xref: /freebsd/usr.sbin/makefs/ffs/ffs_alloc.c (revision 4d65a7c6)
1 /*	$NetBSD: ffs_alloc.c,v 1.14 2004/06/20 22:20:18 jmc Exp $	*/
2 /* From: NetBSD: ffs_alloc.c,v 1.50 2001/09/06 02:16:01 lukem Exp */
3 
4 /*-
5  * SPDX-License-Identifier: BSD-3-Clause
6  *
7  * Copyright (c) 2002 Networks Associates Technology, Inc.
8  * All rights reserved.
9  *
10  * This software was developed for the FreeBSD Project by Marshall
11  * Kirk McKusick and Network Associates Laboratories, the Security
12  * Research Division of Network Associates, Inc. under DARPA/SPAWAR
13  * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS
14  * research program
15  *
16  * Copyright (c) 1982, 1986, 1989, 1993
17  *	The Regents of the University of California.  All rights reserved.
18  *
19  * Redistribution and use in source and binary forms, with or without
20  * modification, are permitted provided that the following conditions
21  * are met:
22  * 1. Redistributions of source code must retain the above copyright
23  *    notice, this list of conditions and the following disclaimer.
24  * 2. Redistributions in binary form must reproduce the above copyright
25  *    notice, this list of conditions and the following disclaimer in the
26  *    documentation and/or other materials provided with the distribution.
27  * 3. Neither the name of the University nor the names of its contributors
28  *    may be used to endorse or promote products derived from this software
29  *    without specific prior written permission.
30  *
31  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
32  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
33  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
34  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
35  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
36  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
37  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
38  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
39  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
40  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
41  * SUCH DAMAGE.
42  */
43 
44 #include <sys/param.h>
45 #include <sys/time.h>
46 
47 #include <errno.h>
48 #include <stdint.h>
49 
50 #include "makefs.h"
51 
52 #include <ufs/ufs/dinode.h>
53 #include <ufs/ffs/fs.h>
54 
55 #include "ffs/ufs_bswap.h"
56 #include "ffs/buf.h"
57 #include "ffs/ufs_inode.h"
58 #include "ffs/ffs_extern.h"
59 
60 static int scanc(u_int, const u_char *, const u_char *, int);
61 
62 static daddr_t ffs_alloccg(struct inode *, int, daddr_t, int);
63 static daddr_t ffs_alloccgblk(struct inode *, struct m_buf *, daddr_t);
64 static daddr_t ffs_hashalloc(struct inode *, u_int, daddr_t, int,
65 		     daddr_t (*)(struct inode *, int, daddr_t, int));
66 static int32_t ffs_mapsearch(struct fs *, struct cg *, daddr_t, int);
67 
68 /*
69  * Allocate a block in the file system.
70  *
71  * The size of the requested block is given, which must be some
72  * multiple of fs_fsize and <= fs_bsize.
73  * A preference may be optionally specified. If a preference is given
74  * the following hierarchy is used to allocate a block:
75  *   1) allocate the requested block.
76  *   2) allocate a rotationally optimal block in the same cylinder.
77  *   3) allocate a block in the same cylinder group.
78  *   4) quadratically rehash into other cylinder groups, until an
79  *      available block is located.
80  * If no block preference is given the following hierarchy is used
81  * to allocate a block:
82  *   1) allocate a block in the cylinder group that contains the
83  *      inode for the file.
84  *   2) quadratically rehash into other cylinder groups, until an
85  *      available block is located.
86  */
87 int
ffs_alloc(struct inode * ip,daddr_t lbn __unused,daddr_t bpref,int size,daddr_t * bnp)88 ffs_alloc(struct inode *ip, daddr_t lbn __unused, daddr_t bpref, int size,
89     daddr_t *bnp)
90 {
91 	struct fs *fs = ip->i_fs;
92 	daddr_t bno;
93 	int cg;
94 
95 	*bnp = 0;
96 	if (size > fs->fs_bsize || fragoff(fs, size) != 0) {
97 		errx(1, "ffs_alloc: bad size: bsize %d size %d",
98 		    fs->fs_bsize, size);
99 	}
100 	if (size == fs->fs_bsize && fs->fs_cstotal.cs_nbfree == 0)
101 		goto nospace;
102 	if (bpref >= fs->fs_size)
103 		bpref = 0;
104 	if (bpref == 0)
105 		cg = ino_to_cg(fs, ip->i_number);
106 	else
107 		cg = dtog(fs, bpref);
108 	bno = ffs_hashalloc(ip, cg, bpref, size, ffs_alloccg);
109 	if (bno > 0) {
110 		if (ip->i_fs->fs_magic == FS_UFS1_MAGIC)
111 			ip->i_ffs1_blocks += size / DEV_BSIZE;
112 		else
113 			ip->i_ffs2_blocks += size / DEV_BSIZE;
114 		*bnp = bno;
115 		return (0);
116 	}
117 nospace:
118 	return (ENOSPC);
119 }
120 
121 /*
122  * Select the desired position for the next block in a file.  The file is
123  * logically divided into sections. The first section is composed of the
124  * direct blocks. Each additional section contains fs_maxbpg blocks.
125  *
126  * If no blocks have been allocated in the first section, the policy is to
127  * request a block in the same cylinder group as the inode that describes
128  * the file. If no blocks have been allocated in any other section, the
129  * policy is to place the section in a cylinder group with a greater than
130  * average number of free blocks.  An appropriate cylinder group is found
131  * by using a rotor that sweeps the cylinder groups. When a new group of
132  * blocks is needed, the sweep begins in the cylinder group following the
133  * cylinder group from which the previous allocation was made. The sweep
134  * continues until a cylinder group with greater than the average number
135  * of free blocks is found. If the allocation is for the first block in an
136  * indirect block, the information on the previous allocation is unavailable;
137  * here a best guess is made based upon the logical block number being
138  * allocated.
139  *
140  * If a section is already partially allocated, the policy is to
141  * contiguously allocate fs_maxcontig blocks.  The end of one of these
142  * contiguous blocks and the beginning of the next is physically separated
143  * so that the disk head will be in transit between them for at least
144  * fs_rotdelay milliseconds.  This is to allow time for the processor to
145  * schedule another I/O transfer.
146  */
147 /* XXX ondisk32 */
148 daddr_t
ffs_blkpref_ufs1(struct inode * ip,daddr_t lbn,int indx,int32_t * bap)149 ffs_blkpref_ufs1(struct inode *ip, daddr_t lbn, int indx, int32_t *bap)
150 {
151 	struct fs *fs;
152 	u_int cg, startcg;
153 	int avgbfree;
154 
155 	fs = ip->i_fs;
156 	if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) {
157 		if (lbn < UFS_NDADDR + NINDIR(fs)) {
158 			cg = ino_to_cg(fs, ip->i_number);
159 			return (fs->fs_fpg * cg + fs->fs_frag);
160 		}
161 		/*
162 		 * Find a cylinder with greater than average number of
163 		 * unused data blocks.
164 		 */
165 		if (indx == 0 || bap[indx - 1] == 0)
166 			startcg =
167 			    ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg;
168 		else
169 			startcg = dtog(fs,
170 				ufs_rw32(bap[indx - 1], UFS_FSNEEDSWAP(fs)) + 1);
171 		startcg %= fs->fs_ncg;
172 		avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
173 		for (cg = startcg; cg < fs->fs_ncg; cg++)
174 			if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree)
175 				return (fs->fs_fpg * cg + fs->fs_frag);
176 		for (cg = 0; cg <= startcg; cg++)
177 			if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree)
178 				return (fs->fs_fpg * cg + fs->fs_frag);
179 		return (0);
180 	}
181 	/*
182 	 * We just always try to lay things out contiguously.
183 	 */
184 	return ufs_rw32(bap[indx - 1], UFS_FSNEEDSWAP(fs)) + fs->fs_frag;
185 }
186 
187 daddr_t
ffs_blkpref_ufs2(struct inode * ip,daddr_t lbn,int indx,int64_t * bap)188 ffs_blkpref_ufs2(struct inode *ip, daddr_t lbn, int indx, int64_t *bap)
189 {
190 	struct fs *fs;
191 	u_int cg, startcg;
192 	int avgbfree;
193 
194 	fs = ip->i_fs;
195 	if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) {
196 		if (lbn < UFS_NDADDR + NINDIR(fs)) {
197 			cg = ino_to_cg(fs, ip->i_number);
198 			return (fs->fs_fpg * cg + fs->fs_frag);
199 		}
200 		/*
201 		 * Find a cylinder with greater than average number of
202 		 * unused data blocks.
203 		 */
204 		if (indx == 0 || bap[indx - 1] == 0)
205 			startcg =
206 			    ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg;
207 		else
208 			startcg = dtog(fs,
209 				ufs_rw64(bap[indx - 1], UFS_FSNEEDSWAP(fs)) + 1);
210 		startcg %= fs->fs_ncg;
211 		avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
212 		for (cg = startcg; cg < fs->fs_ncg; cg++)
213 			if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
214 				return (fs->fs_fpg * cg + fs->fs_frag);
215 			}
216 		for (cg = 0; cg < startcg; cg++)
217 			if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
218 				return (fs->fs_fpg * cg + fs->fs_frag);
219 			}
220 		return (0);
221 	}
222 	/*
223 	 * We just always try to lay things out contiguously.
224 	 */
225 	return ufs_rw64(bap[indx - 1], UFS_FSNEEDSWAP(fs)) + fs->fs_frag;
226 }
227 
228 /*
229  * Implement the cylinder overflow algorithm.
230  *
231  * The policy implemented by this algorithm is:
232  *   1) allocate the block in its requested cylinder group.
233  *   2) quadratically rehash on the cylinder group number.
234  *   3) brute force search for a free block.
235  *
236  * `size':	size for data blocks, mode for inodes
237  */
238 /*VARARGS5*/
239 static daddr_t
ffs_hashalloc(struct inode * ip,u_int cg,daddr_t pref,int size,daddr_t (* allocator)(struct inode *,int,daddr_t,int))240 ffs_hashalloc(struct inode *ip, u_int cg, daddr_t pref, int size,
241     daddr_t (*allocator)(struct inode *, int, daddr_t, int))
242 {
243 	struct fs *fs;
244 	daddr_t result;
245 	u_int i, icg = cg;
246 
247 	fs = ip->i_fs;
248 	/*
249 	 * 1: preferred cylinder group
250 	 */
251 	result = (*allocator)(ip, cg, pref, size);
252 	if (result)
253 		return (result);
254 	/*
255 	 * 2: quadratic rehash
256 	 */
257 	for (i = 1; i < fs->fs_ncg; i *= 2) {
258 		cg += i;
259 		if (cg >= fs->fs_ncg)
260 			cg -= fs->fs_ncg;
261 		result = (*allocator)(ip, cg, 0, size);
262 		if (result)
263 			return (result);
264 	}
265 	/*
266 	 * 3: brute force search
267 	 * Note that we start at i == 2, since 0 was checked initially,
268 	 * and 1 is always checked in the quadratic rehash.
269 	 */
270 	cg = (icg + 2) % fs->fs_ncg;
271 	for (i = 2; i < fs->fs_ncg; i++) {
272 		result = (*allocator)(ip, cg, 0, size);
273 		if (result)
274 			return (result);
275 		cg++;
276 		if (cg == fs->fs_ncg)
277 			cg = 0;
278 	}
279 	return (0);
280 }
281 
282 /*
283  * Determine whether a block can be allocated.
284  *
285  * Check to see if a block of the appropriate size is available,
286  * and if it is, allocate it.
287  */
288 static daddr_t
ffs_alloccg(struct inode * ip,int cg,daddr_t bpref,int size)289 ffs_alloccg(struct inode *ip, int cg, daddr_t bpref, int size)
290 {
291 	struct cg *cgp;
292 	struct m_buf *bp;
293 	daddr_t bno, blkno;
294 	int error, frags, allocsiz, i;
295 	struct fs *fs = ip->i_fs;
296 	const int needswap = UFS_FSNEEDSWAP(fs);
297 
298 	if (fs->fs_cs(fs, cg).cs_nbfree == 0 && size == fs->fs_bsize)
299 		return (0);
300 	error = bread((void *)ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
301 	    (int)fs->fs_cgsize, NULL, &bp);
302 	if (error) {
303 		return (0);
304 	}
305 	cgp = (struct cg *)bp->b_data;
306 	if (!cg_chkmagic_swap(cgp, needswap) ||
307 	    (cgp->cg_cs.cs_nbfree == 0 && size == fs->fs_bsize)) {
308 		brelse(bp);
309 		return (0);
310 	}
311 	if (size == fs->fs_bsize) {
312 		bno = ffs_alloccgblk(ip, bp, bpref);
313 		bdwrite(bp);
314 		return (bno);
315 	}
316 	/*
317 	 * check to see if any fragments are already available
318 	 * allocsiz is the size which will be allocated, hacking
319 	 * it down to a smaller size if necessary
320 	 */
321 	frags = numfrags(fs, size);
322 	for (allocsiz = frags; allocsiz < fs->fs_frag; allocsiz++)
323 		if (cgp->cg_frsum[allocsiz] != 0)
324 			break;
325 	if (allocsiz == fs->fs_frag) {
326 		/*
327 		 * no fragments were available, so a block will be
328 		 * allocated, and hacked up
329 		 */
330 		if (cgp->cg_cs.cs_nbfree == 0) {
331 			brelse(bp);
332 			return (0);
333 		}
334 		bno = ffs_alloccgblk(ip, bp, bpref);
335 		bpref = dtogd(fs, bno);
336 		for (i = frags; i < fs->fs_frag; i++)
337 			setbit(cg_blksfree_swap(cgp, needswap), bpref + i);
338 		i = fs->fs_frag - frags;
339 		ufs_add32(cgp->cg_cs.cs_nffree, i, needswap);
340 		fs->fs_cstotal.cs_nffree += i;
341 		fs->fs_cs(fs, cg).cs_nffree += i;
342 		fs->fs_fmod = 1;
343 		ufs_add32(cgp->cg_frsum[i], 1, needswap);
344 		bdwrite(bp);
345 		return (bno);
346 	}
347 	bno = ffs_mapsearch(fs, cgp, bpref, allocsiz);
348 	for (i = 0; i < frags; i++)
349 		clrbit(cg_blksfree_swap(cgp, needswap), bno + i);
350 	ufs_add32(cgp->cg_cs.cs_nffree, -frags, needswap);
351 	fs->fs_cstotal.cs_nffree -= frags;
352 	fs->fs_cs(fs, cg).cs_nffree -= frags;
353 	fs->fs_fmod = 1;
354 	ufs_add32(cgp->cg_frsum[allocsiz], -1, needswap);
355 	if (frags != allocsiz)
356 		ufs_add32(cgp->cg_frsum[allocsiz - frags], 1, needswap);
357 	blkno = cg * fs->fs_fpg + bno;
358 	bdwrite(bp);
359 	return blkno;
360 }
361 
362 /*
363  * Allocate a block in a cylinder group.
364  *
365  * This algorithm implements the following policy:
366  *   1) allocate the requested block.
367  *   2) allocate a rotationally optimal block in the same cylinder.
368  *   3) allocate the next available block on the block rotor for the
369  *      specified cylinder group.
370  * Note that this routine only allocates fs_bsize blocks; these
371  * blocks may be fragmented by the routine that allocates them.
372  */
373 static daddr_t
ffs_alloccgblk(struct inode * ip,struct m_buf * bp,daddr_t bpref)374 ffs_alloccgblk(struct inode *ip, struct m_buf *bp, daddr_t bpref)
375 {
376 	struct cg *cgp;
377 	daddr_t blkno;
378 	int32_t bno;
379 	struct fs *fs = ip->i_fs;
380 	const int needswap = UFS_FSNEEDSWAP(fs);
381 	u_int8_t *blksfree_swap;
382 
383 	cgp = (struct cg *)bp->b_data;
384 	blksfree_swap = cg_blksfree_swap(cgp, needswap);
385 	if (bpref == 0 || (uint32_t)dtog(fs, bpref) != ufs_rw32(cgp->cg_cgx, needswap)) {
386 		bpref = ufs_rw32(cgp->cg_rotor, needswap);
387 	} else {
388 		bpref = blknum(fs, bpref);
389 		bno = dtogd(fs, bpref);
390 		/*
391 		 * if the requested block is available, use it
392 		 */
393 		if (ffs_isblock(fs, blksfree_swap, fragstoblks(fs, bno)))
394 			goto gotit;
395 	}
396 	/*
397 	 * Take the next available one in this cylinder group.
398 	 */
399 	bno = ffs_mapsearch(fs, cgp, bpref, (int)fs->fs_frag);
400 	if (bno < 0)
401 		return (0);
402 	cgp->cg_rotor = ufs_rw32(bno, needswap);
403 gotit:
404 	blkno = fragstoblks(fs, bno);
405 	ffs_clrblock(fs, blksfree_swap, (long)blkno);
406 	ffs_clusteracct(fs, cgp, blkno, -1);
407 	ufs_add32(cgp->cg_cs.cs_nbfree, -1, needswap);
408 	fs->fs_cstotal.cs_nbfree--;
409 	fs->fs_cs(fs, ufs_rw32(cgp->cg_cgx, needswap)).cs_nbfree--;
410 	fs->fs_fmod = 1;
411 	blkno = ufs_rw32(cgp->cg_cgx, needswap) * fs->fs_fpg + bno;
412 	return (blkno);
413 }
414 
415 /*
416  * Free a block or fragment.
417  *
418  * The specified block or fragment is placed back in the
419  * free map. If a fragment is deallocated, a possible
420  * block reassembly is checked.
421  */
422 void
ffs_blkfree(struct inode * ip,daddr_t bno,long size)423 ffs_blkfree(struct inode *ip, daddr_t bno, long size)
424 {
425 	struct cg *cgp;
426 	struct m_buf *bp;
427 	int32_t fragno, cgbno;
428 	int i, error, cg, blk, frags, bbase;
429 	struct fs *fs = ip->i_fs;
430 	const int needswap = UFS_FSNEEDSWAP(fs);
431 
432 	if (size > fs->fs_bsize || fragoff(fs, size) != 0 ||
433 	    fragnum(fs, bno) + numfrags(fs, size) > fs->fs_frag) {
434 		errx(1, "blkfree: bad size: bno %lld bsize %d size %ld",
435 		    (long long)bno, fs->fs_bsize, size);
436 	}
437 	cg = dtog(fs, bno);
438 	if (bno >= fs->fs_size) {
439 		warnx("bad block %lld, ino %ju", (long long)bno,
440 		    (uintmax_t)ip->i_number);
441 		return;
442 	}
443 	error = bread((void *)ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
444 	    (int)fs->fs_cgsize, NULL, &bp);
445 	if (error) {
446 		return;
447 	}
448 	cgp = (struct cg *)bp->b_data;
449 	if (!cg_chkmagic_swap(cgp, needswap)) {
450 		brelse(bp);
451 		return;
452 	}
453 	cgbno = dtogd(fs, bno);
454 	if (size == fs->fs_bsize) {
455 		fragno = fragstoblks(fs, cgbno);
456 		if (!ffs_isfreeblock(fs, cg_blksfree_swap(cgp, needswap), fragno)) {
457 			errx(1, "blkfree: freeing free block %lld",
458 			    (long long)bno);
459 		}
460 		ffs_setblock(fs, cg_blksfree_swap(cgp, needswap), fragno);
461 		ffs_clusteracct(fs, cgp, fragno, 1);
462 		ufs_add32(cgp->cg_cs.cs_nbfree, 1, needswap);
463 		fs->fs_cstotal.cs_nbfree++;
464 		fs->fs_cs(fs, cg).cs_nbfree++;
465 	} else {
466 		bbase = cgbno - fragnum(fs, cgbno);
467 		/*
468 		 * decrement the counts associated with the old frags
469 		 */
470 		blk = blkmap(fs, cg_blksfree_swap(cgp, needswap), bbase);
471 		ffs_fragacct_swap(fs, blk, cgp->cg_frsum, -1, needswap);
472 		/*
473 		 * deallocate the fragment
474 		 */
475 		frags = numfrags(fs, size);
476 		for (i = 0; i < frags; i++) {
477 			if (isset(cg_blksfree_swap(cgp, needswap), cgbno + i)) {
478 				errx(1, "blkfree: freeing free frag: block %lld",
479 				    (long long)(cgbno + i));
480 			}
481 			setbit(cg_blksfree_swap(cgp, needswap), cgbno + i);
482 		}
483 		ufs_add32(cgp->cg_cs.cs_nffree, i, needswap);
484 		fs->fs_cstotal.cs_nffree += i;
485 		fs->fs_cs(fs, cg).cs_nffree += i;
486 		/*
487 		 * add back in counts associated with the new frags
488 		 */
489 		blk = blkmap(fs, cg_blksfree_swap(cgp, needswap), bbase);
490 		ffs_fragacct_swap(fs, blk, cgp->cg_frsum, 1, needswap);
491 		/*
492 		 * if a complete block has been reassembled, account for it
493 		 */
494 		fragno = fragstoblks(fs, bbase);
495 		if (ffs_isblock(fs, cg_blksfree_swap(cgp, needswap), fragno)) {
496 			ufs_add32(cgp->cg_cs.cs_nffree, -fs->fs_frag, needswap);
497 			fs->fs_cstotal.cs_nffree -= fs->fs_frag;
498 			fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag;
499 			ffs_clusteracct(fs, cgp, fragno, 1);
500 			ufs_add32(cgp->cg_cs.cs_nbfree, 1, needswap);
501 			fs->fs_cstotal.cs_nbfree++;
502 			fs->fs_cs(fs, cg).cs_nbfree++;
503 		}
504 	}
505 	fs->fs_fmod = 1;
506 	bdwrite(bp);
507 }
508 
509 
510 static int
scanc(u_int size,const u_char * cp,const u_char table[],int mask)511 scanc(u_int size, const u_char *cp, const u_char table[], int mask)
512 {
513 	const u_char *end = &cp[size];
514 
515 	while (cp < end && (table[*cp] & mask) == 0)
516 		cp++;
517 	return (end - cp);
518 }
519 
520 /*
521  * Find a block of the specified size in the specified cylinder group.
522  *
523  * It is a panic if a request is made to find a block if none are
524  * available.
525  */
526 static int32_t
ffs_mapsearch(struct fs * fs,struct cg * cgp,daddr_t bpref,int allocsiz)527 ffs_mapsearch(struct fs *fs, struct cg *cgp, daddr_t bpref, int allocsiz)
528 {
529 	int32_t bno;
530 	int start, len, loc, i;
531 	int blk, field, subfield, pos;
532 	int ostart, olen;
533 	const int needswap = UFS_FSNEEDSWAP(fs);
534 
535 	/*
536 	 * find the fragment by searching through the free block
537 	 * map for an appropriate bit pattern
538 	 */
539 	if (bpref)
540 		start = dtogd(fs, bpref) / NBBY;
541 	else
542 		start = ufs_rw32(cgp->cg_frotor, needswap) / NBBY;
543 	len = howmany(fs->fs_fpg, NBBY) - start;
544 	ostart = start;
545 	olen = len;
546 	loc = scanc((u_int)len,
547 		(const u_char *)&cg_blksfree_swap(cgp, needswap)[start],
548 		(const u_char *)fragtbl[fs->fs_frag],
549 		(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
550 	if (loc == 0) {
551 		len = start + 1;
552 		start = 0;
553 		loc = scanc((u_int)len,
554 			(const u_char *)&cg_blksfree_swap(cgp, needswap)[0],
555 			(const u_char *)fragtbl[fs->fs_frag],
556 			(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
557 		if (loc == 0) {
558 			errx(1,
559     "ffs_alloccg: map corrupted: start %d len %d offset %d %ld",
560 				ostart, olen,
561 				ufs_rw32(cgp->cg_freeoff, needswap),
562 				(long)cg_blksfree_swap(cgp, needswap) - (long)cgp);
563 			/* NOTREACHED */
564 		}
565 	}
566 	bno = (start + len - loc) * NBBY;
567 	cgp->cg_frotor = ufs_rw32(bno, needswap);
568 	/*
569 	 * found the byte in the map
570 	 * sift through the bits to find the selected frag
571 	 */
572 	for (i = bno + NBBY; bno < i; bno += fs->fs_frag) {
573 		blk = blkmap(fs, cg_blksfree_swap(cgp, needswap), bno);
574 		blk <<= 1;
575 		field = around[allocsiz];
576 		subfield = inside[allocsiz];
577 		for (pos = 0; pos <= fs->fs_frag - allocsiz; pos++) {
578 			if ((blk & field) == subfield)
579 				return (bno + pos);
580 			field <<= 1;
581 			subfield <<= 1;
582 		}
583 	}
584 	errx(1, "ffs_alloccg: block not in map: bno %lld", (long long)bno);
585 	return (-1);
586 }
587 
588 /*
589  * Update the cluster map because of an allocation or free.
590  *
591  * Cnt == 1 means free; cnt == -1 means allocating.
592  */
593 void
ffs_clusteracct(struct fs * fs,struct cg * cgp,int32_t blkno,int cnt)594 ffs_clusteracct(struct fs *fs, struct cg *cgp, int32_t blkno, int cnt)
595 {
596 	int32_t *sump;
597 	int32_t *lp;
598 	u_char *freemapp, *mapp;
599 	int i, start, end, forw, back, map, bit;
600 	const int needswap = UFS_FSNEEDSWAP(fs);
601 
602 	if (fs->fs_contigsumsize <= 0)
603 		return;
604 	freemapp = cg_clustersfree_swap(cgp, needswap);
605 	sump = cg_clustersum_swap(cgp, needswap);
606 	/*
607 	 * Allocate or clear the actual block.
608 	 */
609 	if (cnt > 0)
610 		setbit(freemapp, blkno);
611 	else
612 		clrbit(freemapp, blkno);
613 	/*
614 	 * Find the size of the cluster going forward.
615 	 */
616 	start = blkno + 1;
617 	end = start + fs->fs_contigsumsize;
618 	if ((unsigned)end >= ufs_rw32(cgp->cg_nclusterblks, needswap))
619 		end = ufs_rw32(cgp->cg_nclusterblks, needswap);
620 	mapp = &freemapp[start / NBBY];
621 	map = *mapp++;
622 	bit = 1 << (start % NBBY);
623 	for (i = start; i < end; i++) {
624 		if ((map & bit) == 0)
625 			break;
626 		if ((i & (NBBY - 1)) != (NBBY - 1)) {
627 			bit <<= 1;
628 		} else {
629 			map = *mapp++;
630 			bit = 1;
631 		}
632 	}
633 	forw = i - start;
634 	/*
635 	 * Find the size of the cluster going backward.
636 	 */
637 	start = blkno - 1;
638 	end = start - fs->fs_contigsumsize;
639 	if (end < 0)
640 		end = -1;
641 	mapp = &freemapp[start / NBBY];
642 	map = *mapp--;
643 	bit = 1 << (start % NBBY);
644 	for (i = start; i > end; i--) {
645 		if ((map & bit) == 0)
646 			break;
647 		if ((i & (NBBY - 1)) != 0) {
648 			bit >>= 1;
649 		} else {
650 			map = *mapp--;
651 			bit = 1 << (NBBY - 1);
652 		}
653 	}
654 	back = start - i;
655 	/*
656 	 * Account for old cluster and the possibly new forward and
657 	 * back clusters.
658 	 */
659 	i = back + forw + 1;
660 	if (i > fs->fs_contigsumsize)
661 		i = fs->fs_contigsumsize;
662 	ufs_add32(sump[i], cnt, needswap);
663 	if (back > 0)
664 		ufs_add32(sump[back], -cnt, needswap);
665 	if (forw > 0)
666 		ufs_add32(sump[forw], -cnt, needswap);
667 
668 	/*
669 	 * Update cluster summary information.
670 	 */
671 	lp = &sump[fs->fs_contigsumsize];
672 	for (i = fs->fs_contigsumsize; i > 0; i--)
673 		if (ufs_rw32(*lp--, needswap) > 0)
674 			break;
675 	fs->fs_maxcluster[ufs_rw32(cgp->cg_cgx, needswap)] = i;
676 }
677