xref: /netbsd-src/sbin/fsck_lfs/segwrite.c (revision d909946ca08dceb44d7d0f22ec9488679695d976)
1 /* $NetBSD: segwrite.c,v 1.45 2015/10/03 08:28:15 dholland Exp $ */
2 /*-
3  * Copyright (c) 2003 The NetBSD Foundation, Inc.
4  * All rights reserved.
5  *
6  * This code is derived from software contributed to The NetBSD Foundation
7  * by Konrad E. Schroder <perseant@hhhh.org>.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
19  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
20  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
21  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
22  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
23  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
24  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
25  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
26  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
27  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
28  * POSSIBILITY OF SUCH DAMAGE.
29  */
30 /*
31  * Copyright (c) 1991, 1993
32  *	The Regents of the University of California.  All rights reserved.
33  *
34  * Redistribution and use in source and binary forms, with or without
35  * modification, are permitted provided that the following conditions
36  * are met:
37  * 1. Redistributions of source code must retain the above copyright
38  *    notice, this list of conditions and the following disclaimer.
39  * 2. Redistributions in binary form must reproduce the above copyright
40  *    notice, this list of conditions and the following disclaimer in the
41  *    documentation and/or other materials provided with the distribution.
42  * 3. Neither the name of the University nor the names of its contributors
43  *    may be used to endorse or promote products derived from this software
44  *    without specific prior written permission.
45  *
46  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
47  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
48  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
49  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
50  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
51  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
52  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
53  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
54  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
55  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
56  * SUCH DAMAGE.
57  *
58  *	@(#)lfs_segment.c	8.10 (Berkeley) 6/10/95
59  */
60 
61 /*
62  * Partial segment writer, taken from the kernel and adapted for userland.
63  */
64 #include <sys/types.h>
65 #include <sys/param.h>
66 #include <sys/time.h>
67 #include <sys/buf.h>
68 #include <sys/mount.h>
69 
70 /* Override certain things to make <ufs/lfs/lfs.h> work */
71 #define VU_DIROP 0x01000000 /* XXX XXX from sys/vnode.h */
72 #define vnode uvnode
73 #define buf ubuf
74 #define panic call_panic
75 
76 #include <ufs/lfs/lfs.h>
77 #include <ufs/lfs/lfs_accessors.h>
78 #include <ufs/lfs/lfs_inode.h>
79 
80 #include <assert.h>
81 #include <stdio.h>
82 #include <stdlib.h>
83 #include <string.h>
84 #include <err.h>
85 #include <errno.h>
86 #include <util.h>
87 
88 #include "bufcache.h"
89 #include "vnode.h"
90 #include "lfs_user.h"
91 #include "segwrite.h"
92 
93 /* Compatibility definitions */
94 extern off_t locked_queue_bytes;
95 int locked_queue_count;
96 off_t written_bytes = 0;
97 off_t written_data = 0;
98 off_t written_indir = 0;
99 off_t written_dev = 0;
100 int written_inodes = 0;
101 
102 /* Global variables */
103 time_t write_time;
104 
105 extern u_int32_t cksum(void *, size_t);
106 extern u_int32_t lfs_sb_cksum(struct lfs *);
107 extern int preen;
108 
109 static void lfs_shellsort(struct lfs *,
110 			  struct ubuf **, union lfs_blocks *, int, int);
111 
112 /*
113  * Logical block number match routines used when traversing the dirty block
114  * chain.
115  */
116 int
117 lfs_match_data(struct lfs * fs, struct ubuf * bp)
118 {
119 	return (bp->b_lblkno >= 0);
120 }
121 
122 int
123 lfs_match_indir(struct lfs * fs, struct ubuf * bp)
124 {
125 	daddr_t lbn;
126 
127 	lbn = bp->b_lblkno;
128 	return (lbn < 0 && (-lbn - ULFS_NDADDR) % LFS_NINDIR(fs) == 0);
129 }
130 
131 int
132 lfs_match_dindir(struct lfs * fs, struct ubuf * bp)
133 {
134 	daddr_t lbn;
135 
136 	lbn = bp->b_lblkno;
137 	return (lbn < 0 && (-lbn - ULFS_NDADDR) % LFS_NINDIR(fs) == 1);
138 }
139 
140 int
141 lfs_match_tindir(struct lfs * fs, struct ubuf * bp)
142 {
143 	daddr_t lbn;
144 
145 	lbn = bp->b_lblkno;
146 	return (lbn < 0 && (-lbn - ULFS_NDADDR) % LFS_NINDIR(fs) == 2);
147 }
148 
149 /*
150  * Do a checkpoint.
151  */
152 int
153 lfs_segwrite(struct lfs * fs, int flags)
154 {
155 	struct inode *ip;
156 	struct segment *sp;
157 	struct uvnode *vp;
158 	SEGSUM *ssp;
159 	int redo;
160 
161 	lfs_seglock(fs, flags | SEGM_CKP);
162 	sp = fs->lfs_sp;
163 
164 	lfs_writevnodes(fs, sp, VN_REG);
165 	lfs_writevnodes(fs, sp, VN_DIROP);
166 	ssp = (SEGSUM *)sp->segsum;
167 	lfs_ss_setflags(fs, ssp, lfs_ss_getflags(fs, ssp) & ~(SS_CONT));
168 
169 	do {
170 		vp = fs->lfs_ivnode;
171 		fs->lfs_flags &= ~LFS_IFDIRTY;
172 		ip = VTOI(vp);
173 		if (LIST_FIRST(&vp->v_dirtyblkhd) != NULL || lfs_sb_getidaddr(fs) <= 0)
174 			lfs_writefile(fs, sp, vp);
175 
176 		redo = lfs_writeinode(fs, sp, ip);
177 		redo += lfs_writeseg(fs, sp);
178 		redo += (fs->lfs_flags & LFS_IFDIRTY);
179 	} while (redo);
180 
181 	lfs_segunlock(fs);
182 #if 0
183 	printf("wrote %" PRId64 " bytes (%" PRId32 " fsb)\n",
184 		written_bytes, (ulfs_daddr_t)lfs_btofsb(fs, written_bytes));
185 	printf("wrote %" PRId64 " bytes data (%" PRId32 " fsb)\n",
186 		written_data, (ulfs_daddr_t)lfs_btofsb(fs, written_data));
187 	printf("wrote %" PRId64 " bytes indir (%" PRId32 " fsb)\n",
188 		written_indir, (ulfs_daddr_t)lfs_btofsb(fs, written_indir));
189 	printf("wrote %" PRId64 " bytes dev (%" PRId32 " fsb)\n",
190 		written_dev, (ulfs_daddr_t)lfs_btofsb(fs, written_dev));
191 	printf("wrote %d inodes (%" PRId32 " fsb)\n",
192 		written_inodes, lfs_btofsb(fs, written_inodes * fs->lfs_ibsize));
193 #endif
194 	return 0;
195 }
196 
197 /*
198  * Write the dirty blocks associated with a vnode.
199  */
200 void
201 lfs_writefile(struct lfs * fs, struct segment * sp, struct uvnode * vp)
202 {
203 	struct ubuf *bp;
204 	FINFO *fip;
205 	struct inode *ip;
206 	IFILE *ifp;
207 	SEGSUM *ssp;
208 
209 	ip = VTOI(vp);
210 
211 	if (sp->seg_bytes_left < lfs_sb_getbsize(fs) ||
212 	    sp->sum_bytes_left < FINFOSIZE(fs) + LFS_BLKPTRSIZE(fs))
213 		(void) lfs_writeseg(fs, sp);
214 
215 	sp->sum_bytes_left -= FINFOSIZE(fs);
216 	ssp = (SEGSUM *)sp->segsum;
217 	lfs_ss_setnfinfo(fs, ssp, lfs_ss_getnfinfo(fs, ssp) + 1);
218 
219 	if (vp->v_uflag & VU_DIROP) {
220 		lfs_ss_setflags(fs, ssp,
221 				lfs_ss_getflags(fs, ssp) | (SS_DIROP | SS_CONT));
222 	}
223 
224 	fip = sp->fip;
225 	lfs_fi_setnblocks(fs, fip, 0);
226 	lfs_fi_setino(fs, fip, ip->i_number);
227 	LFS_IENTRY(ifp, fs, lfs_fi_getino(fs, fip), bp);
228 	lfs_fi_setversion(fs, fip, lfs_if_getversion(fs, ifp));
229 	brelse(bp, 0);
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 (lfs_fi_getnblocks(fs, fip) != 0) {
238 		sp->fip = NEXT_FINFO(fs, fip);
239 		lfs_blocks_fromfinfo(fs, &sp->start_lbp, sp->fip);
240 	} else {
241 		/* XXX shouldn't this update sp->fip? */
242 		sp->sum_bytes_left += FINFOSIZE(fs);
243 		lfs_ss_setnfinfo(fs, ssp, lfs_ss_getnfinfo(fs, ssp) - 1);
244 	}
245 }
246 
247 int
248 lfs_writeinode(struct lfs * fs, struct segment * sp, struct inode * ip)
249 {
250 	struct ubuf *bp, *ibp;
251 	union lfs_dinode *cdp;
252 	IFILE *ifp;
253 	SEGUSE *sup;
254 	SEGSUM *ssp;
255 	daddr_t daddr;
256 	ino_t ino;
257 	IINFO *iip;
258 	int i, fsb = 0;
259 	int redo_ifile = 0;
260 	struct timespec ts;
261 	int gotblk = 0;
262 
263 	/* Allocate a new inode block if necessary. */
264 	if ((ip->i_number != LFS_IFILE_INUM || sp->idp == NULL) &&
265 	    sp->ibp == NULL) {
266 		/* Allocate a new segment if necessary. */
267 		if (sp->seg_bytes_left < lfs_sb_getibsize(fs) ||
268 		    sp->sum_bytes_left < LFS_BLKPTRSIZE(fs))
269 			(void) lfs_writeseg(fs, sp);
270 
271 		/* Get next inode block. */
272 		daddr = lfs_sb_getoffset(fs);
273 		lfs_sb_addoffset(fs, lfs_btofsb(fs, lfs_sb_getibsize(fs)));
274 		sp->ibp = *sp->cbpp++ =
275 		    getblk(fs->lfs_devvp, LFS_FSBTODB(fs, daddr),
276 		    lfs_sb_getibsize(fs));
277 		sp->ibp->b_flags |= B_GATHERED;
278 		gotblk++;
279 
280 		/* Zero out inode numbers */
281 		for (i = 0; i < LFS_INOPB(fs); ++i) {
282 			union lfs_dinode *tmpdip;
283 
284 			tmpdip = DINO_IN_BLOCK(fs, sp->ibp->b_data, i);
285 			lfs_dino_setinumber(fs, tmpdip, 0);
286 		}
287 
288 		++sp->start_bpp;
289 		lfs_sb_subavail(fs, lfs_btofsb(fs, lfs_sb_getibsize(fs)));
290 		/* Set remaining space counters. */
291 		sp->seg_bytes_left -= lfs_sb_getibsize(fs);
292 		sp->sum_bytes_left -= LFS_BLKPTRSIZE(fs);
293 
294 		/* Store the address in the segment summary. */
295 		iip = NTH_IINFO(fs, sp->segsum, sp->ninodes / LFS_INOPB(fs));
296 		lfs_ii_setblock(fs, iip, daddr);
297 	}
298 	/* Update the inode times and copy the inode onto the inode page. */
299 	ts.tv_nsec = 0;
300 	ts.tv_sec = write_time;
301 	/* XXX kludge --- don't redirty the ifile just to put times on it */
302 	if (ip->i_number != LFS_IFILE_INUM)
303 		LFS_ITIMES(ip, &ts, &ts, &ts);
304 
305 	/*
306 	 * If this is the Ifile, and we've already written the Ifile in this
307 	 * partial segment, just overwrite it (it's not on disk yet) and
308 	 * continue.
309 	 *
310 	 * XXX we know that the bp that we get the second time around has
311 	 * already been gathered.
312 	 */
313 	if (ip->i_number == LFS_IFILE_INUM && sp->idp) {
314 		lfs_copy_dinode(fs, sp->idp, ip->i_din);
315 		ip->i_lfs_osize = lfs_dino_getsize(fs, ip->i_din);
316 		return 0;
317 	}
318 	bp = sp->ibp;
319 	cdp = DINO_IN_BLOCK(fs, bp->b_data, sp->ninodes % LFS_INOPB(fs));
320 	lfs_copy_dinode(fs, cdp, ip->i_din);
321 
322 	/* If all blocks are goig to disk, update the "size on disk" */
323 	ip->i_lfs_osize = lfs_dino_getsize(fs, ip->i_din);
324 
325 	if (ip->i_number == LFS_IFILE_INUM)	/* We know sp->idp == NULL */
326 		sp->idp = DINO_IN_BLOCK(fs, bp->b_data, sp->ninodes % LFS_INOPB(fs));
327 	if (gotblk) {
328 		LFS_LOCK_BUF(bp);
329 		assert(!(bp->b_flags & B_INVAL));
330 		brelse(bp, 0);
331 	}
332 	/* Increment inode count in segment summary block. */
333 	ssp = (SEGSUM *)sp->segsum;
334 	lfs_ss_setninos(fs, ssp, lfs_ss_getninos(fs, ssp) + 1);
335 
336 	/* If this page is full, set flag to allocate a new page. */
337 	if (++sp->ninodes % LFS_INOPB(fs) == 0)
338 		sp->ibp = NULL;
339 
340 	/*
341 	 * If updating the ifile, update the super-block.  Update the disk
342 	 * address for this inode in the ifile.
343 	 */
344 	ino = ip->i_number;
345 	if (ino == LFS_IFILE_INUM) {
346 		daddr = lfs_sb_getidaddr(fs);
347 		lfs_sb_setidaddr(fs, LFS_DBTOFSB(fs, bp->b_blkno));
348 		sbdirty();
349 	} else {
350 		LFS_IENTRY(ifp, fs, ino, ibp);
351 		daddr = lfs_if_getdaddr(fs, ifp);
352 		lfs_if_setdaddr(fs, ifp, LFS_DBTOFSB(fs, bp->b_blkno) + fsb);
353 		(void)LFS_BWRITE_LOG(ibp);	/* Ifile */
354 	}
355 
356 	/*
357 	 * Account the inode: it no longer belongs to its former segment,
358 	 * though it will not belong to the new segment until that segment
359 	 * is actually written.
360 	 */
361 	if (daddr != LFS_UNUSED_DADDR) {
362 		u_int32_t oldsn = lfs_dtosn(fs, daddr);
363 		LFS_SEGENTRY(sup, fs, oldsn, bp);
364 		sup->su_nbytes -= DINOSIZE(fs);
365 		redo_ifile =
366 		    (ino == LFS_IFILE_INUM && !(bp->b_flags & B_GATHERED));
367 		if (redo_ifile)
368 			fs->lfs_flags |= LFS_IFDIRTY;
369 		LFS_WRITESEGENTRY(sup, fs, oldsn, bp);	/* Ifile */
370 	}
371 	return redo_ifile;
372 }
373 
374 int
375 lfs_gatherblock(struct segment * sp, struct ubuf * bp)
376 {
377 	struct lfs *fs;
378 	SEGSUM *ssp;
379 	int version;
380 	int j, blksinblk;
381 
382 	/*
383 	 * If full, finish this segment.  We may be doing I/O, so
384 	 * release and reacquire the splbio().
385 	 */
386 	fs = sp->fs;
387 	blksinblk = howmany(bp->b_bcount, lfs_sb_getbsize(fs));
388 	if (sp->sum_bytes_left < LFS_BLKPTRSIZE(fs) * blksinblk ||
389 	    sp->seg_bytes_left < bp->b_bcount) {
390 		lfs_updatemeta(sp);
391 
392 		version = lfs_fi_getversion(fs, sp->fip);
393 		(void) lfs_writeseg(fs, sp);
394 
395 		lfs_fi_setversion(fs, sp->fip, version);
396 		lfs_fi_setino(fs, sp->fip, VTOI(sp->vp)->i_number);
397 		/* Add the current file to the segment summary. */
398 		ssp = (SEGSUM *)sp->segsum;
399 		lfs_ss_setnfinfo(fs, ssp, lfs_ss_getnfinfo(fs, ssp) + 1);
400 		sp->sum_bytes_left -= FINFOSIZE(fs);
401 
402 		return 1;
403 	}
404 	/* Insert into the buffer list, update the FINFO block. */
405 	bp->b_flags |= B_GATHERED;
406 	/* bp->b_flags &= ~B_DONE; */
407 
408 	*sp->cbpp++ = bp;
409 	for (j = 0; j < blksinblk; j++) {
410 		unsigned bn;
411 
412 		bn = lfs_fi_getnblocks(fs, sp->fip);
413 		lfs_fi_setnblocks(fs, sp->fip, bn + 1);
414 		lfs_fi_setblock(fs, sp->fip, bn, bp->b_lblkno + j);;
415 	}
416 
417 	sp->sum_bytes_left -= LFS_BLKPTRSIZE(fs) * blksinblk;
418 	sp->seg_bytes_left -= bp->b_bcount;
419 	return 0;
420 }
421 
422 int
423 lfs_gather(struct lfs * fs, struct segment * sp, struct uvnode * vp, int (*match) (struct lfs *, struct ubuf *))
424 {
425 	struct ubuf *bp, *nbp;
426 	int count = 0;
427 
428 	sp->vp = vp;
429 loop:
430 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
431 		nbp = LIST_NEXT(bp, b_vnbufs);
432 
433 		assert(bp->b_flags & B_DELWRI);
434 		if ((bp->b_flags & (B_BUSY | B_GATHERED)) || !match(fs, bp)) {
435 			continue;
436 		}
437 		if (lfs_gatherblock(sp, bp)) {
438 			goto loop;
439 		}
440 		count++;
441 	}
442 
443 	lfs_updatemeta(sp);
444 	sp->vp = NULL;
445 	return count;
446 }
447 
448 
449 /*
450  * Change the given block's address to ndaddr, finding its previous
451  * location using ulfs_bmaparray().
452  *
453  * Account for this change in the segment table.
454  */
455 static void
456 lfs_update_single(struct lfs * fs, struct segment * sp, daddr_t lbn,
457     daddr_t ndaddr, int size)
458 {
459 	SEGUSE *sup;
460 	struct ubuf *bp;
461 	struct indir a[ULFS_NIADDR + 2], *ap;
462 	struct inode *ip;
463 	struct uvnode *vp;
464 	daddr_t daddr, ooff;
465 	int num, error;
466 	int osize;
467 	int frags, ofrags;
468 
469 	vp = sp->vp;
470 	ip = VTOI(vp);
471 
472 	error = ulfs_bmaparray(fs, vp, lbn, &daddr, a, &num);
473 	if (error)
474 		errx(EXIT_FAILURE, "%s: ulfs_bmaparray returned %d looking up lbn %"
475 		    PRId64 "", __func__, error, lbn);
476 	if (daddr > 0)
477 		daddr = LFS_DBTOFSB(fs, daddr);
478 
479 	frags = lfs_numfrags(fs, size);
480 	switch (num) {
481 	case 0:
482 		ooff = lfs_dino_getdb(fs, ip->i_din, lbn);
483 		if (ooff == UNWRITTEN)
484 			lfs_dino_setblocks(fs, ip->i_din,
485 			    lfs_dino_getblocks(fs, ip->i_din) + frags);
486 		else {
487 			/* possible fragment truncation or extension */
488 			ofrags = lfs_btofsb(fs, ip->i_lfs_fragsize[lbn]);
489 			lfs_dino_setblocks(fs, ip->i_din,
490 			    lfs_dino_getblocks(fs, ip->i_din) + (frags - ofrags));
491 		}
492 		lfs_dino_setdb(fs, ip->i_din, lbn, ndaddr);
493 		break;
494 	case 1:
495 		ooff = lfs_dino_getib(fs, ip->i_din, a[0].in_off);
496 		if (ooff == UNWRITTEN)
497 			lfs_dino_setblocks(fs, ip->i_din,
498 			    lfs_dino_getblocks(fs, ip->i_din) + frags);
499 		lfs_dino_setib(fs, ip->i_din, a[0].in_off, ndaddr);
500 		break;
501 	default:
502 		ap = &a[num - 1];
503 		if (bread(vp, ap->in_lbn, lfs_sb_getbsize(fs), 0, &bp))
504 			errx(EXIT_FAILURE, "%s: bread bno %" PRId64, __func__,
505 			    ap->in_lbn);
506 
507 		ooff = lfs_iblock_get(fs, bp->b_data, ap->in_off);
508 		if (ooff == UNWRITTEN)
509 			lfs_dino_setblocks(fs, ip->i_din,
510 			    lfs_dino_getblocks(fs, ip->i_din) + frags);
511 		lfs_iblock_set(fs, bp->b_data, ap->in_off, ndaddr);
512 		(void) VOP_BWRITE(bp);
513 	}
514 
515 	/*
516 	 * Update segment usage information, based on old size
517 	 * and location.
518 	 */
519 	if (daddr > 0) {
520 		u_int32_t oldsn = lfs_dtosn(fs, daddr);
521 		if (lbn >= 0 && lbn < ULFS_NDADDR)
522 			osize = ip->i_lfs_fragsize[lbn];
523 		else
524 			osize = lfs_sb_getbsize(fs);
525 		LFS_SEGENTRY(sup, fs, oldsn, bp);
526 		sup->su_nbytes -= osize;
527 		if (!(bp->b_flags & B_GATHERED))
528 			fs->lfs_flags |= LFS_IFDIRTY;
529 		LFS_WRITESEGENTRY(sup, fs, oldsn, bp);
530 	}
531 	/*
532 	 * Now that this block has a new address, and its old
533 	 * segment no longer owns it, we can forget about its
534 	 * old size.
535 	 */
536 	if (lbn >= 0 && lbn < ULFS_NDADDR)
537 		ip->i_lfs_fragsize[lbn] = size;
538 }
539 
540 /*
541  * Update the metadata that points to the blocks listed in the FINFO
542  * array.
543  */
544 void
545 lfs_updatemeta(struct segment * sp)
546 {
547 	struct ubuf *sbp;
548 	struct lfs *fs;
549 	struct uvnode *vp;
550 	daddr_t lbn;
551 	int i, nblocks, num;
552 	int frags;
553 	int bytesleft, size;
554 	union lfs_blocks tmpptr;
555 
556 	fs = sp->fs;
557 	vp = sp->vp;
558 
559 	/*
560 	 * This code was cutpasted from the kernel. See the
561 	 * corresponding comment in lfs_segment.c.
562 	 */
563 #if 0
564 	nblocks = &sp->fip->fi_blocks[sp->fip->fi_nblocks] - sp->start_lbp;
565 #else
566 	lfs_blocks_fromvoid(fs, &tmpptr, (void *)NEXT_FINFO(fs, sp->fip));
567 	nblocks = lfs_blocks_sub(fs, &tmpptr, &sp->start_lbp);
568 	//nblocks_orig = nblocks;
569 #endif
570 
571 	if (vp == NULL || nblocks == 0)
572 		return;
573 
574 	/*
575 	 * This count may be high due to oversize blocks from lfs_gop_write.
576 	 * Correct for this. (XXX we should be able to keep track of these.)
577 	 */
578 	for (i = 0; i < nblocks; i++) {
579 		if (sp->start_bpp[i] == NULL) {
580 			printf("nblocks = %d, not %d\n", i, nblocks);
581 			nblocks = i;
582 			break;
583 		}
584 		num = howmany(sp->start_bpp[i]->b_bcount, lfs_sb_getbsize(fs));
585 		nblocks -= num - 1;
586 	}
587 
588 	/*
589 	 * Sort the blocks.
590 	 */
591 	lfs_shellsort(fs, sp->start_bpp, &sp->start_lbp, nblocks, lfs_sb_getbsize(fs));
592 
593 	/*
594 	 * Record the length of the last block in case it's a fragment.
595 	 * If there are indirect blocks present, they sort last.  An
596 	 * indirect block will be lfs_bsize and its presence indicates
597 	 * that you cannot have fragments.
598 	 */
599 	lfs_fi_setlastlength(fs, sp->fip, ((sp->start_bpp[nblocks - 1]->b_bcount - 1) &
600 	    lfs_sb_getbmask(fs)) + 1);
601 
602 	/*
603 	 * Assign disk addresses, and update references to the logical
604 	 * block and the segment usage information.
605 	 */
606 	for (i = nblocks; i--; ++sp->start_bpp) {
607 		sbp = *sp->start_bpp;
608 		lbn = lfs_blocks_get(fs, &sp->start_lbp, 0);
609 
610 		sbp->b_blkno = LFS_FSBTODB(fs, lfs_sb_getoffset(fs));
611 
612 		/*
613 		 * If we write a frag in the wrong place, the cleaner won't
614 		 * be able to correctly identify its size later, and the
615 		 * segment will be uncleanable.	 (Even worse, it will assume
616 		 * that the indirect block that actually ends the list
617 		 * is of a smaller size!)
618 		 */
619 		if ((sbp->b_bcount & lfs_sb_getbmask(fs)) && i != 0)
620 			errx(EXIT_FAILURE, "%s: fragment is not last block", __func__);
621 
622 		/*
623 		 * For each subblock in this possibly oversized block,
624 		 * update its address on disk.
625 		 */
626 		for (bytesleft = sbp->b_bcount; bytesleft > 0;
627 		    bytesleft -= lfs_sb_getbsize(fs)) {
628 			size = MIN(bytesleft, lfs_sb_getbsize(fs));
629 			frags = lfs_numfrags(fs, size);
630 			lbn = lfs_blocks_get(fs, &sp->start_lbp, 0);
631 			lfs_blocks_inc(fs, &sp->start_lbp);
632 			lfs_update_single(fs, sp, lbn, lfs_sb_getoffset(fs), size);
633 			lfs_sb_addoffset(fs, frags);
634 		}
635 
636 	}
637 }
638 
639 /*
640  * Start a new segment.
641  */
642 int
643 lfs_initseg(struct lfs * fs)
644 {
645 	struct segment *sp;
646 	SEGUSE *sup;
647 	SEGSUM *ssp;
648 	struct ubuf *bp, *sbp;
649 	int repeat;
650 
651 	sp = fs->lfs_sp;
652 
653 	repeat = 0;
654 
655 	/* Advance to the next segment. */
656 	if (!LFS_PARTIAL_FITS(fs)) {
657 		/* lfs_avail eats the remaining space */
658 		lfs_sb_subavail(fs, lfs_sb_getfsbpseg(fs) - (lfs_sb_getoffset(fs) -
659 		    lfs_sb_getcurseg(fs)));
660 		lfs_newseg(fs);
661 		repeat = 1;
662 		lfs_sb_setoffset(fs, lfs_sb_getcurseg(fs));
663 
664 		sp->seg_number = lfs_dtosn(fs, lfs_sb_getcurseg(fs));
665 		sp->seg_bytes_left = lfs_fsbtob(fs, lfs_sb_getfsbpseg(fs));
666 
667 		/*
668 		 * If the segment contains a superblock, update the offset
669 		 * and summary address to skip over it.
670 		 */
671 		LFS_SEGENTRY(sup, fs, sp->seg_number, bp);
672 		if (sup->su_flags & SEGUSE_SUPERBLOCK) {
673 			lfs_sb_addoffset(fs, lfs_btofsb(fs, LFS_SBPAD));
674 			sp->seg_bytes_left -= LFS_SBPAD;
675 		}
676 		brelse(bp, 0);
677 		/* Segment zero could also contain the labelpad */
678 		if (lfs_sb_getversion(fs) > 1 && sp->seg_number == 0 &&
679 		    lfs_sb_gets0addr(fs) < lfs_btofsb(fs, LFS_LABELPAD)) {
680 			lfs_sb_addoffset(fs, lfs_btofsb(fs, LFS_LABELPAD) - lfs_sb_gets0addr(fs));
681 			sp->seg_bytes_left -= LFS_LABELPAD - lfs_fsbtob(fs, lfs_sb_gets0addr(fs));
682 		}
683 	} else {
684 		sp->seg_number = lfs_dtosn(fs, lfs_sb_getcurseg(fs));
685 		sp->seg_bytes_left = lfs_fsbtob(fs, lfs_sb_getfsbpseg(fs) -
686 		    (lfs_sb_getoffset(fs) - lfs_sb_getcurseg(fs)));
687 	}
688 	lfs_sb_setlastpseg(fs, lfs_sb_getoffset(fs));
689 
690 	sp->fs = fs;
691 	sp->ibp = NULL;
692 	sp->idp = NULL;
693 	sp->ninodes = 0;
694 	sp->ndupino = 0;
695 
696 	/* Get a new buffer for SEGSUM and enter it into the buffer list. */
697 	sp->cbpp = sp->bpp;
698 	sbp = *sp->cbpp = getblk(fs->lfs_devvp,
699 	    LFS_FSBTODB(fs, lfs_sb_getoffset(fs)), lfs_sb_getsumsize(fs));
700 	sp->segsum = sbp->b_data;
701 	memset(sp->segsum, 0, lfs_sb_getsumsize(fs));
702 	sp->start_bpp = ++sp->cbpp;
703 	lfs_sb_addoffset(fs, lfs_btofsb(fs, lfs_sb_getsumsize(fs)));
704 
705 	/* Set point to SEGSUM, initialize it. */
706 	ssp = sp->segsum;
707 	lfs_ss_setnext(fs, ssp, lfs_sb_getnextseg(fs));
708 	lfs_ss_setnfinfo(fs, ssp, 0);
709 	lfs_ss_setninos(fs, ssp, 0);
710 	lfs_ss_setmagic(fs, ssp, SS_MAGIC);
711 
712 	/* Set pointer to first FINFO, initialize it. */
713 	sp->fip = SEGSUM_FINFOBASE(fs, ssp);
714 	lfs_fi_setnblocks(fs, sp->fip, 0);
715 	lfs_blocks_fromfinfo(fs, &sp->start_lbp, sp->fip);
716 	lfs_fi_setlastlength(fs, sp->fip, 0);
717 
718 	sp->seg_bytes_left -= lfs_sb_getsumsize(fs);
719 	sp->sum_bytes_left = lfs_sb_getsumsize(fs) - SEGSUM_SIZE(fs);
720 
721 	LFS_LOCK_BUF(sbp);
722 	brelse(sbp, 0);
723 	return repeat;
724 }
725 
726 /*
727  * Return the next segment to write.
728  */
729 void
730 lfs_newseg(struct lfs * fs)
731 {
732 	CLEANERINFO *cip;
733 	SEGUSE *sup;
734 	struct ubuf *bp;
735 	int curseg, isdirty, sn;
736 
737 	LFS_SEGENTRY(sup, fs, lfs_dtosn(fs, lfs_sb_getnextseg(fs)), bp);
738 	sup->su_flags |= SEGUSE_DIRTY | SEGUSE_ACTIVE;
739 	sup->su_nbytes = 0;
740 	sup->su_nsums = 0;
741 	sup->su_ninos = 0;
742 	LFS_WRITESEGENTRY(sup, fs, lfs_dtosn(fs, lfs_sb_getnextseg(fs)), bp);
743 
744 	LFS_CLEANERINFO(cip, fs, bp);
745 	lfs_ci_shiftcleantodirty(fs, cip, 1);
746 	lfs_sb_setnclean(fs, lfs_ci_getclean(fs, cip));
747 	LFS_SYNC_CLEANERINFO(cip, fs, bp, 1);
748 
749 	lfs_sb_setlastseg(fs, lfs_sb_getcurseg(fs));
750 	lfs_sb_setcurseg(fs, lfs_sb_getnextseg(fs));
751 	for (sn = curseg = lfs_dtosn(fs, lfs_sb_getcurseg(fs)) + lfs_sb_getinterleave(fs);;) {
752 		sn = (sn + 1) % lfs_sb_getnseg(fs);
753 		if (sn == curseg)
754 			errx(EXIT_FAILURE, "%s: no clean segments", __func__);
755 		LFS_SEGENTRY(sup, fs, sn, bp);
756 		isdirty = sup->su_flags & SEGUSE_DIRTY;
757 		brelse(bp, 0);
758 
759 		if (!isdirty)
760 			break;
761 	}
762 
763 	++fs->lfs_nactive;
764 	lfs_sb_setnextseg(fs, lfs_sntod(fs, sn));
765 }
766 
767 
768 int
769 lfs_writeseg(struct lfs * fs, struct segment * sp)
770 {
771 	struct ubuf **bpp, *bp;
772 	SEGUSE *sup;
773 	SEGSUM *ssp;
774 	char *datap, *dp;
775 	int i;
776 	int do_again, nblocks, byteoffset;
777 	size_t el_size;
778 	u_short ninos;
779 	size_t sumstart;
780 	struct uvnode *devvp;
781 
782 	/*
783 	 * If there are no buffers other than the segment summary to write
784 	 * and it is not a checkpoint, don't do anything.  On a checkpoint,
785 	 * even if there aren't any buffers, you need to write the superblock.
786 	 */
787 	nblocks = sp->cbpp - sp->bpp;
788 #if 0
789 	printf("write %d blocks at 0x%x\n",
790 		nblocks, (int)LFS_DBTOFSB(fs, (*sp->bpp)->b_blkno));
791 #endif
792 	if (nblocks == 1)
793 		return 0;
794 
795 	devvp = fs->lfs_devvp;
796 
797 	/* Update the segment usage information. */
798 	LFS_SEGENTRY(sup, fs, sp->seg_number, bp);
799 	sup->su_flags |= SEGUSE_DIRTY | SEGUSE_ACTIVE;
800 
801 	/* Loop through all blocks, except the segment summary. */
802 	for (bpp = sp->bpp; ++bpp < sp->cbpp;) {
803 		if ((*bpp)->b_vp != devvp) {
804 			sup->su_nbytes += (*bpp)->b_bcount;
805 		}
806 		assert(lfs_dtosn(fs, LFS_DBTOFSB(fs, (*bpp)->b_blkno)) == sp->seg_number);
807 	}
808 
809 	ssp = (SEGSUM *) sp->segsum;
810 	lfs_ss_setflags(fs, ssp, lfs_ss_getflags(fs, ssp) | SS_RFW);
811 
812 	ninos = (lfs_ss_getninos(fs, ssp) + LFS_INOPB(fs) - 1) / LFS_INOPB(fs);
813 	sup->su_nbytes += lfs_ss_getninos(fs, ssp) * DINOSIZE(fs);
814 
815 	if (lfs_sb_getversion(fs) == 1)
816 		sup->su_olastmod = write_time;
817 	else
818 		sup->su_lastmod = write_time;
819 	sup->su_ninos += ninos;
820 	++sup->su_nsums;
821 	lfs_sb_adddmeta(fs, (lfs_btofsb(fs, lfs_sb_getsumsize(fs)) + lfs_btofsb(fs, ninos *
822 		lfs_sb_getibsize(fs))));
823 	lfs_sb_subavail(fs, lfs_btofsb(fs, lfs_sb_getsumsize(fs)));
824 
825 	do_again = !(bp->b_flags & B_GATHERED);
826 	LFS_WRITESEGENTRY(sup, fs, sp->seg_number, bp);	/* Ifile */
827 
828 	/*
829 	 * Compute checksum across data and then across summary; the first
830 	 * block (the summary block) is skipped.  Set the create time here
831 	 * so that it's guaranteed to be later than the inode mod times.
832 	 */
833 	if (lfs_sb_getversion(fs) == 1)
834 		el_size = sizeof(u_long);
835 	else
836 		el_size = sizeof(u_int32_t);
837 	datap = dp = emalloc(nblocks * el_size);
838 	for (bpp = sp->bpp, i = nblocks - 1; i--;) {
839 		++bpp;
840 		/* Loop through gop_write cluster blocks */
841 		for (byteoffset = 0; byteoffset < (*bpp)->b_bcount;
842 		    byteoffset += lfs_sb_getbsize(fs)) {
843 			memcpy(dp, (*bpp)->b_data + byteoffset, el_size);
844 			dp += el_size;
845 		}
846 		bremfree(*bpp);
847 		(*bpp)->b_flags |= B_BUSY;
848 	}
849 	if (lfs_sb_getversion(fs) == 1)
850 		lfs_ss_setocreate(fs, ssp, write_time);
851 	else {
852 		lfs_ss_setcreate(fs, ssp, write_time);
853 		lfs_sb_addserial(fs, 1);
854 		lfs_ss_setserial(fs, ssp, lfs_sb_getserial(fs));
855 		lfs_ss_setident(fs, ssp, lfs_sb_getident(fs));
856 	}
857 	/* Set the summary block busy too */
858 	bremfree(*(sp->bpp));
859 	(*(sp->bpp))->b_flags |= B_BUSY;
860 
861 	lfs_ss_setdatasum(fs, ssp, cksum(datap, (nblocks - 1) * el_size));
862 	sumstart = lfs_ss_getsumstart(fs);
863 	lfs_ss_setsumsum(fs, ssp,
864 	    cksum((char *)ssp + sumstart, lfs_sb_getsumsize(fs) - sumstart));
865 	free(datap);
866 	datap = dp = NULL;
867 	lfs_sb_subbfree(fs, (lfs_btofsb(fs, ninos * lfs_sb_getibsize(fs)) +
868 	    lfs_btofsb(fs, lfs_sb_getsumsize(fs))));
869 
870 	if (devvp == NULL)
871 		errx(EXIT_FAILURE, "devvp is NULL");
872 	for (bpp = sp->bpp, i = nblocks; i; bpp++, i--) {
873 		bp = *bpp;
874 #if 0
875 		printf("i = %d, bp = %p, flags %lx, bn = %" PRIx64 "\n",
876 		       nblocks - i, bp, bp->b_flags, bp->b_blkno);
877 		printf("  vp = %p\n", bp->b_vp);
878 		if (bp->b_vp != fs->lfs_devvp)
879 			printf("  ino = %d lbn = %" PRId64 "\n",
880 			       VTOI(bp->b_vp)->i_number, bp->b_lblkno);
881 #endif
882 		if (bp->b_vp == fs->lfs_devvp)
883 			written_dev += bp->b_bcount;
884 		else {
885 			if (bp->b_lblkno >= 0)
886 				written_data += bp->b_bcount;
887 			else
888 				written_indir += bp->b_bcount;
889 		}
890 		bp->b_flags &= ~(B_DELWRI | B_READ | B_GATHERED | B_ERROR |
891 				 B_LOCKED);
892 		bwrite(bp);
893 		written_bytes += bp->b_bcount;
894 	}
895 	written_inodes += ninos;
896 
897 	return (lfs_initseg(fs) || do_again);
898 }
899 
900 /*
901  * Our own copy of shellsort.  XXX use qsort or heapsort.
902  */
903 static void
904 lfs_shellsort(struct lfs *fs,
905 	      struct ubuf ** bp_array, union lfs_blocks *lb_array, int nmemb, int size)
906 {
907 	static int __rsshell_increments[] = {4, 1, 0};
908 	int incr, *incrp, t1, t2;
909 	struct ubuf *bp_temp;
910 
911 	for (incrp = __rsshell_increments; (incr = *incrp++) != 0;)
912 		for (t1 = incr; t1 < nmemb; ++t1)
913 			for (t2 = t1 - incr; t2 >= 0;)
914 				if ((u_int32_t) bp_array[t2]->b_lblkno >
915 				    (u_int32_t) bp_array[t2 + incr]->b_lblkno) {
916 					bp_temp = bp_array[t2];
917 					bp_array[t2] = bp_array[t2 + incr];
918 					bp_array[t2 + incr] = bp_temp;
919 					t2 -= incr;
920 				} else
921 					break;
922 
923 	/* Reform the list of logical blocks */
924 	incr = 0;
925 	for (t1 = 0; t1 < nmemb; t1++) {
926 		for (t2 = 0; t2 * size < bp_array[t1]->b_bcount; t2++) {
927 			lfs_blocks_set(fs, lb_array, incr++,
928 				       bp_array[t1]->b_lblkno + t2);
929 		}
930 	}
931 }
932 
933 
934 /*
935  * lfs_seglock --
936  *	Single thread the segment writer.
937  */
938 int
939 lfs_seglock(struct lfs * fs, unsigned long flags)
940 {
941 	struct segment *sp;
942 	size_t allocsize;
943 
944 	if (fs->lfs_seglock) {
945 		++fs->lfs_seglock;
946 		fs->lfs_sp->seg_flags |= flags;
947 		return 0;
948 	}
949 	fs->lfs_seglock = 1;
950 
951 	sp = fs->lfs_sp = emalloc(sizeof(*sp));
952 	allocsize = lfs_sb_getssize(fs) * sizeof(struct ubuf *);
953 	sp->bpp = emalloc(allocsize);
954 	if (!sp->bpp)
955 		err(!preen, "Could not allocate %zu bytes", allocsize);
956 	sp->seg_flags = flags;
957 	sp->vp = NULL;
958 	sp->seg_iocount = 0;
959 	(void) lfs_initseg(fs);
960 
961 	return 0;
962 }
963 
964 /*
965  * lfs_segunlock --
966  *	Single thread the segment writer.
967  */
968 void
969 lfs_segunlock(struct lfs * fs)
970 {
971 	struct segment *sp;
972 	struct ubuf *bp;
973 
974 	sp = fs->lfs_sp;
975 
976 	if (fs->lfs_seglock == 1) {
977 		if (sp->bpp != sp->cbpp) {
978 			/* Free allocated segment summary */
979 			lfs_sb_suboffset(fs, lfs_btofsb(fs, lfs_sb_getsumsize(fs)));
980 			bp = *sp->bpp;
981 			bremfree(bp);
982 			bp->b_flags |= B_DONE | B_INVAL;
983 			bp->b_flags &= ~B_DELWRI;
984 			reassignbuf(bp, bp->b_vp);
985 			bp->b_flags |= B_BUSY; /* XXX */
986 			brelse(bp, 0);
987 		} else
988 			printf("unlock to 0 with no summary");
989 
990 		free(sp->bpp);
991 		sp->bpp = NULL;
992 		free(sp);
993 		fs->lfs_sp = NULL;
994 
995 		fs->lfs_nactive = 0;
996 
997 		/* Since we *know* everything's on disk, write both sbs */
998 		lfs_writesuper(fs, lfs_sb_getsboff(fs, 0));
999 		lfs_writesuper(fs, lfs_sb_getsboff(fs, 1));
1000 
1001 		--fs->lfs_seglock;
1002 		fs->lfs_lockpid = 0;
1003 	} else if (fs->lfs_seglock == 0) {
1004 		errx(EXIT_FAILURE, "Seglock not held");
1005 	} else {
1006 		--fs->lfs_seglock;
1007 	}
1008 }
1009 
1010 int
1011 lfs_writevnodes(struct lfs *fs, struct segment *sp, int op)
1012 {
1013 	struct inode *ip;
1014 	struct uvnode *vp;
1015 	int inodes_written = 0;
1016 
1017 	LIST_FOREACH(vp, &vnodelist, v_mntvnodes) {
1018 		if (vp->v_bmap_op != lfs_vop_bmap)
1019 			continue;
1020 
1021 		ip = VTOI(vp);
1022 
1023 		if ((op == VN_DIROP && !(vp->v_uflag & VU_DIROP)) ||
1024 		    (op != VN_DIROP && (vp->v_uflag & VU_DIROP))) {
1025 			continue;
1026 		}
1027 		/*
1028 		 * Write the inode/file if dirty and it's not the IFILE.
1029 		 */
1030 		if (ip->i_flag & IN_ALLMOD || !LIST_EMPTY(&vp->v_dirtyblkhd)) {
1031 			if (ip->i_number != LFS_IFILE_INUM)
1032 				lfs_writefile(fs, sp, vp);
1033 			(void) lfs_writeinode(fs, sp, ip);
1034 			inodes_written++;
1035 		}
1036 	}
1037 	return inodes_written;
1038 }
1039 
1040 void
1041 lfs_writesuper(struct lfs *fs, daddr_t daddr)
1042 {
1043 	struct ubuf *bp;
1044 
1045 	/* Set timestamp of this version of the superblock */
1046 	if (lfs_sb_getversion(fs) == 1)
1047 		lfs_sb_setotstamp(fs, write_time);
1048 	lfs_sb_settstamp(fs, write_time);
1049 
1050 	__CTASSERT(sizeof(struct dlfs) == sizeof(struct dlfs64));
1051 
1052 	/* Checksum the superblock and copy it into a buffer. */
1053 	lfs_sb_setcksum(fs, lfs_sb_cksum(fs));
1054 	assert(daddr > 0);
1055 	bp = getblk(fs->lfs_devvp, LFS_FSBTODB(fs, daddr), LFS_SBPAD);
1056 	memcpy(bp->b_data, &fs->lfs_dlfs_u, sizeof(struct dlfs));
1057 	memset(bp->b_data + sizeof(struct dlfs), 0,
1058 	    LFS_SBPAD - sizeof(struct dlfs));
1059 
1060 	bwrite(bp);
1061 }
1062