762542c4defc4f70e1eb76fb508fa)
1 /* $NetBSD: pass6.c,v 1.20 2007/10/10 20:42:20 ad Exp $	 */
2 
3 /*-
4  * Copyright (c) 2003 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Konrad E. Schroder <perseant@hhhh.org>.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *      This product includes software developed by the NetBSD
21  *      Foundation, Inc. and its contributors.
22  * 4. Neither the name of The NetBSD Foundation nor the names of its
23  *    contributors may be used to endorse or promote products derived
24  *    from this software without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
27  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
28  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
29  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
30  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36  * POSSIBILITY OF SUCH DAMAGE.
37  */
38 
39 #include <sys/types.h>
40 #include <sys/param.h>
41 #include <sys/time.h>
42 #include <sys/buf.h>
43 #include <sys/mount.h>
44 
45 #include <ufs/ufs/ufsmount.h>
46 #include <ufs/ufs/inode.h>
47 #include <ufs/ufs/dir.h>
48 #define vnode uvnode
49 #include <ufs/lfs/lfs.h>
50 #undef vnode
51 
52 #include <assert.h>
53 #include <err.h>
54 #include <signal.h>
55 #include <string.h>
56 #include <stdio.h>
57 #include <stdlib.h>
58 #include <util.h>
59 
60 #include "bufcache.h"
61 #include "vnode.h"
62 #include "lfs_user.h"
63 #include "segwrite.h"
64 
65 #include "fsck.h"
66 #include "extern.h"
67 #include "fsutil.h"
68 
69 extern u_int32_t cksum(void *, size_t);
70 extern u_int32_t lfs_sb_cksum(struct dlfs *);
71 
72 extern ufs_daddr_t badblk;
73 extern SEGUSE *seg_table;
74 
75 static int nnewblocks;
76 
77 /*
78  * Our own copy of lfs_update_single so we can account in seg_table
79  * as well as the Ifile; and so we can add the blocks to their new
80  * segment.
81  *
82  * Change the given block's address to ndaddr, finding its previous
83  * location using ufs_bmaparray().
84  *
85  * Account for this change in the segment table.
86  */
87 static void
88 rfw_update_single(struct uvnode *vp, daddr_t lbn, ufs_daddr_t ndaddr, int size)
89 {
90 	SEGUSE *sup;
91 	struct ubuf *bp;
92 	struct indir a[NIADDR + 2], *ap;
93 	struct inode *ip;
94 	daddr_t daddr, ooff;
95 	int num, error;
96 	int i, bb, osize = 0, obb = 0;
97 	u_int32_t oldsn, sn;
98 
99 	ip = VTOI(vp);
100 	ip->i_flag |= IN_MODIFIED;
101 
102 	error = ufs_bmaparray(fs, vp, lbn, &daddr, a, &num);
103 	if (error)
104 		errx(1, "lfs_updatemeta: ufs_bmaparray returned %d"
105 		     " looking up lbn %" PRId64 "\n", error, lbn);
106 	if (daddr > 0)
107 		daddr = dbtofsb(fs, daddr);
108 
109 	bb = fragstofsb(fs, numfrags(fs, size));
110 	switch (num) {
111 	case 0:
112 		ooff = ip->i_ffs1_db[lbn];
113 		if (ooff <= 0)
114 			ip->i_ffs1_blocks += bb;
115 		else {
116 			/* possible fragment truncation or extension */
117 			obb = btofsb(fs, ip->i_lfs_fragsize[lbn]);
118 			ip->i_ffs1_blocks += (bb - obb);
119 		}
120 		ip->i_ffs1_db[lbn] = ndaddr;
121 		break;
122 	case 1:
123 		ooff = ip->i_ffs1_ib[a[0].in_off];
124 		if (ooff <= 0)
125 			ip->i_ffs1_blocks += bb;
126 		ip->i_ffs1_ib[a[0].in_off] = ndaddr;
127 		break;
128 	default:
129 		ap = &a[num - 1];
130 		if (bread(vp, ap->in_lbn, fs->lfs_bsize, NULL, &bp))
131 			errx(1, "lfs_updatemeta: bread bno %" PRId64,
132 			    ap->in_lbn);
133 
134 		ooff = ((ufs_daddr_t *) bp->b_data)[ap->in_off];
135 		if (ooff <= 0)
136 			ip->i_ffs1_blocks += bb;
137 		((ufs_daddr_t *) bp->b_data)[ap->in_off] = ndaddr;
138 		(void) VOP_BWRITE(bp);
139 	}
140 
141 	/*
142 	 * Update segment usage information, based on old size
143 	 * and location.
144 	 */
145 	if (daddr > 0) {
146 		oldsn = dtosn(fs, daddr);
147 		if (lbn >= 0 && lbn < NDADDR)
148 			osize = ip->i_lfs_fragsize[lbn];
149 		else
150 			osize = fs->lfs_bsize;
151 		LFS_SEGENTRY(sup, fs, oldsn, bp);
152 		seg_table[oldsn].su_nbytes -= osize;
153 		sup->su_nbytes -= osize;
154 		if (!(bp->b_flags & B_GATHERED))
155 			fs->lfs_flags |= LFS_IFDIRTY;
156 		LFS_WRITESEGENTRY(sup, fs, oldsn, bp);
157 		for (i = 0; i < btofsb(fs, osize); i++)
158 			clrbmap(daddr + i);
159 	}
160 
161 	/* If block is beyond EOF, update size */
162 	if (lbn >= 0 && ip->i_ffs1_size <= (lbn << fs->lfs_bshift)) {
163 		ip->i_ffs1_size = (lbn << fs->lfs_bshift) + 1;
164 	}
165 
166 	/* If block frag size is too large for old EOF, update size */
167 	if (lbn < NDADDR) {
168 		off_t minsize;
169 
170 		minsize = (lbn << fs->lfs_bshift);
171 		minsize += (size - fs->lfs_fsize) + 1;
172 		if (ip->i_ffs1_size < minsize)
173 			ip->i_ffs1_size = minsize;
174 	}
175 
176 	/* Count for the user */
177 	++nnewblocks;
178 
179 	/* Add block to its new segment */
180 	sn = dtosn(fs, ndaddr);
181 	LFS_SEGENTRY(sup, fs, sn, bp);
182 	seg_table[sn].su_nbytes += size;
183 	sup->su_nbytes += size;
184 	if (!(bp->b_flags & B_GATHERED))
185 		fs->lfs_flags |= LFS_IFDIRTY;
186 	LFS_WRITESEGENTRY(sup, fs, sn, bp);
187 	for (i = 0; i < btofsb(fs, size); i++)
188 #ifndef VERBOSE_BLOCKMAP
189 		setbmap(daddr + i);
190 #else
191 		setbmap(daddr + i, ip->i_number);
192 #endif
193 
194 	/* Check bfree accounting as well */
195 	if (daddr <= 0) {
196 		fs->lfs_bfree -= btofsb(fs, size);
197 	} else if (size != osize) {
198 		fs->lfs_bfree -= (bb - obb);
199 	}
200 
201 	/*
202 	 * Now that this block has a new address, and its old
203 	 * segment no longer owns it, we can forget about its
204 	 * old size.
205 	 */
206 	if (lbn >= 0 && lbn < NDADDR)
207 		ip->i_lfs_fragsize[lbn] = size;
208 }
209 
210 /*
211  * Remove the vnode from the cache, including any blocks it
212  * may hold.  Account the blocks.  Finally account the removal
213  * of the inode from its segment.
214  */
215 static void
216 remove_ino(struct uvnode *vp, ino_t ino)
217 {
218 	IFILE *ifp;
219 	SEGUSE *sup;
220 	CLEANERINFO *cip;
221 	struct ubuf *bp, *sbp, *cbp;
222 	struct inodesc idesc;
223 	ufs_daddr_t daddr;
224 	int obfree;
225 
226 	if (debug)
227 		pwarn("remove ino %d\n", (int)ino);
228 
229 	obfree = fs->lfs_bfree;
230 	LFS_IENTRY(ifp, fs, ino, bp);
231 	daddr = ifp->if_daddr;
232 	if (daddr > 0) {
233 		ifp->if_daddr = 0x0;
234 
235 		LFS_GET_HEADFREE(fs, cip, cbp, &(ifp->if_nextfree));
236 		VOP_BWRITE(bp);
237 		LFS_PUT_HEADFREE(fs, cip, cbp, ino);
238 		sbdirty();
239 
240 		if (vp == NULL)
241 			vp = lfs_raw_vget(fs, ino, fs->lfs_ivnode->v_fd, daddr);
242 
243 		LFS_SEGENTRY(sup, fs, dtosn(fs, daddr), sbp);
244 		sup->su_nbytes -= DINODE1_SIZE;
245 		VOP_BWRITE(sbp);
246 		seg_table[dtosn(fs, daddr)].su_nbytes -= DINODE1_SIZE;
247 	} else
248 		brelse(bp, 0);
249 
250 	/* Do on-disk accounting */
251 	if (vp) {
252 		idesc.id_number = ino;
253 		idesc.id_func = pass4check; /* Delete dinode and blocks */
254 		idesc.id_type = ADDR;
255 		idesc.id_lblkno = 0;
256 		clri(&idesc, "unknown", 2); /* XXX magic number 2 */
257 		/* vp has been destroyed */
258 	}
259 }
260 
261 /*
262  * Use FIP records to update blocks, if the generation number matches.
263  */
264 static void
265 pass6harvest(ufs_daddr_t daddr, FINFO *fip)
266 {
267 	struct uvnode *vp;
268 	int i;
269 	size_t size;
270 
271 	vp = vget(fs, fip->fi_ino);
272 	if (vp && vp != fs->lfs_ivnode &&
273 	    VTOI(vp)->i_ffs1_gen == fip->fi_version) {
274 		for (i = 0; i < fip->fi_nblocks; i++) {
275 			size = (i == fip->fi_nblocks - 1 ?
276 				fip->fi_lastlength : fs->lfs_bsize);
277 			if (debug)
278 				pwarn("ino %lld lbn %lld -> 0x%lx\n",
279 					(long long)fip->fi_ino,
280 					(long long)fip->fi_blocks[i],
281 					(long)daddr);
282 			rfw_update_single(vp, fip->fi_blocks[i], daddr, size);
283 			daddr += btofsb(fs, size);
284 		}
285 	}
286 }
287 
288 /*
289  * Check validity of blocks on roll-forward inodes.
290  */
291 int
292 pass6check(struct inodesc * idesc)
293 {
294 	int i, sn, anyout, anynew;
295 
296 	/* Brand new blocks are always OK */
297 	if (idesc->id_blkno == UNWRITTEN)
298 		return KEEPON;
299 
300 	/* Check that the blocks do not lie within clean segments. */
301 	anyout = anynew = 0;
302 	for (i = 0; i < fragstofsb(fs, idesc->id_numfrags); i++) {
303 		sn = dtosn(fs, idesc->id_blkno + i);
304 		if (sn < 0 || sn >= fs->lfs_nseg ||
305 		    (seg_table[sn].su_flags & SEGUSE_DIRTY) == 0) {
306 			anyout = 1;
307 			break;
308 		}
309 		if (seg_table[sn].su_flags & SEGUSE_ACTIVE) {
310 			if (sn != dtosn(fs, fs->lfs_offset) ||
311 			    idesc->id_blkno > fs->lfs_offset) {
312 				++anynew;
313 			}
314 		}
315 		if (!anynew) {
316 			/* Clear so pass1check won't be surprised */
317 			clrbmap(idesc->id_blkno + i);
318 			seg_table[sn].su_nbytes -= fsbtob(fs, 1);
319 		}
320 	}
321 	if (anyout) {
322 		blkerror(idesc->id_number, "BAD", idesc->id_blkno);
323 		if (badblk++ >= MAXBAD) {
324 			pwarn("EXCESSIVE BAD BLKS I=%llu",
325 			    (unsigned long long)idesc->id_number);
326 			if (preen)
327 				pwarn(" (SKIPPING)\n");
328 			else if (reply("CONTINUE") == 0)
329 				err(EEXIT, "%s", "");
330 			return (STOP);
331 		}
332 	}
333 
334 	return pass1check(idesc);
335 }
336 
337 static void
338 account_indir(struct uvnode *vp, struct ufs1_dinode *dp, daddr_t ilbn, daddr_t daddr, int lvl)
339 {
340 	struct ubuf *bp;
341 	int32_t *dap, *odap, *buf, *obuf;
342 	daddr_t lbn;
343 
344 	if (lvl == 0)
345 		lbn = -ilbn;
346 	else
347 		lbn = ilbn + 1;
348 	bread(fs->lfs_devvp, fsbtodb(fs, daddr), fs->lfs_bsize, NULL, &bp);
349 	buf = emalloc(fs->lfs_bsize);
350 	memcpy(buf, bp->b_data, fs->lfs_bsize);
351 	brelse(bp, 0);
352 
353 	obuf = emalloc(fs->lfs_bsize);
354 	if (vp) {
355 		bread(vp, ilbn, fs->lfs_bsize, NULL, &bp);
356 		memcpy(obuf, bp->b_data, fs->lfs_bsize);
357 		brelse(bp, 0);
358 	} else
359 		memset(obuf, 0, fs->lfs_bsize);
360 
361 	for (dap = buf, odap = obuf;
362 	     dap < (int32_t *)((char *)buf + fs->lfs_bsize);
363 	     ++dap, ++odap) {
364 		if (*dap > 0 && *dap != *odap) {
365 			rfw_update_single(vp, lbn, *dap, dblksize(fs, dp, lbn));
366 			if (lvl > 0)
367 				account_indir(vp, dp, lbn, *dap, lvl - 1);
368 		}
369 		if (lvl == 0)
370 			++lbn;
371 		else if (lvl == 1)
372 			lbn -= NINDIR(fs);
373 		else if (lvl == 2)
374 			lbn -= NINDIR(fs) * NINDIR(fs);
375 	}
376 
377 	free(obuf);
378 	free(buf);
379 }
380 
381 /*
382  * Account block changes between new found inode and existing inode.
383  */
384 static void
385 account_block_changes(struct ufs1_dinode *dp)
386 {
387 	int i;
388 	daddr_t lbn, off, odaddr;
389 	struct uvnode *vp;
390 	struct inode *ip;
391 
392 	vp = vget(fs, dp->di_inumber);
393 	ip = (vp ? VTOI(vp) : NULL);
394 
395 	/* Check direct block holdings between existing and new */
396 	for (i = 0; i < NDADDR; i++) {
397 		odaddr = (ip ? ip->i_ffs1_db[i] : 0x0);
398 		if (dp->di_db[i] > 0 && dp->di_db[i] != odaddr)
399 			rfw_update_single(vp, i, dp->di_db[i],
400 					  dblksize(fs, dp, i));
401 	}
402 
403 	/* Check indirect block holdings between existing and new */
404 	off = 0;
405 	for (i = 0; i < NIADDR; i++) {
406 		odaddr = (ip ? ip->i_ffs1_ib[i] : 0x0);
407 		if (dp->di_ib[i] > 0 && dp->di_ib[i] != odaddr) {
408 			lbn = -(NDADDR + off + i);
409 			rfw_update_single(vp, i, dp->di_ib[i], fs->lfs_bsize);
410 			account_indir(vp, dp, lbn, dp->di_ib[i], i);
411 		}
412 		if (off == 0)
413 			off = NINDIR(fs);
414 		else
415 			off *= NINDIR(fs);
416 	}
417 }
418 
419 /*
420  * Give a previously allocated inode a new address; do segment
421  * accounting if necessary.
422  *
423  * Caller has ensured that this inode is not on the free list, so no
424  * free list accounting is done.
425  */
426 static void
427 readdress_inode(struct ufs1_dinode *dp, ufs_daddr_t daddr)
428 {
429 	IFILE *ifp;
430 	SEGUSE *sup;
431 	struct ubuf *bp;
432 	int sn;
433 	ufs_daddr_t odaddr;
434 	ino_t thisino = dp->di_inumber;
435 	struct uvnode *vp;
436 
437 	/* Recursively check all block holdings, account changes */
438 	account_block_changes(dp);
439 
440 	/* Move ifile pointer to this location */
441 	LFS_IENTRY(ifp, fs, thisino, bp);
442 	odaddr = ifp->if_daddr;
443 	assert(odaddr != 0);
444 	ifp->if_daddr = daddr;
445 	VOP_BWRITE(bp);
446 
447 	if (debug)
448 		pwarn("readdress ino %d from 0x%x to 0x%x mode %o nlink %d\n",
449 			(int)dp->di_inumber,
450 			(unsigned)odaddr,
451 			(unsigned)daddr,
452 			(int)dp->di_mode, (int)dp->di_nlink);
453 
454 	/* Copy over preexisting in-core inode, if any */
455 	vp = vget(fs, thisino);
456 	memcpy(VTOI(vp)->i_din.ffs1_din, dp, sizeof(*dp));
457 
458 	/* Finally account the inode itself */
459 	sn = dtosn(fs, odaddr);
460 	LFS_SEGENTRY(sup, fs, sn, bp);
461 	sup->su_nbytes -= DINODE1_SIZE;
462 	VOP_BWRITE(bp);
463 	seg_table[sn].su_nbytes -= DINODE1_SIZE;
464 
465 	sn = dtosn(fs, daddr);
466 	LFS_SEGENTRY(sup, fs, sn, bp);
467 	sup->su_nbytes += DINODE1_SIZE;
468 	VOP_BWRITE(bp);
469 	seg_table[sn].su_nbytes += DINODE1_SIZE;
470 }
471 
472 /*
473  * Allocate the given inode from the free list.
474  */
475 static void
476 alloc_inode(ino_t thisino, ufs_daddr_t daddr)
477 {
478 	ino_t ino, nextfree, oldhead;
479 	IFILE *ifp;
480 	SEGUSE *sup;
481 	struct ubuf *bp, *cbp;
482 	CLEANERINFO *cip;
483 
484 	if (debug)
485 		pwarn("allocating ino %d at 0x%x\n", (int)thisino,
486 			(unsigned)daddr);
487 	while (thisino >= maxino) {
488 		extend_ifile(fs);
489 	}
490 
491 	LFS_IENTRY(ifp, fs, thisino, bp);
492 	if (ifp->if_daddr != 0) {
493 		pwarn("allocated inode %lld already allocated\n",
494 			(long long)thisino);
495 	}
496 	nextfree = ifp->if_nextfree;
497 	ifp->if_nextfree = 0;
498 	ifp->if_daddr = daddr;
499 	VOP_BWRITE(bp);
500 
501 	LFS_GET_HEADFREE(fs, cip, cbp, &oldhead);
502 	if (oldhead == thisino) {
503 		LFS_PUT_HEADFREE(fs, cip, cbp, nextfree);
504 		sbdirty();
505 		if (nextfree == 0) {
506 			extend_ifile(fs);
507 		}
508 	} else {
509 		/* Search the free list for this inode */
510 		ino = oldhead;
511 		while (ino) {
512 			LFS_IENTRY(ifp, fs, ino, bp);
513 			assert(ifp->if_nextfree != ino);
514 			if (ifp->if_nextfree == thisino) {
515 				ifp->if_nextfree = nextfree;
516 				VOP_BWRITE(bp);
517 				if (nextfree == 0)
518 					LFS_PUT_TAILFREE(fs, cip, cbp, ino);
519 				break;
520 			} else
521 				ino = ifp->if_nextfree;
522 			brelse(bp, 0);
523 		}
524 	}
525 
526 	/* Account for new location */
527 	LFS_SEGENTRY(sup, fs, dtosn(fs, daddr), bp);
528 	sup->su_nbytes += DINODE1_SIZE;
529 	VOP_BWRITE(bp);
530 	seg_table[dtosn(fs, daddr)].su_nbytes += DINODE1_SIZE;
531 }
532 
533 /*
534  * Roll forward from the last verified checkpoint.
535  *
536  * Basic strategy:
537  *
538  * Run through the summaries finding the last valid partial segment.
539  * Note segment numbers as we go.  For each inode that we find, compare
540  * its generation number; if newer than old inode's (or if old inode is
541  * USTATE), change to that inode.  Recursively look at inode blocks that
542  * do not have their old disk addresses.  These addresses must lie in
543  * segments we have seen already in our roll forward.
544  *
545  * A second pass through the past-checkpoint area verifies the validity
546  * of these new blocks, as well as updating other blocks that do not
547  * have corresponding new inodes (but their generation number must match
548  * the old generation number).
549  */
550 void
551 pass6(void)
552 {
553 	ufs_daddr_t daddr, ibdaddr, odaddr, lastgood, nextseg, *idaddrp;
554 	struct uvnode *vp, *devvp;
555 	CLEANERINFO *cip;
556 	SEGUSE *sup;
557 	SEGSUM *sp;
558 	struct ubuf *bp, *ibp, *sbp, *cbp;
559 	struct ufs1_dinode *dp;
560 	struct inodesc idesc;
561 	int i, j, bc, hassuper;
562 	int nnewfiles, ndelfiles, nmvfiles;
563 	int sn, curseg;
564 	char *ibbuf;
565 	long lastserial;
566 
567 	devvp = fs->lfs_devvp;
568 
569 	/* If we can't roll forward because of created files, don't try */
570 	if (no_roll_forward) {
571 		if (debug)
572 			pwarn("not rolling forward due to possible allocation conflict\n");
573 		return;
574 	}
575 
576 	/* Find last valid partial segment */
577 	lastgood = try_verify(fs, devvp, 0, debug);
578 	if (lastgood == fs->lfs_offset) {
579 		if (debug)
580 			pwarn("not rolling forward, nothing to recover\n");
581 		return;
582 	}
583 
584 	if (debug)
585 		pwarn("could roll forward from 0x%" PRIx32 " to 0x%" PRIx32 "\n",
586 			fs->lfs_offset, lastgood);
587 
588 	if (!preen && reply("ROLL FORWARD") == 0)
589 		return;
590 	/*
591 	 * Pass 1: find inode blocks.  We ignore the Ifile inode but accept
592 	 * changes to any other inode.
593 	 */
594 
595 	ibbuf = emalloc(fs->lfs_ibsize);
596 	nnewfiles = ndelfiles = nmvfiles = nnewblocks = 0;
597 	daddr = fs->lfs_offset;
598 	nextseg = fs->lfs_nextseg;
599 	hassuper = 0;
600 	lastserial = 0;
601 	while (daddr != lastgood) {
602 		seg_table[dtosn(fs, daddr)].su_flags |= SEGUSE_DIRTY | SEGUSE_ACTIVE;
603 		LFS_SEGENTRY(sup, fs, dtosn(fs, daddr), sbp);
604 		sup->su_flags |= SEGUSE_DIRTY;
605 		VOP_BWRITE(sbp);
606 
607 		/* Could be a superblock */
608 		if (sntod(fs, dtosn(fs, daddr)) == daddr) {
609 			if (daddr == fs->lfs_start) {
610 				++hassuper;
611 				daddr += btofsb(fs, LFS_LABELPAD);
612 			}
613 			for (i = 0; i < LFS_MAXNUMSB; i++) {
614 				if (daddr == fs->lfs_sboffs[i]) {
615 					++hassuper;
616 					daddr += btofsb(fs, LFS_SBPAD);
617 				}
618 				if (daddr < fs->lfs_sboffs[i])
619 					break;
620 			}
621 		}
622 
623 		/* Read in summary block */
624 		bread(devvp, fsbtodb(fs, daddr), fs->lfs_sumsize, NULL, &bp);
625 		sp = (SEGSUM *)bp->b_data;
626 		if (debug)
627 			pwarn("sum at 0x%x: ninos=%d nfinfo=%d\n",
628 				(unsigned)daddr, (int)sp->ss_ninos,
629 				(int)sp->ss_nfinfo);
630 
631 		/* We have verified that this is a good summary. */
632 		LFS_SEGENTRY(sup, fs, dtosn(fs, daddr), sbp);
633 		++sup->su_nsums;
634 		VOP_BWRITE(sbp);
635 		fs->lfs_bfree -= btofsb(fs, fs->lfs_sumsize);
636 		fs->lfs_dmeta += btofsb(fs, fs->lfs_sumsize);
637 		sbdirty();
638 		nextseg = sp->ss_next;
639 		if (sntod(fs, dtosn(fs, daddr)) == daddr +
640 		    hassuper * btofsb(fs, LFS_SBPAD) &&
641 		    dtosn(fs, daddr) != dtosn(fs, fs->lfs_offset)) {
642 			--fs->lfs_nclean;
643 			sbdirty();
644 		}
645 
646 		/* Find inodes, look at generation number. */
647 		if (sp->ss_ninos) {
648 			LFS_SEGENTRY(sup, fs, dtosn(fs, daddr), sbp);
649 			sup->su_ninos += howmany(sp->ss_ninos, INOPB(fs));
650 			VOP_BWRITE(sbp);
651 			fs->lfs_dmeta += btofsb(fs, howmany(sp->ss_ninos,
652 							    INOPB(fs)) *
653 						fs->lfs_ibsize);
654 		}
655 		idaddrp = ((ufs_daddr_t *)((char *)bp->b_data + fs->lfs_sumsize));
656 		for (i = 0; i < howmany(sp->ss_ninos, INOPB(fs)); i++) {
657 			ino_t *inums;
658 
659 			inums = ecalloc(INOPB(fs) + 1, sizeof(*inums));
660 			ibdaddr = *--idaddrp;
661 			fs->lfs_bfree -= btofsb(fs, fs->lfs_ibsize);
662 			sbdirty();
663 			bread(devvp, fsbtodb(fs, ibdaddr), fs->lfs_ibsize,
664 			      NOCRED, &ibp);
665 			memcpy(ibbuf, ibp->b_data, fs->lfs_ibsize);
666 			brelse(ibp, 0);
667 
668 			j = 0;
669 			for (dp = (struct ufs1_dinode *)ibbuf;
670 			     dp < (struct ufs1_dinode *)ibbuf + INOPB(fs);
671 			     ++dp) {
672 				if (dp->di_u.inumber == 0 ||
673 				    dp->di_u.inumber == fs->lfs_ifile)
674 					continue;
675 				/* Basic sanity checks */
676 				if (dp->di_nlink < 0
677 #if 0
678 				    || dp->di_u.inumber < 0
679 				    || dp->di_size < 0
680 #endif
681 				) {
682 					pwarn("BAD INODE AT 0x%" PRIx32 "\n",
683 						ibdaddr);
684 					brelse(bp, 0);
685 					free(inums);
686 					goto out;
687 				}
688 
689 				vp = vget(fs, dp->di_u.inumber);
690 
691 				/*
692 				 * Four cases:
693 				 * (1) Invalid inode (nlink == 0).
694 				 *     If currently allocated, remove.
695 				 */
696 				if (dp->di_nlink == 0) {
697 					remove_ino(vp, dp->di_u.inumber);
698 					++ndelfiles;
699 					continue;
700 				}
701 				/*
702 				 * (2) New valid inode, previously free.
703 				 *     Nothing to do except account
704 				 *     the inode itself, done after the
705 				 *     loop.
706 				 */
707 				if (vp == NULL) {
708 					if (!(sp->ss_flags & SS_DIROP))
709 						pfatal("NEW FILE IN NON-DIROP PARTIAL SEGMENT");
710 					else {
711 						inums[j++] = dp->di_u.inumber;
712 						nnewfiles++;
713 					}
714 					continue;
715 				}
716 				/*
717 				 * (3) Valid new version of previously
718 				 *     allocated inode.  Delete old file
719 				 *     and proceed as in (2).
720 				 */
721 				if (vp && VTOI(vp)->i_ffs1_gen < dp->di_gen) {
722 					remove_ino(vp, dp->di_u.inumber);
723 					if (!(sp->ss_flags & SS_DIROP))
724 						pfatal("NEW FILE VERSION IN NON-DIROP PARTIAL SEGMENT");
725 					else {
726 						inums[j++] = dp->di_u.inumber;
727 						ndelfiles++;
728 						nnewfiles++;
729 					}
730 					continue;
731 				}
732 				/*
733 				 * (4) Same version of previously
734 				 *     allocated inode.  Move inode to
735 				 *     this location, account inode change
736 				 *     only.  We'll pick up any new
737 				 *     blocks when we do the block pass.
738 				 */
739 				if (vp && VTOI(vp)->i_ffs1_gen == dp->di_gen) {
740 					nmvfiles++;
741 					readdress_inode(dp, ibdaddr);
742 
743 					/* Update with new info */
744 					VTOD(vp)->di_mode = dp->di_mode;
745 					VTOD(vp)->di_nlink = dp->di_nlink;
746 					/* XXX size is important */
747 					VTOD(vp)->di_size = dp->di_size;
748 					VTOD(vp)->di_atime = dp->di_atime;
749 					VTOD(vp)->di_atimensec = dp->di_atimensec;
750 					VTOD(vp)->di_mtime = dp->di_mtime;
751 					VTOD(vp)->di_mtimensec = dp->di_mtimensec;
752 					VTOD(vp)->di_ctime = dp->di_ctime;
753 					VTOD(vp)->di_ctimensec = dp->di_ctimensec;
754 					VTOD(vp)->di_flags = dp->di_flags;
755 					VTOD(vp)->di_uid = dp->di_uid;
756 					VTOD(vp)->di_gid = dp->di_gid;
757 					inodirty(VTOI(vp));
758 				}
759 			}
760 			for (j = 0; inums[j]; j++) {
761 				alloc_inode(inums[j], ibdaddr);
762 				vp = lfs_raw_vget(fs, inums[j],
763 					      devvp->v_fd, ibdaddr);
764 				/* We'll get the blocks later */
765 				if (debug)
766 					pwarn("alloc ino %d nlink %d\n",
767 						(int)inums[j], VTOD(vp)->di_nlink);
768 				memset(VTOD(vp)->di_db, 0, (NDADDR + NIADDR) *
769 				       sizeof(ufs_daddr_t));
770 				VTOD(vp)->di_blocks = 0;
771 
772 				vp->v_uflag |= VU_DIROP;
773 				inodirty(VTOI(vp));
774 			}
775 			free(inums);
776 		}
777 
778 		bc = check_summary(fs, sp, daddr, debug, devvp, NULL);
779 		if (bc == 0) {
780 			pwarn("unexpected bad seg ptr at 0x%x with serial=%d\n",
781 				(int)daddr, (int)sp->ss_serial);
782 			brelse(bp, 0);
783 			break;
784 		} else {
785 			if (debug)
786 				pwarn("good seg ptr at 0x%x with serial=%d\n",
787 					(int)daddr, (int)sp->ss_serial);
788 			lastserial = sp->ss_serial;
789 		}
790 		odaddr = daddr;
791 		daddr += btofsb(fs, fs->lfs_sumsize + bc);
792 		if (dtosn(fs, odaddr) != dtosn(fs, daddr) ||
793 		    dtosn(fs, daddr) != dtosn(fs, daddr +
794 			btofsb(fs, fs->lfs_sumsize + fs->lfs_bsize) - 1)) {
795 			daddr = ((SEGSUM *)bp->b_data)->ss_next;
796 		}
797 		brelse(bp, 0);
798 	}
799 
800     out:
801 	free(ibbuf);
802 
803 	/* Set serial here, just to be sure (XXX should be right already) */
804 	fs->lfs_serial = lastserial + 1;
805 
806 	/*
807 	 * Check our new vnodes.  Any blocks must lie in segments that
808 	 * we've seen before (SEGUSE_DIRTY or SEGUSE_RFW); and the rest
809 	 * of the pass 1 checks as well.
810 	 */
811 	memset(&idesc, 0, sizeof(struct inodesc));
812 	idesc.id_type = ADDR;
813 	idesc.id_func = pass6check;
814 	idesc.id_lblkno = 0;
815 	LIST_FOREACH(vp, &vnodelist, v_mntvnodes) {
816 		if ((vp->v_uflag & VU_DIROP) == 0)
817 			--n_files; /* Don't double count */
818 		checkinode(VTOI(vp)->i_number, &idesc);
819 	}
820 
821 	/*
822 	 * Second pass.  Run through FINFO entries looking for blocks
823 	 * with the same generation number as files we've seen before.
824 	 * If they have it, pretend like we just wrote them.  We don't
825 	 * do the pretend-write, though, if we've already seen them
826 	 * (the accounting would have been done for us already).
827 	 */
828 	daddr = fs->lfs_offset;
829 	while (daddr != lastgood) {
830 		if (!(seg_table[dtosn(fs, daddr)].su_flags & SEGUSE_DIRTY)) {
831 			seg_table[dtosn(fs, daddr)].su_flags |= SEGUSE_DIRTY;
832 			LFS_SEGENTRY(sup, fs, dtosn(fs, daddr), sbp);
833 			sup->su_flags |= SEGUSE_DIRTY;
834 			VOP_BWRITE(sbp);
835 		}
836 
837 		/* Could be a superblock */
838 		if (sntod(fs, dtosn(fs, daddr)) == daddr) {
839 			if (daddr == fs->lfs_start)
840 				daddr += btofsb(fs, LFS_LABELPAD);
841 			for (i = 0; i < LFS_MAXNUMSB; i++) {
842 				if (daddr == fs->lfs_sboffs[i]) {
843 					daddr += btofsb(fs, LFS_SBPAD);
844 				}
845 				if (daddr < fs->lfs_sboffs[i])
846 					break;
847 			}
848 		}
849 
850 		/* Read in summary block */
851 		bread(devvp, fsbtodb(fs, daddr), fs->lfs_sumsize, NULL, &bp);
852 		sp = (SEGSUM *)bp->b_data;
853 		bc = check_summary(fs, sp, daddr, debug, devvp, pass6harvest);
854 		if (bc == 0) {
855 			pwarn("unexpected bad seg ptr [2] at 0x%x with serial=%d\n",
856 				(int)daddr, (int)sp->ss_serial);
857 			brelse(bp, 0);
858 			break;
859 		}
860 		odaddr = daddr;
861 		daddr += btofsb(fs, fs->lfs_sumsize + bc);
862 		fs->lfs_avail -= btofsb(fs, fs->lfs_sumsize + bc);
863 		if (dtosn(fs, odaddr) != dtosn(fs, daddr) ||
864 		    dtosn(fs, daddr) != dtosn(fs, daddr +
865 			btofsb(fs, fs->lfs_sumsize + fs->lfs_bsize) - 1)) {
866 			fs->lfs_avail -= sntod(fs, dtosn(fs, daddr) + 1) - daddr;
867 			daddr = ((SEGSUM *)bp->b_data)->ss_next;
868 		}
869 		LFS_CLEANERINFO(cip, fs, cbp);
870 		LFS_SYNC_CLEANERINFO(cip, fs, cbp, 0);
871 		bp->b_flags |= B_AGE;
872 		brelse(bp, 0);
873 	}
874 
875 	/* Final address could also be a superblock */
876 	if (sntod(fs, dtosn(fs, lastgood)) == lastgood) {
877 		if (lastgood == fs->lfs_start)
878 			lastgood += btofsb(fs, LFS_LABELPAD);
879 		for (i = 0; i < LFS_MAXNUMSB; i++) {
880 			if (lastgood == fs->lfs_sboffs[i])
881 				lastgood += btofsb(fs, LFS_SBPAD);
882 			if (lastgood < fs->lfs_sboffs[i])
883 				break;
884 		}
885 	}
886 
887 	/* Update offset to point at correct location */
888 	fs->lfs_offset = lastgood;
889 	fs->lfs_curseg = sntod(fs, dtosn(fs, lastgood));
890 	for (sn = curseg = dtosn(fs, fs->lfs_curseg);;) {
891 		sn = (sn + 1) % fs->lfs_nseg;
892 		if (sn == curseg)
893 			errx(1, "no clean segments");
894 		LFS_SEGENTRY(sup, fs, sn, bp);
895 		if ((sup->su_flags & SEGUSE_DIRTY) == 0) {
896 			sup->su_flags |= SEGUSE_DIRTY | SEGUSE_ACTIVE;
897 			VOP_BWRITE(bp);
898 			break;
899 		}
900 		brelse(bp, 0);
901 	}
902 	fs->lfs_nextseg = sntod(fs, sn);
903 
904 	if (preen) {
905 		if (ndelfiles)
906 			pwarn("roll forward deleted %d file%s\n", ndelfiles,
907 				(ndelfiles > 1 ? "s" : ""));
908 		if (nnewfiles)
909 			pwarn("roll forward added %d file%s\n", nnewfiles,
910 				(nnewfiles > 1 ? "s" : ""));
911 		if (nmvfiles)
912 			pwarn("roll forward relocated %d inode%s\n", nmvfiles,
913 				(nmvfiles > 1 ? "s" : ""));
914 		if (nnewblocks)
915 			pwarn("roll forward verified %d data block%s\n", nnewblocks,
916 				(nnewblocks > 1 ? "s" : ""));
917 		if (ndelfiles == 0 && nnewfiles == 0 && nmvfiles == 0 &&
918 		    nnewblocks == 0)
919 			pwarn("roll forward produced nothing new\n");
920 	}
921 
922 	if (!preen) {
923 		/* Run pass 5 again (it's quick anyway). */
924 		pwarn("** Phase 6b - Recheck Segment Block Accounting\n");
925 		pass5();
926 	}
927 
928 	/* Likewise for pass 0 */
929 	if (!preen)
930 		pwarn("** Phase 6c - Recheck Inode Free List\n");
931 	pass0();
932 }
933