xref: /netbsd-src/sbin/fsck_lfs/lfs.c (revision 63aea4bd5b445e491ff0389fe27ec78b3099dba3)
1 /* $NetBSD: lfs.c,v 1.69 2015/10/15 06:24:55 dholland 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 <stdio.h>
84 #include <stdlib.h>
85 #include <string.h>
86 #include <unistd.h>
87 #include <util.h>
88 
89 #include "bufcache.h"
90 #include "vnode.h"
91 #include "lfs_user.h"
92 #include "segwrite.h"
93 #include "kernelops.h"
94 
95 #define panic call_panic
96 
97 extern u_int32_t cksum(void *, size_t);
98 extern u_int32_t lfs_sb_cksum(struct lfs *);
99 extern void pwarn(const char *, ...);
100 
101 extern struct uvnodelst vnodelist;
102 extern struct uvnodelst getvnodelist[VNODE_HASH_MAX];
103 extern int nvnodes;
104 
105 long dev_bsize = DEV_BSIZE;
106 
107 static int
108 lfs_fragextend(struct uvnode *, int, int, daddr_t, struct ubuf **);
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 = kops.ko_pread(bp->b_vp->v_fd, bp->b_data, bp->b_bcount,
124 		    bp->b_blkno * dev_bsize);
125 		if (count == bp->b_bcount)
126 			bp->b_flags |= B_DONE;
127 	} else {
128 		count = kops.ko_pwrite(bp->b_vp->v_fd, bp->b_data, bp->b_bcount,
129 		    bp->b_blkno * dev_bsize);
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 		lfs_sb_subavail(fs, lfs_btofsb(fs, bp->b_bcount));
148 	}
149 	bp->b_flags |= B_DELWRI | B_LOCKED;
150 	reassignbuf(bp, bp->b_vp);
151 	brelse(bp, 0);
152 	return 0;
153 }
154 
155 /*
156  * ulfs_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 ulfs_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[ULFS_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 < ULFS_NDADDR) {
174 		if (nump != NULL)
175 			*nump = 0;
176 		*bnp = LFS_FSBTODB(fs, lfs_dino_getdb(fs, ip->i_din, 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 = ulfs_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 = lfs_dino_getib(fs, ip->i_din, 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, 0);
206 
207 		xap->in_exists = 1;
208 		bp = getblk(vp, metalbn, lfs_sb_getbsize(fs));
209 
210 		if (!(bp->b_flags & (B_DONE | B_DELWRI))) {
211 			bp->b_blkno = LFS_FSBTODB(fs, daddr);
212 			bp->b_flags |= B_READ;
213 			VOP_STRATEGY(bp);
214 		}
215 		daddr = lfs_iblock_get(fs, bp->b_data, xap->in_off);
216 	}
217 	if (bp)
218 		brelse(bp, 0);
219 
220 	daddr = LFS_FSBTODB(fs, 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 di_ib and
232  * once with the offset into the page itself.
233  */
234 int
235 ulfs_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 = lfs_sb_getnindir(fs); 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 ULFS_NIADDR - i is the number of levels
259 	 * of indirection needed to locate the requested block. */
260 
261 	bn -= ULFS_NDADDR;
262 	for (lbc = 0, i = ULFS_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 + ULFS_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 = ULFS_NIADDR - i;
282 	ap->in_exists = 0;
283 	ap++;
284 	for (++numlevels; i <= ULFS_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) & (lfs_sb_getnindir(fs) - 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 ulfs_bmaparray(vp->v_fs, vp, lbn, daddrp, NULL, NULL);
310 }
311 
312 /* Search a block for a specific dinode. */
313 union lfs_dinode *
314 lfs_ifind(struct lfs *fs, ino_t ino, struct ubuf *bp)
315 {
316 	union lfs_dinode *ldip;
317 	unsigned i, num;
318 
319 	num = LFS_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 (i = num; i-- > 0; ) {
327 		ldip = DINO_IN_BLOCK(fs, bp->b_data, i);
328 		if (lfs_dino_getinumber(fs, ldip) == ino)
329 			return (ldip);
330 	}
331 	return NULL;
332 }
333 
334 /*
335  * lfs_raw_vget makes us a new vnode from the inode at the given disk address.
336  * XXX it currently loses atime information.
337  */
338 struct uvnode *
339 lfs_raw_vget(struct lfs * fs, ino_t ino, int fd, daddr_t daddr)
340 {
341 	struct uvnode *vp;
342 	struct inode *ip;
343 	union lfs_dinode *dip;
344 	struct ubuf *bp;
345 	int i, hash;
346 
347 	vp = ecalloc(1, sizeof(*vp));
348 	vp->v_fd = fd;
349 	vp->v_fs = fs;
350 	vp->v_usecount = 0;
351 	vp->v_strategy_op = lfs_vop_strategy;
352 	vp->v_bwrite_op = lfs_vop_bwrite;
353 	vp->v_bmap_op = lfs_vop_bmap;
354 	LIST_INIT(&vp->v_cleanblkhd);
355 	LIST_INIT(&vp->v_dirtyblkhd);
356 
357 	ip = ecalloc(1, sizeof(*ip));
358 
359 	ip->i_din = dip = ecalloc(1, sizeof(*dip));
360 
361 	/* Initialize the inode -- from lfs_vcreate. */
362 	ip->inode_ext.lfs = ecalloc(1, sizeof(*ip->inode_ext.lfs));
363 	vp->v_data = ip;
364 	/* ip->i_vnode = vp; */
365 	ip->i_number = ino;
366 	ip->i_lockf = 0;
367 	ip->i_lfs_effnblks = 0;
368 	ip->i_flag = 0;
369 
370 	/* Load inode block and find inode */
371 	if (daddr > 0) {
372 		bread(fs->lfs_devvp, LFS_FSBTODB(fs, daddr), lfs_sb_getibsize(fs),
373 		    0, &bp);
374 		bp->b_flags |= B_AGE;
375 		dip = lfs_ifind(fs, ino, bp);
376 		if (dip == NULL) {
377 			brelse(bp, 0);
378 			free(ip);
379 			free(vp);
380 			return NULL;
381 		}
382 		lfs_copy_dinode(fs, ip->i_din, dip);
383 		brelse(bp, 0);
384 	}
385 	ip->i_number = ino;
386 	/* ip->i_devvp = fs->lfs_devvp; */
387 	ip->i_lfs = fs;
388 
389 	ip->i_lfs_effnblks = lfs_dino_getblocks(fs, ip->i_din);
390 	ip->i_lfs_osize = lfs_dino_getsize(fs, ip->i_din);
391 #if 0
392 	if (lfs_sb_getversion(fs) > 1) {
393 		lfs_dino_setatime(fs, ip->i_din, ts.tv_sec);
394 		lfs_dino_setatimensec(fs, ip->i_din, ts.tv_nsec);
395 	}
396 #endif
397 
398 	memset(ip->i_lfs_fragsize, 0, ULFS_NDADDR * sizeof(*ip->i_lfs_fragsize));
399 	for (i = 0; i < ULFS_NDADDR; i++)
400 		if (lfs_dino_getdb(fs, ip->i_din, i) != 0)
401 			ip->i_lfs_fragsize[i] = lfs_blksize(fs, ip, i);
402 
403 	++nvnodes;
404 	hash = ((int)(intptr_t)fs + ino) & (VNODE_HASH_MAX - 1);
405 	LIST_INSERT_HEAD(&getvnodelist[hash], vp, v_getvnodes);
406 	LIST_INSERT_HEAD(&vnodelist, vp, v_mntvnodes);
407 
408 	return vp;
409 }
410 
411 static struct uvnode *
412 lfs_vget(void *vfs, ino_t ino)
413 {
414 	struct lfs *fs = (struct lfs *)vfs;
415 	daddr_t daddr;
416 	struct ubuf *bp;
417 	IFILE *ifp;
418 
419 	LFS_IENTRY(ifp, fs, ino, bp);
420 	daddr = lfs_if_getdaddr(fs, ifp);
421 	brelse(bp, 0);
422 	if (daddr <= 0 || lfs_dtosn(fs, daddr) >= lfs_sb_getnseg(fs))
423 		return NULL;
424 	return lfs_raw_vget(fs, ino, fs->lfs_ivnode->v_fd, daddr);
425 }
426 
427 /*
428  * Check superblock magic number and checksum.
429  * Sets lfs_is64 and lfs_dobyteswap.
430  */
431 static int
432 check_sb(struct lfs *fs)
433 {
434 	u_int32_t checksum;
435 	u_int32_t magic;
436 
437 	/* we can read the magic out of either the 32-bit or 64-bit dlfs */
438 	magic = fs->lfs_dlfs_u.u_32.dlfs_magic;
439 
440 	switch (magic) {
441 	    case LFS_MAGIC:
442 		fs->lfs_is64 = false;
443 		fs->lfs_dobyteswap = false;
444 		break;
445 	    case LFS_MAGIC_SWAPPED:
446 		fs->lfs_is64 = false;
447 		fs->lfs_dobyteswap = true;
448 		break;
449 	    case LFS64_MAGIC:
450 		fs->lfs_is64 = true;
451 		fs->lfs_dobyteswap = false;
452 		break;
453 	    case LFS64_MAGIC_SWAPPED:
454 		fs->lfs_is64 = true;
455 		fs->lfs_dobyteswap = true;
456 		break;
457 	    default:
458 		printf("Superblock magic number (0x%lx) does not match "
459 		       "expected 0x%lx\n", (unsigned long) magic,
460 		       (unsigned long) LFS_MAGIC);
461 		return 1;
462 	}
463 
464 	/* checksum */
465 	checksum = lfs_sb_cksum(fs);
466 	if (lfs_sb_getcksum(fs) != checksum) {
467 		printf("Superblock checksum (%lx) does not match computed checksum (%lx)\n",
468 		    (unsigned long) lfs_sb_getcksum(fs), (unsigned long) checksum);
469 		return 1;
470 	}
471 	return 0;
472 }
473 
474 /* Initialize LFS library; load superblocks and choose which to use. */
475 struct lfs *
476 lfs_init(int devfd, daddr_t sblkno, daddr_t idaddr, int dummy_read, int debug)
477 {
478 	struct uvnode *devvp;
479 	struct ubuf *bp;
480 	int tryalt;
481 	struct lfs *fs, *altfs;
482 
483 	vfs_init();
484 
485 	devvp = ecalloc(1, sizeof(*devvp));
486 	devvp->v_fs = NULL;
487 	devvp->v_fd = devfd;
488 	devvp->v_strategy_op = raw_vop_strategy;
489 	devvp->v_bwrite_op = raw_vop_bwrite;
490 	devvp->v_bmap_op = raw_vop_bmap;
491 	LIST_INIT(&devvp->v_cleanblkhd);
492 	LIST_INIT(&devvp->v_dirtyblkhd);
493 
494 	tryalt = 0;
495 	if (dummy_read) {
496 		if (sblkno == 0)
497 			sblkno = LFS_LABELPAD / dev_bsize;
498 		fs = ecalloc(1, sizeof(*fs));
499 		fs->lfs_devvp = devvp;
500 	} else {
501 		if (sblkno == 0) {
502 			sblkno = LFS_LABELPAD / dev_bsize;
503 			tryalt = 1;
504 		} else if (debug) {
505 			printf("No -b flag given, not attempting to verify checkpoint\n");
506 		}
507 
508 		dev_bsize = DEV_BSIZE;
509 
510 		(void)bread(devvp, sblkno, LFS_SBPAD, 0, &bp);
511 		fs = ecalloc(1, sizeof(*fs));
512 		__CTASSERT(sizeof(struct dlfs) == sizeof(struct dlfs64));
513 		memcpy(&fs->lfs_dlfs_u, bp->b_data, sizeof(struct dlfs));
514 		fs->lfs_devvp = devvp;
515 		bp->b_flags |= B_INVAL;
516 		brelse(bp, 0);
517 
518 		dev_bsize = lfs_sb_getfsize(fs) >> lfs_sb_getfsbtodb(fs);
519 
520 		if (tryalt) {
521 			(void)bread(devvp, LFS_FSBTODB(fs, lfs_sb_getsboff(fs, 1)),
522 		    	LFS_SBPAD, 0, &bp);
523 			altfs = ecalloc(1, sizeof(*altfs));
524 			memcpy(&altfs->lfs_dlfs_u, bp->b_data,
525 			       sizeof(struct dlfs));
526 			altfs->lfs_devvp = devvp;
527 			bp->b_flags |= B_INVAL;
528 			brelse(bp, 0);
529 
530 			if (check_sb(fs) || lfs_sb_getidaddr(fs) <= 0) {
531 				if (debug)
532 					printf("Primary superblock is no good, using first alternate\n");
533 				free(fs);
534 				fs = altfs;
535 			} else {
536 				/* If both superblocks check out, try verification */
537 				if (check_sb(altfs)) {
538 					if (debug)
539 						printf("First alternate superblock is no good, using primary\n");
540 					free(altfs);
541 				} else {
542 					if (lfs_verify(fs, altfs, devvp, debug) == fs) {
543 						free(altfs);
544 					} else {
545 						free(fs);
546 						fs = altfs;
547 					}
548 				}
549 			}
550 		}
551 		if (check_sb(fs)) {
552 			free(fs);
553 			return NULL;
554 		}
555 	}
556 
557 	/* Compatibility */
558 	if (lfs_sb_getversion(fs) < 2) {
559 		lfs_sb_setsumsize(fs, LFS_V1_SUMMARY_SIZE);
560 		lfs_sb_setibsize(fs, lfs_sb_getbsize(fs));
561 		lfs_sb_sets0addr(fs, lfs_sb_getsboff(fs, 0));
562 		lfs_sb_settstamp(fs, lfs_sb_getotstamp(fs));
563 		lfs_sb_setfsbtodb(fs, 0);
564 	}
565 
566 	if (!dummy_read) {
567 		fs->lfs_suflags = emalloc(2 * sizeof(u_int32_t *));
568 		fs->lfs_suflags[0] = emalloc(lfs_sb_getnseg(fs) * sizeof(u_int32_t));
569 		fs->lfs_suflags[1] = emalloc(lfs_sb_getnseg(fs) * sizeof(u_int32_t));
570 	}
571 
572 	if (idaddr == 0)
573 		idaddr = lfs_sb_getidaddr(fs);
574 	else
575 		lfs_sb_setidaddr(fs, idaddr);
576 	/* NB: If dummy_read!=0, idaddr==0 here so we get a fake inode. */
577 	fs->lfs_ivnode = lfs_raw_vget(fs, LFS_IFILE_INUM,
578 		devvp->v_fd, idaddr);
579 	if (fs->lfs_ivnode == NULL)
580 		return NULL;
581 
582 	register_vget((void *)fs, lfs_vget);
583 
584 	return fs;
585 }
586 
587 /*
588  * Check partial segment validity between fs->lfs_offset and the given goal.
589  *
590  * If goal == 0, just keep on going until the segments stop making sense,
591  * and return the address of the last valid partial segment.
592  *
593  * If goal != 0, return the address of the first partial segment that failed,
594  * or "goal" if we reached it without failure (the partial segment *at* goal
595  * need not be valid).
596  */
597 daddr_t
598 try_verify(struct lfs *osb, struct uvnode *devvp, daddr_t goal, int debug)
599 {
600 	daddr_t daddr, odaddr;
601 	SEGSUM *sp;
602 	int i, bc, hitclean;
603 	struct ubuf *bp;
604 	daddr_t nodirop_daddr;
605 	u_int64_t serial;
606 
607 	bc = 0;
608 	hitclean = 0;
609 	odaddr = -1;
610 	daddr = lfs_sb_getoffset(osb);
611 	nodirop_daddr = daddr;
612 	serial = lfs_sb_getserial(osb);
613 	while (daddr != goal) {
614 		/*
615 		 * Don't mistakenly read a superblock, if there is one here.
616 		 */
617 		if (lfs_sntod(osb, lfs_dtosn(osb, daddr)) == daddr) {
618 			if (daddr == lfs_sb_gets0addr(osb))
619 				daddr += lfs_btofsb(osb, LFS_LABELPAD);
620 			for (i = 0; i < LFS_MAXNUMSB; i++) {
621 				/* XXX dholland 20150828 I think this is wrong */
622 				if (lfs_sb_getsboff(osb, i) < daddr)
623 					break;
624 				if (lfs_sb_getsboff(osb, i) == daddr)
625 					daddr += lfs_btofsb(osb, LFS_SBPAD);
626 			}
627 		}
628 
629 		/* Read in summary block */
630 		bread(devvp, LFS_FSBTODB(osb, daddr), lfs_sb_getsumsize(osb),
631 		    0, &bp);
632 		sp = (SEGSUM *)bp->b_data;
633 
634 		/*
635 		 * Check for a valid segment summary belonging to our fs.
636 		 */
637 		if (lfs_ss_getmagic(osb, sp) != SS_MAGIC ||
638 		    lfs_ss_getident(osb, sp) != lfs_sb_getident(osb) ||
639 		    lfs_ss_getserial(osb, sp) < serial ||	/* XXX strengthen this */
640 		    lfs_ss_getsumsum(osb, sp) !=
641 		            cksum((char *)sp + lfs_ss_getsumstart(osb),
642 				  lfs_sb_getsumsize(osb) - lfs_ss_getsumstart(osb))) {
643 			brelse(bp, 0);
644 			if (debug) {
645 				if (lfs_ss_getmagic(osb, sp) != SS_MAGIC)
646 					pwarn("pseg at 0x%jx: "
647 					      "wrong magic number\n",
648 					      (uintmax_t)daddr);
649 				else if (lfs_ss_getident(osb, sp) != lfs_sb_getident(osb))
650 					pwarn("pseg at 0x%jx: "
651 					      "expected ident %jx, got %jx\n",
652 					      (uintmax_t)daddr,
653 					      (uintmax_t)lfs_ss_getident(osb, sp),
654 					      (uintmax_t)lfs_sb_getident(osb));
655 				else if (lfs_ss_getserial(osb, sp) >= serial)
656 					pwarn("pseg at 0x%jx: "
657 					      "serial %d < %d\n",
658 					      (uintmax_t)daddr,
659 					      (int)lfs_ss_getserial(osb, sp), (int)serial);
660 				else
661 					pwarn("pseg at 0x%jx: "
662 					      "summary checksum wrong\n",
663 					      (uintmax_t)daddr);
664 			}
665 			break;
666 		}
667 		if (debug && lfs_ss_getserial(osb, sp) != serial)
668 			pwarn("warning, serial=%d ss_serial=%d\n",
669 				(int)serial, (int)lfs_ss_getserial(osb, sp));
670 		++serial;
671 		bc = check_summary(osb, sp, daddr, debug, devvp, NULL);
672 		if (bc == 0) {
673 			brelse(bp, 0);
674 			break;
675 		}
676 		if (debug)
677 			pwarn("summary good: 0x%x/%d\n", (uintmax_t)daddr,
678 			      (int)lfs_ss_getserial(osb, sp));
679 		assert (bc > 0);
680 		odaddr = daddr;
681 		daddr += lfs_btofsb(osb, lfs_sb_getsumsize(osb) + bc);
682 		if (lfs_dtosn(osb, odaddr) != lfs_dtosn(osb, daddr) ||
683 		    lfs_dtosn(osb, daddr) != lfs_dtosn(osb, daddr +
684 			lfs_btofsb(osb, lfs_sb_getsumsize(osb) + lfs_sb_getbsize(osb)) - 1)) {
685 			daddr = lfs_ss_getnext(osb, sp);
686 		}
687 
688 		/*
689 		 * Check for the beginning and ending of a sequence of
690 		 * dirops.  Writes from the cleaner never involve new
691 		 * information, and are always checkpoints; so don't try
692 		 * to roll forward through them.  Likewise, psegs written
693 		 * by a previous roll-forward attempt are not interesting.
694 		 */
695 		if (lfs_ss_getflags(osb, sp) & (SS_CLEAN | SS_RFW))
696 			hitclean = 1;
697 		if (hitclean == 0 && (lfs_ss_getflags(osb, sp) & SS_CONT) == 0)
698 			nodirop_daddr = daddr;
699 
700 		brelse(bp, 0);
701 	}
702 
703 	if (goal == 0)
704 		return nodirop_daddr;
705 	else
706 		return daddr;
707 }
708 
709 /* Use try_verify to check whether the newer superblock is valid. */
710 struct lfs *
711 lfs_verify(struct lfs *sb0, struct lfs *sb1, struct uvnode *devvp, int debug)
712 {
713 	daddr_t daddr;
714 	struct lfs *osb, *nsb;
715 
716 	/*
717 	 * Verify the checkpoint of the newer superblock,
718 	 * if the timestamp/serial number of the two superblocks is
719 	 * different.
720 	 */
721 
722 	osb = NULL;
723 	if (debug)
724 		pwarn("sb0 %ju, sb1 %ju",
725 		      (uintmax_t) lfs_sb_getserial(sb0),
726 		      (uintmax_t) lfs_sb_getserial(sb1));
727 
728 	if ((lfs_sb_getversion(sb0) == 1 &&
729 		lfs_sb_getotstamp(sb0) != lfs_sb_getotstamp(sb1)) ||
730 	    (lfs_sb_getversion(sb0) > 1 &&
731 		lfs_sb_getserial(sb0) != lfs_sb_getserial(sb1))) {
732 		if (lfs_sb_getversion(sb0) == 1) {
733 			if (lfs_sb_getotstamp(sb0) > lfs_sb_getotstamp(sb1)) {
734 				osb = sb1;
735 				nsb = sb0;
736 			} else {
737 				osb = sb0;
738 				nsb = sb1;
739 			}
740 		} else {
741 			if (lfs_sb_getserial(sb0) > lfs_sb_getserial(sb1)) {
742 				osb = sb1;
743 				nsb = sb0;
744 			} else {
745 				osb = sb0;
746 				nsb = sb1;
747 			}
748 		}
749 		if (debug) {
750 			printf("Attempting to verify newer checkpoint...");
751 			fflush(stdout);
752 		}
753 		daddr = try_verify(osb, devvp, lfs_sb_getoffset(nsb), debug);
754 
755 		if (debug)
756 			printf("done.\n");
757 		if (daddr == lfs_sb_getoffset(nsb)) {
758 			pwarn("** Newer checkpoint verified; recovered %jd seconds of data\n",
759 			    (intmax_t)(lfs_sb_gettstamp(nsb) - lfs_sb_gettstamp(osb)));
760 			sbdirty();
761 		} else {
762 			pwarn("** Newer checkpoint invalid; lost %jd seconds of data\n", (intmax_t)(lfs_sb_gettstamp(nsb) - lfs_sb_gettstamp(osb)));
763 		}
764 		return (daddr == lfs_sb_getoffset(nsb) ? nsb : osb);
765 	}
766 	/* Nothing to check */
767 	return osb;
768 }
769 
770 /* Verify a partial-segment summary; return the number of bytes on disk. */
771 int
772 check_summary(struct lfs *fs, SEGSUM *sp, daddr_t pseg_addr, int debug,
773 	      struct uvnode *devvp, void (func(daddr_t, FINFO *)))
774 {
775 	FINFO *fp;
776 	int bc;			/* Bytes in partial segment */
777 	int nblocks;
778 	daddr_t daddr;
779 	IINFO *iibase, *iip;
780 	struct ubuf *bp;
781 	int i, j, k, datac, len;
782 	lfs_checkword *datap;
783 	u_int32_t ccksum;
784 
785 	/* We've already checked the sumsum, just do the data bounds and sum */
786 
787 	/* Count the blocks. */
788 	nblocks = howmany(lfs_ss_getninos(fs, sp), LFS_INOPB(fs));
789 	bc = nblocks << (lfs_sb_getversion(fs) > 1 ? lfs_sb_getffshift(fs) : lfs_sb_getbshift(fs));
790 	assert(bc >= 0);
791 
792 	fp = SEGSUM_FINFOBASE(fs, sp);
793 	for (i = 0; i < lfs_ss_getnfinfo(fs, sp); i++) {
794 		nblocks += lfs_fi_getnblocks(fs, fp);
795 		bc += lfs_fi_getlastlength(fs, fp) + ((lfs_fi_getnblocks(fs, fp) - 1)
796 					   << lfs_sb_getbshift(fs));
797 		assert(bc >= 0);
798 		fp = NEXT_FINFO(fs, fp);
799 		if (((char *)fp) - (char *)sp > lfs_sb_getsumsize(fs))
800 			return 0;
801 	}
802 	datap = emalloc(nblocks * sizeof(*datap));
803 	datac = 0;
804 
805 	iibase = SEGSUM_IINFOSTART(fs, sp);
806 
807 	iip = iibase;
808 	daddr = pseg_addr + lfs_btofsb(fs, lfs_sb_getsumsize(fs));
809 	fp = SEGSUM_FINFOBASE(fs, sp);
810 	for (i = 0, j = 0;
811 	     i < lfs_ss_getnfinfo(fs, sp) || j < howmany(lfs_ss_getninos(fs, sp), LFS_INOPB(fs)); i++) {
812 		if (i >= lfs_ss_getnfinfo(fs, sp) && lfs_ii_getblock(fs, iip) != daddr) {
813 			pwarn("Not enough inode blocks in pseg at 0x%jx: "
814 			      "found %d, wanted %d\n",
815 			      pseg_addr, j, howmany(lfs_ss_getninos(fs, sp),
816 						    LFS_INOPB(fs)));
817 			if (debug)
818 				pwarn("iip=0x%jx, daddr=0x%jx\n",
819 				    (uintmax_t)lfs_ii_getblock(fs, iip),
820 				    (intmax_t)daddr);
821 			break;
822 		}
823 		while (j < howmany(lfs_ss_getninos(fs, sp), LFS_INOPB(fs)) && lfs_ii_getblock(fs, iip) == daddr) {
824 			bread(devvp, LFS_FSBTODB(fs, daddr), lfs_sb_getibsize(fs),
825 			    0, &bp);
826 			datap[datac++] = ((lfs_checkword *)bp->b_data)[0];
827 			brelse(bp, 0);
828 
829 			++j;
830 			daddr += lfs_btofsb(fs, lfs_sb_getibsize(fs));
831 			iip = NEXTLOWER_IINFO(fs, iip);
832 		}
833 		if (i < lfs_ss_getnfinfo(fs, sp)) {
834 			if (func)
835 				func(daddr, fp);
836 			for (k = 0; k < lfs_fi_getnblocks(fs, fp); k++) {
837 				len = (k == lfs_fi_getnblocks(fs, fp) - 1 ?
838 				       lfs_fi_getlastlength(fs, fp)
839 				       : lfs_sb_getbsize(fs));
840 				bread(devvp, LFS_FSBTODB(fs, daddr), len,
841 				    0, &bp);
842 				datap[datac++] = ((lfs_checkword *)bp->b_data)[0];
843 				brelse(bp, 0);
844 				daddr += lfs_btofsb(fs, len);
845 			}
846 			fp = NEXT_FINFO(fs, fp);
847 		}
848 	}
849 
850 	if (datac != nblocks) {
851 		pwarn("Partial segment at 0x%jx expected %d blocks counted %d\n",
852 		    (intmax_t)pseg_addr, nblocks, datac);
853 	}
854 	ccksum = cksum(datap, nblocks * sizeof(datap[0]));
855 	/* Check the data checksum */
856 	if (ccksum != lfs_ss_getdatasum(fs, sp)) {
857 		pwarn("Partial segment at 0x%jx data checksum"
858 		      " mismatch: given 0x%x, computed 0x%x\n",
859 		      (uintmax_t)pseg_addr, lfs_ss_getdatasum(fs, sp), ccksum);
860 		free(datap);
861 		return 0;
862 	}
863 	free(datap);
864 	assert(bc >= 0);
865 	return bc;
866 }
867 
868 /* print message and exit */
869 void
870 my_vpanic(int fatal, const char *fmt, va_list ap)
871 {
872         (void) vprintf(fmt, ap);
873 	exit(8);
874 }
875 
876 void
877 call_panic(const char *fmt, ...)
878 {
879 	va_list ap;
880 
881 	va_start(ap, fmt);
882         panic_func(1, fmt, ap);
883 	va_end(ap);
884 }
885 
886 /* Allocate a new inode. */
887 struct uvnode *
888 lfs_valloc(struct lfs *fs, ino_t ino)
889 {
890 	struct ubuf *bp, *cbp;
891 	IFILE *ifp;
892 	ino_t new_ino;
893 	int error;
894 	CLEANERINFO *cip;
895 
896 	/* Get the head of the freelist. */
897 	LFS_GET_HEADFREE(fs, cip, cbp, &new_ino);
898 
899 	/*
900 	 * Remove the inode from the free list and write the new start
901 	 * of the free list into the superblock.
902 	 */
903 	LFS_IENTRY(ifp, fs, new_ino, bp);
904 	if (lfs_if_getdaddr(fs, ifp) != LFS_UNUSED_DADDR)
905 		panic("lfs_valloc: inuse inode %d on the free list", new_ino);
906 	LFS_PUT_HEADFREE(fs, cip, cbp, lfs_if_getnextfree(fs, ifp));
907 
908 	brelse(bp, 0);
909 
910 	/* Extend IFILE so that the next lfs_valloc will succeed. */
911 	if (lfs_sb_getfreehd(fs) == LFS_UNUSED_INUM) {
912 		if ((error = extend_ifile(fs)) != 0) {
913 			LFS_PUT_HEADFREE(fs, cip, cbp, new_ino);
914 			return NULL;
915 		}
916 	}
917 
918 	/* Set superblock modified bit and increment file count. */
919         sbdirty();
920 	lfs_sb_addnfiles(fs, 1);
921 
922         return lfs_raw_vget(fs, ino, fs->lfs_devvp->v_fd, 0x0);
923 }
924 
925 #ifdef IN_FSCK_LFS
926 void reset_maxino(ino_t);
927 #endif
928 
929 /*
930  * Add a new block to the Ifile, to accommodate future file creations.
931  */
932 int
933 extend_ifile(struct lfs *fs)
934 {
935 	struct uvnode *vp;
936 	struct inode *ip;
937 	IFILE64 *ifp64;
938 	IFILE32 *ifp32;
939 	IFILE_V1 *ifp_v1;
940 	struct ubuf *bp, *cbp;
941 	daddr_t i, blkno, max;
942 	ino_t oldlast;
943 	CLEANERINFO *cip;
944 
945 	vp = fs->lfs_ivnode;
946 	ip = VTOI(vp);
947 	blkno = lfs_lblkno(fs, lfs_dino_getsize(fs, ip->i_din));
948 
949 	lfs_balloc(vp, lfs_dino_getsize(fs, ip->i_din), lfs_sb_getbsize(fs), &bp);
950 	lfs_dino_setsize(fs, ip->i_din,
951 	    lfs_dino_getsize(fs, ip->i_din) + lfs_sb_getbsize(fs));
952 	ip->i_flag |= IN_MODIFIED;
953 
954 	i = (blkno - lfs_sb_getsegtabsz(fs) - lfs_sb_getcleansz(fs)) *
955 		lfs_sb_getifpb(fs);
956 	LFS_GET_HEADFREE(fs, cip, cbp, &oldlast);
957 	LFS_PUT_HEADFREE(fs, cip, cbp, i);
958 	max = i + lfs_sb_getifpb(fs);
959 	lfs_sb_subbfree(fs, lfs_btofsb(fs, lfs_sb_getbsize(fs)));
960 
961 	if (fs->lfs_is64) {
962 		for (ifp64 = (IFILE64 *)bp->b_data; i < max; ++ifp64) {
963 			ifp64->if_version = 1;
964 			ifp64->if_daddr = LFS_UNUSED_DADDR;
965 			ifp64->if_nextfree = ++i;
966 		}
967 		ifp64--;
968 		ifp64->if_nextfree = oldlast;
969 	} else if (lfs_sb_getversion(fs) > 1) {
970 		for (ifp32 = (IFILE32 *)bp->b_data; i < max; ++ifp32) {
971 			ifp32->if_version = 1;
972 			ifp32->if_daddr = LFS_UNUSED_DADDR;
973 			ifp32->if_nextfree = ++i;
974 		}
975 		ifp32--;
976 		ifp32->if_nextfree = oldlast;
977 	} else {
978 		for (ifp_v1 = (IFILE_V1 *)bp->b_data; i < max; ++ifp_v1) {
979 			ifp_v1->if_version = 1;
980 			ifp_v1->if_daddr = LFS_UNUSED_DADDR;
981 			ifp_v1->if_nextfree = ++i;
982 		}
983 		ifp_v1--;
984 		ifp_v1->if_nextfree = oldlast;
985 	}
986 	LFS_PUT_TAILFREE(fs, cip, cbp, max - 1);
987 
988 	LFS_BWRITE_LOG(bp);
989 
990 #ifdef IN_FSCK_LFS
991 	reset_maxino(((lfs_dino_getsize(fs, ip->i_din) >> lfs_sb_getbshift(fs))
992 		      - lfs_sb_getsegtabsz(fs)
993 		      - lfs_sb_getcleansz(fs)) * lfs_sb_getifpb(fs));
994 #endif
995 	return 0;
996 }
997 
998 /*
999  * Allocate a block, and to inode and filesystem block accounting for it
1000  * and for any indirect blocks the may need to be created in order for
1001  * this block to be created.
1002  *
1003  * Blocks which have never been accounted for (i.e., which "do not exist")
1004  * have disk address 0, which is translated by ulfs_bmap to the special value
1005  * UNASSIGNED == -1, as in the historical ULFS.
1006  *
1007  * Blocks which have been accounted for but which have not yet been written
1008  * to disk are given the new special disk address UNWRITTEN == -2, so that
1009  * they can be differentiated from completely new blocks.
1010  */
1011 int
1012 lfs_balloc(struct uvnode *vp, off_t startoffset, int iosize, struct ubuf **bpp)
1013 {
1014 	int offset;
1015 	daddr_t daddr, idaddr;
1016 	struct ubuf *ibp, *bp;
1017 	struct inode *ip;
1018 	struct lfs *fs;
1019 	struct indir indirs[ULFS_NIADDR+2], *idp;
1020 	daddr_t	lbn, lastblock;
1021 	int bcount;
1022 	int error, frags, i, nsize, osize, num;
1023 
1024 	ip = VTOI(vp);
1025 	fs = ip->i_lfs;
1026 	offset = lfs_blkoff(fs, startoffset);
1027 	lbn = lfs_lblkno(fs, startoffset);
1028 
1029 	/*
1030 	 * Three cases: it's a block beyond the end of file, it's a block in
1031 	 * the file that may or may not have been assigned a disk address or
1032 	 * we're writing an entire block.
1033 	 *
1034 	 * Note, if the daddr is UNWRITTEN, the block already exists in
1035 	 * the cache (it was read or written earlier).	If so, make sure
1036 	 * we don't count it as a new block or zero out its contents. If
1037 	 * it did not, make sure we allocate any necessary indirect
1038 	 * blocks.
1039 	 *
1040 	 * If we are writing a block beyond the end of the file, we need to
1041 	 * check if the old last block was a fragment.	If it was, we need
1042 	 * to rewrite it.
1043 	 */
1044 
1045 	if (bpp)
1046 		*bpp = NULL;
1047 
1048 	/* Check for block beyond end of file and fragment extension needed. */
1049 	lastblock = lfs_lblkno(fs, lfs_dino_getsize(fs, ip->i_din));
1050 	if (lastblock < ULFS_NDADDR && lastblock < lbn) {
1051 		osize = lfs_blksize(fs, ip, lastblock);
1052 		if (osize < lfs_sb_getbsize(fs) && osize > 0) {
1053 			if ((error = lfs_fragextend(vp, osize, lfs_sb_getbsize(fs),
1054 						    lastblock,
1055 						    (bpp ? &bp : NULL))))
1056 				return (error);
1057 			lfs_dino_setsize(fs, ip->i_din, (lastblock + 1) * lfs_sb_getbsize(fs));
1058 			ip->i_flag |= IN_CHANGE | IN_UPDATE;
1059 			if (bpp)
1060 				(void) VOP_BWRITE(bp);
1061 		}
1062 	}
1063 
1064 	/*
1065 	 * If the block we are writing is a direct block, it's the last
1066 	 * block in the file, and offset + iosize is less than a full
1067 	 * block, we can write one or more fragments.  There are two cases:
1068 	 * the block is brand new and we should allocate it the correct
1069 	 * size or it already exists and contains some fragments and
1070 	 * may need to extend it.
1071 	 */
1072 	if (lbn < ULFS_NDADDR && lfs_lblkno(fs, lfs_dino_getsize(fs, ip->i_din)) <= lbn) {
1073 		osize = lfs_blksize(fs, ip, lbn);
1074 		nsize = lfs_fragroundup(fs, offset + iosize);
1075 		if (lfs_lblktosize(fs, lbn) >= lfs_dino_getsize(fs, ip->i_din)) {
1076 			/* Brand new block or fragment */
1077 			frags = lfs_numfrags(fs, nsize);
1078 			if (bpp) {
1079 				*bpp = bp = getblk(vp, lbn, nsize);
1080 				bp->b_blkno = UNWRITTEN;
1081 			}
1082 			ip->i_lfs_effnblks += frags;
1083 			lfs_sb_subbfree(fs, frags);
1084 			lfs_dino_setdb(fs, ip->i_din, lbn, UNWRITTEN);
1085 		} else {
1086 			if (nsize <= osize) {
1087 				/* No need to extend */
1088 				if (bpp && (error = bread(vp, lbn, osize,
1089 				    0, &bp)))
1090 					return error;
1091 			} else {
1092 				/* Extend existing block */
1093 				if ((error =
1094 				     lfs_fragextend(vp, osize, nsize, lbn,
1095 						    (bpp ? &bp : NULL))))
1096 					return error;
1097 			}
1098 			if (bpp)
1099 				*bpp = bp;
1100 		}
1101 		return 0;
1102 	}
1103 
1104 	error = ulfs_bmaparray(fs, vp, lbn, &daddr, &indirs[0], &num);
1105 	if (error)
1106 		return (error);
1107 
1108 	/*
1109 	 * Do byte accounting all at once, so we can gracefully fail *before*
1110 	 * we start assigning blocks.
1111 	 */
1112         frags = LFS_FSBTODB(fs, 1); /* frags = VFSTOULFS(vp->v_mount)->um_seqinc; */
1113 	bcount = 0;
1114 	if (daddr == UNASSIGNED) {
1115 		bcount = frags;
1116 	}
1117 	for (i = 1; i < num; ++i) {
1118 		if (!indirs[i].in_exists) {
1119 			bcount += frags;
1120 		}
1121 	}
1122 	lfs_sb_subbfree(fs, bcount);
1123 	ip->i_lfs_effnblks += bcount;
1124 
1125 	if (daddr == UNASSIGNED) {
1126 		if (num > 0 && lfs_dino_getib(fs, ip->i_din, indirs[0].in_off) == 0) {
1127 			lfs_dino_setib(fs, ip->i_din, indirs[0].in_off,
1128 				       UNWRITTEN);
1129 		}
1130 
1131 		/*
1132 		 * Create new indirect blocks if necessary
1133 		 */
1134 		if (num > 1) {
1135 			idaddr = lfs_dino_getib(fs, ip->i_din, indirs[0].in_off);
1136 			for (i = 1; i < num; ++i) {
1137 				ibp = getblk(vp, indirs[i].in_lbn,
1138 				    lfs_sb_getbsize(fs));
1139 				if (!indirs[i].in_exists) {
1140 					memset(ibp->b_data, 0, ibp->b_bufsize);
1141 					ibp->b_blkno = UNWRITTEN;
1142 				} else if (!(ibp->b_flags & (B_DELWRI | B_DONE))) {
1143 					ibp->b_blkno = LFS_FSBTODB(fs, idaddr);
1144 					ibp->b_flags |= B_READ;
1145 					VOP_STRATEGY(ibp);
1146 				}
1147 				/*
1148 				 * This block exists, but the next one may not.
1149 				 * If that is the case mark it UNWRITTEN to
1150                                  * keep the accounting straight.
1151 				 */
1152 				if (lfs_iblock_get(fs, ibp->b_data,
1153 						indirs[i].in_off) == 0)
1154 					lfs_iblock_set(fs, ibp->b_data,
1155 						indirs[i].in_off, UNWRITTEN);
1156 				idaddr = lfs_iblock_get(fs, ibp->b_data,
1157 						indirs[i].in_off);
1158 				if ((error = VOP_BWRITE(ibp)))
1159 					return error;
1160 			}
1161 		}
1162 	}
1163 
1164 
1165 	/*
1166 	 * Get the existing block from the cache, if requested.
1167 	 */
1168 	if (bpp)
1169 		*bpp = bp = getblk(vp, lbn, lfs_blksize(fs, ip, lbn));
1170 
1171 	/*
1172 	 * The block we are writing may be a brand new block
1173 	 * in which case we need to do accounting.
1174 	 *
1175 	 * We can tell a truly new block because ulfs_bmaparray will say
1176 	 * it is UNASSIGNED.  Once we allocate it we will assign it the
1177 	 * disk address UNWRITTEN.
1178 	 */
1179 	if (daddr == UNASSIGNED) {
1180 		if (bpp) {
1181 			/* Note the new address */
1182 			bp->b_blkno = UNWRITTEN;
1183 		}
1184 
1185 		switch (num) {
1186 		    case 0:
1187 			lfs_dino_setdb(fs, ip->i_din, lbn, UNWRITTEN);
1188 			break;
1189 		    case 1:
1190 			lfs_dino_setib(fs, ip->i_din, indirs[0].in_off,
1191 				       UNWRITTEN);
1192 			break;
1193 		    default:
1194 			idp = &indirs[num - 1];
1195 			if (bread(vp, idp->in_lbn, lfs_sb_getbsize(fs), 0, &ibp))
1196 				panic("lfs_balloc: bread bno %lld",
1197 				    (long long)idp->in_lbn);
1198 			lfs_iblock_set(fs, ibp->b_data, idp->in_off,
1199 				       UNWRITTEN);
1200 			VOP_BWRITE(ibp);
1201 		}
1202 	} else if (bpp && !(bp->b_flags & (B_DONE|B_DELWRI))) {
1203 		/*
1204 		 * Not a brand new block, also not in the cache;
1205 		 * read it in from disk.
1206 		 */
1207 		if (iosize == lfs_sb_getbsize(fs))
1208 			/* Optimization: I/O is unnecessary. */
1209 			bp->b_blkno = daddr;
1210 		else {
1211 			/*
1212 			 * We need to read the block to preserve the
1213 			 * existing bytes.
1214 			 */
1215 			bp->b_blkno = daddr;
1216 			bp->b_flags |= B_READ;
1217 			VOP_STRATEGY(bp);
1218 			return 0;
1219 		}
1220 	}
1221 
1222 	return (0);
1223 }
1224 
1225 int
1226 lfs_fragextend(struct uvnode *vp, int osize, int nsize, daddr_t lbn,
1227                struct ubuf **bpp)
1228 {
1229 	struct inode *ip;
1230 	struct lfs *fs;
1231 	int frags;
1232 	int error;
1233 
1234 	ip = VTOI(vp);
1235 	fs = ip->i_lfs;
1236 	frags = (long)lfs_numfrags(fs, nsize - osize);
1237 	error = 0;
1238 
1239 	/*
1240 	 * If we are not asked to actually return the block, all we need
1241 	 * to do is allocate space for it.  UBC will handle dirtying the
1242 	 * appropriate things and making sure it all goes to disk.
1243 	 * Don't bother to read in that case.
1244 	 */
1245 	if (bpp && (error = bread(vp, lbn, osize, 0, bpp))) {
1246 		brelse(*bpp, 0);
1247 		goto out;
1248 	}
1249 
1250 	lfs_sb_subbfree(fs, frags);
1251 	ip->i_lfs_effnblks += frags;
1252 	ip->i_flag |= IN_CHANGE | IN_UPDATE;
1253 
1254 	if (bpp) {
1255 		(*bpp)->b_data = erealloc((*bpp)->b_data, nsize);
1256 		(void)memset((*bpp)->b_data + osize, 0, nsize - osize);
1257 	}
1258 
1259     out:
1260 	return (error);
1261 }
1262