xref: /netbsd-src/sbin/fsck_lfs/segwrite.c (revision fad4c9f71477ae11cea2ee75ec82151ac770a534)
1 /* $NetBSD: segwrite.c,v 1.12 2006/05/23 22:35:20 jnemeth 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. Neither the name of the University nor the names of its contributors
50  *    may be used to endorse or promote products derived from this software
51  *    without specific prior written permission.
52  *
53  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
54  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
55  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
56  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
57  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
58  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
59  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
60  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
61  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
62  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
63  * SUCH DAMAGE.
64  *
65  *	@(#)lfs_segment.c	8.10 (Berkeley) 6/10/95
66  */
67 
68 /*
69  * Partial segment writer, taken from the kernel and adapted for userland.
70  */
71 #include <sys/types.h>
72 #include <sys/param.h>
73 #include <sys/time.h>
74 #include <sys/buf.h>
75 #include <sys/mount.h>
76 
77 #include <ufs/ufs/inode.h>
78 #include <ufs/ufs/ufsmount.h>
79 
80 /* Override certain things to make <ufs/lfs/lfs.h> work */
81 #define vnode uvnode
82 #define buf ubuf
83 #define panic call_panic
84 
85 #include <ufs/lfs/lfs.h>
86 
87 #include <assert.h>
88 #include <stdio.h>
89 #include <stdlib.h>
90 #include <string.h>
91 #include <err.h>
92 #include <errno.h>
93 
94 #include "bufcache.h"
95 #include "vnode.h"
96 #include "lfs_user.h"
97 #include "segwrite.h"
98 
99 /* Compatibility definitions */
100 extern off_t locked_queue_bytes;
101 int locked_queue_count;
102 off_t written_bytes = 0;
103 off_t written_data = 0;
104 off_t written_indir = 0;
105 off_t written_dev = 0;
106 int written_inodes = 0;
107 
108 /* Global variables */
109 time_t write_time;
110 
111 extern u_int32_t cksum(void *, size_t);
112 extern u_int32_t lfs_sb_cksum(struct dlfs *);
113 extern int preen;
114 
115 /*
116  * Logical block number match routines used when traversing the dirty block
117  * chain.
118  */
119 int
120 lfs_match_data(struct lfs * fs, struct ubuf * bp)
121 {
122 	return (bp->b_lblkno >= 0);
123 }
124 
125 int
126 lfs_match_indir(struct lfs * fs, struct ubuf * bp)
127 {
128 	daddr_t lbn;
129 
130 	lbn = bp->b_lblkno;
131 	return (lbn < 0 && (-lbn - NDADDR) % NINDIR(fs) == 0);
132 }
133 
134 int
135 lfs_match_dindir(struct lfs * fs, struct ubuf * bp)
136 {
137 	daddr_t lbn;
138 
139 	lbn = bp->b_lblkno;
140 	return (lbn < 0 && (-lbn - NDADDR) % NINDIR(fs) == 1);
141 }
142 
143 int
144 lfs_match_tindir(struct lfs * fs, struct ubuf * bp)
145 {
146 	daddr_t lbn;
147 
148 	lbn = bp->b_lblkno;
149 	return (lbn < 0 && (-lbn - NDADDR) % NINDIR(fs) == 2);
150 }
151 
152 /*
153  * Do a checkpoint.
154  */
155 int
156 lfs_segwrite(struct lfs * fs, int flags)
157 {
158 	struct inode *ip;
159 	struct segment *sp;
160 	struct uvnode *vp;
161 	int redo;
162 
163 	lfs_seglock(fs, flags | SEGM_CKP);
164 	sp = fs->lfs_sp;
165 
166 	lfs_writevnodes(fs, sp, VN_REG);
167 	lfs_writevnodes(fs, sp, VN_DIROP);
168 	((SEGSUM *) (sp->segsum))->ss_flags &= ~(SS_CONT);
169 
170 	do {
171 		vp = fs->lfs_ivnode;
172 		fs->lfs_flags &= ~LFS_IFDIRTY;
173 		ip = VTOI(vp);
174 		if (LIST_FIRST(&vp->v_dirtyblkhd) != NULL || fs->lfs_idaddr <= 0)
175 			lfs_writefile(fs, sp, vp);
176 
177 		redo = lfs_writeinode(fs, sp, ip);
178 		redo += lfs_writeseg(fs, sp);
179 		redo += (fs->lfs_flags & LFS_IFDIRTY);
180 	} while (redo);
181 
182 	lfs_segunlock(fs);
183 #if 0
184 	printf("wrote %" PRId64 " bytes (%" PRId32 " fsb)\n",
185 		written_bytes, (ufs_daddr_t)btofsb(fs, written_bytes));
186 	printf("wrote %" PRId64 " bytes data (%" PRId32 " fsb)\n",
187 		written_data, (ufs_daddr_t)btofsb(fs, written_data));
188 	printf("wrote %" PRId64 " bytes indir (%" PRId32 " fsb)\n",
189 		written_indir, (ufs_daddr_t)btofsb(fs, written_indir));
190 	printf("wrote %" PRId64 " bytes dev (%" PRId32 " fsb)\n",
191 		written_dev, (ufs_daddr_t)btofsb(fs, written_dev));
192 	printf("wrote %d inodes (%" PRId32 " fsb)\n",
193 		written_inodes, btofsb(fs, written_inodes * fs->lfs_ibsize));
194 #endif
195 	return 0;
196 }
197 
198 /*
199  * Write the dirty blocks associated with a vnode.
200  */
201 void
202 lfs_writefile(struct lfs * fs, struct segment * sp, struct uvnode * vp)
203 {
204 	struct ubuf *bp;
205 	struct finfo *fip;
206 	struct inode *ip;
207 	IFILE *ifp;
208 
209 	ip = VTOI(vp);
210 
211 	if (sp->seg_bytes_left < fs->lfs_bsize ||
212 	    sp->sum_bytes_left < sizeof(struct finfo))
213 		(void) lfs_writeseg(fs, sp);
214 
215 	sp->sum_bytes_left -= FINFOSIZE;
216 	++((SEGSUM *) (sp->segsum))->ss_nfinfo;
217 
218 	if (vp->v_flag & VDIROP)
219 		((SEGSUM *) (sp->segsum))->ss_flags |= (SS_DIROP | SS_CONT);
220 
221 	fip = sp->fip;
222 	fip->fi_nblocks = 0;
223 	fip->fi_ino = ip->i_number;
224 	LFS_IENTRY(ifp, fs, fip->fi_ino, bp);
225 	fip->fi_version = ifp->if_version;
226 	brelse(bp);
227 
228 	lfs_gather(fs, sp, vp, lfs_match_data);
229 	lfs_gather(fs, sp, vp, lfs_match_indir);
230 	lfs_gather(fs, sp, vp, lfs_match_dindir);
231 	lfs_gather(fs, sp, vp, lfs_match_tindir);
232 
233 	fip = sp->fip;
234 	if (fip->fi_nblocks != 0) {
235 		sp->fip = (FINFO *) ((caddr_t) fip + FINFOSIZE +
236 		    sizeof(ufs_daddr_t) * (fip->fi_nblocks));
237 		sp->start_lbp = &sp->fip->fi_blocks[0];
238 	} else {
239 		sp->sum_bytes_left += FINFOSIZE;
240 		--((SEGSUM *) (sp->segsum))->ss_nfinfo;
241 	}
242 }
243 
244 int
245 lfs_writeinode(struct lfs * fs, struct segment * sp, struct inode * ip)
246 {
247 	struct ubuf *bp, *ibp;
248 	struct ufs1_dinode *cdp;
249 	IFILE *ifp;
250 	SEGUSE *sup;
251 	daddr_t daddr;
252 	ino_t ino;
253 	int error, i, ndx, fsb = 0;
254 	int redo_ifile = 0;
255 	struct timespec ts;
256 	int gotblk = 0;
257 
258 	/* Allocate a new inode block if necessary. */
259 	if ((ip->i_number != LFS_IFILE_INUM || sp->idp == NULL) &&
260 	    sp->ibp == NULL) {
261 		/* Allocate a new segment if necessary. */
262 		if (sp->seg_bytes_left < fs->lfs_ibsize ||
263 		    sp->sum_bytes_left < sizeof(ufs_daddr_t))
264 			(void) lfs_writeseg(fs, sp);
265 
266 		/* Get next inode block. */
267 		daddr = fs->lfs_offset;
268 		fs->lfs_offset += btofsb(fs, fs->lfs_ibsize);
269 		sp->ibp = *sp->cbpp++ =
270 		    getblk(fs->lfs_devvp, fsbtodb(fs, daddr),
271 		    fs->lfs_ibsize);
272 		sp->ibp->b_flags |= B_GATHERED;
273 		gotblk++;
274 
275 		/* Zero out inode numbers */
276 		for (i = 0; i < INOPB(fs); ++i)
277 			((struct ufs1_dinode *) sp->ibp->b_data)[i].di_inumber = 0;
278 
279 		++sp->start_bpp;
280 		fs->lfs_avail -= btofsb(fs, fs->lfs_ibsize);
281 		/* Set remaining space counters. */
282 		sp->seg_bytes_left -= fs->lfs_ibsize;
283 		sp->sum_bytes_left -= sizeof(ufs_daddr_t);
284 		ndx = fs->lfs_sumsize / sizeof(ufs_daddr_t) -
285 		    sp->ninodes / INOPB(fs) - 1;
286 		((ufs_daddr_t *) (sp->segsum))[ndx] = daddr;
287 	}
288 	/* Update the inode times and copy the inode onto the inode page. */
289 	ts.tv_nsec = 0;
290 	ts.tv_sec = write_time;
291 	/* XXX kludge --- don't redirty the ifile just to put times on it */
292 	if (ip->i_number != LFS_IFILE_INUM)
293 		LFS_ITIMES(ip, &ts, &ts, &ts);
294 
295 	/*
296 	 * If this is the Ifile, and we've already written the Ifile in this
297 	 * partial segment, just overwrite it (it's not on disk yet) and
298 	 * continue.
299 	 *
300 	 * XXX we know that the bp that we get the second time around has
301 	 * already been gathered.
302 	 */
303 	if (ip->i_number == LFS_IFILE_INUM && sp->idp) {
304 		*(sp->idp) = *ip->i_din.ffs1_din;
305 		ip->i_lfs_osize = ip->i_ffs1_size;
306 		return 0;
307 	}
308 	bp = sp->ibp;
309 	cdp = ((struct ufs1_dinode *) bp->b_data) + (sp->ninodes % INOPB(fs));
310 	*cdp = *ip->i_din.ffs1_din;
311 
312 	/* If all blocks are goig to disk, update the "size on disk" */
313 	ip->i_lfs_osize = ip->i_ffs1_size;
314 
315 	if (ip->i_number == LFS_IFILE_INUM)	/* We know sp->idp == NULL */
316 		sp->idp = ((struct ufs1_dinode *) bp->b_data) +
317 		    (sp->ninodes % INOPB(fs));
318 	if (gotblk) {
319 		LFS_LOCK_BUF(bp);
320 		assert(!(bp->b_flags & B_INVAL));
321 		brelse(bp);
322 	}
323 	/* Increment inode count in segment summary block. */
324 	++((SEGSUM *) (sp->segsum))->ss_ninos;
325 
326 	/* If this page is full, set flag to allocate a new page. */
327 	if (++sp->ninodes % INOPB(fs) == 0)
328 		sp->ibp = NULL;
329 
330 	/*
331 	 * If updating the ifile, update the super-block.  Update the disk
332 	 * address and access times for this inode in the ifile.
333 	 */
334 	ino = ip->i_number;
335 	if (ino == LFS_IFILE_INUM) {
336 		daddr = fs->lfs_idaddr;
337 		fs->lfs_idaddr = dbtofsb(fs, bp->b_blkno);
338 	} else {
339 		LFS_IENTRY(ifp, fs, ino, ibp);
340 		daddr = ifp->if_daddr;
341 		ifp->if_daddr = dbtofsb(fs, bp->b_blkno) + fsb;
342 		error = LFS_BWRITE_LOG(ibp);	/* Ifile */
343 	}
344 
345 	/*
346 	 * Account the inode: it no longer belongs to its former segment,
347 	 * though it will not belong to the new segment until that segment
348 	 * is actually written.
349 	 */
350 	if (daddr != LFS_UNUSED_DADDR) {
351 		u_int32_t oldsn = dtosn(fs, daddr);
352 		LFS_SEGENTRY(sup, fs, oldsn, bp);
353 		sup->su_nbytes -= DINODE1_SIZE;
354 		redo_ifile =
355 		    (ino == LFS_IFILE_INUM && !(bp->b_flags & B_GATHERED));
356 		if (redo_ifile)
357 			fs->lfs_flags |= LFS_IFDIRTY;
358 		LFS_WRITESEGENTRY(sup, fs, oldsn, bp);	/* Ifile */
359 	}
360 	return redo_ifile;
361 }
362 
363 int
364 lfs_gatherblock(struct segment * sp, struct ubuf * bp)
365 {
366 	struct lfs *fs;
367 	int version;
368 	int j, blksinblk;
369 
370 	/*
371 	 * If full, finish this segment.  We may be doing I/O, so
372 	 * release and reacquire the splbio().
373 	 */
374 	fs = sp->fs;
375 	blksinblk = howmany(bp->b_bcount, fs->lfs_bsize);
376 	if (sp->sum_bytes_left < sizeof(ufs_daddr_t) * blksinblk ||
377 	    sp->seg_bytes_left < bp->b_bcount) {
378 		lfs_updatemeta(sp);
379 
380 		version = sp->fip->fi_version;
381 		(void) lfs_writeseg(fs, sp);
382 
383 		sp->fip->fi_version = version;
384 		sp->fip->fi_ino = VTOI(sp->vp)->i_number;
385 		/* Add the current file to the segment summary. */
386 		++((SEGSUM *) (sp->segsum))->ss_nfinfo;
387 		sp->sum_bytes_left -= FINFOSIZE;
388 
389 		return 1;
390 	}
391 	/* Insert into the buffer list, update the FINFO block. */
392 	bp->b_flags |= B_GATHERED;
393 	/* bp->b_flags &= ~B_DONE; */
394 
395 	*sp->cbpp++ = bp;
396 	for (j = 0; j < blksinblk; j++)
397 		sp->fip->fi_blocks[sp->fip->fi_nblocks++] = bp->b_lblkno + j;
398 
399 	sp->sum_bytes_left -= sizeof(ufs_daddr_t) * blksinblk;
400 	sp->seg_bytes_left -= bp->b_bcount;
401 	return 0;
402 }
403 
404 int
405 lfs_gather(struct lfs * fs, struct segment * sp, struct uvnode * vp, int (*match) (struct lfs *, struct ubuf *))
406 {
407 	struct ubuf *bp, *nbp;
408 	int count = 0;
409 
410 	sp->vp = vp;
411 loop:
412 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
413 		nbp = LIST_NEXT(bp, b_vnbufs);
414 
415 		assert(bp->b_flags & B_DELWRI);
416 		if ((bp->b_flags & (B_BUSY | B_GATHERED)) || !match(fs, bp)) {
417 			continue;
418 		}
419 		if (lfs_gatherblock(sp, bp)) {
420 			goto loop;
421 		}
422 		count++;
423 	}
424 
425 	lfs_updatemeta(sp);
426 	sp->vp = NULL;
427 	return count;
428 }
429 
430 
431 /*
432  * Change the given block's address to ndaddr, finding its previous
433  * location using ufs_bmaparray().
434  *
435  * Account for this change in the segment table.
436  */
437 void
438 lfs_update_single(struct lfs * fs, struct segment * sp, daddr_t lbn,
439     ufs_daddr_t ndaddr, int size)
440 {
441 	SEGUSE *sup;
442 	struct ubuf *bp;
443 	struct indir a[NIADDR + 2], *ap;
444 	struct inode *ip;
445 	struct uvnode *vp;
446 	daddr_t daddr, ooff;
447 	int num, error;
448 	int bb, osize, obb;
449 
450 	vp = sp->vp;
451 	ip = VTOI(vp);
452 
453 	error = ufs_bmaparray(fs, vp, lbn, &daddr, a, &num);
454 	if (error)
455 		errx(1, "lfs_updatemeta: ufs_bmaparray returned %d looking up lbn %" PRId64 "\n", error, lbn);
456 	if (daddr > 0)
457 		daddr = dbtofsb(fs, daddr);
458 
459 	bb = fragstofsb(fs, numfrags(fs, size));
460 	switch (num) {
461 	case 0:
462 		ooff = ip->i_ffs1_db[lbn];
463 		if (ooff == UNWRITTEN)
464 			ip->i_ffs1_blocks += bb;
465 		else {
466 			/* possible fragment truncation or extension */
467 			obb = btofsb(fs, ip->i_lfs_fragsize[lbn]);
468 			ip->i_ffs1_blocks += (bb - obb);
469 		}
470 		ip->i_ffs1_db[lbn] = ndaddr;
471 		break;
472 	case 1:
473 		ooff = ip->i_ffs1_ib[a[0].in_off];
474 		if (ooff == UNWRITTEN)
475 			ip->i_ffs1_blocks += bb;
476 		ip->i_ffs1_ib[a[0].in_off] = ndaddr;
477 		break;
478 	default:
479 		ap = &a[num - 1];
480 		if (bread(vp, ap->in_lbn, fs->lfs_bsize, NULL, &bp))
481 			errx(1, "lfs_updatemeta: bread bno %" PRId64,
482 			    ap->in_lbn);
483 
484 		ooff = ((ufs_daddr_t *) bp->b_data)[ap->in_off];
485 		if (ooff == UNWRITTEN)
486 			ip->i_ffs1_blocks += bb;
487 		((ufs_daddr_t *) bp->b_data)[ap->in_off] = ndaddr;
488 		(void) VOP_BWRITE(bp);
489 	}
490 
491 	/*
492 	 * Update segment usage information, based on old size
493 	 * and location.
494 	 */
495 	if (daddr > 0) {
496 		u_int32_t oldsn = dtosn(fs, daddr);
497 		if (lbn >= 0 && lbn < NDADDR)
498 			osize = ip->i_lfs_fragsize[lbn];
499 		else
500 			osize = fs->lfs_bsize;
501 		LFS_SEGENTRY(sup, fs, oldsn, bp);
502 		sup->su_nbytes -= osize;
503 		if (!(bp->b_flags & B_GATHERED))
504 			fs->lfs_flags |= LFS_IFDIRTY;
505 		LFS_WRITESEGENTRY(sup, fs, oldsn, bp);
506 	}
507 	/*
508 	 * Now that this block has a new address, and its old
509 	 * segment no longer owns it, we can forget about its
510 	 * old size.
511 	 */
512 	if (lbn >= 0 && lbn < NDADDR)
513 		ip->i_lfs_fragsize[lbn] = size;
514 }
515 
516 /*
517  * Update the metadata that points to the blocks listed in the FINFO
518  * array.
519  */
520 void
521 lfs_updatemeta(struct segment * sp)
522 {
523 	struct ubuf *sbp;
524 	struct lfs *fs;
525 	struct uvnode *vp;
526 	daddr_t lbn;
527 	int i, nblocks, num;
528 	int bb;
529 	int bytesleft, size;
530 
531 	vp = sp->vp;
532 	nblocks = &sp->fip->fi_blocks[sp->fip->fi_nblocks] - sp->start_lbp;
533 
534 	if (vp == NULL || nblocks == 0)
535 		return;
536 
537 	/*
538 	 * This count may be high due to oversize blocks from lfs_gop_write.
539 	 * Correct for this. (XXX we should be able to keep track of these.)
540 	 */
541 	fs = sp->fs;
542 	for (i = 0; i < nblocks; i++) {
543 		if (sp->start_bpp[i] == NULL) {
544 			printf("nblocks = %d, not %d\n", i, nblocks);
545 			nblocks = i;
546 			break;
547 		}
548 		num = howmany(sp->start_bpp[i]->b_bcount, fs->lfs_bsize);
549 		nblocks -= num - 1;
550 	}
551 
552 	/*
553 	 * Sort the blocks.
554 	 */
555 	lfs_shellsort(sp->start_bpp, sp->start_lbp, nblocks, fs->lfs_bsize);
556 
557 	/*
558 	 * Record the length of the last block in case it's a fragment.
559 	 * If there are indirect blocks present, they sort last.  An
560 	 * indirect block will be lfs_bsize and its presence indicates
561 	 * that you cannot have fragments.
562 	 */
563 	sp->fip->fi_lastlength = ((sp->start_bpp[nblocks - 1]->b_bcount - 1) &
564 	    fs->lfs_bmask) + 1;
565 
566 	/*
567 	 * Assign disk addresses, and update references to the logical
568 	 * block and the segment usage information.
569 	 */
570 	for (i = nblocks; i--; ++sp->start_bpp) {
571 		sbp = *sp->start_bpp;
572 		lbn = *sp->start_lbp;
573 
574 		sbp->b_blkno = fsbtodb(fs, fs->lfs_offset);
575 
576 		/*
577 		 * If we write a frag in the wrong place, the cleaner won't
578 		 * be able to correctly identify its size later, and the
579 		 * segment will be uncleanable.	 (Even worse, it will assume
580 		 * that the indirect block that actually ends the list
581 		 * is of a smaller size!)
582 		 */
583 		if ((sbp->b_bcount & fs->lfs_bmask) && i != 0)
584 			errx(1, "lfs_updatemeta: fragment is not last block");
585 
586 		/*
587 		 * For each subblock in this possibly oversized block,
588 		 * update its address on disk.
589 		 */
590 		for (bytesleft = sbp->b_bcount; bytesleft > 0;
591 		    bytesleft -= fs->lfs_bsize) {
592 			size = MIN(bytesleft, fs->lfs_bsize);
593 			bb = fragstofsb(fs, numfrags(fs, size));
594 			lbn = *sp->start_lbp++;
595 			lfs_update_single(fs, sp, lbn, fs->lfs_offset, size);
596 			fs->lfs_offset += bb;
597 		}
598 
599 	}
600 }
601 
602 /*
603  * Start a new segment.
604  */
605 int
606 lfs_initseg(struct lfs * fs)
607 {
608 	struct segment *sp;
609 	SEGUSE *sup;
610 	SEGSUM *ssp;
611 	struct ubuf *bp, *sbp;
612 	int repeat;
613 
614 	sp = fs->lfs_sp;
615 
616 	repeat = 0;
617 
618 	/* Advance to the next segment. */
619 	if (!LFS_PARTIAL_FITS(fs)) {
620 		/* lfs_avail eats the remaining space */
621 		fs->lfs_avail -= fs->lfs_fsbpseg - (fs->lfs_offset -
622 		    fs->lfs_curseg);
623 		lfs_newseg(fs);
624 		repeat = 1;
625 		fs->lfs_offset = fs->lfs_curseg;
626 
627 		sp->seg_number = dtosn(fs, fs->lfs_curseg);
628 		sp->seg_bytes_left = fsbtob(fs, fs->lfs_fsbpseg);
629 
630 		/*
631 		 * If the segment contains a superblock, update the offset
632 		 * and summary address to skip over it.
633 		 */
634 		LFS_SEGENTRY(sup, fs, sp->seg_number, bp);
635 		if (sup->su_flags & SEGUSE_SUPERBLOCK) {
636 			fs->lfs_offset += btofsb(fs, LFS_SBPAD);
637 			sp->seg_bytes_left -= LFS_SBPAD;
638 		}
639 		brelse(bp);
640 		/* Segment zero could also contain the labelpad */
641 		if (fs->lfs_version > 1 && sp->seg_number == 0 &&
642 		    fs->lfs_start < btofsb(fs, LFS_LABELPAD)) {
643 			fs->lfs_offset += btofsb(fs, LFS_LABELPAD) - fs->lfs_start;
644 			sp->seg_bytes_left -= LFS_LABELPAD - fsbtob(fs, fs->lfs_start);
645 		}
646 	} else {
647 		sp->seg_number = dtosn(fs, fs->lfs_curseg);
648 		sp->seg_bytes_left = fsbtob(fs, fs->lfs_fsbpseg -
649 		    (fs->lfs_offset - fs->lfs_curseg));
650 	}
651 	fs->lfs_lastpseg = fs->lfs_offset;
652 
653 	sp->fs = fs;
654 	sp->ibp = NULL;
655 	sp->idp = NULL;
656 	sp->ninodes = 0;
657 	sp->ndupino = 0;
658 
659 	/* Get a new buffer for SEGSUM and enter it into the buffer list. */
660 	sp->cbpp = sp->bpp;
661 	sbp = *sp->cbpp = getblk(fs->lfs_devvp,
662 	    fsbtodb(fs, fs->lfs_offset), fs->lfs_sumsize);
663 	sp->segsum = sbp->b_data;
664 	memset(sp->segsum, 0, fs->lfs_sumsize);
665 	sp->start_bpp = ++sp->cbpp;
666 	fs->lfs_offset += btofsb(fs, fs->lfs_sumsize);
667 
668 	/* Set point to SEGSUM, initialize it. */
669 	ssp = sp->segsum;
670 	ssp->ss_next = fs->lfs_nextseg;
671 	ssp->ss_nfinfo = ssp->ss_ninos = 0;
672 	ssp->ss_magic = SS_MAGIC;
673 
674 	/* Set pointer to first FINFO, initialize it. */
675 	sp->fip = (struct finfo *) ((caddr_t) sp->segsum + SEGSUM_SIZE(fs));
676 	sp->fip->fi_nblocks = 0;
677 	sp->start_lbp = &sp->fip->fi_blocks[0];
678 	sp->fip->fi_lastlength = 0;
679 
680 	sp->seg_bytes_left -= fs->lfs_sumsize;
681 	sp->sum_bytes_left = fs->lfs_sumsize - SEGSUM_SIZE(fs);
682 
683 	LFS_LOCK_BUF(sbp);
684 	brelse(sbp);
685 	return repeat;
686 }
687 
688 /*
689  * Return the next segment to write.
690  */
691 void
692 lfs_newseg(struct lfs * fs)
693 {
694 	CLEANERINFO *cip;
695 	SEGUSE *sup;
696 	struct ubuf *bp;
697 	int curseg, isdirty, sn;
698 
699 	LFS_SEGENTRY(sup, fs, dtosn(fs, fs->lfs_nextseg), bp);
700 	sup->su_flags |= SEGUSE_DIRTY | SEGUSE_ACTIVE;
701 	sup->su_nbytes = 0;
702 	sup->su_nsums = 0;
703 	sup->su_ninos = 0;
704 	LFS_WRITESEGENTRY(sup, fs, dtosn(fs, fs->lfs_nextseg), bp);
705 
706 	LFS_CLEANERINFO(cip, fs, bp);
707 	--cip->clean;
708 	++cip->dirty;
709 	fs->lfs_nclean = cip->clean;
710 	LFS_SYNC_CLEANERINFO(cip, fs, bp, 1);
711 
712 	fs->lfs_lastseg = fs->lfs_curseg;
713 	fs->lfs_curseg = fs->lfs_nextseg;
714 	for (sn = curseg = dtosn(fs, fs->lfs_curseg) + fs->lfs_interleave;;) {
715 		sn = (sn + 1) % fs->lfs_nseg;
716 		if (sn == curseg)
717 			errx(1, "lfs_nextseg: no clean segments");
718 		LFS_SEGENTRY(sup, fs, sn, bp);
719 		isdirty = sup->su_flags & SEGUSE_DIRTY;
720 		brelse(bp);
721 
722 		if (!isdirty)
723 			break;
724 	}
725 
726 	++fs->lfs_nactive;
727 	fs->lfs_nextseg = sntod(fs, sn);
728 }
729 
730 
731 int
732 lfs_writeseg(struct lfs * fs, struct segment * sp)
733 {
734 	struct ubuf **bpp, *bp;
735 	SEGUSE *sup;
736 	SEGSUM *ssp;
737 	char *datap, *dp;
738 	int i;
739 	int do_again, nblocks, byteoffset;
740 	size_t el_size;
741 	u_short ninos;
742 	struct uvnode *devvp;
743 
744 	/*
745 	 * If there are no buffers other than the segment summary to write
746 	 * and it is not a checkpoint, don't do anything.  On a checkpoint,
747 	 * even if there aren't any buffers, you need to write the superblock.
748 	 */
749 	if ((nblocks = sp->cbpp - sp->bpp) == 1)
750 		return 0;
751 
752 	devvp = fs->lfs_devvp;
753 
754 	/* Update the segment usage information. */
755 	LFS_SEGENTRY(sup, fs, sp->seg_number, bp);
756 
757 	/* Loop through all blocks, except the segment summary. */
758 	for (bpp = sp->bpp; ++bpp < sp->cbpp;) {
759 		if ((*bpp)->b_vp != devvp) {
760 			sup->su_nbytes += (*bpp)->b_bcount;
761 		}
762 	}
763 
764 	ssp = (SEGSUM *) sp->segsum;
765 
766 	ninos = (ssp->ss_ninos + INOPB(fs) - 1) / INOPB(fs);
767 	sup->su_nbytes += ssp->ss_ninos * DINODE1_SIZE;
768 
769 	if (fs->lfs_version == 1)
770 		sup->su_olastmod = write_time;
771 	else
772 		sup->su_lastmod = write_time;
773 	sup->su_ninos += ninos;
774 	++sup->su_nsums;
775 	fs->lfs_dmeta += (btofsb(fs, fs->lfs_sumsize) + btofsb(fs, ninos *
776 		fs->lfs_ibsize));
777 	fs->lfs_avail -= btofsb(fs, fs->lfs_sumsize);
778 
779 	do_again = !(bp->b_flags & B_GATHERED);
780 	LFS_WRITESEGENTRY(sup, fs, sp->seg_number, bp);	/* Ifile */
781 
782 	/*
783 	 * Compute checksum across data and then across summary; the first
784 	 * block (the summary block) is skipped.  Set the create time here
785 	 * so that it's guaranteed to be later than the inode mod times.
786 	 */
787 	if (fs->lfs_version == 1)
788 		el_size = sizeof(u_long);
789 	else
790 		el_size = sizeof(u_int32_t);
791 	datap = dp = malloc(nblocks * el_size);
792 	if (dp == NULL)
793 		err(1, NULL);
794 	for (bpp = sp->bpp, i = nblocks - 1; i--;) {
795 		++bpp;
796 		/* Loop through gop_write cluster blocks */
797 		for (byteoffset = 0; byteoffset < (*bpp)->b_bcount;
798 		    byteoffset += fs->lfs_bsize) {
799 			memcpy(dp, (*bpp)->b_data + byteoffset, el_size);
800 			dp += el_size;
801 		}
802 		bremfree(*bpp);
803 		(*bpp)->b_flags |= B_BUSY;
804 	}
805 	if (fs->lfs_version == 1)
806 		ssp->ss_ocreate = write_time;
807 	else {
808 		ssp->ss_create = write_time;
809 		ssp->ss_serial = ++fs->lfs_serial;
810 		ssp->ss_ident = fs->lfs_ident;
811 	}
812 	/* Set the summary block busy too */
813 	bremfree(*(sp->bpp));
814 	(*(sp->bpp))->b_flags |= B_BUSY;
815 
816 	ssp->ss_datasum = cksum(datap, (nblocks - 1) * el_size);
817 	ssp->ss_sumsum =
818 	    cksum(&ssp->ss_datasum, fs->lfs_sumsize - sizeof(ssp->ss_sumsum));
819 	free(datap);
820 	datap = dp = NULL;
821 	fs->lfs_bfree -= (btofsb(fs, ninos * fs->lfs_ibsize) +
822 	    btofsb(fs, fs->lfs_sumsize));
823 
824 	if (devvp == NULL)
825 		errx(1, "devvp is NULL");
826 	for (bpp = sp->bpp, i = nblocks; i; bpp++, i--) {
827 		bp = *bpp;
828 #if 0
829 		printf("i = %d, bp = %p, flags %lx, bn = %" PRIx64 "\n",
830 		       nblocks - i, bp, bp->b_flags, bp->b_blkno);
831 		printf("  vp = %p\n", bp->b_vp);
832 		if (bp->b_vp != fs->lfs_devvp)
833 			printf("  ino = %d lbn = %" PRId64 "\n",
834 			       VTOI(bp->b_vp)->i_number, bp->b_lblkno);
835 #endif
836 		if (bp->b_vp == fs->lfs_devvp)
837 			written_dev += bp->b_bcount;
838 		else {
839 			if (bp->b_lblkno >= 0)
840 				written_data += bp->b_bcount;
841 			else
842 				written_indir += bp->b_bcount;
843 		}
844 		bp->b_flags &= ~(B_DELWRI | B_READ | B_GATHERED | B_ERROR |
845 				 B_LOCKED);
846 		bwrite(bp);
847 		written_bytes += bp->b_bcount;
848 	}
849 	written_inodes += ninos;
850 
851 	return (lfs_initseg(fs) || do_again);
852 }
853 
854 /*
855  * Our own copy of shellsort.  XXX use qsort or heapsort.
856  */
857 void
858 lfs_shellsort(struct ubuf ** bp_array, ufs_daddr_t * lb_array, int nmemb, int size)
859 {
860 	static int __rsshell_increments[] = {4, 1, 0};
861 	int incr, *incrp, t1, t2;
862 	struct ubuf *bp_temp;
863 
864 	for (incrp = __rsshell_increments; (incr = *incrp++) != 0;)
865 		for (t1 = incr; t1 < nmemb; ++t1)
866 			for (t2 = t1 - incr; t2 >= 0;)
867 				if ((u_int32_t) bp_array[t2]->b_lblkno >
868 				    (u_int32_t) bp_array[t2 + incr]->b_lblkno) {
869 					bp_temp = bp_array[t2];
870 					bp_array[t2] = bp_array[t2 + incr];
871 					bp_array[t2 + incr] = bp_temp;
872 					t2 -= incr;
873 				} else
874 					break;
875 
876 	/* Reform the list of logical blocks */
877 	incr = 0;
878 	for (t1 = 0; t1 < nmemb; t1++) {
879 		for (t2 = 0; t2 * size < bp_array[t1]->b_bcount; t2++) {
880 			lb_array[incr++] = bp_array[t1]->b_lblkno + t2;
881 		}
882 	}
883 }
884 
885 
886 /*
887  * lfs_seglock --
888  *	Single thread the segment writer.
889  */
890 int
891 lfs_seglock(struct lfs * fs, unsigned long flags)
892 {
893 	struct segment *sp;
894 
895 	if (fs->lfs_seglock) {
896 		++fs->lfs_seglock;
897 		fs->lfs_sp->seg_flags |= flags;
898 		return 0;
899 	}
900 	fs->lfs_seglock = 1;
901 
902 	sp = fs->lfs_sp = (struct segment *) malloc(sizeof(*sp));
903 	if (sp == NULL)
904 		err(1, NULL);
905 	sp->bpp = (struct ubuf **) malloc(fs->lfs_ssize * sizeof(struct ubuf *));
906 	if (!sp->bpp)
907 		errx(!preen, "Could not allocate %zu bytes: %s",
908 			(size_t)(fs->lfs_ssize * sizeof(struct ubuf *)),
909 			strerror(errno));
910 	sp->seg_flags = flags;
911 	sp->vp = NULL;
912 	sp->seg_iocount = 0;
913 	(void) lfs_initseg(fs);
914 
915 	return 0;
916 }
917 
918 /*
919  * lfs_segunlock --
920  *	Single thread the segment writer.
921  */
922 void
923 lfs_segunlock(struct lfs * fs)
924 {
925 	struct segment *sp;
926 	struct ubuf *bp;
927 
928 	sp = fs->lfs_sp;
929 
930 	if (fs->lfs_seglock == 1) {
931 		if (sp->bpp != sp->cbpp) {
932 			/* Free allocated segment summary */
933 			fs->lfs_offset -= btofsb(fs, fs->lfs_sumsize);
934 			bp = *sp->bpp;
935 			bremfree(bp);
936 			bp->b_flags |= B_DONE | B_INVAL;
937 			bp->b_flags &= ~B_DELWRI;
938 			reassignbuf(bp, bp->b_vp);
939 			bp->b_flags |= B_BUSY; /* XXX */
940 			brelse(bp);
941 		} else
942 			printf("unlock to 0 with no summary");
943 
944 		free(sp->bpp);
945 		sp->bpp = NULL;
946 		free(sp);
947 		fs->lfs_sp = NULL;
948 
949 		fs->lfs_nactive = 0;
950 
951 		/* Since we *know* everything's on disk, write both sbs */
952 		lfs_writesuper(fs, fs->lfs_sboffs[0]);
953 		lfs_writesuper(fs, fs->lfs_sboffs[1]);
954 
955 		--fs->lfs_seglock;
956 		fs->lfs_lockpid = 0;
957 	} else if (fs->lfs_seglock == 0) {
958 		errx(1, "Seglock not held");
959 	} else {
960 		--fs->lfs_seglock;
961 	}
962 }
963 
964 int
965 lfs_writevnodes(struct lfs *fs, struct segment *sp, int op)
966 {
967 	struct inode *ip;
968 	struct uvnode *vp;
969 	int inodes_written = 0;
970 
971 	LIST_FOREACH(vp, &vnodelist, v_mntvnodes) {
972 		if (vp->v_bmap_op != lfs_vop_bmap)
973 			continue;
974 
975 		ip = VTOI(vp);
976 
977 		if ((op == VN_DIROP && !(vp->v_flag & VDIROP)) ||
978 		    (op != VN_DIROP && (vp->v_flag & VDIROP))) {
979 			continue;
980 		}
981 		/*
982 		 * Write the inode/file if dirty and it's not the IFILE.
983 		 */
984 		if (ip->i_flag & IN_ALLMOD || !LIST_EMPTY(&vp->v_dirtyblkhd)) {
985 			if (ip->i_number != LFS_IFILE_INUM)
986 				lfs_writefile(fs, sp, vp);
987 			(void) lfs_writeinode(fs, sp, ip);
988 			inodes_written++;
989 		}
990 	}
991 	return inodes_written;
992 }
993 
994 void
995 lfs_writesuper(struct lfs *fs, ufs_daddr_t daddr)
996 {
997 	struct ubuf *bp;
998 
999 	/* Set timestamp of this version of the superblock */
1000 	if (fs->lfs_version == 1)
1001 		fs->lfs_otstamp = write_time;
1002 	fs->lfs_tstamp = write_time;
1003 
1004 	/* Checksum the superblock and copy it into a buffer. */
1005 	fs->lfs_cksum = lfs_sb_cksum(&(fs->lfs_dlfs));
1006 	assert(daddr > 0);
1007 	bp = getblk(fs->lfs_devvp, fsbtodb(fs, daddr), LFS_SBPAD);
1008 	memset(bp->b_data + sizeof(struct dlfs), 0,
1009 	    LFS_SBPAD - sizeof(struct dlfs));
1010 	*(struct dlfs *) bp->b_data = fs->lfs_dlfs;
1011 
1012 	bwrite(bp);
1013 }
1014