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 */
5 * SPDX-License-Identifier: BSD-3-Clause
7 * Copyright (c) 2002 Networks Associates Technology, Inc.
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
16 * Copyright (c) 1982, 1986, 1989, 1993
17 * The Regents of the University of California. All rights reserved.
19 * Redistribution and use in source and binary forms, with or without
20 * modification, are permitted provided that the following conditions
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.
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
43 * @(#)ffs_alloc.c 8.19 (Berkeley) 7/13/95
46 #include <sys/cdefs.h>
47 __FBSDID("$FreeBSD$");
49 #include <sys/param.h>
57 #include <ufs/ufs/dinode.h>
58 #include <ufs/ffs/fs.h>
60 #include "ffs/ufs_bswap.h"
62 #include "ffs/ufs_inode.h"
63 #include "ffs/ffs_extern.h"
65 static int scanc(u_int, const u_char *, const u_char *, int);
67 static daddr_t ffs_alloccg(struct inode *, int, daddr_t, int);
68 static daddr_t ffs_alloccgblk(struct inode *, struct buf *, daddr_t);
69 static daddr_t ffs_hashalloc(struct inode *, u_int, daddr_t, int,
70 daddr_t (*)(struct inode *, int, daddr_t, int));
71 static int32_t ffs_mapsearch(struct fs *, struct cg *, daddr_t, int);
74 * Allocate a block in the file system.
76 * The size of the requested block is given, which must be some
77 * multiple of fs_fsize and <= fs_bsize.
78 * A preference may be optionally specified. If a preference is given
79 * the following hierarchy is used to allocate a block:
80 * 1) allocate the requested block.
81 * 2) allocate a rotationally optimal block in the same cylinder.
82 * 3) allocate a block in the same cylinder group.
83 * 4) quadradically rehash into other cylinder groups, until an
84 * available block is located.
85 * If no block preference is given the following hierarchy is used
86 * to allocate a block:
87 * 1) allocate a block in the cylinder group that contains the
89 * 2) quadradically rehash into other cylinder groups, until an
90 * available block is located.
93 ffs_alloc(struct inode *ip, daddr_t lbn __unused, daddr_t bpref, int size,
96 struct fs *fs = ip->i_fs;
101 if (size > fs->fs_bsize || fragoff(fs, size) != 0) {
102 errx(1, "ffs_alloc: bad size: bsize %d size %d",
105 if (size == fs->fs_bsize && fs->fs_cstotal.cs_nbfree == 0)
107 if (bpref >= fs->fs_size)
110 cg = ino_to_cg(fs, ip->i_number);
112 cg = dtog(fs, bpref);
113 bno = ffs_hashalloc(ip, cg, bpref, size, ffs_alloccg);
115 if (ip->i_fs->fs_magic == FS_UFS1_MAGIC)
116 ip->i_ffs1_blocks += size / DEV_BSIZE;
118 ip->i_ffs2_blocks += size / DEV_BSIZE;
127 * Select the desired position for the next block in a file. The file is
128 * logically divided into sections. The first section is composed of the
129 * direct blocks. Each additional section contains fs_maxbpg blocks.
131 * If no blocks have been allocated in the first section, the policy is to
132 * request a block in the same cylinder group as the inode that describes
133 * the file. If no blocks have been allocated in any other section, the
134 * policy is to place the section in a cylinder group with a greater than
135 * average number of free blocks. An appropriate cylinder group is found
136 * by using a rotor that sweeps the cylinder groups. When a new group of
137 * blocks is needed, the sweep begins in the cylinder group following the
138 * cylinder group from which the previous allocation was made. The sweep
139 * continues until a cylinder group with greater than the average number
140 * of free blocks is found. If the allocation is for the first block in an
141 * indirect block, the information on the previous allocation is unavailable;
142 * here a best guess is made based upon the logical block number being
145 * If a section is already partially allocated, the policy is to
146 * contiguously allocate fs_maxcontig blocks. The end of one of these
147 * contiguous blocks and the beginning of the next is physically separated
148 * so that the disk head will be in transit between them for at least
149 * fs_rotdelay milliseconds. This is to allow time for the processor to
150 * schedule another I/O transfer.
154 ffs_blkpref_ufs1(struct inode *ip, daddr_t lbn, int indx, int32_t *bap)
161 if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) {
162 if (lbn < UFS_NDADDR + NINDIR(fs)) {
163 cg = ino_to_cg(fs, ip->i_number);
164 return (fs->fs_fpg * cg + fs->fs_frag);
167 * Find a cylinder with greater than average number of
168 * unused data blocks.
170 if (indx == 0 || bap[indx - 1] == 0)
172 ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg;
175 ufs_rw32(bap[indx - 1], UFS_FSNEEDSWAP(fs)) + 1);
176 startcg %= fs->fs_ncg;
177 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
178 for (cg = startcg; cg < fs->fs_ncg; cg++)
179 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree)
180 return (fs->fs_fpg * cg + fs->fs_frag);
181 for (cg = 0; cg <= startcg; cg++)
182 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree)
183 return (fs->fs_fpg * cg + fs->fs_frag);
187 * We just always try to lay things out contiguously.
189 return ufs_rw32(bap[indx - 1], UFS_FSNEEDSWAP(fs)) + fs->fs_frag;
193 ffs_blkpref_ufs2(struct inode *ip, daddr_t lbn, int indx, int64_t *bap)
200 if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) {
201 if (lbn < UFS_NDADDR + NINDIR(fs)) {
202 cg = ino_to_cg(fs, ip->i_number);
203 return (fs->fs_fpg * cg + fs->fs_frag);
206 * Find a cylinder with greater than average number of
207 * unused data blocks.
209 if (indx == 0 || bap[indx - 1] == 0)
211 ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg;
214 ufs_rw64(bap[indx - 1], UFS_FSNEEDSWAP(fs)) + 1);
215 startcg %= fs->fs_ncg;
216 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
217 for (cg = startcg; cg < fs->fs_ncg; cg++)
218 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
219 return (fs->fs_fpg * cg + fs->fs_frag);
221 for (cg = 0; cg < startcg; cg++)
222 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
223 return (fs->fs_fpg * cg + fs->fs_frag);
228 * We just always try to lay things out contiguously.
230 return ufs_rw64(bap[indx - 1], UFS_FSNEEDSWAP(fs)) + fs->fs_frag;
234 * Implement the cylinder overflow algorithm.
236 * The policy implemented by this algorithm is:
237 * 1) allocate the block in its requested cylinder group.
238 * 2) quadradically rehash on the cylinder group number.
239 * 3) brute force search for a free block.
241 * `size': size for data blocks, mode for inodes
245 ffs_hashalloc(struct inode *ip, u_int cg, daddr_t pref, int size,
246 daddr_t (*allocator)(struct inode *, int, daddr_t, int))
254 * 1: preferred cylinder group
256 result = (*allocator)(ip, cg, pref, size);
260 * 2: quadratic rehash
262 for (i = 1; i < fs->fs_ncg; i *= 2) {
264 if (cg >= fs->fs_ncg)
266 result = (*allocator)(ip, cg, 0, size);
271 * 3: brute force search
272 * Note that we start at i == 2, since 0 was checked initially,
273 * and 1 is always checked in the quadratic rehash.
275 cg = (icg + 2) % fs->fs_ncg;
276 for (i = 2; i < fs->fs_ncg; i++) {
277 result = (*allocator)(ip, cg, 0, size);
281 if (cg == fs->fs_ncg)
288 * Determine whether a block can be allocated.
290 * Check to see if a block of the appropriate size is available,
291 * and if it is, allocate it.
294 ffs_alloccg(struct inode *ip, int cg, daddr_t bpref, int size)
299 int error, frags, allocsiz, i;
300 struct fs *fs = ip->i_fs;
301 const int needswap = UFS_FSNEEDSWAP(fs);
303 if (fs->fs_cs(fs, cg).cs_nbfree == 0 && size == fs->fs_bsize)
305 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), (int)fs->fs_cgsize,
311 cgp = (struct cg *)bp->b_data;
312 if (!cg_chkmagic_swap(cgp, needswap) ||
313 (cgp->cg_cs.cs_nbfree == 0 && size == fs->fs_bsize)) {
317 if (size == fs->fs_bsize) {
318 bno = ffs_alloccgblk(ip, bp, bpref);
323 * check to see if any fragments are already available
324 * allocsiz is the size which will be allocated, hacking
325 * it down to a smaller size if necessary
327 frags = numfrags(fs, size);
328 for (allocsiz = frags; allocsiz < fs->fs_frag; allocsiz++)
329 if (cgp->cg_frsum[allocsiz] != 0)
331 if (allocsiz == fs->fs_frag) {
333 * no fragments were available, so a block will be
334 * allocated, and hacked up
336 if (cgp->cg_cs.cs_nbfree == 0) {
340 bno = ffs_alloccgblk(ip, bp, bpref);
341 bpref = dtogd(fs, bno);
342 for (i = frags; i < fs->fs_frag; i++)
343 setbit(cg_blksfree_swap(cgp, needswap), bpref + i);
344 i = fs->fs_frag - frags;
345 ufs_add32(cgp->cg_cs.cs_nffree, i, needswap);
346 fs->fs_cstotal.cs_nffree += i;
347 fs->fs_cs(fs, cg).cs_nffree += i;
349 ufs_add32(cgp->cg_frsum[i], 1, needswap);
353 bno = ffs_mapsearch(fs, cgp, bpref, allocsiz);
354 for (i = 0; i < frags; i++)
355 clrbit(cg_blksfree_swap(cgp, needswap), bno + i);
356 ufs_add32(cgp->cg_cs.cs_nffree, -frags, needswap);
357 fs->fs_cstotal.cs_nffree -= frags;
358 fs->fs_cs(fs, cg).cs_nffree -= frags;
360 ufs_add32(cgp->cg_frsum[allocsiz], -1, needswap);
361 if (frags != allocsiz)
362 ufs_add32(cgp->cg_frsum[allocsiz - frags], 1, needswap);
363 blkno = cg * fs->fs_fpg + bno;
369 * Allocate a block in a cylinder group.
371 * This algorithm implements the following policy:
372 * 1) allocate the requested block.
373 * 2) allocate a rotationally optimal block in the same cylinder.
374 * 3) allocate the next available block on the block rotor for the
375 * specified cylinder group.
376 * Note that this routine only allocates fs_bsize blocks; these
377 * blocks may be fragmented by the routine that allocates them.
380 ffs_alloccgblk(struct inode *ip, struct buf *bp, daddr_t bpref)
385 struct fs *fs = ip->i_fs;
386 const int needswap = UFS_FSNEEDSWAP(fs);
387 u_int8_t *blksfree_swap;
389 cgp = (struct cg *)bp->b_data;
390 blksfree_swap = cg_blksfree_swap(cgp, needswap);
391 if (bpref == 0 || (uint32_t)dtog(fs, bpref) != ufs_rw32(cgp->cg_cgx, needswap)) {
392 bpref = ufs_rw32(cgp->cg_rotor, needswap);
394 bpref = blknum(fs, bpref);
395 bno = dtogd(fs, bpref);
397 * if the requested block is available, use it
399 if (ffs_isblock(fs, blksfree_swap, fragstoblks(fs, bno)))
403 * Take the next available one in this cylinder group.
405 bno = ffs_mapsearch(fs, cgp, bpref, (int)fs->fs_frag);
408 cgp->cg_rotor = ufs_rw32(bno, needswap);
410 blkno = fragstoblks(fs, bno);
411 ffs_clrblock(fs, blksfree_swap, (long)blkno);
412 ffs_clusteracct(fs, cgp, blkno, -1);
413 ufs_add32(cgp->cg_cs.cs_nbfree, -1, needswap);
414 fs->fs_cstotal.cs_nbfree--;
415 fs->fs_cs(fs, ufs_rw32(cgp->cg_cgx, needswap)).cs_nbfree--;
417 blkno = ufs_rw32(cgp->cg_cgx, needswap) * fs->fs_fpg + bno;
422 * Free a block or fragment.
424 * The specified block or fragment is placed back in the
425 * free map. If a fragment is deallocated, a possible
426 * block reassembly is checked.
429 ffs_blkfree(struct inode *ip, daddr_t bno, long size)
433 int32_t fragno, cgbno;
434 int i, error, cg, blk, frags, bbase;
435 struct fs *fs = ip->i_fs;
436 const int needswap = UFS_FSNEEDSWAP(fs);
438 if (size > fs->fs_bsize || fragoff(fs, size) != 0 ||
439 fragnum(fs, bno) + numfrags(fs, size) > fs->fs_frag) {
440 errx(1, "blkfree: bad size: bno %lld bsize %d size %ld",
441 (long long)bno, fs->fs_bsize, size);
444 if (bno >= fs->fs_size) {
445 warnx("bad block %lld, ino %ju", (long long)bno,
446 (uintmax_t)ip->i_number);
449 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), (int)fs->fs_cgsize,
455 cgp = (struct cg *)bp->b_data;
456 if (!cg_chkmagic_swap(cgp, needswap)) {
460 cgbno = dtogd(fs, bno);
461 if (size == fs->fs_bsize) {
462 fragno = fragstoblks(fs, cgbno);
463 if (!ffs_isfreeblock(fs, cg_blksfree_swap(cgp, needswap), fragno)) {
464 errx(1, "blkfree: freeing free block %lld",
467 ffs_setblock(fs, cg_blksfree_swap(cgp, needswap), fragno);
468 ffs_clusteracct(fs, cgp, fragno, 1);
469 ufs_add32(cgp->cg_cs.cs_nbfree, 1, needswap);
470 fs->fs_cstotal.cs_nbfree++;
471 fs->fs_cs(fs, cg).cs_nbfree++;
473 bbase = cgbno - fragnum(fs, cgbno);
475 * decrement the counts associated with the old frags
477 blk = blkmap(fs, cg_blksfree_swap(cgp, needswap), bbase);
478 ffs_fragacct_swap(fs, blk, cgp->cg_frsum, -1, needswap);
480 * deallocate the fragment
482 frags = numfrags(fs, size);
483 for (i = 0; i < frags; i++) {
484 if (isset(cg_blksfree_swap(cgp, needswap), cgbno + i)) {
485 errx(1, "blkfree: freeing free frag: block %lld",
486 (long long)(cgbno + i));
488 setbit(cg_blksfree_swap(cgp, needswap), cgbno + i);
490 ufs_add32(cgp->cg_cs.cs_nffree, i, needswap);
491 fs->fs_cstotal.cs_nffree += i;
492 fs->fs_cs(fs, cg).cs_nffree += i;
494 * add back in counts associated with the new frags
496 blk = blkmap(fs, cg_blksfree_swap(cgp, needswap), bbase);
497 ffs_fragacct_swap(fs, blk, cgp->cg_frsum, 1, needswap);
499 * if a complete block has been reassembled, account for it
501 fragno = fragstoblks(fs, bbase);
502 if (ffs_isblock(fs, cg_blksfree_swap(cgp, needswap), fragno)) {
503 ufs_add32(cgp->cg_cs.cs_nffree, -fs->fs_frag, needswap);
504 fs->fs_cstotal.cs_nffree -= fs->fs_frag;
505 fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag;
506 ffs_clusteracct(fs, cgp, fragno, 1);
507 ufs_add32(cgp->cg_cs.cs_nbfree, 1, needswap);
508 fs->fs_cstotal.cs_nbfree++;
509 fs->fs_cs(fs, cg).cs_nbfree++;
518 scanc(u_int size, const u_char *cp, const u_char table[], int mask)
520 const u_char *end = &cp[size];
522 while (cp < end && (table[*cp] & mask) == 0)
528 * Find a block of the specified size in the specified cylinder group.
530 * It is a panic if a request is made to find a block if none are
534 ffs_mapsearch(struct fs *fs, struct cg *cgp, daddr_t bpref, int allocsiz)
537 int start, len, loc, i;
538 int blk, field, subfield, pos;
540 const int needswap = UFS_FSNEEDSWAP(fs);
543 * find the fragment by searching through the free block
544 * map for an appropriate bit pattern
547 start = dtogd(fs, bpref) / NBBY;
549 start = ufs_rw32(cgp->cg_frotor, needswap) / NBBY;
550 len = howmany(fs->fs_fpg, NBBY) - start;
553 loc = scanc((u_int)len,
554 (const u_char *)&cg_blksfree_swap(cgp, needswap)[start],
555 (const u_char *)fragtbl[fs->fs_frag],
556 (1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
560 loc = scanc((u_int)len,
561 (const u_char *)&cg_blksfree_swap(cgp, needswap)[0],
562 (const u_char *)fragtbl[fs->fs_frag],
563 (1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
566 "ffs_alloccg: map corrupted: start %d len %d offset %d %ld",
568 ufs_rw32(cgp->cg_freeoff, needswap),
569 (long)cg_blksfree_swap(cgp, needswap) - (long)cgp);
573 bno = (start + len - loc) * NBBY;
574 cgp->cg_frotor = ufs_rw32(bno, needswap);
576 * found the byte in the map
577 * sift through the bits to find the selected frag
579 for (i = bno + NBBY; bno < i; bno += fs->fs_frag) {
580 blk = blkmap(fs, cg_blksfree_swap(cgp, needswap), bno);
582 field = around[allocsiz];
583 subfield = inside[allocsiz];
584 for (pos = 0; pos <= fs->fs_frag - allocsiz; pos++) {
585 if ((blk & field) == subfield)
591 errx(1, "ffs_alloccg: block not in map: bno %lld", (long long)bno);
596 * Update the cluster map because of an allocation or free.
598 * Cnt == 1 means free; cnt == -1 means allocating.
601 ffs_clusteracct(struct fs *fs, struct cg *cgp, int32_t blkno, int cnt)
605 u_char *freemapp, *mapp;
606 int i, start, end, forw, back, map, bit;
607 const int needswap = UFS_FSNEEDSWAP(fs);
609 if (fs->fs_contigsumsize <= 0)
611 freemapp = cg_clustersfree_swap(cgp, needswap);
612 sump = cg_clustersum_swap(cgp, needswap);
614 * Allocate or clear the actual block.
617 setbit(freemapp, blkno);
619 clrbit(freemapp, blkno);
621 * Find the size of the cluster going forward.
624 end = start + fs->fs_contigsumsize;
625 if ((unsigned)end >= ufs_rw32(cgp->cg_nclusterblks, needswap))
626 end = ufs_rw32(cgp->cg_nclusterblks, needswap);
627 mapp = &freemapp[start / NBBY];
629 bit = 1 << (start % NBBY);
630 for (i = start; i < end; i++) {
631 if ((map & bit) == 0)
633 if ((i & (NBBY - 1)) != (NBBY - 1)) {
642 * Find the size of the cluster going backward.
645 end = start - fs->fs_contigsumsize;
648 mapp = &freemapp[start / NBBY];
650 bit = 1 << (start % NBBY);
651 for (i = start; i > end; i--) {
652 if ((map & bit) == 0)
654 if ((i & (NBBY - 1)) != 0) {
658 bit = 1 << (NBBY - 1);
663 * Account for old cluster and the possibly new forward and
667 if (i > fs->fs_contigsumsize)
668 i = fs->fs_contigsumsize;
669 ufs_add32(sump[i], cnt, needswap);
671 ufs_add32(sump[back], -cnt, needswap);
673 ufs_add32(sump[forw], -cnt, needswap);
676 * Update cluster summary information.
678 lp = &sump[fs->fs_contigsumsize];
679 for (i = fs->fs_contigsumsize; i > 0; i--)
680 if (ufs_rw32(*lp--, needswap) > 0)
682 fs->fs_maxcluster[ufs_rw32(cgp->cg_cgx, needswap)] = i;