xref: /netbsd-src/sbin/fsck_lfs/lfs.c (revision e89934bbf778a6d6d6894877c4da59d0c7835b0f)
1 /* $NetBSD: lfs.c,v 1.72 2016/09/16 11:13:47 christos 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  *
18  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
19  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
20  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
21  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
22  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
23  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
24  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
25  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
26  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
27  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
28  * POSSIBILITY OF SUCH DAMAGE.
29  */
30 /*
31  * Copyright (c) 1989, 1991, 1993
32  *	The Regents of the University of California.  All rights reserved.
33  * (c) UNIX System Laboratories, Inc.
34  * All or some portions of this file are derived from material licensed
35  * to the University of California by American Telephone and Telegraph
36  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
37  * the permission of UNIX System Laboratories, Inc.
38  *
39  * Redistribution and use in source and binary forms, with or without
40  * modification, are permitted provided that the following conditions
41  * are met:
42  * 1. Redistributions of source code must retain the above copyright
43  *    notice, this list of conditions and the following disclaimer.
44  * 2. Redistributions in binary form must reproduce the above copyright
45  *    notice, this list of conditions and the following disclaimer in the
46  *    documentation and/or other materials provided with the distribution.
47  * 3. Neither the name of the University nor the names of its contributors
48  *    may be used to endorse or promote products derived from this software
49  *    without specific prior written permission.
50  *
51  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
52  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
53  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
54  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
55  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
56  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
57  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
58  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
59  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
60  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
61  * SUCH DAMAGE.
62  *
63  *	@(#)ufs_bmap.c	8.8 (Berkeley) 8/11/95
64  */
65 
66 
67 #include <sys/types.h>
68 #include <sys/param.h>
69 #include <sys/time.h>
70 #include <sys/buf.h>
71 #include <sys/mount.h>
72 
73 #define vnode uvnode
74 #include <ufs/lfs/lfs.h>
75 #include <ufs/lfs/lfs_inode.h>
76 #include <ufs/lfs/lfs_accessors.h>
77 #undef vnode
78 
79 #include <assert.h>
80 #include <err.h>
81 #include <errno.h>
82 #include <stdarg.h>
83 #include <stdbool.h>
84 #include <stdio.h>
85 #include <stdlib.h>
86 #include <string.h>
87 #include <unistd.h>
88 #include <util.h>
89 
90 #include "bufcache.h"
91 #include "vnode.h"
92 #include "lfs_user.h"
93 #include "segwrite.h"
94 #include "kernelops.h"
95 
96 #define panic call_panic
97 
98 extern u_int32_t cksum(void *, size_t);
99 extern u_int32_t lfs_sb_cksum(struct lfs *);
100 extern void pwarn(const char *, ...);
101 
102 extern struct uvnodelst vnodelist;
103 extern struct uvnodelst getvnodelist[VNODE_HASH_MAX];
104 extern int nvnodes;
105 
106 long dev_bsize = DEV_BSIZE;
107 
108 static int
109 lfs_fragextend(struct uvnode *, int, int, daddr_t, struct ubuf **);
110 
111 int fsdirty = 0;
112 void (*panic_func)(int, const char *, va_list) = my_vpanic;
113 
114 /*
115  * LFS buffer and uvnode operations
116  */
117 
118 int
119 lfs_vop_strategy(struct ubuf * bp)
120 {
121 	int count;
122 
123 	if (bp->b_flags & B_READ) {
124 		count = kops.ko_pread(bp->b_vp->v_fd, bp->b_data, bp->b_bcount,
125 		    bp->b_blkno * dev_bsize);
126 		if (count == bp->b_bcount)
127 			bp->b_flags |= B_DONE;
128 	} else {
129 		count = kops.ko_pwrite(bp->b_vp->v_fd, bp->b_data, bp->b_bcount,
130 		    bp->b_blkno * dev_bsize);
131 		if (count == 0) {
132 			perror("pwrite");
133 			return -1;
134 		}
135 		bp->b_flags &= ~B_DELWRI;
136 		reassignbuf(bp, bp->b_vp);
137 	}
138 	return 0;
139 }
140 
141 int
142 lfs_vop_bwrite(struct ubuf * bp)
143 {
144 	struct lfs *fs;
145 
146 	fs = bp->b_vp->v_fs;
147 	if (!(bp->b_flags & B_DELWRI)) {
148 		lfs_sb_subavail(fs, lfs_btofsb(fs, bp->b_bcount));
149 	}
150 	bp->b_flags |= B_DELWRI | B_LOCKED;
151 	reassignbuf(bp, bp->b_vp);
152 	brelse(bp, 0);
153 	return 0;
154 }
155 
156 /*
157  * ulfs_bmaparray does the bmap conversion, and if requested returns the
158  * array of logical blocks which must be traversed to get to a block.
159  * Each entry contains the offset into that block that gets you to the
160  * next block and the disk address of the block (if it is assigned).
161  */
162 int
163 ulfs_bmaparray(struct lfs * fs, struct uvnode * vp, daddr_t bn, daddr_t * bnp, struct indir * ap, int *nump)
164 {
165 	struct inode *ip;
166 	struct ubuf *bp;
167 	struct indir a[ULFS_NIADDR + 1], *xap;
168 	daddr_t daddr;
169 	daddr_t metalbn;
170 	int error, num;
171 
172 	ip = VTOI(vp);
173 
174 	if (bn >= 0 && bn < ULFS_NDADDR) {
175 		if (nump != NULL)
176 			*nump = 0;
177 		*bnp = LFS_FSBTODB(fs, lfs_dino_getdb(fs, ip->i_din, bn));
178 		if (*bnp == 0)
179 			*bnp = -1;
180 		return (0);
181 	}
182 	xap = ap == NULL ? a : ap;
183 	if (!nump)
184 		nump = &num;
185 	if ((error = ulfs_getlbns(fs, vp, bn, xap, nump)) != 0)
186 		return (error);
187 
188 	num = *nump;
189 
190 	/* Get disk address out of indirect block array */
191 	daddr = lfs_dino_getib(fs, ip->i_din, xap->in_off);
192 
193 	for (bp = NULL, ++xap; --num; ++xap) {
194 		/* Exit the loop if there is no disk address assigned yet and
195 		 * the indirect block isn't in the cache, or if we were
196 		 * looking for an indirect block and we've found it. */
197 
198 		metalbn = xap->in_lbn;
199 		if ((daddr == 0 && !incore(vp, metalbn)) || metalbn == bn)
200 			break;
201 		/*
202 		 * If we get here, we've either got the block in the cache
203 		 * or we have a disk address for it, go fetch it.
204 		 */
205 		if (bp)
206 			brelse(bp, 0);
207 
208 		xap->in_exists = 1;
209 		bp = getblk(vp, metalbn, lfs_sb_getbsize(fs));
210 
211 		if (!(bp->b_flags & (B_DONE | B_DELWRI))) {
212 			bp->b_blkno = LFS_FSBTODB(fs, daddr);
213 			bp->b_flags |= B_READ;
214 			VOP_STRATEGY(bp);
215 		}
216 		daddr = lfs_iblock_get(fs, bp->b_data, xap->in_off);
217 	}
218 	if (bp)
219 		brelse(bp, 0);
220 
221 	daddr = LFS_FSBTODB(fs, daddr);
222 	*bnp = daddr == 0 ? -1 : daddr;
223 	return (0);
224 }
225 
226 /*
227  * Create an array of logical block number/offset pairs which represent the
228  * path of indirect blocks required to access a data block.  The first "pair"
229  * contains the logical block number of the appropriate single, double or
230  * triple indirect block and the offset into the inode indirect block array.
231  * Note, the logical block number of the inode single/double/triple indirect
232  * block appears twice in the array, once with the offset into di_ib and
233  * once with the offset into the page itself.
234  */
235 int
236 ulfs_getlbns(struct lfs * fs, struct uvnode * vp, daddr_t bn, struct indir * ap, int *nump)
237 {
238 	daddr_t metalbn, realbn;
239 	int64_t blockcnt;
240 	int lbc;
241 	int i, numlevels, off;
242 	int lognindir, indir;
243 
244 	metalbn = 0;    /* XXXGCC -Wuninitialized [sh3] */
245 
246 	if (nump)
247 		*nump = 0;
248 	numlevels = 0;
249 	realbn = bn;
250 	if (bn < 0)
251 		bn = -bn;
252 
253 	lognindir = -1;
254 	for (indir = lfs_sb_getnindir(fs); indir; indir >>= 1)
255 		++lognindir;
256 
257 	/* Determine the number of levels of indirection.  After this loop is
258 	 * done, blockcnt indicates the number of data blocks possible at the
259 	 * given level of indirection, and ULFS_NIADDR - i is the number of levels
260 	 * of indirection needed to locate the requested block. */
261 
262 	bn -= ULFS_NDADDR;
263 	for (lbc = 0, i = ULFS_NIADDR;; i--, bn -= blockcnt) {
264 		if (i == 0)
265 			return (EFBIG);
266 
267 		lbc += lognindir;
268 		blockcnt = (int64_t) 1 << lbc;
269 
270 		if (bn < blockcnt)
271 			break;
272 	}
273 
274 	/* Calculate the address of the first meta-block. */
275 	metalbn = -((realbn >= 0 ? realbn : -realbn) - bn + ULFS_NIADDR - i);
276 
277 	/* At each iteration, off is the offset into the bap array which is an
278 	 * array of disk addresses at the current level of indirection. The
279 	 * logical block number and the offset in that block are stored into
280 	 * the argument array. */
281 	ap->in_lbn = metalbn;
282 	ap->in_off = off = ULFS_NIADDR - i;
283 	ap->in_exists = 0;
284 	ap++;
285 	for (++numlevels; i <= ULFS_NIADDR; i++) {
286 		/* If searching for a meta-data block, quit when found. */
287 		if (metalbn == realbn)
288 			break;
289 
290 		lbc -= lognindir;
291 		/*blockcnt = (int64_t) 1 << lbc;*/
292 		off = (bn >> lbc) & (lfs_sb_getnindir(fs) - 1);
293 
294 		++numlevels;
295 		ap->in_lbn = metalbn;
296 		ap->in_off = off;
297 		ap->in_exists = 0;
298 		++ap;
299 
300 		metalbn -= -1 + (off << lbc);
301 	}
302 	if (nump)
303 		*nump = numlevels;
304 	return (0);
305 }
306 
307 int
308 lfs_vop_bmap(struct uvnode * vp, daddr_t lbn, daddr_t * daddrp)
309 {
310 	return ulfs_bmaparray(vp->v_fs, vp, lbn, daddrp, NULL, NULL);
311 }
312 
313 /* Search a block for a specific dinode. */
314 union lfs_dinode *
315 lfs_ifind(struct lfs *fs, ino_t ino, struct ubuf *bp)
316 {
317 	union lfs_dinode *ldip;
318 	unsigned i, num;
319 
320 	num = LFS_INOPB(fs);
321 
322 	/*
323 	 * Read the inode block backwards, since later versions of the
324 	 * inode will supercede earlier ones.  Though it is unlikely, it is
325 	 * possible that the same inode will appear in the same inode block.
326 	 */
327 	for (i = num; i-- > 0; ) {
328 		ldip = DINO_IN_BLOCK(fs, bp->b_data, i);
329 		if (lfs_dino_getinumber(fs, ldip) == ino)
330 			return (ldip);
331 	}
332 	return NULL;
333 }
334 
335 /*
336  * lfs_raw_vget makes us a new vnode from the inode at the given disk address.
337  * XXX it currently loses atime information.
338  */
339 struct uvnode *
340 lfs_raw_vget(struct lfs * fs, ino_t ino, int fd, daddr_t daddr)
341 {
342 	struct uvnode *vp;
343 	struct inode *ip;
344 	union lfs_dinode *dip;
345 	struct ubuf *bp;
346 	int i, hash;
347 
348 	vp = ecalloc(1, sizeof(*vp));
349 	vp->v_fd = fd;
350 	vp->v_fs = fs;
351 	vp->v_usecount = 0;
352 	vp->v_strategy_op = lfs_vop_strategy;
353 	vp->v_bwrite_op = lfs_vop_bwrite;
354 	vp->v_bmap_op = lfs_vop_bmap;
355 	LIST_INIT(&vp->v_cleanblkhd);
356 	LIST_INIT(&vp->v_dirtyblkhd);
357 
358 	ip = ecalloc(1, sizeof(*ip));
359 
360 	ip->i_din = dip = ecalloc(1, sizeof(*dip));
361 
362 	/* Initialize the inode -- from lfs_vcreate. */
363 	ip->inode_ext.lfs = ecalloc(1, sizeof(*ip->inode_ext.lfs));
364 	vp->v_data = ip;
365 	/* ip->i_vnode = vp; */
366 	ip->i_number = ino;
367 	ip->i_lockf = 0;
368 	ip->i_lfs_effnblks = 0;
369 	ip->i_flag = 0;
370 
371 	/* Load inode block and find inode */
372 	if (daddr > 0) {
373 		bread(fs->lfs_devvp, LFS_FSBTODB(fs, daddr), lfs_sb_getibsize(fs),
374 		    0, &bp);
375 		bp->b_flags |= B_AGE;
376 		dip = lfs_ifind(fs, ino, bp);
377 		if (dip == NULL) {
378 			brelse(bp, 0);
379 			free(ip->i_din);
380 			free(ip->inode_ext.lfs);
381 			free(ip);
382 			free(vp);
383 			return NULL;
384 		}
385 		lfs_copy_dinode(fs, ip->i_din, dip);
386 		brelse(bp, 0);
387 	}
388 	ip->i_number = ino;
389 	/* ip->i_devvp = fs->lfs_devvp; */
390 	ip->i_lfs = fs;
391 
392 	ip->i_lfs_effnblks = lfs_dino_getblocks(fs, ip->i_din);
393 	ip->i_lfs_osize = lfs_dino_getsize(fs, ip->i_din);
394 #if 0
395 	if (lfs_sb_getversion(fs) > 1) {
396 		lfs_dino_setatime(fs, ip->i_din, ts.tv_sec);
397 		lfs_dino_setatimensec(fs, ip->i_din, ts.tv_nsec);
398 	}
399 #endif
400 
401 	memset(ip->i_lfs_fragsize, 0, ULFS_NDADDR * sizeof(*ip->i_lfs_fragsize));
402 	for (i = 0; i < ULFS_NDADDR; i++)
403 		if (lfs_dino_getdb(fs, ip->i_din, i) != 0)
404 			ip->i_lfs_fragsize[i] = lfs_blksize(fs, ip, i);
405 
406 	++nvnodes;
407 	hash = ((int)(intptr_t)fs + ino) & (VNODE_HASH_MAX - 1);
408 	LIST_INSERT_HEAD(&getvnodelist[hash], vp, v_getvnodes);
409 	LIST_INSERT_HEAD(&vnodelist, vp, v_mntvnodes);
410 
411 	return vp;
412 }
413 
414 static struct uvnode *
415 lfs_vget(void *vfs, ino_t ino)
416 {
417 	struct lfs *fs = (struct lfs *)vfs;
418 	daddr_t daddr;
419 	struct ubuf *bp;
420 	IFILE *ifp;
421 
422 	LFS_IENTRY(ifp, fs, ino, bp);
423 	daddr = lfs_if_getdaddr(fs, ifp);
424 	brelse(bp, 0);
425 	if (daddr <= 0 || lfs_dtosn(fs, daddr) >= lfs_sb_getnseg(fs))
426 		return NULL;
427 	return lfs_raw_vget(fs, ino, fs->lfs_ivnode->v_fd, daddr);
428 }
429 
430 /*
431  * Check superblock magic number and checksum.
432  * Sets lfs_is64 and lfs_dobyteswap.
433  */
434 static int
435 check_sb(struct lfs *fs)
436 {
437 	u_int32_t checksum;
438 	u_int32_t magic;
439 
440 	/* we can read the magic out of either the 32-bit or 64-bit dlfs */
441 	magic = fs->lfs_dlfs_u.u_32.dlfs_magic;
442 
443 	switch (magic) {
444 	    case LFS_MAGIC:
445 		fs->lfs_is64 = false;
446 		fs->lfs_dobyteswap = false;
447 		break;
448 	    case LFS_MAGIC_SWAPPED:
449 		fs->lfs_is64 = false;
450 		fs->lfs_dobyteswap = true;
451 		break;
452 	    case LFS64_MAGIC:
453 		fs->lfs_is64 = true;
454 		fs->lfs_dobyteswap = false;
455 		break;
456 	    case LFS64_MAGIC_SWAPPED:
457 		fs->lfs_is64 = true;
458 		fs->lfs_dobyteswap = true;
459 		break;
460 	    default:
461 		printf("Superblock magic number (0x%lx) does not match "
462 		       "expected 0x%lx\n", (unsigned long) magic,
463 		       (unsigned long) LFS_MAGIC);
464 		return 1;
465 	}
466 
467 	/* checksum */
468 	checksum = lfs_sb_cksum(fs);
469 	if (lfs_sb_getcksum(fs) != checksum) {
470 		printf("Superblock checksum (%lx) does not match computed checksum (%lx)\n",
471 		    (unsigned long) lfs_sb_getcksum(fs), (unsigned long) checksum);
472 		return 1;
473 	}
474 	return 0;
475 }
476 
477 /* Initialize LFS library; load superblocks and choose which to use. */
478 struct lfs *
479 lfs_init(int devfd, daddr_t sblkno, daddr_t idaddr, int dummy_read, int debug)
480 {
481 	struct uvnode *devvp;
482 	struct ubuf *bp;
483 	int tryalt;
484 	struct lfs *fs, *altfs;
485 
486 	vfs_init();
487 
488 	devvp = ecalloc(1, sizeof(*devvp));
489 	devvp->v_fs = NULL;
490 	devvp->v_fd = devfd;
491 	devvp->v_strategy_op = raw_vop_strategy;
492 	devvp->v_bwrite_op = raw_vop_bwrite;
493 	devvp->v_bmap_op = raw_vop_bmap;
494 	LIST_INIT(&devvp->v_cleanblkhd);
495 	LIST_INIT(&devvp->v_dirtyblkhd);
496 
497 	tryalt = 0;
498 	if (dummy_read) {
499 		if (sblkno == 0)
500 			sblkno = LFS_LABELPAD / dev_bsize;
501 		fs = ecalloc(1, sizeof(*fs));
502 		fs->lfs_devvp = devvp;
503 	} else {
504 		if (sblkno == 0) {
505 			sblkno = LFS_LABELPAD / dev_bsize;
506 			tryalt = 1;
507 		} else if (debug) {
508 			printf("No -b flag given, not attempting to verify checkpoint\n");
509 		}
510 
511 		dev_bsize = DEV_BSIZE;
512 
513 		(void)bread(devvp, sblkno, LFS_SBPAD, 0, &bp);
514 		fs = ecalloc(1, sizeof(*fs));
515 		__CTASSERT(sizeof(struct dlfs) == sizeof(struct dlfs64));
516 		memcpy(&fs->lfs_dlfs_u, bp->b_data, sizeof(struct dlfs));
517 		fs->lfs_devvp = devvp;
518 		bp->b_flags |= B_INVAL;
519 		brelse(bp, 0);
520 
521 		dev_bsize = lfs_sb_getfsize(fs) >> lfs_sb_getfsbtodb(fs);
522 
523 		if (tryalt) {
524 			(void)bread(devvp, LFS_FSBTODB(fs, lfs_sb_getsboff(fs, 1)),
525 		    	LFS_SBPAD, 0, &bp);
526 			altfs = ecalloc(1, sizeof(*altfs));
527 			memcpy(&altfs->lfs_dlfs_u, bp->b_data,
528 			       sizeof(struct dlfs));
529 			altfs->lfs_devvp = devvp;
530 			bp->b_flags |= B_INVAL;
531 			brelse(bp, 0);
532 
533 			if (check_sb(fs) || lfs_sb_getidaddr(fs) <= 0) {
534 				if (debug)
535 					printf("Primary superblock is no good, using first alternate\n");
536 				free(fs);
537 				fs = altfs;
538 			} else {
539 				/* If both superblocks check out, try verification */
540 				if (check_sb(altfs)) {
541 					if (debug)
542 						printf("First alternate superblock is no good, using primary\n");
543 					free(altfs);
544 				} else {
545 					if (lfs_verify(fs, altfs, devvp, debug) == fs) {
546 						free(altfs);
547 					} else {
548 						free(fs);
549 						fs = altfs;
550 					}
551 				}
552 			}
553 		}
554 		if (check_sb(fs)) {
555 			free(fs);
556 			return NULL;
557 		}
558 	}
559 
560 	/* Compatibility */
561 	if (lfs_sb_getversion(fs) < 2) {
562 		lfs_sb_setsumsize(fs, LFS_V1_SUMMARY_SIZE);
563 		lfs_sb_setibsize(fs, lfs_sb_getbsize(fs));
564 		lfs_sb_sets0addr(fs, lfs_sb_getsboff(fs, 0));
565 		lfs_sb_settstamp(fs, lfs_sb_getotstamp(fs));
566 		lfs_sb_setfsbtodb(fs, 0);
567 	}
568 
569 	if (!dummy_read) {
570 		fs->lfs_suflags = emalloc(2 * sizeof(u_int32_t *));
571 		fs->lfs_suflags[0] = emalloc(lfs_sb_getnseg(fs) * sizeof(u_int32_t));
572 		fs->lfs_suflags[1] = emalloc(lfs_sb_getnseg(fs) * sizeof(u_int32_t));
573 	}
574 
575 	if (idaddr == 0)
576 		idaddr = lfs_sb_getidaddr(fs);
577 	else
578 		lfs_sb_setidaddr(fs, idaddr);
579 	/* NB: If dummy_read!=0, idaddr==0 here so we get a fake inode. */
580 	fs->lfs_ivnode = lfs_raw_vget(fs, LFS_IFILE_INUM,
581 		devvp->v_fd, idaddr);
582 	if (fs->lfs_ivnode == NULL)
583 		return NULL;
584 
585 	register_vget((void *)fs, lfs_vget);
586 
587 	return fs;
588 }
589 
590 /*
591  * Check partial segment validity between fs->lfs_offset and the given goal.
592  *
593  * If goal == 0, just keep on going until the segments stop making sense,
594  * and return the address of the last valid partial segment.
595  *
596  * If goal != 0, return the address of the first partial segment that failed,
597  * or "goal" if we reached it without failure (the partial segment *at* goal
598  * need not be valid).
599  */
600 daddr_t
601 try_verify(struct lfs *osb, struct uvnode *devvp, daddr_t goal, int debug)
602 {
603 	daddr_t daddr, odaddr;
604 	SEGSUM *sp;
605 	int i, bc, hitclean;
606 	struct ubuf *bp;
607 	daddr_t nodirop_daddr;
608 	u_int64_t serial;
609 
610 	bc = 0;
611 	hitclean = 0;
612 	odaddr = -1;
613 	daddr = lfs_sb_getoffset(osb);
614 	nodirop_daddr = daddr;
615 	serial = lfs_sb_getserial(osb);
616 	while (daddr != goal) {
617 		/*
618 		 * Don't mistakenly read a superblock, if there is one here.
619 		 */
620 		if (lfs_sntod(osb, lfs_dtosn(osb, daddr)) == daddr) {
621 			if (daddr == lfs_sb_gets0addr(osb))
622 				daddr += lfs_btofsb(osb, LFS_LABELPAD);
623 			for (i = 0; i < LFS_MAXNUMSB; i++) {
624 				/* XXX dholland 20150828 I think this is wrong */
625 				if (lfs_sb_getsboff(osb, i) < daddr)
626 					break;
627 				if (lfs_sb_getsboff(osb, i) == daddr)
628 					daddr += lfs_btofsb(osb, LFS_SBPAD);
629 			}
630 		}
631 
632 		/* Read in summary block */
633 		bread(devvp, LFS_FSBTODB(osb, daddr), lfs_sb_getsumsize(osb),
634 		    0, &bp);
635 		sp = (SEGSUM *)bp->b_data;
636 
637 		/*
638 		 * Check for a valid segment summary belonging to our fs.
639 		 */
640 		if (lfs_ss_getmagic(osb, sp) != SS_MAGIC ||
641 		    lfs_ss_getident(osb, sp) != lfs_sb_getident(osb) ||
642 		    lfs_ss_getserial(osb, sp) < serial ||	/* XXX strengthen this */
643 		    lfs_ss_getsumsum(osb, sp) !=
644 		            cksum((char *)sp + lfs_ss_getsumstart(osb),
645 				  lfs_sb_getsumsize(osb) - lfs_ss_getsumstart(osb))) {
646 			brelse(bp, 0);
647 			if (debug) {
648 				if (lfs_ss_getmagic(osb, sp) != SS_MAGIC)
649 					pwarn("pseg at 0x%jx: "
650 					      "wrong magic number\n",
651 					      (uintmax_t)daddr);
652 				else if (lfs_ss_getident(osb, sp) != lfs_sb_getident(osb))
653 					pwarn("pseg at 0x%jx: "
654 					      "expected ident %jx, got %jx\n",
655 					      (uintmax_t)daddr,
656 					      (uintmax_t)lfs_ss_getident(osb, sp),
657 					      (uintmax_t)lfs_sb_getident(osb));
658 				else if (lfs_ss_getserial(osb, sp) >= serial)
659 					pwarn("pseg at 0x%jx: "
660 					      "serial %d < %d\n",
661 					      (uintmax_t)daddr,
662 					      (int)lfs_ss_getserial(osb, sp), (int)serial);
663 				else
664 					pwarn("pseg at 0x%jx: "
665 					      "summary checksum wrong\n",
666 					      (uintmax_t)daddr);
667 			}
668 			break;
669 		}
670 		if (debug && lfs_ss_getserial(osb, sp) != serial)
671 			pwarn("warning, serial=%d ss_serial=%d\n",
672 				(int)serial, (int)lfs_ss_getserial(osb, sp));
673 		++serial;
674 		bc = check_summary(osb, sp, daddr, debug, devvp, NULL);
675 		if (bc == 0) {
676 			brelse(bp, 0);
677 			break;
678 		}
679 		if (debug)
680 			pwarn("summary good: 0x%x/%d\n", (uintmax_t)daddr,
681 			      (int)lfs_ss_getserial(osb, sp));
682 		assert (bc > 0);
683 		odaddr = daddr;
684 		daddr += lfs_btofsb(osb, lfs_sb_getsumsize(osb) + bc);
685 		if (lfs_dtosn(osb, odaddr) != lfs_dtosn(osb, daddr) ||
686 		    lfs_dtosn(osb, daddr) != lfs_dtosn(osb, daddr +
687 			lfs_btofsb(osb, lfs_sb_getsumsize(osb) + lfs_sb_getbsize(osb)) - 1)) {
688 			daddr = lfs_ss_getnext(osb, sp);
689 		}
690 
691 		/*
692 		 * Check for the beginning and ending of a sequence of
693 		 * dirops.  Writes from the cleaner never involve new
694 		 * information, and are always checkpoints; so don't try
695 		 * to roll forward through them.  Likewise, psegs written
696 		 * by a previous roll-forward attempt are not interesting.
697 		 */
698 		if (lfs_ss_getflags(osb, sp) & (SS_CLEAN | SS_RFW))
699 			hitclean = 1;
700 		if (hitclean == 0 && (lfs_ss_getflags(osb, sp) & SS_CONT) == 0)
701 			nodirop_daddr = daddr;
702 
703 		brelse(bp, 0);
704 	}
705 
706 	if (goal == 0)
707 		return nodirop_daddr;
708 	else
709 		return daddr;
710 }
711 
712 /* Use try_verify to check whether the newer superblock is valid. */
713 struct lfs *
714 lfs_verify(struct lfs *sb0, struct lfs *sb1, struct uvnode *devvp, int debug)
715 {
716 	daddr_t daddr;
717 	struct lfs *osb, *nsb;
718 
719 	/*
720 	 * Verify the checkpoint of the newer superblock,
721 	 * if the timestamp/serial number of the two superblocks is
722 	 * different.
723 	 */
724 
725 	osb = NULL;
726 	if (debug)
727 		pwarn("sb0 %ju, sb1 %ju",
728 		      (uintmax_t) lfs_sb_getserial(sb0),
729 		      (uintmax_t) lfs_sb_getserial(sb1));
730 
731 	if ((lfs_sb_getversion(sb0) == 1 &&
732 		lfs_sb_getotstamp(sb0) != lfs_sb_getotstamp(sb1)) ||
733 	    (lfs_sb_getversion(sb0) > 1 &&
734 		lfs_sb_getserial(sb0) != lfs_sb_getserial(sb1))) {
735 		if (lfs_sb_getversion(sb0) == 1) {
736 			if (lfs_sb_getotstamp(sb0) > lfs_sb_getotstamp(sb1)) {
737 				osb = sb1;
738 				nsb = sb0;
739 			} else {
740 				osb = sb0;
741 				nsb = sb1;
742 			}
743 		} else {
744 			if (lfs_sb_getserial(sb0) > lfs_sb_getserial(sb1)) {
745 				osb = sb1;
746 				nsb = sb0;
747 			} else {
748 				osb = sb0;
749 				nsb = sb1;
750 			}
751 		}
752 		if (debug) {
753 			printf("Attempting to verify newer checkpoint...");
754 			fflush(stdout);
755 		}
756 		daddr = try_verify(osb, devvp, lfs_sb_getoffset(nsb), debug);
757 
758 		if (debug)
759 			printf("done.\n");
760 		if (daddr == lfs_sb_getoffset(nsb)) {
761 			pwarn("** Newer checkpoint verified; recovered %jd seconds of data\n",
762 			    (intmax_t)(lfs_sb_gettstamp(nsb) - lfs_sb_gettstamp(osb)));
763 			sbdirty();
764 		} else {
765 			pwarn("** Newer checkpoint invalid; lost %jd seconds of data\n", (intmax_t)(lfs_sb_gettstamp(nsb) - lfs_sb_gettstamp(osb)));
766 		}
767 		return (daddr == lfs_sb_getoffset(nsb) ? nsb : osb);
768 	}
769 	/* Nothing to check */
770 	return osb;
771 }
772 
773 /* Verify a partial-segment summary; return the number of bytes on disk. */
774 int
775 check_summary(struct lfs *fs, SEGSUM *sp, daddr_t pseg_addr, int debug,
776 	      struct uvnode *devvp, void (func(daddr_t, FINFO *)))
777 {
778 	FINFO *fp;
779 	int bc;			/* Bytes in partial segment */
780 	int nblocks;
781 	daddr_t daddr;
782 	IINFO *iibase, *iip;
783 	struct ubuf *bp;
784 	int i, j, k, datac, len;
785 	lfs_checkword *datap;
786 	u_int32_t ccksum;
787 
788 	/* We've already checked the sumsum, just do the data bounds and sum */
789 
790 	/* Count the blocks. */
791 	nblocks = howmany(lfs_ss_getninos(fs, sp), LFS_INOPB(fs));
792 	bc = nblocks << (lfs_sb_getversion(fs) > 1 ? lfs_sb_getffshift(fs) : lfs_sb_getbshift(fs));
793 	assert(bc >= 0);
794 
795 	fp = SEGSUM_FINFOBASE(fs, sp);
796 	for (i = 0; i < lfs_ss_getnfinfo(fs, sp); i++) {
797 		nblocks += lfs_fi_getnblocks(fs, fp);
798 		bc += lfs_fi_getlastlength(fs, fp) + ((lfs_fi_getnblocks(fs, fp) - 1)
799 					   << lfs_sb_getbshift(fs));
800 		assert(bc >= 0);
801 		fp = NEXT_FINFO(fs, fp);
802 		if (((char *)fp) - (char *)sp > lfs_sb_getsumsize(fs))
803 			return 0;
804 	}
805 	datap = emalloc(nblocks * sizeof(*datap));
806 	datac = 0;
807 
808 	iibase = SEGSUM_IINFOSTART(fs, sp);
809 
810 	iip = iibase;
811 	daddr = pseg_addr + lfs_btofsb(fs, lfs_sb_getsumsize(fs));
812 	fp = SEGSUM_FINFOBASE(fs, sp);
813 	for (i = 0, j = 0;
814 	     i < lfs_ss_getnfinfo(fs, sp) || j < howmany(lfs_ss_getninos(fs, sp), LFS_INOPB(fs)); i++) {
815 		if (i >= lfs_ss_getnfinfo(fs, sp) && lfs_ii_getblock(fs, iip) != daddr) {
816 			pwarn("Not enough inode blocks in pseg at 0x%jx: "
817 			      "found %d, wanted %d\n",
818 			      pseg_addr, j, howmany(lfs_ss_getninos(fs, sp),
819 						    LFS_INOPB(fs)));
820 			if (debug)
821 				pwarn("iip=0x%jx, daddr=0x%jx\n",
822 				    (uintmax_t)lfs_ii_getblock(fs, iip),
823 				    (intmax_t)daddr);
824 			break;
825 		}
826 		while (j < howmany(lfs_ss_getninos(fs, sp), LFS_INOPB(fs)) && lfs_ii_getblock(fs, iip) == daddr) {
827 			bread(devvp, LFS_FSBTODB(fs, daddr), lfs_sb_getibsize(fs),
828 			    0, &bp);
829 			datap[datac++] = ((lfs_checkword *)bp->b_data)[0];
830 			brelse(bp, 0);
831 
832 			++j;
833 			daddr += lfs_btofsb(fs, lfs_sb_getibsize(fs));
834 			iip = NEXTLOWER_IINFO(fs, iip);
835 		}
836 		if (i < lfs_ss_getnfinfo(fs, sp)) {
837 			if (func)
838 				func(daddr, fp);
839 			for (k = 0; k < lfs_fi_getnblocks(fs, fp); k++) {
840 				len = (k == lfs_fi_getnblocks(fs, fp) - 1 ?
841 				       lfs_fi_getlastlength(fs, fp)
842 				       : lfs_sb_getbsize(fs));
843 				bread(devvp, LFS_FSBTODB(fs, daddr), len,
844 				    0, &bp);
845 				datap[datac++] = ((lfs_checkword *)bp->b_data)[0];
846 				brelse(bp, 0);
847 				daddr += lfs_btofsb(fs, len);
848 			}
849 			fp = NEXT_FINFO(fs, fp);
850 		}
851 	}
852 
853 	if (datac != nblocks) {
854 		pwarn("Partial segment at 0x%jx expected %d blocks counted %d\n",
855 		    (intmax_t)pseg_addr, nblocks, datac);
856 	}
857 	ccksum = cksum(datap, nblocks * sizeof(datap[0]));
858 	/* Check the data checksum */
859 	if (ccksum != lfs_ss_getdatasum(fs, sp)) {
860 		pwarn("Partial segment at 0x%jx data checksum"
861 		      " mismatch: given 0x%x, computed 0x%x\n",
862 		      (uintmax_t)pseg_addr, lfs_ss_getdatasum(fs, sp), ccksum);
863 		free(datap);
864 		return 0;
865 	}
866 	free(datap);
867 	assert(bc >= 0);
868 	return bc;
869 }
870 
871 /* print message and exit */
872 void
873 my_vpanic(int fatal, const char *fmt, va_list ap)
874 {
875         (void) vprintf(fmt, ap);
876 	exit(8);
877 }
878 
879 void
880 call_panic(const char *fmt, ...)
881 {
882 	va_list ap;
883 
884 	va_start(ap, fmt);
885         panic_func(1, fmt, ap);
886 	va_end(ap);
887 }
888 
889 /* Allocate a new inode. */
890 struct uvnode *
891 lfs_valloc(struct lfs *fs, ino_t ino)
892 {
893 	struct ubuf *bp, *cbp;
894 	IFILE *ifp;
895 	ino_t new_ino;
896 	int error;
897 	CLEANERINFO *cip;
898 
899 	/* Get the head of the freelist. */
900 	LFS_GET_HEADFREE(fs, cip, cbp, &new_ino);
901 
902 	/*
903 	 * Remove the inode from the free list and write the new start
904 	 * of the free list into the superblock.
905 	 */
906 	LFS_IENTRY(ifp, fs, new_ino, bp);
907 	if (lfs_if_getdaddr(fs, ifp) != LFS_UNUSED_DADDR)
908 		panic("lfs_valloc: inuse inode %d on the free list", new_ino);
909 	LFS_PUT_HEADFREE(fs, cip, cbp, lfs_if_getnextfree(fs, ifp));
910 
911 	brelse(bp, 0);
912 
913 	/* Extend IFILE so that the next lfs_valloc will succeed. */
914 	if (lfs_sb_getfreehd(fs) == LFS_UNUSED_INUM) {
915 		if ((error = extend_ifile(fs)) != 0) {
916 			LFS_PUT_HEADFREE(fs, cip, cbp, new_ino);
917 			return NULL;
918 		}
919 	}
920 
921 	/* Set superblock modified bit and increment file count. */
922         sbdirty();
923 	lfs_sb_addnfiles(fs, 1);
924 
925         return lfs_raw_vget(fs, ino, fs->lfs_devvp->v_fd, 0x0);
926 }
927 
928 #ifdef IN_FSCK_LFS
929 void reset_maxino(ino_t);
930 #endif
931 
932 /*
933  * Add a new block to the Ifile, to accommodate future file creations.
934  */
935 int
936 extend_ifile(struct lfs *fs)
937 {
938 	struct uvnode *vp;
939 	struct inode *ip;
940 	IFILE64 *ifp64;
941 	IFILE32 *ifp32;
942 	IFILE_V1 *ifp_v1;
943 	struct ubuf *bp, *cbp;
944 	daddr_t i, blkno, max;
945 	ino_t oldlast;
946 	CLEANERINFO *cip;
947 
948 	vp = fs->lfs_ivnode;
949 	ip = VTOI(vp);
950 	blkno = lfs_lblkno(fs, lfs_dino_getsize(fs, ip->i_din));
951 
952 	lfs_balloc(vp, lfs_dino_getsize(fs, ip->i_din), lfs_sb_getbsize(fs), &bp);
953 	lfs_dino_setsize(fs, ip->i_din,
954 	    lfs_dino_getsize(fs, ip->i_din) + lfs_sb_getbsize(fs));
955 	ip->i_flag |= IN_MODIFIED;
956 
957 	i = (blkno - lfs_sb_getsegtabsz(fs) - lfs_sb_getcleansz(fs)) *
958 		lfs_sb_getifpb(fs);
959 	LFS_GET_HEADFREE(fs, cip, cbp, &oldlast);
960 	LFS_PUT_HEADFREE(fs, cip, cbp, i);
961 	max = i + lfs_sb_getifpb(fs);
962 	lfs_sb_subbfree(fs, lfs_btofsb(fs, lfs_sb_getbsize(fs)));
963 
964 	if (fs->lfs_is64) {
965 		for (ifp64 = (IFILE64 *)bp->b_data; i < max; ++ifp64) {
966 			ifp64->if_version = 1;
967 			ifp64->if_daddr = LFS_UNUSED_DADDR;
968 			ifp64->if_nextfree = ++i;
969 		}
970 		ifp64--;
971 		ifp64->if_nextfree = oldlast;
972 	} else if (lfs_sb_getversion(fs) > 1) {
973 		for (ifp32 = (IFILE32 *)bp->b_data; i < max; ++ifp32) {
974 			ifp32->if_version = 1;
975 			ifp32->if_daddr = LFS_UNUSED_DADDR;
976 			ifp32->if_nextfree = ++i;
977 		}
978 		ifp32--;
979 		ifp32->if_nextfree = oldlast;
980 	} else {
981 		for (ifp_v1 = (IFILE_V1 *)bp->b_data; i < max; ++ifp_v1) {
982 			ifp_v1->if_version = 1;
983 			ifp_v1->if_daddr = LFS_UNUSED_DADDR;
984 			ifp_v1->if_nextfree = ++i;
985 		}
986 		ifp_v1--;
987 		ifp_v1->if_nextfree = oldlast;
988 	}
989 	LFS_PUT_TAILFREE(fs, cip, cbp, max - 1);
990 
991 	LFS_BWRITE_LOG(bp);
992 
993 #ifdef IN_FSCK_LFS
994 	reset_maxino(((lfs_dino_getsize(fs, ip->i_din) >> lfs_sb_getbshift(fs))
995 		      - lfs_sb_getsegtabsz(fs)
996 		      - lfs_sb_getcleansz(fs)) * lfs_sb_getifpb(fs));
997 #endif
998 	return 0;
999 }
1000 
1001 /*
1002  * Allocate a block, and to inode and filesystem block accounting for it
1003  * and for any indirect blocks the may need to be created in order for
1004  * this block to be created.
1005  *
1006  * Blocks which have never been accounted for (i.e., which "do not exist")
1007  * have disk address 0, which is translated by ulfs_bmap to the special value
1008  * UNASSIGNED == -1, as in the historical ULFS.
1009  *
1010  * Blocks which have been accounted for but which have not yet been written
1011  * to disk are given the new special disk address UNWRITTEN == -2, so that
1012  * they can be differentiated from completely new blocks.
1013  */
1014 int
1015 lfs_balloc(struct uvnode *vp, off_t startoffset, int iosize, struct ubuf **bpp)
1016 {
1017 	int offset;
1018 	daddr_t daddr, idaddr;
1019 	struct ubuf *ibp, *bp;
1020 	struct inode *ip;
1021 	struct lfs *fs;
1022 	struct indir indirs[ULFS_NIADDR+2], *idp;
1023 	daddr_t	lbn, lastblock;
1024 	int bcount;
1025 	int error, frags, i, nsize, osize, num;
1026 
1027 	ip = VTOI(vp);
1028 	fs = ip->i_lfs;
1029 	offset = lfs_blkoff(fs, startoffset);
1030 	lbn = lfs_lblkno(fs, startoffset);
1031 
1032 	/*
1033 	 * Three cases: it's a block beyond the end of file, it's a block in
1034 	 * the file that may or may not have been assigned a disk address or
1035 	 * we're writing an entire block.
1036 	 *
1037 	 * Note, if the daddr is UNWRITTEN, the block already exists in
1038 	 * the cache (it was read or written earlier).	If so, make sure
1039 	 * we don't count it as a new block or zero out its contents. If
1040 	 * it did not, make sure we allocate any necessary indirect
1041 	 * blocks.
1042 	 *
1043 	 * If we are writing a block beyond the end of the file, we need to
1044 	 * check if the old last block was a fragment.	If it was, we need
1045 	 * to rewrite it.
1046 	 */
1047 
1048 	if (bpp)
1049 		*bpp = NULL;
1050 
1051 	/* Check for block beyond end of file and fragment extension needed. */
1052 	lastblock = lfs_lblkno(fs, lfs_dino_getsize(fs, ip->i_din));
1053 	if (lastblock < ULFS_NDADDR && lastblock < lbn) {
1054 		osize = lfs_blksize(fs, ip, lastblock);
1055 		if (osize < lfs_sb_getbsize(fs) && osize > 0) {
1056 			if ((error = lfs_fragextend(vp, osize, lfs_sb_getbsize(fs),
1057 						    lastblock,
1058 						    (bpp ? &bp : NULL))))
1059 				return (error);
1060 			lfs_dino_setsize(fs, ip->i_din, (lastblock + 1) * lfs_sb_getbsize(fs));
1061 			ip->i_flag |= IN_CHANGE | IN_UPDATE;
1062 			if (bpp)
1063 				(void) VOP_BWRITE(bp);
1064 		}
1065 	}
1066 
1067 	/*
1068 	 * If the block we are writing is a direct block, it's the last
1069 	 * block in the file, and offset + iosize is less than a full
1070 	 * block, we can write one or more fragments.  There are two cases:
1071 	 * the block is brand new and we should allocate it the correct
1072 	 * size or it already exists and contains some fragments and
1073 	 * may need to extend it.
1074 	 */
1075 	if (lbn < ULFS_NDADDR && lfs_lblkno(fs, lfs_dino_getsize(fs, ip->i_din)) <= lbn) {
1076 		osize = lfs_blksize(fs, ip, lbn);
1077 		nsize = lfs_fragroundup(fs, offset + iosize);
1078 		if (lfs_lblktosize(fs, lbn) >= lfs_dino_getsize(fs, ip->i_din)) {
1079 			/* Brand new block or fragment */
1080 			frags = lfs_numfrags(fs, nsize);
1081 			if (bpp) {
1082 				*bpp = bp = getblk(vp, lbn, nsize);
1083 				bp->b_blkno = UNWRITTEN;
1084 			}
1085 			ip->i_lfs_effnblks += frags;
1086 			lfs_sb_subbfree(fs, frags);
1087 			lfs_dino_setdb(fs, ip->i_din, lbn, UNWRITTEN);
1088 		} else {
1089 			if (nsize <= osize) {
1090 				/* No need to extend */
1091 				if (bpp && (error = bread(vp, lbn, osize,
1092 				    0, &bp)))
1093 					return error;
1094 			} else {
1095 				/* Extend existing block */
1096 				if ((error =
1097 				     lfs_fragextend(vp, osize, nsize, lbn,
1098 						    (bpp ? &bp : NULL))))
1099 					return error;
1100 			}
1101 			if (bpp)
1102 				*bpp = bp;
1103 		}
1104 		return 0;
1105 	}
1106 
1107 	error = ulfs_bmaparray(fs, vp, lbn, &daddr, &indirs[0], &num);
1108 	if (error)
1109 		return (error);
1110 
1111 	/*
1112 	 * Do byte accounting all at once, so we can gracefully fail *before*
1113 	 * we start assigning blocks.
1114 	 */
1115         frags = LFS_FSBTODB(fs, 1); /* frags = VFSTOULFS(vp->v_mount)->um_seqinc; */
1116 	bcount = 0;
1117 	if (daddr == UNASSIGNED) {
1118 		bcount = frags;
1119 	}
1120 	for (i = 1; i < num; ++i) {
1121 		if (!indirs[i].in_exists) {
1122 			bcount += frags;
1123 		}
1124 	}
1125 	lfs_sb_subbfree(fs, bcount);
1126 	ip->i_lfs_effnblks += bcount;
1127 
1128 	if (daddr == UNASSIGNED) {
1129 		if (num > 0 && lfs_dino_getib(fs, ip->i_din, indirs[0].in_off) == 0) {
1130 			lfs_dino_setib(fs, ip->i_din, indirs[0].in_off,
1131 				       UNWRITTEN);
1132 		}
1133 
1134 		/*
1135 		 * Create new indirect blocks if necessary
1136 		 */
1137 		if (num > 1) {
1138 			idaddr = lfs_dino_getib(fs, ip->i_din, indirs[0].in_off);
1139 			for (i = 1; i < num; ++i) {
1140 				ibp = getblk(vp, indirs[i].in_lbn,
1141 				    lfs_sb_getbsize(fs));
1142 				if (!indirs[i].in_exists) {
1143 					memset(ibp->b_data, 0, ibp->b_bufsize);
1144 					ibp->b_blkno = UNWRITTEN;
1145 				} else if (!(ibp->b_flags & (B_DELWRI | B_DONE))) {
1146 					ibp->b_blkno = LFS_FSBTODB(fs, idaddr);
1147 					ibp->b_flags |= B_READ;
1148 					VOP_STRATEGY(ibp);
1149 				}
1150 				/*
1151 				 * This block exists, but the next one may not.
1152 				 * If that is the case mark it UNWRITTEN to
1153                                  * keep the accounting straight.
1154 				 */
1155 				if (lfs_iblock_get(fs, ibp->b_data,
1156 						indirs[i].in_off) == 0)
1157 					lfs_iblock_set(fs, ibp->b_data,
1158 						indirs[i].in_off, UNWRITTEN);
1159 				idaddr = lfs_iblock_get(fs, ibp->b_data,
1160 						indirs[i].in_off);
1161 				if ((error = VOP_BWRITE(ibp)))
1162 					return error;
1163 			}
1164 		}
1165 	}
1166 
1167 
1168 	/*
1169 	 * Get the existing block from the cache, if requested.
1170 	 */
1171 	if (bpp)
1172 		*bpp = bp = getblk(vp, lbn, lfs_blksize(fs, ip, lbn));
1173 
1174 	/*
1175 	 * The block we are writing may be a brand new block
1176 	 * in which case we need to do accounting.
1177 	 *
1178 	 * We can tell a truly new block because ulfs_bmaparray will say
1179 	 * it is UNASSIGNED.  Once we allocate it we will assign it the
1180 	 * disk address UNWRITTEN.
1181 	 */
1182 	if (daddr == UNASSIGNED) {
1183 		if (bpp) {
1184 			/* Note the new address */
1185 			bp->b_blkno = UNWRITTEN;
1186 		}
1187 
1188 		switch (num) {
1189 		    case 0:
1190 			lfs_dino_setdb(fs, ip->i_din, lbn, UNWRITTEN);
1191 			break;
1192 		    case 1:
1193 			lfs_dino_setib(fs, ip->i_din, indirs[0].in_off,
1194 				       UNWRITTEN);
1195 			break;
1196 		    default:
1197 			idp = &indirs[num - 1];
1198 			if (bread(vp, idp->in_lbn, lfs_sb_getbsize(fs), 0, &ibp))
1199 				panic("lfs_balloc: bread bno %lld",
1200 				    (long long)idp->in_lbn);
1201 			lfs_iblock_set(fs, ibp->b_data, idp->in_off,
1202 				       UNWRITTEN);
1203 			VOP_BWRITE(ibp);
1204 		}
1205 	} else if (bpp && !(bp->b_flags & (B_DONE|B_DELWRI))) {
1206 		/*
1207 		 * Not a brand new block, also not in the cache;
1208 		 * read it in from disk.
1209 		 */
1210 		if (iosize == lfs_sb_getbsize(fs))
1211 			/* Optimization: I/O is unnecessary. */
1212 			bp->b_blkno = daddr;
1213 		else {
1214 			/*
1215 			 * We need to read the block to preserve the
1216 			 * existing bytes.
1217 			 */
1218 			bp->b_blkno = daddr;
1219 			bp->b_flags |= B_READ;
1220 			VOP_STRATEGY(bp);
1221 			return 0;
1222 		}
1223 	}
1224 
1225 	return (0);
1226 }
1227 
1228 int
1229 lfs_fragextend(struct uvnode *vp, int osize, int nsize, daddr_t lbn,
1230                struct ubuf **bpp)
1231 {
1232 	struct inode *ip;
1233 	struct lfs *fs;
1234 	int frags;
1235 	int error;
1236 
1237 	ip = VTOI(vp);
1238 	fs = ip->i_lfs;
1239 	frags = (long)lfs_numfrags(fs, nsize - osize);
1240 	error = 0;
1241 
1242 	/*
1243 	 * If we are not asked to actually return the block, all we need
1244 	 * to do is allocate space for it.  UBC will handle dirtying the
1245 	 * appropriate things and making sure it all goes to disk.
1246 	 * Don't bother to read in that case.
1247 	 */
1248 	if (bpp && (error = bread(vp, lbn, osize, 0, bpp))) {
1249 		brelse(*bpp, 0);
1250 		goto out;
1251 	}
1252 
1253 	lfs_sb_subbfree(fs, frags);
1254 	ip->i_lfs_effnblks += frags;
1255 	ip->i_flag |= IN_CHANGE | IN_UPDATE;
1256 
1257 	if (bpp) {
1258 		(*bpp)->b_data = erealloc((*bpp)->b_data, nsize);
1259 		(void)memset((*bpp)->b_data + osize, 0, nsize - osize);
1260 	}
1261 
1262     out:
1263 	return (error);
1264 }
1265