xref: /netbsd-src/sbin/fsck_lfs/segwrite.c (revision de1dfb1250df962f1ff3a011772cf58e605aed11)
1 /* $NetBSD: segwrite.c,v 1.6 2003/12/24 01:39:27 heas 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 #undef simple_lock
82 #define simple_lock(x)
83 #undef simple_unlock
84 #define simple_unlock(x)
85 #define vnode uvnode
86 #define buf ubuf
87 #define panic call_panic
88 
89 #include <ufs/lfs/lfs.h>
90 
91 #include <assert.h>
92 #include <stdio.h>
93 #include <stdlib.h>
94 #include <string.h>
95 #include <err.h>
96 #include <errno.h>
97 
98 #include "bufcache.h"
99 #include "vnode.h"
100 #include "lfs.h"
101 #include "segwrite.h"
102 
103 /* Compatibility definitions */
104 extern off_t locked_queue_bytes;
105 int locked_queue_count;
106 off_t written_bytes = 0;
107 off_t written_data = 0;
108 off_t written_indir = 0;
109 off_t written_dev = 0;
110 int written_inodes = 0;
111 
112 /* Global variables */
113 time_t write_time;
114 
115 extern u_int32_t cksum(void *, size_t);
116 extern u_int32_t lfs_sb_cksum(struct dlfs *);
117 
118 /*
119  * Logical block number match routines used when traversing the dirty block
120  * chain.
121  */
122 int
123 lfs_match_data(struct lfs * fs, struct ubuf * bp)
124 {
125 	return (bp->b_lblkno >= 0);
126 }
127 
128 int
129 lfs_match_indir(struct lfs * fs, struct ubuf * bp)
130 {
131 	daddr_t lbn;
132 
133 	lbn = bp->b_lblkno;
134 	return (lbn < 0 && (-lbn - NDADDR) % NINDIR(fs) == 0);
135 }
136 
137 int
138 lfs_match_dindir(struct lfs * fs, struct ubuf * bp)
139 {
140 	daddr_t lbn;
141 
142 	lbn = bp->b_lblkno;
143 	return (lbn < 0 && (-lbn - NDADDR) % NINDIR(fs) == 1);
144 }
145 
146 int
147 lfs_match_tindir(struct lfs * fs, struct ubuf * bp)
148 {
149 	daddr_t lbn;
150 
151 	lbn = bp->b_lblkno;
152 	return (lbn < 0 && (-lbn - NDADDR) % NINDIR(fs) == 2);
153 }
154 
155 /*
156  * Do a checkpoint.
157  */
158 int
159 lfs_segwrite(struct lfs * fs, int flags)
160 {
161 	struct inode *ip;
162 	struct segment *sp;
163 	struct uvnode *vp;
164 	int redo;
165 
166 	lfs_seglock(fs, flags | SEGM_CKP);
167 	sp = fs->lfs_sp;
168 
169 	lfs_writevnodes(fs, sp, VN_REG);
170 	lfs_writevnodes(fs, sp, VN_DIROP);
171 	((SEGSUM *) (sp->segsum))->ss_flags &= ~(SS_CONT);
172 
173 	do {
174 		vp = fs->lfs_ivnode;
175 		fs->lfs_flags &= ~LFS_IFDIRTY;
176 		ip = VTOI(vp);
177 		if (LIST_FIRST(&vp->v_dirtyblkhd) != NULL)
178 			lfs_writefile(fs, sp, vp);
179 
180 		redo = lfs_writeinode(fs, sp, ip);
181 		redo += lfs_writeseg(fs, sp);
182 		redo += (fs->lfs_flags & LFS_IFDIRTY);
183 	} while (redo);
184 
185 	lfs_segunlock(fs);
186 #if 0
187 	printf("wrote %" PRId64 " bytes (%" PRId32 " fsb)\n",
188 		written_bytes, (ufs_daddr_t)btofsb(fs, written_bytes));
189 	printf("wrote %" PRId64 " bytes data (%" PRId32 " fsb)\n",
190 		written_data, (ufs_daddr_t)btofsb(fs, written_data));
191 	printf("wrote %" PRId64 " bytes indir (%" PRId32 " fsb)\n",
192 		written_indir, (ufs_daddr_t)btofsb(fs, written_indir));
193 	printf("wrote %" PRId64 " bytes dev (%" PRId32 " fsb)\n",
194 		written_dev, (ufs_daddr_t)btofsb(fs, written_dev));
195 	printf("wrote %d inodes (%" PRId32 " fsb)\n",
196 		written_inodes, btofsb(fs, written_inodes * fs->lfs_ibsize));
197 #endif
198 	return 0;
199 }
200 
201 /*
202  * Write the dirty blocks associated with a vnode.
203  */
204 void
205 lfs_writefile(struct lfs * fs, struct segment * sp, struct uvnode * vp)
206 {
207 	struct ubuf *bp;
208 	struct finfo *fip;
209 	struct inode *ip;
210 	IFILE *ifp;
211 
212 	ip = VTOI(vp);
213 
214 	if (sp->seg_bytes_left < fs->lfs_bsize ||
215 	    sp->sum_bytes_left < sizeof(struct finfo))
216 		(void) lfs_writeseg(fs, sp);
217 
218 	sp->sum_bytes_left -= FINFOSIZE;
219 	++((SEGSUM *) (sp->segsum))->ss_nfinfo;
220 
221 	if (vp->v_flag & VDIROP)
222 		((SEGSUM *) (sp->segsum))->ss_flags |= (SS_DIROP | SS_CONT);
223 
224 	fip = sp->fip;
225 	fip->fi_nblocks = 0;
226 	fip->fi_ino = ip->i_number;
227 	LFS_IENTRY(ifp, fs, fip->fi_ino, bp);
228 	fip->fi_version = ifp->if_version;
229 	brelse(bp);
230 
231 	lfs_gather(fs, sp, vp, lfs_match_data);
232 	lfs_gather(fs, sp, vp, lfs_match_indir);
233 	lfs_gather(fs, sp, vp, lfs_match_dindir);
234 	lfs_gather(fs, sp, vp, lfs_match_tindir);
235 
236 	fip = sp->fip;
237 	if (fip->fi_nblocks != 0) {
238 		sp->fip = (FINFO *) ((caddr_t) fip + FINFOSIZE +
239 		    sizeof(ufs_daddr_t) * (fip->fi_nblocks));
240 		sp->start_lbp = &sp->fip->fi_blocks[0];
241 	} else {
242 		sp->sum_bytes_left += FINFOSIZE;
243 		--((SEGSUM *) (sp->segsum))->ss_nfinfo;
244 	}
245 }
246 
247 int
248 lfs_writeinode(struct lfs * fs, struct segment * sp, struct inode * ip)
249 {
250 	struct ubuf *bp, *ibp;
251 	struct ufs1_dinode *cdp;
252 	IFILE *ifp;
253 	SEGUSE *sup;
254 	daddr_t daddr;
255 	ino_t ino;
256 	int error, i, ndx, fsb = 0;
257 	int redo_ifile = 0;
258 	struct timespec ts;
259 	int gotblk = 0;
260 
261 	/* Allocate a new inode block if necessary. */
262 	if ((ip->i_number != LFS_IFILE_INUM || sp->idp == NULL) &&
263 	    sp->ibp == NULL) {
264 		/* Allocate a new segment if necessary. */
265 		if (sp->seg_bytes_left < fs->lfs_ibsize ||
266 		    sp->sum_bytes_left < sizeof(ufs_daddr_t))
267 			(void) lfs_writeseg(fs, sp);
268 
269 		/* Get next inode block. */
270 		daddr = fs->lfs_offset;
271 		fs->lfs_offset += btofsb(fs, fs->lfs_ibsize);
272 		sp->ibp = *sp->cbpp++ =
273 		    getblk(fs->lfs_unlockvp, fsbtodb(fs, daddr),
274 		    fs->lfs_ibsize);
275 		sp->ibp->b_flags |= B_GATHERED;
276 		gotblk++;
277 
278 		/* Zero out inode numbers */
279 		for (i = 0; i < INOPB(fs); ++i)
280 			((struct ufs1_dinode *) sp->ibp->b_data)[i].di_inumber = 0;
281 
282 		++sp->start_bpp;
283 		fs->lfs_avail -= btofsb(fs, fs->lfs_ibsize);
284 		/* Set remaining space counters. */
285 		sp->seg_bytes_left -= fs->lfs_ibsize;
286 		sp->sum_bytes_left -= sizeof(ufs_daddr_t);
287 		ndx = fs->lfs_sumsize / sizeof(ufs_daddr_t) -
288 		    sp->ninodes / INOPB(fs) - 1;
289 		((ufs_daddr_t *) (sp->segsum))[ndx] = daddr;
290 	}
291 	/* Update the inode times and copy the inode onto the inode page. */
292 	ts.tv_nsec = 0;
293 	ts.tv_sec = write_time;
294 	/* XXX kludge --- don't redirty the ifile just to put times on it */
295 	if (ip->i_number != LFS_IFILE_INUM)
296 		LFS_ITIMES(ip, &ts, &ts, &ts);
297 
298 	/*
299 	 * If this is the Ifile, and we've already written the Ifile in this
300 	 * partial segment, just overwrite it (it's not on disk yet) and
301 	 * continue.
302 	 *
303 	 * XXX we know that the bp that we get the second time around has
304 	 * already been gathered.
305 	 */
306 	if (ip->i_number == LFS_IFILE_INUM && sp->idp) {
307 		*(sp->idp) = *ip->i_din.ffs1_din;
308 		ip->i_lfs_osize = ip->i_ffs1_size;
309 		return 0;
310 	}
311 	bp = sp->ibp;
312 	cdp = ((struct ufs1_dinode *) bp->b_data) + (sp->ninodes % INOPB(fs));
313 	*cdp = *ip->i_din.ffs1_din;
314 
315 	/* If all blocks are goig to disk, update the "size on disk" */
316 	ip->i_lfs_osize = ip->i_ffs1_size;
317 
318 	if (ip->i_number == LFS_IFILE_INUM)	/* We know sp->idp == NULL */
319 		sp->idp = ((struct ufs1_dinode *) bp->b_data) +
320 		    (sp->ninodes % INOPB(fs));
321 	if (gotblk) {
322 		LFS_LOCK_BUF(bp);
323 		brelse(bp);
324 	}
325 	/* Increment inode count in segment summary block. */
326 	++((SEGSUM *) (sp->segsum))->ss_ninos;
327 
328 	/* If this page is full, set flag to allocate a new page. */
329 	if (++sp->ninodes % INOPB(fs) == 0)
330 		sp->ibp = NULL;
331 
332 	/*
333 	 * If updating the ifile, update the super-block.  Update the disk
334 	 * address and access times for this inode in the ifile.
335 	 */
336 	ino = ip->i_number;
337 	if (ino == LFS_IFILE_INUM) {
338 		daddr = fs->lfs_idaddr;
339 		fs->lfs_idaddr = dbtofsb(fs, bp->b_blkno);
340 	} else {
341 		LFS_IENTRY(ifp, fs, ino, ibp);
342 		daddr = ifp->if_daddr;
343 		ifp->if_daddr = dbtofsb(fs, bp->b_blkno) + fsb;
344 		error = LFS_BWRITE_LOG(ibp);	/* Ifile */
345 	}
346 
347 	/*
348 	 * Account the inode: it no longer belongs to its former segment,
349 	 * though it will not belong to the new segment until that segment
350 	 * is actually written.
351 	 */
352 	if (daddr != LFS_UNUSED_DADDR) {
353 		u_int32_t oldsn = dtosn(fs, daddr);
354 		LFS_SEGENTRY(sup, fs, oldsn, bp);
355 		sup->su_nbytes -= DINODE1_SIZE;
356 		redo_ifile =
357 		    (ino == LFS_IFILE_INUM && !(bp->b_flags & B_GATHERED));
358 		if (redo_ifile)
359 			fs->lfs_flags |= LFS_IFDIRTY;
360 		LFS_WRITESEGENTRY(sup, fs, oldsn, bp);	/* Ifile */
361 	}
362 	return redo_ifile;
363 }
364 
365 int
366 lfs_gatherblock(struct segment * sp, struct ubuf * bp)
367 {
368 	struct lfs *fs;
369 	int version;
370 	int j, blksinblk;
371 
372 	/*
373 	 * If full, finish this segment.  We may be doing I/O, so
374 	 * release and reacquire the splbio().
375 	 */
376 	fs = sp->fs;
377 	blksinblk = howmany(bp->b_bcount, fs->lfs_bsize);
378 	if (sp->sum_bytes_left < sizeof(ufs_daddr_t) * blksinblk ||
379 	    sp->seg_bytes_left < bp->b_bcount) {
380 		lfs_updatemeta(sp);
381 
382 		version = sp->fip->fi_version;
383 		(void) lfs_writeseg(fs, sp);
384 
385 		sp->fip->fi_version = version;
386 		sp->fip->fi_ino = VTOI(sp->vp)->i_number;
387 		/* Add the current file to the segment summary. */
388 		++((SEGSUM *) (sp->segsum))->ss_nfinfo;
389 		sp->sum_bytes_left -= FINFOSIZE;
390 
391 		return 1;
392 	}
393 	/* Insert into the buffer list, update the FINFO block. */
394 	bp->b_flags |= B_GATHERED;
395 	/* bp->b_flags &= ~B_DONE; */
396 
397 	*sp->cbpp++ = bp;
398 	for (j = 0; j < blksinblk; j++)
399 		sp->fip->fi_blocks[sp->fip->fi_nblocks++] = bp->b_lblkno + j;
400 
401 	sp->sum_bytes_left -= sizeof(ufs_daddr_t) * blksinblk;
402 	sp->seg_bytes_left -= bp->b_bcount;
403 	return 0;
404 }
405 
406 int
407 lfs_gather(struct lfs * fs, struct segment * sp, struct uvnode * vp, int (*match) (struct lfs *, struct ubuf *))
408 {
409 	struct ubuf *bp, *nbp;
410 	int count = 0;
411 
412 	sp->vp = vp;
413 loop:
414 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
415 		nbp = LIST_NEXT(bp, b_vnbufs);
416 
417 		assert(bp->b_flags & B_DELWRI);
418 		if ((bp->b_flags & (B_BUSY | B_GATHERED)) || !match(fs, bp)) {
419 			continue;
420 		}
421 		if (lfs_gatherblock(sp, bp)) {
422 			goto loop;
423 		}
424 		count++;
425 	}
426 
427 	lfs_updatemeta(sp);
428 	sp->vp = NULL;
429 	return count;
430 }
431 
432 
433 /*
434  * Change the given block's address to ndaddr, finding its previous
435  * location using ufs_bmaparray().
436  *
437  * Account for this change in the segment table.
438  */
439 void
440 lfs_update_single(struct lfs * fs, struct segment * sp, daddr_t lbn,
441     ufs_daddr_t ndaddr, int size)
442 {
443 	SEGUSE *sup;
444 	struct ubuf *bp;
445 	struct indir a[NIADDR + 2], *ap;
446 	struct inode *ip;
447 	struct uvnode *vp;
448 	daddr_t daddr, ooff;
449 	int num, error;
450 	int bb, osize, obb;
451 
452 	vp = sp->vp;
453 	ip = VTOI(vp);
454 
455 	error = ufs_bmaparray(fs, vp, lbn, &daddr, a, &num);
456 	if (error)
457 		errx(1, "lfs_updatemeta: ufs_bmaparray returned %d looking up lbn %" PRId64 "\n", error, lbn);
458 	if (daddr > 0)
459 		daddr = dbtofsb(fs, daddr);
460 
461 	bb = fragstofsb(fs, numfrags(fs, size));
462 	switch (num) {
463 	case 0:
464 		ooff = ip->i_ffs1_db[lbn];
465 		if (ooff == UNWRITTEN)
466 			ip->i_ffs1_blocks += bb;
467 		else {
468 			/* possible fragment truncation or extension */
469 			obb = btofsb(fs, ip->i_lfs_fragsize[lbn]);
470 			ip->i_ffs1_blocks += (bb - obb);
471 		}
472 		ip->i_ffs1_db[lbn] = ndaddr;
473 		break;
474 	case 1:
475 		ooff = ip->i_ffs1_ib[a[0].in_off];
476 		if (ooff == UNWRITTEN)
477 			ip->i_ffs1_blocks += bb;
478 		ip->i_ffs1_ib[a[0].in_off] = ndaddr;
479 		break;
480 	default:
481 		ap = &a[num - 1];
482 		if (bread(vp, ap->in_lbn, fs->lfs_bsize, NULL, &bp))
483 			errx(1, "lfs_updatemeta: bread bno %" PRId64,
484 			    ap->in_lbn);
485 
486 		ooff = ((ufs_daddr_t *) bp->b_data)[ap->in_off];
487 		if (ooff == UNWRITTEN)
488 			ip->i_ffs1_blocks += bb;
489 		((ufs_daddr_t *) bp->b_data)[ap->in_off] = ndaddr;
490 		(void) VOP_BWRITE(bp);
491 	}
492 
493 	/*
494 	 * Update segment usage information, based on old size
495 	 * and location.
496 	 */
497 	if (daddr > 0) {
498 		u_int32_t oldsn = dtosn(fs, daddr);
499 		if (lbn >= 0 && lbn < NDADDR)
500 			osize = ip->i_lfs_fragsize[lbn];
501 		else
502 			osize = fs->lfs_bsize;
503 		LFS_SEGENTRY(sup, fs, oldsn, bp);
504 		sup->su_nbytes -= osize;
505 		if (!(bp->b_flags & B_GATHERED))
506 			fs->lfs_flags |= LFS_IFDIRTY;
507 		LFS_WRITESEGENTRY(sup, fs, oldsn, bp);
508 	}
509 	/*
510 	 * Now that this block has a new address, and its old
511 	 * segment no longer owns it, we can forget about its
512 	 * old size.
513 	 */
514 	if (lbn >= 0 && lbn < NDADDR)
515 		ip->i_lfs_fragsize[lbn] = size;
516 }
517 
518 /*
519  * Update the metadata that points to the blocks listed in the FINFO
520  * array.
521  */
522 void
523 lfs_updatemeta(struct segment * sp)
524 {
525 	struct ubuf *sbp;
526 	struct lfs *fs;
527 	struct uvnode *vp;
528 	daddr_t lbn;
529 	int i, nblocks, num;
530 	int bb;
531 	int bytesleft, size;
532 
533 	vp = sp->vp;
534 	nblocks = &sp->fip->fi_blocks[sp->fip->fi_nblocks] - sp->start_lbp;
535 
536 	if (vp == NULL || nblocks == 0)
537 		return;
538 
539 	/*
540 	 * This count may be high due to oversize blocks from lfs_gop_write.
541 	 * Correct for this. (XXX we should be able to keep track of these.)
542 	 */
543 	fs = sp->fs;
544 	for (i = 0; i < nblocks; i++) {
545 		if (sp->start_bpp[i] == NULL) {
546 			printf("nblocks = %d, not %d\n", i, nblocks);
547 			nblocks = i;
548 			break;
549 		}
550 		num = howmany(sp->start_bpp[i]->b_bcount, fs->lfs_bsize);
551 		nblocks -= num - 1;
552 	}
553 
554 	/*
555 	 * Sort the blocks.
556 	 */
557 	lfs_shellsort(sp->start_bpp, sp->start_lbp, nblocks, fs->lfs_bsize);
558 
559 	/*
560 	 * Record the length of the last block in case it's a fragment.
561 	 * If there are indirect blocks present, they sort last.  An
562 	 * indirect block will be lfs_bsize and its presence indicates
563 	 * that you cannot have fragments.
564 	 */
565 	sp->fip->fi_lastlength = ((sp->start_bpp[nblocks - 1]->b_bcount - 1) &
566 	    fs->lfs_bmask) + 1;
567 
568 	/*
569 	 * Assign disk addresses, and update references to the logical
570 	 * block and the segment usage information.
571 	 */
572 	for (i = nblocks; i--; ++sp->start_bpp) {
573 		sbp = *sp->start_bpp;
574 		lbn = *sp->start_lbp;
575 
576 		sbp->b_blkno = fsbtodb(fs, fs->lfs_offset);
577 
578 		/*
579 		 * If we write a frag in the wrong place, the cleaner won't
580 		 * be able to correctly identify its size later, and the
581 		 * segment will be uncleanable.	 (Even worse, it will assume
582 		 * that the indirect block that actually ends the list
583 		 * is of a smaller size!)
584 		 */
585 		if ((sbp->b_bcount & fs->lfs_bmask) && i != 0)
586 			errx(1, "lfs_updatemeta: fragment is not last block");
587 
588 		/*
589 		 * For each subblock in this possibly oversized block,
590 		 * update its address on disk.
591 		 */
592 		for (bytesleft = sbp->b_bcount; bytesleft > 0;
593 		    bytesleft -= fs->lfs_bsize) {
594 			size = MIN(bytesleft, fs->lfs_bsize);
595 			bb = fragstofsb(fs, numfrags(fs, size));
596 			lbn = *sp->start_lbp++;
597 			lfs_update_single(fs, sp, lbn, fs->lfs_offset, size);
598 			fs->lfs_offset += bb;
599 		}
600 
601 	}
602 }
603 
604 /*
605  * Start a new segment.
606  */
607 int
608 lfs_initseg(struct lfs * fs)
609 {
610 	struct segment *sp;
611 	SEGUSE *sup;
612 	SEGSUM *ssp;
613 	struct ubuf *bp, *sbp;
614 	int repeat;
615 
616 	sp = fs->lfs_sp;
617 
618 	repeat = 0;
619 
620 	/* Advance to the next segment. */
621 	if (!LFS_PARTIAL_FITS(fs)) {
622 		/* lfs_avail eats the remaining space */
623 		fs->lfs_avail -= fs->lfs_fsbpseg - (fs->lfs_offset -
624 		    fs->lfs_curseg);
625 		lfs_newseg(fs);
626 		repeat = 1;
627 		fs->lfs_offset = fs->lfs_curseg;
628 
629 		sp->seg_number = dtosn(fs, fs->lfs_curseg);
630 		sp->seg_bytes_left = fsbtob(fs, fs->lfs_fsbpseg);
631 
632 		/*
633 		 * If the segment contains a superblock, update the offset
634 		 * and summary address to skip over it.
635 		 */
636 		LFS_SEGENTRY(sup, fs, sp->seg_number, bp);
637 		if (sup->su_flags & SEGUSE_SUPERBLOCK) {
638 			fs->lfs_offset += btofsb(fs, LFS_SBPAD);
639 			sp->seg_bytes_left -= LFS_SBPAD;
640 		}
641 		brelse(bp);
642 		/* Segment zero could also contain the labelpad */
643 		if (fs->lfs_version > 1 && sp->seg_number == 0 &&
644 		    fs->lfs_start < btofsb(fs, LFS_LABELPAD)) {
645 			fs->lfs_offset += btofsb(fs, LFS_LABELPAD) - fs->lfs_start;
646 			sp->seg_bytes_left -= LFS_LABELPAD - fsbtob(fs, fs->lfs_start);
647 		}
648 	} else {
649 		sp->seg_number = dtosn(fs, fs->lfs_curseg);
650 		sp->seg_bytes_left = fsbtob(fs, fs->lfs_fsbpseg -
651 		    (fs->lfs_offset - fs->lfs_curseg));
652 	}
653 	fs->lfs_lastpseg = fs->lfs_offset;
654 
655 	sp->fs = fs;
656 	sp->ibp = NULL;
657 	sp->idp = NULL;
658 	sp->ninodes = 0;
659 	sp->ndupino = 0;
660 
661 	/* Get a new buffer for SEGSUM and enter it into the buffer list. */
662 	sp->cbpp = sp->bpp;
663 	sbp = *sp->cbpp = getblk(fs->lfs_unlockvp,
664 	    fsbtodb(fs, fs->lfs_offset), fs->lfs_sumsize);
665 	sp->segsum = sbp->b_data;
666 	memset(sp->segsum, 0, fs->lfs_sumsize);
667 	sp->start_bpp = ++sp->cbpp;
668 	fs->lfs_offset += btofsb(fs, fs->lfs_sumsize);
669 
670 	/* Set point to SEGSUM, initialize it. */
671 	ssp = sp->segsum;
672 	ssp->ss_next = fs->lfs_nextseg;
673 	ssp->ss_nfinfo = ssp->ss_ninos = 0;
674 	ssp->ss_magic = SS_MAGIC;
675 
676 	/* Set pointer to first FINFO, initialize it. */
677 	sp->fip = (struct finfo *) ((caddr_t) sp->segsum + SEGSUM_SIZE(fs));
678 	sp->fip->fi_nblocks = 0;
679 	sp->start_lbp = &sp->fip->fi_blocks[0];
680 	sp->fip->fi_lastlength = 0;
681 
682 	sp->seg_bytes_left -= fs->lfs_sumsize;
683 	sp->sum_bytes_left = fs->lfs_sumsize - SEGSUM_SIZE(fs);
684 
685 	LFS_LOCK_BUF(sbp);
686 	brelse(sbp);
687 	return repeat;
688 }
689 
690 /*
691  * Return the next segment to write.
692  */
693 void
694 lfs_newseg(struct lfs * fs)
695 {
696 	CLEANERINFO *cip;
697 	SEGUSE *sup;
698 	struct ubuf *bp;
699 	int curseg, isdirty, sn;
700 
701 	LFS_SEGENTRY(sup, fs, dtosn(fs, fs->lfs_nextseg), bp);
702 	sup->su_flags |= SEGUSE_DIRTY | SEGUSE_ACTIVE;
703 	sup->su_nbytes = 0;
704 	sup->su_nsums = 0;
705 	sup->su_ninos = 0;
706 	LFS_WRITESEGENTRY(sup, fs, dtosn(fs, fs->lfs_nextseg), bp);
707 
708 	LFS_CLEANERINFO(cip, fs, bp);
709 	--cip->clean;
710 	++cip->dirty;
711 	fs->lfs_nclean = cip->clean;
712 	LFS_SYNC_CLEANERINFO(cip, fs, bp, 1);
713 
714 	fs->lfs_lastseg = fs->lfs_curseg;
715 	fs->lfs_curseg = fs->lfs_nextseg;
716 	for (sn = curseg = dtosn(fs, fs->lfs_curseg) + fs->lfs_interleave;;) {
717 		sn = (sn + 1) % fs->lfs_nseg;
718 		if (sn == curseg)
719 			errx(1, "lfs_nextseg: no clean segments");
720 		LFS_SEGENTRY(sup, fs, sn, bp);
721 		isdirty = sup->su_flags & SEGUSE_DIRTY;
722 		brelse(bp);
723 
724 		if (!isdirty)
725 			break;
726 	}
727 
728 	++fs->lfs_nactive;
729 	fs->lfs_nextseg = sntod(fs, sn);
730 }
731 
732 
733 int
734 lfs_writeseg(struct lfs * fs, struct segment * sp)
735 {
736 	struct ubuf **bpp, *bp;
737 	SEGUSE *sup;
738 	SEGSUM *ssp;
739 	char *datap, *dp;
740 	int i;
741 	int do_again, nblocks, byteoffset;
742 	size_t el_size;
743 	u_short ninos;
744 	struct uvnode *devvp;
745 
746 	/*
747 	 * If there are no buffers other than the segment summary to write
748 	 * and it is not a checkpoint, don't do anything.  On a checkpoint,
749 	 * even if there aren't any buffers, you need to write the superblock.
750 	 */
751 	if ((nblocks = sp->cbpp - sp->bpp) == 1)
752 		return 0;
753 
754 	devvp = fs->lfs_unlockvp;
755 
756 	/* Update the segment usage information. */
757 	LFS_SEGENTRY(sup, fs, sp->seg_number, bp);
758 
759 	/* Loop through all blocks, except the segment summary. */
760 	for (bpp = sp->bpp; ++bpp < sp->cbpp;) {
761 		if ((*bpp)->b_vp != devvp) {
762 			sup->su_nbytes += (*bpp)->b_bcount;
763 		}
764 	}
765 
766 	ssp = (SEGSUM *) sp->segsum;
767 
768 	ninos = (ssp->ss_ninos + INOPB(fs) - 1) / INOPB(fs);
769 	sup->su_nbytes += ssp->ss_ninos * DINODE1_SIZE;
770 
771 	if (fs->lfs_version == 1)
772 		sup->su_olastmod = write_time;
773 	else
774 		sup->su_lastmod = write_time;
775 	sup->su_ninos += ninos;
776 	++sup->su_nsums;
777 	fs->lfs_dmeta += (btofsb(fs, fs->lfs_sumsize) + btofsb(fs, ninos *
778 		fs->lfs_ibsize));
779 	fs->lfs_avail -= btofsb(fs, fs->lfs_sumsize);
780 
781 	do_again = !(bp->b_flags & B_GATHERED);
782 	LFS_WRITESEGENTRY(sup, fs, sp->seg_number, bp);	/* Ifile */
783 
784 	/*
785 	 * Compute checksum across data and then across summary; the first
786 	 * block (the summary block) is skipped.  Set the create time here
787 	 * so that it's guaranteed to be later than the inode mod times.
788 	 */
789 	if (fs->lfs_version == 1)
790 		el_size = sizeof(u_long);
791 	else
792 		el_size = sizeof(u_int32_t);
793 	datap = dp = malloc(nblocks * el_size);
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_unlockvp)
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_unlockvp)
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 	sp->bpp = (struct ubuf **) malloc(fs->lfs_ssize * sizeof(struct ubuf *));
904 	if (!sp->bpp)
905 		errx(1, "Could not allocate %zu bytes: %s",
906 			(size_t)(fs->lfs_ssize * sizeof(struct ubuf *)),
907 			strerror(errno));
908 	sp->seg_flags = flags;
909 	sp->vp = NULL;
910 	sp->seg_iocount = 0;
911 	(void) lfs_initseg(fs);
912 
913 	return 0;
914 }
915 
916 /*
917  * lfs_segunlock --
918  *	Single thread the segment writer.
919  */
920 void
921 lfs_segunlock(struct lfs * fs)
922 {
923 	struct segment *sp;
924 	struct ubuf *bp;
925 
926 	sp = fs->lfs_sp;
927 
928 	if (fs->lfs_seglock == 1) {
929 		if (sp->bpp != sp->cbpp) {
930 			/* Free allocated segment summary */
931 			fs->lfs_offset -= btofsb(fs, fs->lfs_sumsize);
932 			bp = *sp->bpp;
933 			bremfree(bp);
934 			bp->b_flags |= B_DONE | B_INVAL;
935 			bp->b_flags &= ~B_DELWRI;
936 			reassignbuf(bp, bp->b_vp);
937 			bp->b_flags |= B_BUSY; /* XXX */
938 			brelse(bp);
939 		} else
940 			printf("unlock to 0 with no summary");
941 
942 		free(sp->bpp);
943 		sp->bpp = NULL;
944 		free(sp);
945 		fs->lfs_sp = NULL;
946 
947 		fs->lfs_nactive = 0;
948 
949 		/* Since we *know* everything's on disk, write both sbs */
950 		lfs_writesuper(fs, fs->lfs_sboffs[0]);
951 		lfs_writesuper(fs, fs->lfs_sboffs[1]);
952 
953 		--fs->lfs_seglock;
954 		fs->lfs_lockpid = 0;
955 	} else if (fs->lfs_seglock == 0) {
956 		errx(1, "Seglock not held");
957 	} else {
958 		--fs->lfs_seglock;
959 	}
960 }
961 
962 int
963 lfs_writevnodes(struct lfs *fs, struct segment *sp, int op)
964 {
965 	struct inode *ip;
966 	struct uvnode *vp;
967 	int inodes_written = 0;
968 
969 	LIST_FOREACH(vp, &vnodelist, v_mntvnodes) {
970 		if (vp->v_bmap_op != lfs_vop_bmap)
971 			continue;
972 
973 		ip = VTOI(vp);
974 
975 		if ((op == VN_DIROP && !(vp->v_flag & VDIROP)) ||
976 		    (op != VN_DIROP && (vp->v_flag & VDIROP))) {
977 			continue;
978 		}
979 		/*
980 		 * Write the inode/file if dirty and it's not the IFILE.
981 		 */
982 		if (ip->i_flag & IN_ALLMOD || !LIST_EMPTY(&vp->v_dirtyblkhd)) {
983 			if (ip->i_number != LFS_IFILE_INUM)
984 				lfs_writefile(fs, sp, vp);
985 			(void) lfs_writeinode(fs, sp, ip);
986 			inodes_written++;
987 		}
988 	}
989 	return inodes_written;
990 }
991 
992 void
993 lfs_writesuper(struct lfs *fs, ufs_daddr_t daddr)
994 {
995 	struct ubuf *bp;
996 
997 	/* Set timestamp of this version of the superblock */
998 	if (fs->lfs_version == 1)
999 		fs->lfs_otstamp = write_time;
1000 	fs->lfs_tstamp = write_time;
1001 
1002 	/* Checksum the superblock and copy it into a buffer. */
1003 	fs->lfs_cksum = lfs_sb_cksum(&(fs->lfs_dlfs));
1004 	assert(daddr > 0);
1005 	bp = getblk(fs->lfs_unlockvp, fsbtodb(fs, daddr), LFS_SBPAD);
1006 	memset(bp->b_data + sizeof(struct dlfs), 0,
1007 	    LFS_SBPAD - sizeof(struct dlfs));
1008 	*(struct dlfs *) bp->b_data = fs->lfs_dlfs;
1009 
1010 	bwrite(bp);
1011 }
1012