xref: /netbsd-src/sbin/fsck_lfs/lfs.c (revision e5548b402ae4c44fb816de42c7bba9581ce23ef5)
1 /* $NetBSD: lfs.c,v 1.19 2005/10/13 21:14:45 jmc Exp $ */
2 /*-
3  * Copyright (c) 2003 The NetBSD Foundation, Inc.
4  * All rights reserved.
5  *
6  * This code is derived from software contributed to The NetBSD Foundation
7  * by Konrad E. Schroder <perseant@hhhh.org>.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  * 3. All advertising materials mentioning features or use of this software
18  *    must display the following acknowledgement:
19  *	This product includes software developed by the NetBSD
20  *	Foundation, Inc. and its contributors.
21  * 4. Neither the name of The NetBSD Foundation nor the names of its
22  *    contributors may be used to endorse or promote products derived
23  *    from this software without specific prior written permission.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
26  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
27  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
28  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
29  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
30  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
31  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
32  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
33  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
34  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
35  * POSSIBILITY OF SUCH DAMAGE.
36  */
37 /*
38  * Copyright (c) 1989, 1991, 1993
39  *	The Regents of the University of California.  All rights reserved.
40  * (c) UNIX System Laboratories, Inc.
41  * All or some portions of this file are derived from material licensed
42  * to the University of California by American Telephone and Telegraph
43  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
44  * the permission of UNIX System Laboratories, Inc.
45  *
46  * Redistribution and use in source and binary forms, with or without
47  * modification, are permitted provided that the following conditions
48  * are met:
49  * 1. Redistributions of source code must retain the above copyright
50  *    notice, this list of conditions and the following disclaimer.
51  * 2. Redistributions in binary form must reproduce the above copyright
52  *    notice, this list of conditions and the following disclaimer in the
53  *    documentation and/or other materials provided with the distribution.
54  * 3. Neither the name of the University nor the names of its contributors
55  *    may be used to endorse or promote products derived from this software
56  *    without specific prior written permission.
57  *
58  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
59  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
60  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
61  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
62  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
63  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
64  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
65  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
66  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
67  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
68  * SUCH DAMAGE.
69  *
70  *	@(#)ufs_bmap.c	8.8 (Berkeley) 8/11/95
71  */
72 
73 
74 #include <sys/types.h>
75 #include <sys/param.h>
76 #include <sys/time.h>
77 #include <sys/buf.h>
78 #include <sys/mount.h>
79 
80 #include <ufs/ufs/inode.h>
81 #include <ufs/ufs/ufsmount.h>
82 #define vnode uvnode
83 #include <ufs/lfs/lfs.h>
84 #undef vnode
85 
86 #include <assert.h>
87 #include <err.h>
88 #include <errno.h>
89 #include <stdarg.h>
90 #include <stdio.h>
91 #include <stdlib.h>
92 #include <string.h>
93 #include <unistd.h>
94 
95 #include "bufcache.h"
96 #include "vnode.h"
97 #include "lfs_user.h"
98 #include "segwrite.h"
99 
100 #define panic call_panic
101 
102 extern u_int32_t cksum(void *, size_t);
103 extern u_int32_t lfs_sb_cksum(struct dlfs *);
104 extern void pwarn(const char *, ...);
105 
106 extern struct uvnodelst vnodelist;
107 extern struct uvnodelst getvnodelist[VNODE_HASH_MAX];
108 extern int nvnodes;
109 
110 int fsdirty = 0;
111 void (*panic_func)(int, const char *, va_list) = my_vpanic;
112 
113 /*
114  * LFS buffer and uvnode operations
115  */
116 
117 int
118 lfs_vop_strategy(struct ubuf * bp)
119 {
120 	int count;
121 
122 	if (bp->b_flags & B_READ) {
123 		count = pread(bp->b_vp->v_fd, bp->b_data, bp->b_bcount,
124 		    dbtob(bp->b_blkno));
125 		if (count == bp->b_bcount)
126 			bp->b_flags |= B_DONE;
127 	} else {
128 		count = pwrite(bp->b_vp->v_fd, bp->b_data, bp->b_bcount,
129 		    dbtob(bp->b_blkno));
130 		if (count == 0) {
131 			perror("pwrite");
132 			return -1;
133 		}
134 		bp->b_flags &= ~B_DELWRI;
135 		reassignbuf(bp, bp->b_vp);
136 	}
137 	return 0;
138 }
139 
140 int
141 lfs_vop_bwrite(struct ubuf * bp)
142 {
143 	struct lfs *fs;
144 
145 	fs = bp->b_vp->v_fs;
146 	if (!(bp->b_flags & B_DELWRI)) {
147 		fs->lfs_avail -= btofsb(fs, bp->b_bcount);
148 	}
149 	bp->b_flags |= B_DELWRI | B_LOCKED;
150 	reassignbuf(bp, bp->b_vp);
151 	brelse(bp);
152 	return 0;
153 }
154 
155 /*
156  * ufs_bmaparray does the bmap conversion, and if requested returns the
157  * array of logical blocks which must be traversed to get to a block.
158  * Each entry contains the offset into that block that gets you to the
159  * next block and the disk address of the block (if it is assigned).
160  */
161 int
162 ufs_bmaparray(struct lfs * fs, struct uvnode * vp, daddr_t bn, daddr_t * bnp, struct indir * ap, int *nump)
163 {
164 	struct inode *ip;
165 	struct ubuf *bp;
166 	struct indir a[NIADDR + 1], *xap;
167 	daddr_t daddr;
168 	daddr_t metalbn;
169 	int error, num;
170 
171 	ip = VTOI(vp);
172 
173 	if (bn >= 0 && bn < NDADDR) {
174 		if (nump != NULL)
175 			*nump = 0;
176 		*bnp = fsbtodb(fs, ip->i_ffs1_db[bn]);
177 		if (*bnp == 0)
178 			*bnp = -1;
179 		return (0);
180 	}
181 	xap = ap == NULL ? a : ap;
182 	if (!nump)
183 		nump = &num;
184 	if ((error = ufs_getlbns(fs, vp, bn, xap, nump)) != 0)
185 		return (error);
186 
187 	num = *nump;
188 
189 	/* Get disk address out of indirect block array */
190 	daddr = ip->i_ffs1_ib[xap->in_off];
191 
192 	for (bp = NULL, ++xap; --num; ++xap) {
193 		/* Exit the loop if there is no disk address assigned yet and
194 		 * the indirect block isn't in the cache, or if we were
195 		 * looking for an indirect block and we've found it. */
196 
197 		metalbn = xap->in_lbn;
198 		if ((daddr == 0 && !incore(vp, metalbn)) || metalbn == bn)
199 			break;
200 		/*
201 		 * If we get here, we've either got the block in the cache
202 		 * or we have a disk address for it, go fetch it.
203 		 */
204 		if (bp)
205 			brelse(bp);
206 
207 		xap->in_exists = 1;
208 		bp = getblk(vp, metalbn, fs->lfs_bsize);
209 
210 		if (!(bp->b_flags & (B_DONE | B_DELWRI))) {
211 			bp->b_blkno = fsbtodb(fs, daddr);
212 			bp->b_flags |= B_READ;
213 			VOP_STRATEGY(bp);
214 		}
215 		daddr = ((ufs_daddr_t *) bp->b_data)[xap->in_off];
216 	}
217 	if (bp)
218 		brelse(bp);
219 
220 	daddr = fsbtodb(fs, (ufs_daddr_t) daddr);
221 	*bnp = daddr == 0 ? -1 : daddr;
222 	return (0);
223 }
224 
225 /*
226  * Create an array of logical block number/offset pairs which represent the
227  * path of indirect blocks required to access a data block.  The first "pair"
228  * contains the logical block number of the appropriate single, double or
229  * triple indirect block and the offset into the inode indirect block array.
230  * Note, the logical block number of the inode single/double/triple indirect
231  * block appears twice in the array, once with the offset into the i_ffs1_ib and
232  * once with the offset into the page itself.
233  */
234 int
235 ufs_getlbns(struct lfs * fs, struct uvnode * vp, daddr_t bn, struct indir * ap, int *nump)
236 {
237 	daddr_t metalbn, realbn;
238 	int64_t blockcnt;
239 	int lbc;
240 	int i, numlevels, off;
241 	int lognindir, indir;
242 
243 	metalbn = 0;    /* XXXGCC -Wuninitialized [sh3] */
244 
245 	if (nump)
246 		*nump = 0;
247 	numlevels = 0;
248 	realbn = bn;
249 	if (bn < 0)
250 		bn = -bn;
251 
252 	lognindir = -1;
253 	for (indir = fs->lfs_nindir; indir; indir >>= 1)
254 		++lognindir;
255 
256 	/* Determine the number of levels of indirection.  After this loop is
257 	 * done, blockcnt indicates the number of data blocks possible at the
258 	 * given level of indirection, and NIADDR - i is the number of levels
259 	 * of indirection needed to locate the requested block. */
260 
261 	bn -= NDADDR;
262 	for (lbc = 0, i = NIADDR;; i--, bn -= blockcnt) {
263 		if (i == 0)
264 			return (EFBIG);
265 
266 		lbc += lognindir;
267 		blockcnt = (int64_t) 1 << lbc;
268 
269 		if (bn < blockcnt)
270 			break;
271 	}
272 
273 	/* Calculate the address of the first meta-block. */
274 	metalbn = -((realbn >= 0 ? realbn : -realbn) - bn + NIADDR - i);
275 
276 	/* At each iteration, off is the offset into the bap array which is an
277 	 * array of disk addresses at the current level of indirection. The
278 	 * logical block number and the offset in that block are stored into
279 	 * the argument array. */
280 	ap->in_lbn = metalbn;
281 	ap->in_off = off = NIADDR - i;
282 	ap->in_exists = 0;
283 	ap++;
284 	for (++numlevels; i <= NIADDR; i++) {
285 		/* If searching for a meta-data block, quit when found. */
286 		if (metalbn == realbn)
287 			break;
288 
289 		lbc -= lognindir;
290 		blockcnt = (int64_t) 1 << lbc;
291 		off = (bn >> lbc) & (fs->lfs_nindir - 1);
292 
293 		++numlevels;
294 		ap->in_lbn = metalbn;
295 		ap->in_off = off;
296 		ap->in_exists = 0;
297 		++ap;
298 
299 		metalbn -= -1 + (off << lbc);
300 	}
301 	if (nump)
302 		*nump = numlevels;
303 	return (0);
304 }
305 
306 int
307 lfs_vop_bmap(struct uvnode * vp, daddr_t lbn, daddr_t * daddrp)
308 {
309 	return ufs_bmaparray(vp->v_fs, vp, lbn, daddrp, NULL, NULL);
310 }
311 
312 /* Search a block for a specific dinode. */
313 struct ufs1_dinode *
314 lfs_ifind(struct lfs * fs, ino_t ino, struct ubuf * bp)
315 {
316 	struct ufs1_dinode *dip = (struct ufs1_dinode *) bp->b_data;
317 	struct ufs1_dinode *ldip, *fin;
318 
319 	fin = dip + INOPB(fs);
320 
321 	/*
322 	 * Read the inode block backwards, since later versions of the
323 	 * inode will supercede earlier ones.  Though it is unlikely, it is
324 	 * possible that the same inode will appear in the same inode block.
325 	 */
326 	for (ldip = fin - 1; ldip >= dip; --ldip)
327 		if (ldip->di_inumber == ino)
328 			return (ldip);
329 	return NULL;
330 }
331 
332 /*
333  * lfs_raw_vget makes us a new vnode from the inode at the given disk address.
334  * XXX it currently loses atime information.
335  */
336 struct uvnode *
337 lfs_raw_vget(struct lfs * fs, ino_t ino, int fd, ufs_daddr_t daddr)
338 {
339 	struct uvnode *vp;
340 	struct inode *ip;
341 	struct ufs1_dinode *dip;
342 	struct ubuf *bp;
343 	int i, hash;
344 
345 	vp = (struct uvnode *) malloc(sizeof(*vp));
346 	memset(vp, 0, sizeof(*vp));
347 	vp->v_fd = fd;
348 	vp->v_fs = fs;
349 	vp->v_usecount = 0;
350 	vp->v_strategy_op = lfs_vop_strategy;
351 	vp->v_bwrite_op = lfs_vop_bwrite;
352 	vp->v_bmap_op = lfs_vop_bmap;
353 	LIST_INIT(&vp->v_cleanblkhd);
354 	LIST_INIT(&vp->v_dirtyblkhd);
355 
356 	ip = (struct inode *) malloc(sizeof(*ip));
357 	memset(ip, 0, sizeof(*ip));
358 
359 	ip->i_din.ffs1_din = (struct ufs1_dinode *)
360 	    malloc(sizeof(struct ufs1_dinode));
361 	memset(ip->i_din.ffs1_din, 0, sizeof (struct ufs1_dinode));
362 
363 	/* Initialize the inode -- from lfs_vcreate. */
364 	ip->inode_ext.lfs = malloc(sizeof(struct lfs_inode_ext));
365 	memset(ip->inode_ext.lfs, 0, sizeof(struct lfs_inode_ext));
366 	vp->v_data = ip;
367 	/* ip->i_vnode = vp; */
368 	ip->i_number = ino;
369 	ip->i_lockf = 0;
370 	ip->i_diroff = 0;
371 	ip->i_lfs_effnblks = 0;
372 	ip->i_flag = 0;
373 
374 	/* Load inode block and find inode */
375 	if (daddr > 0) {
376 		bread(fs->lfs_devvp, fsbtodb(fs, daddr), fs->lfs_ibsize, NULL, &bp);
377 		bp->b_flags |= B_AGE;
378 		dip = lfs_ifind(fs, ino, bp);
379 		if (dip == NULL) {
380 			brelse(bp);
381 			free(ip);
382 			free(vp);
383 			return NULL;
384 		}
385 		memcpy(ip->i_din.ffs1_din, dip, sizeof(*dip));
386 		brelse(bp);
387 	}
388 	ip->i_number = ino;
389 	/* ip->i_devvp = fs->lfs_devvp; */
390 	ip->i_lfs = fs;
391 
392 	ip->i_ffs_effnlink = ip->i_ffs1_nlink;
393 	ip->i_lfs_effnblks = ip->i_ffs1_blocks;
394 	ip->i_lfs_osize = ip->i_ffs1_size;
395 #if 0
396 	if (fs->lfs_version > 1) {
397 		ip->i_ffs1_atime = ts.tv_sec;
398 		ip->i_ffs1_atimensec = ts.tv_nsec;
399 	}
400 #endif
401 
402 	memset(ip->i_lfs_fragsize, 0, NDADDR * sizeof(*ip->i_lfs_fragsize));
403 	for (i = 0; i < NDADDR; i++)
404 		if (ip->i_ffs1_db[i] != 0)
405 			ip->i_lfs_fragsize[i] = blksize(fs, ip, i);
406 
407 	++nvnodes;
408 	hash = ((int)(intptr_t)fs + ino) & (VNODE_HASH_MAX - 1);
409 	LIST_INSERT_HEAD(&getvnodelist[hash], vp, v_getvnodes);
410 	LIST_INSERT_HEAD(&vnodelist, vp, v_mntvnodes);
411 
412 	return vp;
413 }
414 
415 static struct uvnode *
416 lfs_vget(void *vfs, ino_t ino)
417 {
418 	struct lfs *fs = (struct lfs *)vfs;
419 	ufs_daddr_t daddr;
420 	struct ubuf *bp;
421 	IFILE *ifp;
422 
423 	LFS_IENTRY(ifp, fs, ino, bp);
424 	daddr = ifp->if_daddr;
425 	brelse(bp);
426 	if (daddr <= 0 || dtosn(fs, daddr) >= fs->lfs_nseg)
427 		return NULL;
428 	return lfs_raw_vget(fs, ino, fs->lfs_ivnode->v_fd, daddr);
429 }
430 
431 /* Check superblock magic number and checksum */
432 static int
433 check_sb(struct lfs *fs)
434 {
435 	u_int32_t checksum;
436 
437 	if (fs->lfs_magic != LFS_MAGIC) {
438 		printf("Superblock magic number (0x%lx) does not match "
439 		       "expected 0x%lx\n", (unsigned long) fs->lfs_magic,
440 		       (unsigned long) LFS_MAGIC);
441 		return 1;
442 	}
443 	/* checksum */
444 	checksum = lfs_sb_cksum(&(fs->lfs_dlfs));
445 	if (fs->lfs_cksum != checksum) {
446 		printf("Superblock checksum (%lx) does not match computed checksum (%lx)\n",
447 		    (unsigned long) fs->lfs_cksum, (unsigned long) checksum);
448 		return 1;
449 	}
450 	return 0;
451 }
452 
453 /* Initialize LFS library; load superblocks and choose which to use. */
454 struct lfs *
455 lfs_init(int devfd, daddr_t sblkno, daddr_t idaddr, int dummy_read, int debug)
456 {
457 	struct uvnode *devvp;
458 	struct ubuf *bp;
459 	int tryalt;
460 	struct lfs *fs, *altfs;
461 	int error;
462 
463 	vfs_init();
464 
465 	devvp = (struct uvnode *) malloc(sizeof(*devvp));
466 	memset(devvp, 0, sizeof(*devvp));
467 	devvp->v_fs = NULL;
468 	devvp->v_fd = devfd;
469 	devvp->v_strategy_op = raw_vop_strategy;
470 	devvp->v_bwrite_op = raw_vop_bwrite;
471 	devvp->v_bmap_op = raw_vop_bmap;
472 	LIST_INIT(&devvp->v_cleanblkhd);
473 	LIST_INIT(&devvp->v_dirtyblkhd);
474 
475 	tryalt = 0;
476 	if (dummy_read) {
477 		if (sblkno == 0)
478 			sblkno = btodb(LFS_LABELPAD);
479 		fs = (struct lfs *) malloc(sizeof(*fs));
480 		memset(fs, 0, sizeof(*fs));
481 		fs->lfs_devvp = devvp;
482 	} else {
483 		if (sblkno == 0) {
484 			sblkno = btodb(LFS_LABELPAD);
485 			tryalt = 1;
486 		} else if (debug) {
487 			printf("No -b flag given, not attempting to verify checkpoint\n");
488 		}
489 		error = bread(devvp, sblkno, LFS_SBPAD, NOCRED, &bp);
490 		fs = (struct lfs *) malloc(sizeof(*fs));
491 		memset(fs, 0, sizeof(*fs));
492 		fs->lfs_dlfs = *((struct dlfs *) bp->b_data);
493 		fs->lfs_devvp = devvp;
494 		bp->b_flags |= B_INVAL;
495 		brelse(bp);
496 
497 		if (tryalt) {
498 			error = bread(devvp, fsbtodb(fs, fs->lfs_sboffs[1]),
499 		    	LFS_SBPAD, NOCRED, &bp);
500 			altfs = (struct lfs *) malloc(sizeof(*altfs));
501 			memset(altfs, 0, sizeof(*altfs));
502 			altfs->lfs_dlfs = *((struct dlfs *) bp->b_data);
503 			altfs->lfs_devvp = devvp;
504 			bp->b_flags |= B_INVAL;
505 			brelse(bp);
506 
507 			if (check_sb(fs) || fs->lfs_idaddr <= 0) {
508 				if (debug)
509 					printf("Primary superblock is no good, using first alternate\n");
510 				free(fs);
511 				fs = altfs;
512 			} else {
513 				/* If both superblocks check out, try verification */
514 				if (check_sb(altfs)) {
515 					if (debug)
516 						printf("First alternate superblock is no good, using primary\n");
517 					free(altfs);
518 				} else {
519 					if (lfs_verify(fs, altfs, devvp, debug) == fs) {
520 						free(altfs);
521 					} else {
522 						free(fs);
523 						fs = altfs;
524 					}
525 				}
526 			}
527 		}
528 		if (check_sb(fs)) {
529 			free(fs);
530 			return NULL;
531 		}
532 	}
533 
534 	/* Compatibility */
535 	if (fs->lfs_version < 2) {
536 		fs->lfs_sumsize = LFS_V1_SUMMARY_SIZE;
537 		fs->lfs_ibsize = fs->lfs_bsize;
538 		fs->lfs_start = fs->lfs_sboffs[0];
539 		fs->lfs_tstamp = fs->lfs_otstamp;
540 		fs->lfs_fsbtodb = 0;
541 	}
542 
543 	if (!dummy_read) {
544 		fs->lfs_suflags = (u_int32_t **) malloc(2 * sizeof(u_int32_t *));
545 		fs->lfs_suflags[0] = (u_int32_t *) malloc(fs->lfs_nseg * sizeof(u_int32_t));
546 		fs->lfs_suflags[1] = (u_int32_t *) malloc(fs->lfs_nseg * sizeof(u_int32_t));
547 	}
548 
549 	if (idaddr == 0)
550 		idaddr = fs->lfs_idaddr;
551 	else
552 		fs->lfs_idaddr = idaddr;
553 	/* NB: If dummy_read!=0, idaddr==0 here so we get a fake inode. */
554 	fs->lfs_ivnode = lfs_raw_vget(fs,
555 		(dummy_read ? LFS_IFILE_INUM : fs->lfs_ifile), devvp->v_fd,
556 		idaddr);
557 
558 	register_vget((void *)fs, lfs_vget);
559 
560 	return fs;
561 }
562 
563 /*
564  * Check partial segment validity between fs->lfs_offset and the given goal.
565  *
566  * If goal == 0, just keep on going until the segments stop making sense,
567  * and return the address of the last valid partial segment.
568  *
569  * If goal != 0, return the address of the first partial segment that failed,
570  * or "goal" if we reached it without failure (the partial segment *at* goal
571  * need not be valid).
572  */
573 ufs_daddr_t
574 try_verify(struct lfs *osb, struct uvnode *devvp, ufs_daddr_t goal, int debug)
575 {
576 	ufs_daddr_t daddr, odaddr;
577 	SEGSUM *sp;
578 	int bc, flag;
579 	struct ubuf *bp;
580 	ufs_daddr_t nodirop_daddr;
581 	u_int64_t serial;
582 
583 	odaddr = -1;
584 	daddr = osb->lfs_offset;
585 	nodirop_daddr = daddr;
586 	serial = osb->lfs_serial;
587 	while (daddr != goal) {
588 		flag = 0;
589 oncemore:
590 		/* Read in summary block */
591 		bread(devvp, fsbtodb(osb, daddr), osb->lfs_sumsize, NULL, &bp);
592 		sp = (SEGSUM *)bp->b_data;
593 
594 		/*
595 		 * Could be a superblock instead of a segment summary.
596 		 * XXX should use gseguse, but right now we need to do more
597 		 * setup before we can...fix this
598 		 */
599 		if (sp->ss_magic != SS_MAGIC ||
600 		    sp->ss_ident != osb->lfs_ident ||
601 		    sp->ss_serial < serial ||
602 		    sp->ss_sumsum != cksum(&sp->ss_datasum, osb->lfs_sumsize -
603 			sizeof(sp->ss_sumsum))) {
604 			brelse(bp);
605 			if (flag == 0) {
606 				flag = 1;
607 				daddr += btofsb(osb, LFS_SBPAD);
608 				goto oncemore;
609 			}
610 			break;
611 		}
612 		++serial;
613 		bc = check_summary(osb, sp, daddr, debug, devvp, NULL);
614 		if (bc == 0) {
615 			brelse(bp);
616 			break;
617 		}
618 		assert (bc > 0);
619 		odaddr = daddr;
620 		daddr += btofsb(osb, osb->lfs_sumsize + bc);
621 		if (dtosn(osb, odaddr) != dtosn(osb, daddr) ||
622 		    dtosn(osb, daddr) != dtosn(osb, daddr +
623 			btofsb(osb, osb->lfs_sumsize + osb->lfs_bsize))) {
624 			daddr = sp->ss_next;
625 		}
626 		if (!(sp->ss_flags & SS_CONT))
627 			nodirop_daddr = daddr;
628 		brelse(bp);
629 	}
630 
631 	if (goal == 0)
632 		return nodirop_daddr;
633 	else
634 		return daddr;
635 }
636 
637 /* Use try_verify to check whether the newer superblock is valid. */
638 struct lfs *
639 lfs_verify(struct lfs *sb0, struct lfs *sb1, struct uvnode *devvp, int debug)
640 {
641 	ufs_daddr_t daddr;
642 	struct lfs *osb, *nsb;
643 
644 	/*
645 	 * Verify the checkpoint of the newer superblock,
646 	 * if the timestamp/serial number of the two superblocks is
647 	 * different.
648 	 */
649 
650 	osb = NULL;
651 	if (debug)
652 		printf("sb0 %lld, sb1 %lld\n", (long long) sb0->lfs_serial,
653 		    (long long) sb1->lfs_serial);
654 
655 	if ((sb0->lfs_version == 1 &&
656 		sb0->lfs_otstamp != sb1->lfs_otstamp) ||
657 	    (sb0->lfs_version > 1 &&
658 		sb0->lfs_serial != sb1->lfs_serial)) {
659 		if (sb0->lfs_version == 1) {
660 			if (sb0->lfs_otstamp > sb1->lfs_otstamp) {
661 				osb = sb1;
662 				nsb = sb0;
663 			} else {
664 				osb = sb0;
665 				nsb = sb1;
666 			}
667 		} else {
668 			if (sb0->lfs_serial > sb1->lfs_serial) {
669 				osb = sb1;
670 				nsb = sb0;
671 			} else {
672 				osb = sb0;
673 				nsb = sb1;
674 			}
675 		}
676 		if (debug) {
677 			printf("Attempting to verify newer checkpoint...");
678 			fflush(stdout);
679 		}
680 		daddr = try_verify(osb, devvp, nsb->lfs_offset, debug);
681 
682 		if (debug)
683 			printf("done.\n");
684 		if (daddr == nsb->lfs_offset) {
685 			pwarn("** Newer checkpoint verified, recovered %lld seconds of data\n",
686 			    (long long) nsb->lfs_tstamp - (long long) osb->lfs_tstamp);
687 			sbdirty();
688 		} else {
689 			pwarn("** Newer checkpoint invalid, lost %lld seconds of data\n", (long long) nsb->lfs_tstamp - (long long) osb->lfs_tstamp);
690 		}
691 		return (daddr == nsb->lfs_offset ? nsb : osb);
692 	}
693 	/* Nothing to check */
694 	return osb;
695 }
696 
697 /* Verify a partial-segment summary; return the number of bytes on disk. */
698 int
699 check_summary(struct lfs *fs, SEGSUM *sp, ufs_daddr_t pseg_addr, int debug,
700 	      struct uvnode *devvp, void (func(ufs_daddr_t, FINFO *)))
701 {
702 	FINFO *fp;
703 	int bc;			/* Bytes in partial segment */
704 	int nblocks;
705 	ufs_daddr_t seg_addr, daddr;
706 	ufs_daddr_t *dp, *idp;
707 	struct ubuf *bp;
708 	int i, j, k, datac, len;
709 	long sn;
710 	u_int32_t *datap;
711 	u_int32_t ccksum;
712 
713 	sn = dtosn(fs, pseg_addr);
714 	seg_addr = sntod(fs, sn);
715 
716 	/* We've already checked the sumsum, just do the data bounds and sum */
717 
718 	/* Count the blocks. */
719 	nblocks = howmany(sp->ss_ninos, INOPB(fs));
720 	bc = nblocks << (fs->lfs_version > 1 ? fs->lfs_ffshift : fs->lfs_bshift);
721 	assert(bc >= 0);
722 
723 	fp = (FINFO *) (sp + 1);
724 	for (i = 0; i < sp->ss_nfinfo; i++) {
725 		nblocks += fp->fi_nblocks;
726 		bc += fp->fi_lastlength + ((fp->fi_nblocks - 1)
727 					   << fs->lfs_bshift);
728 		assert(bc >= 0);
729 		fp = (FINFO *) (fp->fi_blocks + fp->fi_nblocks);
730 	}
731 	datap = (u_int32_t *) malloc(nblocks * sizeof(*datap));
732 	datac = 0;
733 
734 	dp = (ufs_daddr_t *) sp;
735 	dp += fs->lfs_sumsize / sizeof(ufs_daddr_t);
736 	dp--;
737 
738 	idp = dp;
739 	daddr = pseg_addr + btofsb(fs, fs->lfs_sumsize);
740 	fp = (FINFO *) (sp + 1);
741 	for (i = 0, j = 0;
742 	     i < sp->ss_nfinfo || j < howmany(sp->ss_ninos, INOPB(fs)); i++) {
743 		if (i >= sp->ss_nfinfo && *idp != daddr) {
744 			pwarn("Not enough inode blocks in pseg at 0x%" PRIx32
745 			      ": found %d, wanted %d\n",
746 			      pseg_addr, j, howmany(sp->ss_ninos, INOPB(fs)));
747 			if (debug)
748 				pwarn("*idp=%x, daddr=%" PRIx32 "\n", *idp,
749 				      daddr);
750 			break;
751 		}
752 		while (j < howmany(sp->ss_ninos, INOPB(fs)) && *idp == daddr) {
753 			bread(devvp, fsbtodb(fs, daddr), fs->lfs_ibsize, NOCRED, &bp);
754 			datap[datac++] = ((u_int32_t *) (bp->b_data))[0];
755 			brelse(bp);
756 
757 			++j;
758 			daddr += btofsb(fs, fs->lfs_ibsize);
759 			--idp;
760 		}
761 		if (i < sp->ss_nfinfo) {
762 			if (func)
763 				func(daddr, fp);
764 			for (k = 0; k < fp->fi_nblocks; k++) {
765 				len = (k == fp->fi_nblocks - 1 ?
766 				       fp->fi_lastlength
767 				       : fs->lfs_bsize);
768 				bread(devvp, fsbtodb(fs, daddr), len, NOCRED, &bp);
769 				datap[datac++] = ((u_int32_t *) (bp->b_data))[0];
770 				brelse(bp);
771 				daddr += btofsb(fs, len);
772 			}
773 			fp = (FINFO *) (fp->fi_blocks + fp->fi_nblocks);
774 		}
775 	}
776 
777 	if (datac != nblocks) {
778 		pwarn("Partial segment at 0x%llx expected %d blocks counted %d\n",
779 		    (long long) pseg_addr, nblocks, datac);
780 	}
781 	ccksum = cksum(datap, nblocks * sizeof(u_int32_t));
782 	/* Check the data checksum */
783 	if (ccksum != sp->ss_datasum) {
784 		pwarn("Partial segment at 0x%" PRIx32 " data checksum"
785 		      " mismatch: given 0x%x, computed 0x%x\n",
786 		      pseg_addr, sp->ss_datasum, ccksum);
787 		free(datap);
788 		return 0;
789 	}
790 	free(datap);
791 	assert(bc >= 0);
792 	return bc;
793 }
794 
795 /* print message and exit */
796 void
797 my_vpanic(int fatal, const char *fmt, va_list ap)
798 {
799         (void) vprintf(fmt, ap);
800 	exit(8);
801 }
802 
803 void
804 call_panic(const char *fmt, ...)
805 {
806 	va_list ap;
807 
808 	va_start(ap, fmt);
809         panic_func(1, fmt, ap);
810 	va_end(ap);
811 }
812 
813 /* Allocate a new inode. */
814 struct uvnode *
815 lfs_valloc(struct lfs *fs, ino_t ino)
816 {
817 	struct ubuf *bp, *cbp;
818 	struct ifile *ifp;
819 	ino_t new_ino;
820 	int error;
821 	int new_gen;
822 	CLEANERINFO *cip;
823 
824 	/* Get the head of the freelist. */
825 	LFS_GET_HEADFREE(fs, cip, cbp, &new_ino);
826 
827 	/*
828 	 * Remove the inode from the free list and write the new start
829 	 * of the free list into the superblock.
830 	 */
831 	LFS_IENTRY(ifp, fs, new_ino, bp);
832 	if (ifp->if_daddr != LFS_UNUSED_DADDR)
833 		panic("lfs_valloc: inuse inode %d on the free list", new_ino);
834 	LFS_PUT_HEADFREE(fs, cip, cbp, ifp->if_nextfree);
835 
836 	new_gen = ifp->if_version; /* version was updated by vfree */
837 	brelse(bp);
838 
839 	/* Extend IFILE so that the next lfs_valloc will succeed. */
840 	if (fs->lfs_freehd == LFS_UNUSED_INUM) {
841 		if ((error = extend_ifile(fs)) != 0) {
842 			LFS_PUT_HEADFREE(fs, cip, cbp, new_ino);
843 			return NULL;
844 		}
845 	}
846 
847 	/* Set superblock modified bit and increment file count. */
848         sbdirty();
849 	++fs->lfs_nfiles;
850 
851         return lfs_raw_vget(fs, ino, fs->lfs_devvp->v_fd, 0x0);
852 }
853 
854 /*
855  * Add a new block to the Ifile, to accommodate future file creations.
856  */
857 int
858 extend_ifile(struct lfs *fs)
859 {
860 	struct uvnode *vp;
861 	struct inode *ip;
862 	IFILE *ifp;
863 	IFILE_V1 *ifp_v1;
864 	struct ubuf *bp, *cbp;
865 	daddr_t i, blkno, max;
866 	ino_t oldlast;
867 	CLEANERINFO *cip;
868 
869 	vp = fs->lfs_ivnode;
870 	ip = VTOI(vp);
871 	blkno = lblkno(fs, ip->i_ffs1_size);
872 
873 	bp = getblk(vp, blkno, fs->lfs_bsize);	/* XXX VOP_BALLOC() */
874 	ip->i_ffs1_size += fs->lfs_bsize;
875 
876 	i = (blkno - fs->lfs_segtabsz - fs->lfs_cleansz) *
877 		fs->lfs_ifpb;
878 	LFS_GET_HEADFREE(fs, cip, cbp, &oldlast);
879 	LFS_PUT_HEADFREE(fs, cip, cbp, i);
880 	max = i + fs->lfs_ifpb;
881 	fs->lfs_bfree -= btofsb(fs, fs->lfs_bsize);
882 
883 	if (fs->lfs_version == 1) {
884 		for (ifp_v1 = (IFILE_V1 *)bp->b_data; i < max; ++ifp_v1) {
885 			ifp_v1->if_version = 1;
886 			ifp_v1->if_daddr = LFS_UNUSED_DADDR;
887 			ifp_v1->if_nextfree = ++i;
888 		}
889 		ifp_v1--;
890 		ifp_v1->if_nextfree = oldlast;
891 	} else {
892 		for (ifp = (IFILE *)bp->b_data; i < max; ++ifp) {
893 			ifp->if_version = 1;
894 			ifp->if_daddr = LFS_UNUSED_DADDR;
895 			ifp->if_nextfree = ++i;
896 		}
897 		ifp--;
898 		ifp->if_nextfree = oldlast;
899 	}
900 	LFS_PUT_TAILFREE(fs, cip, cbp, max - 1);
901 
902 	LFS_BWRITE_LOG(bp);
903 
904 	return 0;
905 }
906 
907