xref: /netbsd-src/sbin/fsck_lfs/segwrite.c (revision d710132b4b8ce7f7cccaaf660cb16aa16b4077a0)
1 /* $NetBSD: segwrite.c,v 1.3 2003/04/02 10:39:28 fvdl 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) 1991, 1993
39  *	The Regents of the University of California.  All rights reserved.
40  *
41  * Redistribution and use in source and binary forms, with or without
42  * modification, are permitted provided that the following conditions
43  * are met:
44  * 1. Redistributions of source code must retain the above copyright
45  *    notice, this list of conditions and the following disclaimer.
46  * 2. Redistributions in binary form must reproduce the above copyright
47  *    notice, this list of conditions and the following disclaimer in the
48  *    documentation and/or other materials provided with the distribution.
49  * 3. All advertising materials mentioning features or use of this software
50  *    must display the following acknowledgement:
51  *	This product includes software developed by the University of
52  *	California, Berkeley and its contributors.
53  * 4. Neither the name of the University nor the names of its contributors
54  *    may be used to endorse or promote products derived from this software
55  *    without specific prior written permission.
56  *
57  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
58  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
59  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
60  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
61  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
62  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
63  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
64  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
65  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
66  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
67  * SUCH DAMAGE.
68  *
69  *	@(#)lfs_segment.c	8.10 (Berkeley) 6/10/95
70  */
71 
72 /*
73  * Partial segment writer, taken from the kernel and adapted for userland.
74  */
75 #include <sys/types.h>
76 #include <sys/param.h>
77 #include <sys/time.h>
78 #include <sys/buf.h>
79 #include <sys/mount.h>
80 
81 #include <ufs/ufs/inode.h>
82 #include <ufs/ufs/ufsmount.h>
83 
84 /* Override certain things to make <ufs/lfs/lfs.h> work */
85 #undef simple_lock
86 #define simple_lock(x)
87 #undef simple_unlock
88 #define simple_unlock(x)
89 #define vnode uvnode
90 #define buf ubuf
91 #define panic call_panic
92 
93 #include <ufs/lfs/lfs.h>
94 
95 #include <assert.h>
96 #include <stdio.h>
97 #include <stdlib.h>
98 #include <string.h>
99 #include <err.h>
100 #include <errno.h>
101 
102 #include "bufcache.h"
103 #include "vnode.h"
104 #include "lfs.h"
105 #include "segwrite.h"
106 
107 /* Compatibility definitions */
108 extern off_t locked_queue_bytes;
109 int locked_queue_count;
110 off_t written_bytes = 0;
111 off_t written_data = 0;
112 off_t written_indir = 0;
113 off_t written_dev = 0;
114 int written_inodes = 0;
115 
116 /* Global variables */
117 time_t write_time;
118 
119 extern u_int32_t cksum(void *, size_t);
120 extern u_int32_t lfs_sb_cksum(struct dlfs *);
121 
122 /*
123  * Logical block number match routines used when traversing the dirty block
124  * chain.
125  */
126 int
127 lfs_match_data(struct lfs * fs, struct ubuf * bp)
128 {
129 	return (bp->b_lblkno >= 0);
130 }
131 
132 int
133 lfs_match_indir(struct lfs * fs, struct ubuf * bp)
134 {
135 	daddr_t lbn;
136 
137 	lbn = bp->b_lblkno;
138 	return (lbn < 0 && (-lbn - NDADDR) % NINDIR(fs) == 0);
139 }
140 
141 int
142 lfs_match_dindir(struct lfs * fs, struct ubuf * bp)
143 {
144 	daddr_t lbn;
145 
146 	lbn = bp->b_lblkno;
147 	return (lbn < 0 && (-lbn - NDADDR) % NINDIR(fs) == 1);
148 }
149 
150 int
151 lfs_match_tindir(struct lfs * fs, struct ubuf * bp)
152 {
153 	daddr_t lbn;
154 
155 	lbn = bp->b_lblkno;
156 	return (lbn < 0 && (-lbn - NDADDR) % NINDIR(fs) == 2);
157 }
158 
159 /*
160  * Do a checkpoint.
161  */
162 int
163 lfs_segwrite(struct lfs * fs, int flags)
164 {
165 	struct inode *ip;
166 	struct segment *sp;
167 	struct uvnode *vp;
168 	int redo;
169 
170 	lfs_seglock(fs, flags | SEGM_CKP);
171 	sp = fs->lfs_sp;
172 
173 	lfs_writevnodes(fs, sp, VN_REG);
174 	lfs_writevnodes(fs, sp, VN_DIROP);
175 	((SEGSUM *) (sp->segsum))->ss_flags &= ~(SS_CONT);
176 
177 	do {
178 		vp = fs->lfs_ivnode;
179 		fs->lfs_flags &= ~LFS_IFDIRTY;
180 		ip = VTOI(vp);
181 		if (LIST_FIRST(&vp->v_dirtyblkhd) != NULL)
182 			lfs_writefile(fs, sp, vp);
183 
184 		redo = lfs_writeinode(fs, sp, ip);
185 		redo += lfs_writeseg(fs, sp);
186 		redo += (fs->lfs_flags & LFS_IFDIRTY);
187 	} while (redo);
188 
189 	lfs_segunlock(fs);
190 #if 0
191 	printf("wrote %" PRId64 " bytes (%" PRId32 " fsb)\n",
192 		written_bytes, (ufs_daddr_t)btofsb(fs, written_bytes));
193 	printf("wrote %" PRId64 " bytes data (%" PRId32 " fsb)\n",
194 		written_data, (ufs_daddr_t)btofsb(fs, written_data));
195 	printf("wrote %" PRId64 " bytes indir (%" PRId32 " fsb)\n",
196 		written_indir, (ufs_daddr_t)btofsb(fs, written_indir));
197 	printf("wrote %" PRId64 " bytes dev (%" PRId32 " fsb)\n",
198 		written_dev, (ufs_daddr_t)btofsb(fs, written_dev));
199 	printf("wrote %d inodes (%" PRId32 " fsb)\n",
200 		written_inodes, btofsb(fs, written_inodes * fs->lfs_ibsize));
201 #endif
202 	return 0;
203 }
204 
205 /*
206  * Write the dirty blocks associated with a vnode.
207  */
208 void
209 lfs_writefile(struct lfs * fs, struct segment * sp, struct uvnode * vp)
210 {
211 	struct ubuf *bp;
212 	struct finfo *fip;
213 	struct inode *ip;
214 	IFILE *ifp;
215 
216 	ip = VTOI(vp);
217 
218 	if (sp->seg_bytes_left < fs->lfs_bsize ||
219 	    sp->sum_bytes_left < sizeof(struct finfo))
220 		(void) lfs_writeseg(fs, sp);
221 
222 	sp->sum_bytes_left -= FINFOSIZE;
223 	++((SEGSUM *) (sp->segsum))->ss_nfinfo;
224 
225 	if (vp->v_flag & VDIROP)
226 		((SEGSUM *) (sp->segsum))->ss_flags |= (SS_DIROP | SS_CONT);
227 
228 	fip = sp->fip;
229 	fip->fi_nblocks = 0;
230 	fip->fi_ino = ip->i_number;
231 	LFS_IENTRY(ifp, fs, fip->fi_ino, bp);
232 	fip->fi_version = ifp->if_version;
233 	brelse(bp);
234 
235 	lfs_gather(fs, sp, vp, lfs_match_data);
236 	lfs_gather(fs, sp, vp, lfs_match_indir);
237 	lfs_gather(fs, sp, vp, lfs_match_dindir);
238 	lfs_gather(fs, sp, vp, lfs_match_tindir);
239 
240 	fip = sp->fip;
241 	if (fip->fi_nblocks != 0) {
242 		sp->fip = (FINFO *) ((caddr_t) fip + FINFOSIZE +
243 		    sizeof(ufs_daddr_t) * (fip->fi_nblocks));
244 		sp->start_lbp = &sp->fip->fi_blocks[0];
245 	} else {
246 		sp->sum_bytes_left += FINFOSIZE;
247 		--((SEGSUM *) (sp->segsum))->ss_nfinfo;
248 	}
249 }
250 
251 int
252 lfs_writeinode(struct lfs * fs, struct segment * sp, struct inode * ip)
253 {
254 	struct ubuf *bp, *ibp;
255 	struct ufs1_dinode *cdp;
256 	IFILE *ifp;
257 	SEGUSE *sup;
258 	daddr_t daddr;
259 	ino_t ino;
260 	int error, i, ndx, fsb = 0;
261 	int redo_ifile = 0;
262 	struct timespec ts;
263 	int gotblk = 0;
264 
265 	/* Allocate a new inode block if necessary. */
266 	if ((ip->i_number != LFS_IFILE_INUM || sp->idp == NULL) &&
267 	    sp->ibp == NULL) {
268 		/* Allocate a new segment if necessary. */
269 		if (sp->seg_bytes_left < fs->lfs_ibsize ||
270 		    sp->sum_bytes_left < sizeof(ufs_daddr_t))
271 			(void) lfs_writeseg(fs, sp);
272 
273 		/* Get next inode block. */
274 		daddr = fs->lfs_offset;
275 		fs->lfs_offset += btofsb(fs, fs->lfs_ibsize);
276 		sp->ibp = *sp->cbpp++ =
277 		    getblk(fs->lfs_unlockvp, fsbtodb(fs, daddr),
278 		    fs->lfs_ibsize);
279 		sp->ibp->b_flags |= B_GATHERED;
280 		gotblk++;
281 
282 		/* Zero out inode numbers */
283 		for (i = 0; i < INOPB(fs); ++i)
284 			((struct ufs1_dinode *) sp->ibp->b_data)[i].di_inumber = 0;
285 
286 		++sp->start_bpp;
287 		fs->lfs_avail -= btofsb(fs, fs->lfs_ibsize);
288 		/* Set remaining space counters. */
289 		sp->seg_bytes_left -= fs->lfs_ibsize;
290 		sp->sum_bytes_left -= sizeof(ufs_daddr_t);
291 		ndx = fs->lfs_sumsize / sizeof(ufs_daddr_t) -
292 		    sp->ninodes / INOPB(fs) - 1;
293 		((ufs_daddr_t *) (sp->segsum))[ndx] = daddr;
294 	}
295 	/* Update the inode times and copy the inode onto the inode page. */
296 	ts.tv_nsec = 0;
297 	ts.tv_sec = write_time;
298 	/* XXX kludge --- don't redirty the ifile just to put times on it */
299 	if (ip->i_number != LFS_IFILE_INUM)
300 		LFS_ITIMES(ip, &ts, &ts, &ts);
301 
302 	/*
303 	 * If this is the Ifile, and we've already written the Ifile in this
304 	 * partial segment, just overwrite it (it's not on disk yet) and
305 	 * continue.
306 	 *
307 	 * XXX we know that the bp that we get the second time around has
308 	 * already been gathered.
309 	 */
310 	if (ip->i_number == LFS_IFILE_INUM && sp->idp) {
311 		*(sp->idp) = *ip->i_din.ffs1_din;
312 		ip->i_lfs_osize = ip->i_ffs1_size;
313 		return 0;
314 	}
315 	bp = sp->ibp;
316 	cdp = ((struct ufs1_dinode *) bp->b_data) + (sp->ninodes % INOPB(fs));
317 	*cdp = *ip->i_din.ffs1_din;
318 
319 	/* If all blocks are goig to disk, update the "size on disk" */
320 	ip->i_lfs_osize = ip->i_ffs1_size;
321 
322 	if (ip->i_number == LFS_IFILE_INUM)	/* We know sp->idp == NULL */
323 		sp->idp = ((struct ufs1_dinode *) bp->b_data) +
324 		    (sp->ninodes % INOPB(fs));
325 	if (gotblk) {
326 		LFS_LOCK_BUF(bp);
327 		brelse(bp);
328 	}
329 	/* Increment inode count in segment summary block. */
330 	++((SEGSUM *) (sp->segsum))->ss_ninos;
331 
332 	/* If this page is full, set flag to allocate a new page. */
333 	if (++sp->ninodes % INOPB(fs) == 0)
334 		sp->ibp = NULL;
335 
336 	/*
337 	 * If updating the ifile, update the super-block.  Update the disk
338 	 * address and access times for this inode in the ifile.
339 	 */
340 	ino = ip->i_number;
341 	if (ino == LFS_IFILE_INUM) {
342 		daddr = fs->lfs_idaddr;
343 		fs->lfs_idaddr = dbtofsb(fs, bp->b_blkno);
344 	} else {
345 		LFS_IENTRY(ifp, fs, ino, ibp);
346 		daddr = ifp->if_daddr;
347 		ifp->if_daddr = dbtofsb(fs, bp->b_blkno) + fsb;
348 		error = LFS_BWRITE_LOG(ibp);	/* Ifile */
349 	}
350 
351 	/*
352 	 * Account the inode: it no longer belongs to its former segment,
353 	 * though it will not belong to the new segment until that segment
354 	 * is actually written.
355 	 */
356 	if (daddr != LFS_UNUSED_DADDR) {
357 		u_int32_t oldsn = dtosn(fs, daddr);
358 		LFS_SEGENTRY(sup, fs, oldsn, bp);
359 		sup->su_nbytes -= DINODE1_SIZE;
360 		redo_ifile =
361 		    (ino == LFS_IFILE_INUM && !(bp->b_flags & B_GATHERED));
362 		if (redo_ifile)
363 			fs->lfs_flags |= LFS_IFDIRTY;
364 		LFS_WRITESEGENTRY(sup, fs, oldsn, bp);	/* Ifile */
365 	}
366 	return redo_ifile;
367 }
368 
369 int
370 lfs_gatherblock(struct segment * sp, struct ubuf * bp)
371 {
372 	struct lfs *fs;
373 	int version;
374 	int j, blksinblk;
375 
376 	/*
377 	 * If full, finish this segment.  We may be doing I/O, so
378 	 * release and reacquire the splbio().
379 	 */
380 	fs = sp->fs;
381 	blksinblk = howmany(bp->b_bcount, fs->lfs_bsize);
382 	if (sp->sum_bytes_left < sizeof(ufs_daddr_t) * blksinblk ||
383 	    sp->seg_bytes_left < bp->b_bcount) {
384 		lfs_updatemeta(sp);
385 
386 		version = sp->fip->fi_version;
387 		(void) lfs_writeseg(fs, sp);
388 
389 		sp->fip->fi_version = version;
390 		sp->fip->fi_ino = VTOI(sp->vp)->i_number;
391 		/* Add the current file to the segment summary. */
392 		++((SEGSUM *) (sp->segsum))->ss_nfinfo;
393 		sp->sum_bytes_left -= FINFOSIZE;
394 
395 		return 1;
396 	}
397 	/* Insert into the buffer list, update the FINFO block. */
398 	bp->b_flags |= B_GATHERED;
399 	/* bp->b_flags &= ~B_DONE; */
400 
401 	*sp->cbpp++ = bp;
402 	for (j = 0; j < blksinblk; j++)
403 		sp->fip->fi_blocks[sp->fip->fi_nblocks++] = bp->b_lblkno + j;
404 
405 	sp->sum_bytes_left -= sizeof(ufs_daddr_t) * blksinblk;
406 	sp->seg_bytes_left -= bp->b_bcount;
407 	return 0;
408 }
409 
410 int
411 lfs_gather(struct lfs * fs, struct segment * sp, struct uvnode * vp, int (*match) (struct lfs *, struct ubuf *))
412 {
413 	struct ubuf *bp, *nbp;
414 	int count = 0;
415 
416 	sp->vp = vp;
417 loop:
418 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
419 		nbp = LIST_NEXT(bp, b_vnbufs);
420 
421 		assert(bp->b_flags & B_DELWRI);
422 		if ((bp->b_flags & (B_BUSY | B_GATHERED)) || !match(fs, bp)) {
423 			continue;
424 		}
425 		if (lfs_gatherblock(sp, bp)) {
426 			goto loop;
427 		}
428 		count++;
429 	}
430 
431 	lfs_updatemeta(sp);
432 	sp->vp = NULL;
433 	return count;
434 }
435 
436 
437 /*
438  * Change the given block's address to ndaddr, finding its previous
439  * location using ufs_bmaparray().
440  *
441  * Account for this change in the segment table.
442  */
443 void
444 lfs_update_single(struct lfs * fs, struct segment * sp, daddr_t lbn,
445     ufs_daddr_t ndaddr, int size)
446 {
447 	SEGUSE *sup;
448 	struct ubuf *bp;
449 	struct indir a[NIADDR + 2], *ap;
450 	struct inode *ip;
451 	struct uvnode *vp;
452 	daddr_t daddr, ooff;
453 	int num, error;
454 	int bb, osize, obb;
455 
456 	vp = sp->vp;
457 	ip = VTOI(vp);
458 
459 	error = ufs_bmaparray(fs, vp, lbn, &daddr, a, &num);
460 	if (error)
461 		errx(1, "lfs_updatemeta: ufs_bmaparray returned %d looking up lbn %" PRId64 "\n", error, lbn);
462 	if (daddr > 0)
463 		daddr = dbtofsb(fs, daddr);
464 
465 	bb = fragstofsb(fs, numfrags(fs, size));
466 	switch (num) {
467 	case 0:
468 		ooff = ip->i_ffs1_db[lbn];
469 		if (ooff == UNWRITTEN)
470 			ip->i_ffs1_blocks += bb;
471 		else {
472 			/* possible fragment truncation or extension */
473 			obb = btofsb(fs, ip->i_lfs_fragsize[lbn]);
474 			ip->i_ffs1_blocks += (bb - obb);
475 		}
476 		ip->i_ffs1_db[lbn] = ndaddr;
477 		break;
478 	case 1:
479 		ooff = ip->i_ffs1_ib[a[0].in_off];
480 		if (ooff == UNWRITTEN)
481 			ip->i_ffs1_blocks += bb;
482 		ip->i_ffs1_ib[a[0].in_off] = ndaddr;
483 		break;
484 	default:
485 		ap = &a[num - 1];
486 		if (bread(vp, ap->in_lbn, fs->lfs_bsize, NULL, &bp))
487 			errx(1, "lfs_updatemeta: bread bno %" PRId64,
488 			    ap->in_lbn);
489 
490 		ooff = ((ufs_daddr_t *) bp->b_data)[ap->in_off];
491 		if (ooff == UNWRITTEN)
492 			ip->i_ffs1_blocks += bb;
493 		((ufs_daddr_t *) bp->b_data)[ap->in_off] = ndaddr;
494 		(void) VOP_BWRITE(bp);
495 	}
496 
497 	/*
498 	 * Update segment usage information, based on old size
499 	 * and location.
500 	 */
501 	if (daddr > 0) {
502 		u_int32_t oldsn = dtosn(fs, daddr);
503 		if (lbn >= 0 && lbn < NDADDR)
504 			osize = ip->i_lfs_fragsize[lbn];
505 		else
506 			osize = fs->lfs_bsize;
507 		LFS_SEGENTRY(sup, fs, oldsn, bp);
508 		sup->su_nbytes -= osize;
509 		if (!(bp->b_flags & B_GATHERED))
510 			fs->lfs_flags |= LFS_IFDIRTY;
511 		LFS_WRITESEGENTRY(sup, fs, oldsn, bp);
512 	}
513 	/*
514 	 * Now that this block has a new address, and its old
515 	 * segment no longer owns it, we can forget about its
516 	 * old size.
517 	 */
518 	if (lbn >= 0 && lbn < NDADDR)
519 		ip->i_lfs_fragsize[lbn] = size;
520 }
521 
522 /*
523  * Update the metadata that points to the blocks listed in the FINFO
524  * array.
525  */
526 void
527 lfs_updatemeta(struct segment * sp)
528 {
529 	struct ubuf *sbp;
530 	struct lfs *fs;
531 	struct uvnode *vp;
532 	daddr_t lbn;
533 	int i, nblocks, num;
534 	int bb;
535 	int bytesleft, size;
536 
537 	vp = sp->vp;
538 	nblocks = &sp->fip->fi_blocks[sp->fip->fi_nblocks] - sp->start_lbp;
539 
540 	if (vp == NULL || nblocks == 0)
541 		return;
542 
543 	/*
544 	 * This count may be high due to oversize blocks from lfs_gop_write.
545 	 * Correct for this. (XXX we should be able to keep track of these.)
546 	 */
547 	fs = sp->fs;
548 	for (i = 0; i < nblocks; i++) {
549 		if (sp->start_bpp[i] == NULL) {
550 			printf("nblocks = %d, not %d\n", i, nblocks);
551 			nblocks = i;
552 			break;
553 		}
554 		num = howmany(sp->start_bpp[i]->b_bcount, fs->lfs_bsize);
555 		nblocks -= num - 1;
556 	}
557 
558 	/*
559 	 * Sort the blocks.
560 	 */
561 	lfs_shellsort(sp->start_bpp, sp->start_lbp, nblocks, fs->lfs_bsize);
562 
563 	/*
564 	 * Record the length of the last block in case it's a fragment.
565 	 * If there are indirect blocks present, they sort last.  An
566 	 * indirect block will be lfs_bsize and its presence indicates
567 	 * that you cannot have fragments.
568 	 */
569 	sp->fip->fi_lastlength = ((sp->start_bpp[nblocks - 1]->b_bcount - 1) &
570 	    fs->lfs_bmask) + 1;
571 
572 	/*
573 	 * Assign disk addresses, and update references to the logical
574 	 * block and the segment usage information.
575 	 */
576 	for (i = nblocks; i--; ++sp->start_bpp) {
577 		sbp = *sp->start_bpp;
578 		lbn = *sp->start_lbp;
579 
580 		sbp->b_blkno = fsbtodb(fs, fs->lfs_offset);
581 
582 		/*
583 		 * If we write a frag in the wrong place, the cleaner won't
584 		 * be able to correctly identify its size later, and the
585 		 * segment will be uncleanable.	 (Even worse, it will assume
586 		 * that the indirect block that actually ends the list
587 		 * is of a smaller size!)
588 		 */
589 		if ((sbp->b_bcount & fs->lfs_bmask) && i != 0)
590 			errx(1, "lfs_updatemeta: fragment is not last block");
591 
592 		/*
593 		 * For each subblock in this possibly oversized block,
594 		 * update its address on disk.
595 		 */
596 		for (bytesleft = sbp->b_bcount; bytesleft > 0;
597 		    bytesleft -= fs->lfs_bsize) {
598 			size = MIN(bytesleft, fs->lfs_bsize);
599 			bb = fragstofsb(fs, numfrags(fs, size));
600 			lbn = *sp->start_lbp++;
601 			lfs_update_single(fs, sp, lbn, fs->lfs_offset, size);
602 			fs->lfs_offset += bb;
603 		}
604 
605 	}
606 }
607 
608 /*
609  * Start a new segment.
610  */
611 int
612 lfs_initseg(struct lfs * fs)
613 {
614 	struct segment *sp;
615 	SEGUSE *sup;
616 	SEGSUM *ssp;
617 	struct ubuf *bp, *sbp;
618 	int repeat;
619 
620 	sp = fs->lfs_sp;
621 
622 	repeat = 0;
623 
624 	/* Advance to the next segment. */
625 	if (!LFS_PARTIAL_FITS(fs)) {
626 		/* lfs_avail eats the remaining space */
627 		fs->lfs_avail -= fs->lfs_fsbpseg - (fs->lfs_offset -
628 		    fs->lfs_curseg);
629 		lfs_newseg(fs);
630 		repeat = 1;
631 		fs->lfs_offset = fs->lfs_curseg;
632 
633 		sp->seg_number = dtosn(fs, fs->lfs_curseg);
634 		sp->seg_bytes_left = fsbtob(fs, fs->lfs_fsbpseg);
635 
636 		/*
637 		 * If the segment contains a superblock, update the offset
638 		 * and summary address to skip over it.
639 		 */
640 		LFS_SEGENTRY(sup, fs, sp->seg_number, bp);
641 		if (sup->su_flags & SEGUSE_SUPERBLOCK) {
642 			fs->lfs_offset += btofsb(fs, LFS_SBPAD);
643 			sp->seg_bytes_left -= LFS_SBPAD;
644 		}
645 		brelse(bp);
646 		/* Segment zero could also contain the labelpad */
647 		if (fs->lfs_version > 1 && sp->seg_number == 0 &&
648 		    fs->lfs_start < btofsb(fs, LFS_LABELPAD)) {
649 			fs->lfs_offset += btofsb(fs, LFS_LABELPAD) - fs->lfs_start;
650 			sp->seg_bytes_left -= LFS_LABELPAD - fsbtob(fs, fs->lfs_start);
651 		}
652 	} else {
653 		sp->seg_number = dtosn(fs, fs->lfs_curseg);
654 		sp->seg_bytes_left = fsbtob(fs, fs->lfs_fsbpseg -
655 		    (fs->lfs_offset - fs->lfs_curseg));
656 	}
657 	fs->lfs_lastpseg = fs->lfs_offset;
658 
659 	sp->fs = fs;
660 	sp->ibp = NULL;
661 	sp->idp = NULL;
662 	sp->ninodes = 0;
663 	sp->ndupino = 0;
664 
665 	/* Get a new buffer for SEGSUM and enter it into the buffer list. */
666 	sp->cbpp = sp->bpp;
667 	sbp = *sp->cbpp = getblk(fs->lfs_unlockvp,
668 	    fsbtodb(fs, fs->lfs_offset), fs->lfs_sumsize);
669 	sp->segsum = sbp->b_data;
670 	memset(sp->segsum, 0, fs->lfs_sumsize);
671 	sp->start_bpp = ++sp->cbpp;
672 	fs->lfs_offset += btofsb(fs, fs->lfs_sumsize);
673 
674 	/* Set point to SEGSUM, initialize it. */
675 	ssp = sp->segsum;
676 	ssp->ss_next = fs->lfs_nextseg;
677 	ssp->ss_nfinfo = ssp->ss_ninos = 0;
678 	ssp->ss_magic = SS_MAGIC;
679 
680 	/* Set pointer to first FINFO, initialize it. */
681 	sp->fip = (struct finfo *) ((caddr_t) sp->segsum + SEGSUM_SIZE(fs));
682 	sp->fip->fi_nblocks = 0;
683 	sp->start_lbp = &sp->fip->fi_blocks[0];
684 	sp->fip->fi_lastlength = 0;
685 
686 	sp->seg_bytes_left -= fs->lfs_sumsize;
687 	sp->sum_bytes_left = fs->lfs_sumsize - SEGSUM_SIZE(fs);
688 
689 	LFS_LOCK_BUF(sbp);
690 	brelse(sbp);
691 	return repeat;
692 }
693 
694 /*
695  * Return the next segment to write.
696  */
697 void
698 lfs_newseg(struct lfs * fs)
699 {
700 	CLEANERINFO *cip;
701 	SEGUSE *sup;
702 	struct ubuf *bp;
703 	int curseg, isdirty, sn;
704 
705 	LFS_SEGENTRY(sup, fs, dtosn(fs, fs->lfs_nextseg), bp);
706 	sup->su_flags |= SEGUSE_DIRTY | SEGUSE_ACTIVE;
707 	sup->su_nbytes = 0;
708 	sup->su_nsums = 0;
709 	sup->su_ninos = 0;
710 	LFS_WRITESEGENTRY(sup, fs, dtosn(fs, fs->lfs_nextseg), bp);
711 
712 	LFS_CLEANERINFO(cip, fs, bp);
713 	--cip->clean;
714 	++cip->dirty;
715 	fs->lfs_nclean = cip->clean;
716 	LFS_SYNC_CLEANERINFO(cip, fs, bp, 1);
717 
718 	fs->lfs_lastseg = fs->lfs_curseg;
719 	fs->lfs_curseg = fs->lfs_nextseg;
720 	for (sn = curseg = dtosn(fs, fs->lfs_curseg) + fs->lfs_interleave;;) {
721 		sn = (sn + 1) % fs->lfs_nseg;
722 		if (sn == curseg)
723 			errx(1, "lfs_nextseg: no clean segments");
724 		LFS_SEGENTRY(sup, fs, sn, bp);
725 		isdirty = sup->su_flags & SEGUSE_DIRTY;
726 		brelse(bp);
727 
728 		if (!isdirty)
729 			break;
730 	}
731 
732 	++fs->lfs_nactive;
733 	fs->lfs_nextseg = sntod(fs, sn);
734 }
735 
736 
737 int
738 lfs_writeseg(struct lfs * fs, struct segment * sp)
739 {
740 	struct ubuf **bpp, *bp;
741 	SEGUSE *sup;
742 	SEGSUM *ssp;
743 	char *datap, *dp;
744 	int i;
745 	int do_again, nblocks, byteoffset;
746 	size_t el_size;
747 	u_short ninos;
748 	struct uvnode *devvp;
749 
750 	/*
751 	 * If there are no buffers other than the segment summary to write
752 	 * and it is not a checkpoint, don't do anything.  On a checkpoint,
753 	 * even if there aren't any buffers, you need to write the superblock.
754 	 */
755 	if ((nblocks = sp->cbpp - sp->bpp) == 1)
756 		return 0;
757 
758 	devvp = fs->lfs_unlockvp;
759 
760 	/* Update the segment usage information. */
761 	LFS_SEGENTRY(sup, fs, sp->seg_number, bp);
762 
763 	/* Loop through all blocks, except the segment summary. */
764 	for (bpp = sp->bpp; ++bpp < sp->cbpp;) {
765 		if ((*bpp)->b_vp != devvp) {
766 			sup->su_nbytes += (*bpp)->b_bcount;
767 		}
768 	}
769 
770 	ssp = (SEGSUM *) sp->segsum;
771 
772 	ninos = (ssp->ss_ninos + INOPB(fs) - 1) / INOPB(fs);
773 	sup->su_nbytes += ssp->ss_ninos * DINODE1_SIZE;
774 
775 	if (fs->lfs_version == 1)
776 		sup->su_olastmod = write_time;
777 	else
778 		sup->su_lastmod = write_time;
779 	sup->su_ninos += ninos;
780 	++sup->su_nsums;
781 	fs->lfs_dmeta += (btofsb(fs, fs->lfs_sumsize) + btofsb(fs, ninos *
782 		fs->lfs_ibsize));
783 	fs->lfs_avail -= btofsb(fs, fs->lfs_sumsize);
784 
785 	do_again = !(bp->b_flags & B_GATHERED);
786 	LFS_WRITESEGENTRY(sup, fs, sp->seg_number, bp);	/* Ifile */
787 
788 	/*
789 	 * Compute checksum across data and then across summary; the first
790 	 * block (the summary block) is skipped.  Set the create time here
791 	 * so that it's guaranteed to be later than the inode mod times.
792 	 */
793 	if (fs->lfs_version == 1)
794 		el_size = sizeof(u_long);
795 	else
796 		el_size = sizeof(u_int32_t);
797 	datap = dp = malloc(nblocks * el_size);
798 	for (bpp = sp->bpp, i = nblocks - 1; i--;) {
799 		++bpp;
800 		/* Loop through gop_write cluster blocks */
801 		for (byteoffset = 0; byteoffset < (*bpp)->b_bcount;
802 		    byteoffset += fs->lfs_bsize) {
803 			memcpy(dp, (*bpp)->b_data + byteoffset, el_size);
804 			dp += el_size;
805 		}
806 		bremfree(*bpp);
807 		(*bpp)->b_flags |= B_BUSY;
808 	}
809 	if (fs->lfs_version == 1)
810 		ssp->ss_ocreate = write_time;
811 	else {
812 		ssp->ss_create = write_time;
813 		ssp->ss_serial = ++fs->lfs_serial;
814 		ssp->ss_ident = fs->lfs_ident;
815 	}
816 	/* Set the summary block busy too */
817 	bremfree(*(sp->bpp));
818 	(*(sp->bpp))->b_flags |= B_BUSY;
819 
820 	ssp->ss_datasum = cksum(datap, (nblocks - 1) * el_size);
821 	ssp->ss_sumsum =
822 	    cksum(&ssp->ss_datasum, fs->lfs_sumsize - sizeof(ssp->ss_sumsum));
823 	free(datap);
824 	datap = dp = NULL;
825 	fs->lfs_bfree -= (btofsb(fs, ninos * fs->lfs_ibsize) +
826 	    btofsb(fs, fs->lfs_sumsize));
827 
828 	if (devvp == NULL)
829 		errx(1, "devvp is NULL");
830 	for (bpp = sp->bpp, i = nblocks; i; bpp++, i--) {
831 		bp = *bpp;
832 #if 0
833 		printf("i = %d, bp = %p, flags %lx, bn = %" PRIx64 "\n",
834 		       nblocks - i, bp, bp->b_flags, bp->b_blkno);
835 		printf("  vp = %p\n", bp->b_vp);
836 		if (bp->b_vp != fs->lfs_unlockvp)
837 			printf("  ino = %d lbn = %" PRId64 "\n",
838 			       VTOI(bp->b_vp)->i_number, bp->b_lblkno);
839 #endif
840 		if (bp->b_vp == fs->lfs_unlockvp)
841 			written_dev += bp->b_bcount;
842 		else {
843 			if (bp->b_lblkno >= 0)
844 				written_data += bp->b_bcount;
845 			else
846 				written_indir += bp->b_bcount;
847 		}
848 		bp->b_flags &= ~(B_DELWRI | B_READ | B_GATHERED | B_ERROR |
849 				 B_LOCKED);
850 		bwrite(bp);
851 		written_bytes += bp->b_bcount;
852 	}
853 	written_inodes += ninos;
854 
855 	return (lfs_initseg(fs) || do_again);
856 }
857 
858 /*
859  * Our own copy of shellsort.  XXX use qsort or heapsort.
860  */
861 void
862 lfs_shellsort(struct ubuf ** bp_array, ufs_daddr_t * lb_array, int nmemb, int size)
863 {
864 	static int __rsshell_increments[] = {4, 1, 0};
865 	int incr, *incrp, t1, t2;
866 	struct ubuf *bp_temp;
867 
868 	for (incrp = __rsshell_increments; (incr = *incrp++) != 0;)
869 		for (t1 = incr; t1 < nmemb; ++t1)
870 			for (t2 = t1 - incr; t2 >= 0;)
871 				if ((u_int32_t) bp_array[t2]->b_lblkno >
872 				    (u_int32_t) bp_array[t2 + incr]->b_lblkno) {
873 					bp_temp = bp_array[t2];
874 					bp_array[t2] = bp_array[t2 + incr];
875 					bp_array[t2 + incr] = bp_temp;
876 					t2 -= incr;
877 				} else
878 					break;
879 
880 	/* Reform the list of logical blocks */
881 	incr = 0;
882 	for (t1 = 0; t1 < nmemb; t1++) {
883 		for (t2 = 0; t2 * size < bp_array[t1]->b_bcount; t2++) {
884 			lb_array[incr++] = bp_array[t1]->b_lblkno + t2;
885 		}
886 	}
887 }
888 
889 
890 /*
891  * lfs_seglock --
892  *	Single thread the segment writer.
893  */
894 int
895 lfs_seglock(struct lfs * fs, unsigned long flags)
896 {
897 	struct segment *sp;
898 
899 	if (fs->lfs_seglock) {
900 		++fs->lfs_seglock;
901 		fs->lfs_sp->seg_flags |= flags;
902 		return 0;
903 	}
904 	fs->lfs_seglock = 1;
905 
906 	sp = fs->lfs_sp = (struct segment *) malloc(sizeof(*sp));
907 	sp->bpp = (struct ubuf **) malloc(fs->lfs_ssize * sizeof(struct ubuf *));
908 	sp->seg_flags = flags;
909 	sp->vp = NULL;
910 	sp->seg_iocount = 0;
911 	(void) lfs_initseg(fs);
912 
913 	/*
914 	 * Keep a cumulative count of the outstanding I/O operations.  If the
915 	 * disk drive catches up with us it could go to zero before we finish,
916 	 * so we artificially increment it by one until we've scheduled all of
917 	 * the writes we intend to do.
918 	 */
919 	++fs->lfs_iocount;
920 	return 0;
921 }
922 
923 /*
924  * lfs_segunlock --
925  *	Single thread the segment writer.
926  */
927 void
928 lfs_segunlock(struct lfs * fs)
929 {
930 	struct segment *sp;
931 	struct ubuf *bp;
932 
933 	sp = fs->lfs_sp;
934 
935 	if (fs->lfs_seglock == 1) {
936 		if (sp->bpp != sp->cbpp) {
937 			/* Free allocated segment summary */
938 			fs->lfs_offset -= btofsb(fs, fs->lfs_sumsize);
939 			bp = *sp->bpp;
940 			bremfree(bp);
941 			bp->b_flags |= B_DONE | B_INVAL;
942 			bp->b_flags &= ~B_DELWRI;
943 			reassignbuf(bp, bp->b_vp);
944 			bp->b_flags |= B_BUSY; /* XXX */
945 			brelse(bp);
946 		} else
947 			printf("unlock to 0 with no summary");
948 
949 		free(sp->bpp);
950 		sp->bpp = NULL;
951 		free(sp);
952 		fs->lfs_sp = NULL;
953 
954 		fs->lfs_nactive = 0;
955 
956 		/* Since we *know* everything's on disk, write both sbs */
957 		lfs_writesuper(fs, fs->lfs_sboffs[fs->lfs_activesb]);
958 		lfs_writesuper(fs, fs->lfs_sboffs[1 - fs->lfs_activesb]);
959 
960 		--fs->lfs_seglock;
961 		fs->lfs_lockpid = 0;
962 	} else if (fs->lfs_seglock == 0) {
963 		errx(1, "Seglock not held");
964 	} else {
965 		--fs->lfs_seglock;
966 	}
967 }
968 
969 int
970 lfs_writevnodes(struct lfs *fs, struct segment *sp, int op)
971 {
972 	struct inode *ip;
973 	struct uvnode *vp;
974 	int inodes_written = 0;
975 
976 	LIST_FOREACH(vp, &vnodelist, v_mntvnodes) {
977 		if (vp->v_bmap_op != lfs_vop_bmap)
978 			continue;
979 
980 		ip = VTOI(vp);
981 
982 		if ((op == VN_DIROP && !(vp->v_flag & VDIROP)) ||
983 		    (op != VN_DIROP && (vp->v_flag & VDIROP))) {
984 			continue;
985 		}
986 		/*
987 		 * Write the inode/file if dirty and it's not the IFILE.
988 		 */
989 		if (ip->i_flag & IN_ALLMOD || !LIST_EMPTY(&vp->v_dirtyblkhd)) {
990 			if (ip->i_number != LFS_IFILE_INUM)
991 				lfs_writefile(fs, sp, vp);
992 			(void) lfs_writeinode(fs, sp, ip);
993 			inodes_written++;
994 		}
995 	}
996 	return inodes_written;
997 }
998 
999 void
1000 lfs_writesuper(struct lfs *fs, ufs_daddr_t daddr)
1001 {
1002 	struct ubuf *bp;
1003 
1004 	/* Set timestamp of this version of the superblock */
1005 	if (fs->lfs_version == 1)
1006 		fs->lfs_otstamp = write_time;
1007 	fs->lfs_tstamp = write_time;
1008 
1009 	/* Checksum the superblock and copy it into a buffer. */
1010 	fs->lfs_cksum = lfs_sb_cksum(&(fs->lfs_dlfs));
1011 	assert(daddr > 0);
1012 	bp = getblk(fs->lfs_unlockvp, fsbtodb(fs, daddr), LFS_SBPAD);
1013 	memset(bp->b_data + sizeof(struct dlfs), 0,
1014 	    LFS_SBPAD - sizeof(struct dlfs));
1015 	*(struct dlfs *) bp->b_data = fs->lfs_dlfs;
1016 
1017 	bwrite(bp);
1018 }
1019