ffs_alloc.c revision 1.12 1 1.12 fvdl /* $NetBSD: ffs_alloc.c,v 1.12 2003/04/02 10:39:49 fvdl Exp $ */
2 1.1 lukem /* From: NetBSD: ffs_alloc.c,v 1.50 2001/09/06 02:16:01 lukem Exp */
3 1.1 lukem
4 1.1 lukem /*
5 1.12 fvdl * Copyright (c) 2002 Networks Associates Technology, Inc.
6 1.12 fvdl * All rights reserved.
7 1.12 fvdl *
8 1.12 fvdl * This software was developed for the FreeBSD Project by Marshall
9 1.12 fvdl * Kirk McKusick and Network Associates Laboratories, the Security
10 1.12 fvdl * Research Division of Network Associates, Inc. under DARPA/SPAWAR
11 1.12 fvdl * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS
12 1.12 fvdl * research program
13 1.12 fvdl *
14 1.1 lukem * Copyright (c) 1982, 1986, 1989, 1993
15 1.1 lukem * The Regents of the University of California. All rights reserved.
16 1.1 lukem *
17 1.1 lukem * Redistribution and use in source and binary forms, with or without
18 1.1 lukem * modification, are permitted provided that the following conditions
19 1.1 lukem * are met:
20 1.1 lukem * 1. Redistributions of source code must retain the above copyright
21 1.1 lukem * notice, this list of conditions and the following disclaimer.
22 1.1 lukem * 2. Redistributions in binary form must reproduce the above copyright
23 1.1 lukem * notice, this list of conditions and the following disclaimer in the
24 1.1 lukem * documentation and/or other materials provided with the distribution.
25 1.1 lukem * 3. All advertising materials mentioning features or use of this software
26 1.1 lukem * must display the following acknowledgement:
27 1.1 lukem * This product includes software developed by the University of
28 1.1 lukem * California, Berkeley and its contributors.
29 1.1 lukem * 4. Neither the name of the University nor the names of its contributors
30 1.1 lukem * may be used to endorse or promote products derived from this software
31 1.1 lukem * without specific prior written permission.
32 1.1 lukem *
33 1.1 lukem * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
34 1.1 lukem * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
35 1.1 lukem * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
36 1.1 lukem * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
37 1.1 lukem * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
38 1.1 lukem * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
39 1.1 lukem * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
40 1.1 lukem * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
41 1.1 lukem * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
42 1.1 lukem * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
43 1.1 lukem * SUCH DAMAGE.
44 1.1 lukem *
45 1.1 lukem * @(#)ffs_alloc.c 8.19 (Berkeley) 7/13/95
46 1.1 lukem */
47 1.2 lukem
48 1.2 lukem #include <sys/cdefs.h>
49 1.8 tv #if defined(__RCSID) && !defined(__lint)
50 1.12 fvdl __RCSID("$NetBSD: ffs_alloc.c,v 1.12 2003/04/02 10:39:49 fvdl Exp $");
51 1.2 lukem #endif /* !__lint */
52 1.1 lukem
53 1.1 lukem #include <sys/param.h>
54 1.1 lukem #include <sys/time.h>
55 1.1 lukem
56 1.1 lukem #include <errno.h>
57 1.4 lukem
58 1.4 lukem #include "makefs.h"
59 1.1 lukem
60 1.7 lukem #include <ufs/ufs/dinode.h>
61 1.5 lukem #include <ufs/ufs/ufs_bswap.h>
62 1.5 lukem #include <ufs/ffs/fs.h>
63 1.1 lukem
64 1.1 lukem #include "ffs/buf.h"
65 1.6 lukem #include "ffs/ufs_inode.h"
66 1.1 lukem #include "ffs/ffs_extern.h"
67 1.1 lukem
68 1.1 lukem
69 1.1 lukem static int scanc(u_int, const u_char *, const u_char *, int);
70 1.1 lukem
71 1.11 fvdl static daddr_t ffs_alloccg(struct inode *, int, daddr_t, int);
72 1.11 fvdl static daddr_t ffs_alloccgblk(struct inode *, struct buf *, daddr_t);
73 1.11 fvdl static daddr_t ffs_hashalloc(struct inode *, int, daddr_t, int,
74 1.11 fvdl daddr_t (*)(struct inode *, int, daddr_t, int));
75 1.12 fvdl static int32_t ffs_mapsearch(struct fs *, struct cg *, daddr_t, int);
76 1.1 lukem
77 1.1 lukem /* in ffs_tables.c */
78 1.1 lukem extern const int inside[], around[];
79 1.1 lukem extern const u_char * const fragtbl[];
80 1.1 lukem
81 1.1 lukem /*
82 1.1 lukem * Allocate a block in the file system.
83 1.1 lukem *
84 1.1 lukem * The size of the requested block is given, which must be some
85 1.1 lukem * multiple of fs_fsize and <= fs_bsize.
86 1.1 lukem * A preference may be optionally specified. If a preference is given
87 1.1 lukem * the following hierarchy is used to allocate a block:
88 1.1 lukem * 1) allocate the requested block.
89 1.1 lukem * 2) allocate a rotationally optimal block in the same cylinder.
90 1.1 lukem * 3) allocate a block in the same cylinder group.
91 1.1 lukem * 4) quadradically rehash into other cylinder groups, until an
92 1.1 lukem * available block is located.
93 1.1 lukem * If no block preference is given the following hierarchy is used
94 1.1 lukem * to allocate a block:
95 1.1 lukem * 1) allocate a block in the cylinder group that contains the
96 1.1 lukem * inode for the file.
97 1.1 lukem * 2) quadradically rehash into other cylinder groups, until an
98 1.1 lukem * available block is located.
99 1.1 lukem */
100 1.1 lukem int
101 1.11 fvdl ffs_alloc(struct inode *ip, daddr_t lbn, daddr_t bpref, int size,
102 1.11 fvdl daddr_t *bnp)
103 1.1 lukem {
104 1.1 lukem struct fs *fs = ip->i_fs;
105 1.11 fvdl daddr_t bno;
106 1.1 lukem int cg;
107 1.1 lukem
108 1.1 lukem *bnp = 0;
109 1.1 lukem if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) {
110 1.1 lukem errx(1, "ffs_alloc: bad size: bsize %d size %d",
111 1.1 lukem fs->fs_bsize, size);
112 1.1 lukem }
113 1.1 lukem if (size == fs->fs_bsize && fs->fs_cstotal.cs_nbfree == 0)
114 1.1 lukem goto nospace;
115 1.1 lukem if (bpref >= fs->fs_size)
116 1.1 lukem bpref = 0;
117 1.1 lukem if (bpref == 0)
118 1.1 lukem cg = ino_to_cg(fs, ip->i_number);
119 1.1 lukem else
120 1.1 lukem cg = dtog(fs, bpref);
121 1.11 fvdl bno = ffs_hashalloc(ip, cg, bpref, size, ffs_alloccg);
122 1.1 lukem if (bno > 0) {
123 1.12 fvdl DIP(ip, blocks) += size / DEV_BSIZE;
124 1.1 lukem *bnp = bno;
125 1.1 lukem return (0);
126 1.1 lukem }
127 1.1 lukem nospace:
128 1.1 lukem return (ENOSPC);
129 1.1 lukem }
130 1.1 lukem
131 1.1 lukem /*
132 1.1 lukem * Select the desired position for the next block in a file. The file is
133 1.1 lukem * logically divided into sections. The first section is composed of the
134 1.1 lukem * direct blocks. Each additional section contains fs_maxbpg blocks.
135 1.1 lukem *
136 1.1 lukem * If no blocks have been allocated in the first section, the policy is to
137 1.1 lukem * request a block in the same cylinder group as the inode that describes
138 1.1 lukem * the file. If no blocks have been allocated in any other section, the
139 1.1 lukem * policy is to place the section in a cylinder group with a greater than
140 1.1 lukem * average number of free blocks. An appropriate cylinder group is found
141 1.1 lukem * by using a rotor that sweeps the cylinder groups. When a new group of
142 1.1 lukem * blocks is needed, the sweep begins in the cylinder group following the
143 1.1 lukem * cylinder group from which the previous allocation was made. The sweep
144 1.1 lukem * continues until a cylinder group with greater than the average number
145 1.1 lukem * of free blocks is found. If the allocation is for the first block in an
146 1.1 lukem * indirect block, the information on the previous allocation is unavailable;
147 1.1 lukem * here a best guess is made based upon the logical block number being
148 1.1 lukem * allocated.
149 1.1 lukem *
150 1.1 lukem * If a section is already partially allocated, the policy is to
151 1.1 lukem * contiguously allocate fs_maxcontig blocks. The end of one of these
152 1.1 lukem * contiguous blocks and the beginning of the next is physically separated
153 1.1 lukem * so that the disk head will be in transit between them for at least
154 1.1 lukem * fs_rotdelay milliseconds. This is to allow time for the processor to
155 1.1 lukem * schedule another I/O transfer.
156 1.1 lukem */
157 1.11 fvdl /* XXX ondisk32 */
158 1.11 fvdl daddr_t
159 1.12 fvdl ffs_blkpref_ufs1(struct inode *ip, daddr_t lbn, int indx, int32_t *bap)
160 1.1 lukem {
161 1.1 lukem struct fs *fs;
162 1.1 lukem int cg;
163 1.1 lukem int avgbfree, startcg;
164 1.1 lukem
165 1.1 lukem fs = ip->i_fs;
166 1.1 lukem if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) {
167 1.1 lukem if (lbn < NDADDR + NINDIR(fs)) {
168 1.1 lukem cg = ino_to_cg(fs, ip->i_number);
169 1.1 lukem return (fs->fs_fpg * cg + fs->fs_frag);
170 1.1 lukem }
171 1.1 lukem /*
172 1.1 lukem * Find a cylinder with greater than average number of
173 1.1 lukem * unused data blocks.
174 1.1 lukem */
175 1.1 lukem if (indx == 0 || bap[indx - 1] == 0)
176 1.1 lukem startcg =
177 1.1 lukem ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg;
178 1.1 lukem else
179 1.1 lukem startcg = dtog(fs,
180 1.1 lukem ufs_rw32(bap[indx - 1], UFS_FSNEEDSWAP(fs)) + 1);
181 1.1 lukem startcg %= fs->fs_ncg;
182 1.1 lukem avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
183 1.1 lukem for (cg = startcg; cg < fs->fs_ncg; cg++)
184 1.12 fvdl if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree)
185 1.12 fvdl return (fs->fs_fpg * cg + fs->fs_frag);
186 1.12 fvdl for (cg = 0; cg <= startcg; cg++)
187 1.12 fvdl if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree)
188 1.12 fvdl return (fs->fs_fpg * cg + fs->fs_frag);
189 1.12 fvdl return (0);
190 1.12 fvdl }
191 1.12 fvdl /*
192 1.12 fvdl * We just always try to lay things out contiguously.
193 1.12 fvdl */
194 1.12 fvdl return ufs_rw32(bap[indx - 1], UFS_FSNEEDSWAP(fs)) + fs->fs_frag;
195 1.12 fvdl }
196 1.12 fvdl
197 1.12 fvdl daddr_t
198 1.12 fvdl ffs_blkpref_ufs2(ip, lbn, indx, bap)
199 1.12 fvdl struct inode *ip;
200 1.12 fvdl daddr_t lbn;
201 1.12 fvdl int indx;
202 1.12 fvdl int64_t *bap;
203 1.12 fvdl {
204 1.12 fvdl struct fs *fs;
205 1.12 fvdl int cg;
206 1.12 fvdl int avgbfree, startcg;
207 1.12 fvdl
208 1.12 fvdl fs = ip->i_fs;
209 1.12 fvdl if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) {
210 1.12 fvdl if (lbn < NDADDR + NINDIR(fs)) {
211 1.12 fvdl cg = ino_to_cg(fs, ip->i_number);
212 1.12 fvdl return (fs->fs_fpg * cg + fs->fs_frag);
213 1.12 fvdl }
214 1.12 fvdl /*
215 1.12 fvdl * Find a cylinder with greater than average number of
216 1.12 fvdl * unused data blocks.
217 1.12 fvdl */
218 1.12 fvdl if (indx == 0 || bap[indx - 1] == 0)
219 1.12 fvdl startcg =
220 1.12 fvdl ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg;
221 1.12 fvdl else
222 1.12 fvdl startcg = dtog(fs,
223 1.12 fvdl ufs_rw64(bap[indx - 1], UFS_FSNEEDSWAP(fs)) + 1);
224 1.12 fvdl startcg %= fs->fs_ncg;
225 1.12 fvdl avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
226 1.12 fvdl for (cg = startcg; cg < fs->fs_ncg; cg++)
227 1.1 lukem if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
228 1.1 lukem return (fs->fs_fpg * cg + fs->fs_frag);
229 1.1 lukem }
230 1.12 fvdl for (cg = 0; cg < startcg; cg++)
231 1.1 lukem if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
232 1.1 lukem return (fs->fs_fpg * cg + fs->fs_frag);
233 1.1 lukem }
234 1.1 lukem return (0);
235 1.1 lukem }
236 1.1 lukem /*
237 1.12 fvdl * We just always try to lay things out contiguously.
238 1.1 lukem */
239 1.12 fvdl return ufs_rw64(bap[indx - 1], UFS_FSNEEDSWAP(fs)) + fs->fs_frag;
240 1.1 lukem }
241 1.1 lukem
242 1.1 lukem /*
243 1.1 lukem * Implement the cylinder overflow algorithm.
244 1.1 lukem *
245 1.1 lukem * The policy implemented by this algorithm is:
246 1.1 lukem * 1) allocate the block in its requested cylinder group.
247 1.1 lukem * 2) quadradically rehash on the cylinder group number.
248 1.1 lukem * 3) brute force search for a free block.
249 1.1 lukem *
250 1.1 lukem * `size': size for data blocks, mode for inodes
251 1.1 lukem */
252 1.1 lukem /*VARARGS5*/
253 1.11 fvdl static daddr_t
254 1.11 fvdl ffs_hashalloc(struct inode *ip, int cg, daddr_t pref, int size,
255 1.11 fvdl daddr_t (*allocator)(struct inode *, int, daddr_t, int))
256 1.1 lukem {
257 1.1 lukem struct fs *fs;
258 1.11 fvdl daddr_t result;
259 1.1 lukem int i, icg = cg;
260 1.1 lukem
261 1.1 lukem fs = ip->i_fs;
262 1.1 lukem /*
263 1.1 lukem * 1: preferred cylinder group
264 1.1 lukem */
265 1.1 lukem result = (*allocator)(ip, cg, pref, size);
266 1.1 lukem if (result)
267 1.1 lukem return (result);
268 1.1 lukem /*
269 1.1 lukem * 2: quadratic rehash
270 1.1 lukem */
271 1.1 lukem for (i = 1; i < fs->fs_ncg; i *= 2) {
272 1.1 lukem cg += i;
273 1.1 lukem if (cg >= fs->fs_ncg)
274 1.1 lukem cg -= fs->fs_ncg;
275 1.1 lukem result = (*allocator)(ip, cg, 0, size);
276 1.1 lukem if (result)
277 1.1 lukem return (result);
278 1.1 lukem }
279 1.1 lukem /*
280 1.1 lukem * 3: brute force search
281 1.1 lukem * Note that we start at i == 2, since 0 was checked initially,
282 1.1 lukem * and 1 is always checked in the quadratic rehash.
283 1.1 lukem */
284 1.1 lukem cg = (icg + 2) % fs->fs_ncg;
285 1.1 lukem for (i = 2; i < fs->fs_ncg; i++) {
286 1.1 lukem result = (*allocator)(ip, cg, 0, size);
287 1.1 lukem if (result)
288 1.1 lukem return (result);
289 1.1 lukem cg++;
290 1.1 lukem if (cg == fs->fs_ncg)
291 1.1 lukem cg = 0;
292 1.1 lukem }
293 1.1 lukem return (0);
294 1.1 lukem }
295 1.1 lukem
296 1.1 lukem /*
297 1.1 lukem * Determine whether a block can be allocated.
298 1.1 lukem *
299 1.1 lukem * Check to see if a block of the appropriate size is available,
300 1.1 lukem * and if it is, allocate it.
301 1.1 lukem */
302 1.11 fvdl static daddr_t
303 1.11 fvdl ffs_alloccg(struct inode *ip, int cg, daddr_t bpref, int size)
304 1.1 lukem {
305 1.1 lukem struct cg *cgp;
306 1.1 lukem struct buf *bp;
307 1.11 fvdl daddr_t bno, blkno;
308 1.1 lukem int error, frags, allocsiz, i;
309 1.1 lukem struct fs *fs = ip->i_fs;
310 1.1 lukem const int needswap = UFS_FSNEEDSWAP(fs);
311 1.1 lukem
312 1.1 lukem if (fs->fs_cs(fs, cg).cs_nbfree == 0 && size == fs->fs_bsize)
313 1.1 lukem return (0);
314 1.1 lukem error = bread(ip->i_fd, ip->i_fs, fsbtodb(fs, cgtod(fs, cg)),
315 1.1 lukem (int)fs->fs_cgsize, &bp);
316 1.1 lukem if (error) {
317 1.1 lukem brelse(bp);
318 1.1 lukem return (0);
319 1.1 lukem }
320 1.1 lukem cgp = (struct cg *)bp->b_data;
321 1.1 lukem if (!cg_chkmagic(cgp, needswap) ||
322 1.1 lukem (cgp->cg_cs.cs_nbfree == 0 && size == fs->fs_bsize)) {
323 1.1 lukem brelse(bp);
324 1.1 lukem return (0);
325 1.1 lukem }
326 1.1 lukem if (size == fs->fs_bsize) {
327 1.1 lukem bno = ffs_alloccgblk(ip, bp, bpref);
328 1.1 lukem bdwrite(bp);
329 1.1 lukem return (bno);
330 1.1 lukem }
331 1.1 lukem /*
332 1.1 lukem * check to see if any fragments are already available
333 1.1 lukem * allocsiz is the size which will be allocated, hacking
334 1.1 lukem * it down to a smaller size if necessary
335 1.1 lukem */
336 1.1 lukem frags = numfrags(fs, size);
337 1.1 lukem for (allocsiz = frags; allocsiz < fs->fs_frag; allocsiz++)
338 1.1 lukem if (cgp->cg_frsum[allocsiz] != 0)
339 1.1 lukem break;
340 1.1 lukem if (allocsiz == fs->fs_frag) {
341 1.1 lukem /*
342 1.1 lukem * no fragments were available, so a block will be
343 1.1 lukem * allocated, and hacked up
344 1.1 lukem */
345 1.1 lukem if (cgp->cg_cs.cs_nbfree == 0) {
346 1.1 lukem brelse(bp);
347 1.1 lukem return (0);
348 1.1 lukem }
349 1.1 lukem bno = ffs_alloccgblk(ip, bp, bpref);
350 1.1 lukem bpref = dtogd(fs, bno);
351 1.1 lukem for (i = frags; i < fs->fs_frag; i++)
352 1.1 lukem setbit(cg_blksfree(cgp, needswap), bpref + i);
353 1.1 lukem i = fs->fs_frag - frags;
354 1.1 lukem ufs_add32(cgp->cg_cs.cs_nffree, i, needswap);
355 1.1 lukem fs->fs_cstotal.cs_nffree += i;
356 1.1 lukem fs->fs_cs(fs, cg).cs_nffree += i;
357 1.1 lukem fs->fs_fmod = 1;
358 1.1 lukem ufs_add32(cgp->cg_frsum[i], 1, needswap);
359 1.1 lukem bdwrite(bp);
360 1.1 lukem return (bno);
361 1.1 lukem }
362 1.1 lukem bno = ffs_mapsearch(fs, cgp, bpref, allocsiz);
363 1.1 lukem for (i = 0; i < frags; i++)
364 1.1 lukem clrbit(cg_blksfree(cgp, needswap), bno + i);
365 1.1 lukem ufs_add32(cgp->cg_cs.cs_nffree, -frags, needswap);
366 1.1 lukem fs->fs_cstotal.cs_nffree -= frags;
367 1.1 lukem fs->fs_cs(fs, cg).cs_nffree -= frags;
368 1.1 lukem fs->fs_fmod = 1;
369 1.1 lukem ufs_add32(cgp->cg_frsum[allocsiz], -1, needswap);
370 1.1 lukem if (frags != allocsiz)
371 1.1 lukem ufs_add32(cgp->cg_frsum[allocsiz - frags], 1, needswap);
372 1.1 lukem blkno = cg * fs->fs_fpg + bno;
373 1.1 lukem bdwrite(bp);
374 1.1 lukem return blkno;
375 1.1 lukem }
376 1.1 lukem
377 1.1 lukem /*
378 1.1 lukem * Allocate a block in a cylinder group.
379 1.1 lukem *
380 1.1 lukem * This algorithm implements the following policy:
381 1.1 lukem * 1) allocate the requested block.
382 1.1 lukem * 2) allocate a rotationally optimal block in the same cylinder.
383 1.1 lukem * 3) allocate the next available block on the block rotor for the
384 1.1 lukem * specified cylinder group.
385 1.1 lukem * Note that this routine only allocates fs_bsize blocks; these
386 1.1 lukem * blocks may be fragmented by the routine that allocates them.
387 1.1 lukem */
388 1.11 fvdl static daddr_t
389 1.11 fvdl ffs_alloccgblk(struct inode *ip, struct buf *bp, daddr_t bpref)
390 1.1 lukem {
391 1.1 lukem struct cg *cgp;
392 1.12 fvdl daddr_t blkno;
393 1.12 fvdl int32_t bno;
394 1.1 lukem struct fs *fs = ip->i_fs;
395 1.1 lukem const int needswap = UFS_FSNEEDSWAP(fs);
396 1.12 fvdl u_int8_t *blksfree;
397 1.1 lukem
398 1.1 lukem cgp = (struct cg *)bp->b_data;
399 1.12 fvdl blksfree = cg_blksfree(cgp, needswap);
400 1.1 lukem if (bpref == 0 || dtog(fs, bpref) != ufs_rw32(cgp->cg_cgx, needswap)) {
401 1.1 lukem bpref = ufs_rw32(cgp->cg_rotor, needswap);
402 1.12 fvdl } else {
403 1.12 fvdl bpref = blknum(fs, bpref);
404 1.12 fvdl bno = dtogd(fs, bpref);
405 1.1 lukem /*
406 1.12 fvdl * if the requested block is available, use it
407 1.1 lukem */
408 1.12 fvdl if (ffs_isblock(fs, blksfree, fragstoblks(fs, bno)))
409 1.12 fvdl goto gotit;
410 1.1 lukem }
411 1.1 lukem /*
412 1.12 fvdl * Take the next available one in this cylinder group.
413 1.1 lukem */
414 1.1 lukem bno = ffs_mapsearch(fs, cgp, bpref, (int)fs->fs_frag);
415 1.1 lukem if (bno < 0)
416 1.1 lukem return (0);
417 1.12 fvdl cgp->cg_rotor = ufs_rw32(bno, needswap);
418 1.1 lukem gotit:
419 1.1 lukem blkno = fragstoblks(fs, bno);
420 1.12 fvdl ffs_clrblock(fs, blksfree, (long)blkno);
421 1.1 lukem ffs_clusteracct(fs, cgp, blkno, -1);
422 1.1 lukem ufs_add32(cgp->cg_cs.cs_nbfree, -1, needswap);
423 1.1 lukem fs->fs_cstotal.cs_nbfree--;
424 1.1 lukem fs->fs_cs(fs, ufs_rw32(cgp->cg_cgx, needswap)).cs_nbfree--;
425 1.1 lukem fs->fs_fmod = 1;
426 1.1 lukem blkno = ufs_rw32(cgp->cg_cgx, needswap) * fs->fs_fpg + bno;
427 1.1 lukem return (blkno);
428 1.1 lukem }
429 1.1 lukem
430 1.1 lukem /*
431 1.1 lukem * Free a block or fragment.
432 1.1 lukem *
433 1.1 lukem * The specified block or fragment is placed back in the
434 1.1 lukem * free map. If a fragment is deallocated, a possible
435 1.1 lukem * block reassembly is checked.
436 1.1 lukem */
437 1.1 lukem void
438 1.11 fvdl ffs_blkfree(struct inode *ip, daddr_t bno, long size)
439 1.1 lukem {
440 1.1 lukem struct cg *cgp;
441 1.1 lukem struct buf *bp;
442 1.12 fvdl int32_t fragno, cgbno;
443 1.1 lukem int i, error, cg, blk, frags, bbase;
444 1.1 lukem struct fs *fs = ip->i_fs;
445 1.1 lukem const int needswap = UFS_FSNEEDSWAP(fs);
446 1.1 lukem
447 1.1 lukem if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0 ||
448 1.1 lukem fragnum(fs, bno) + numfrags(fs, size) > fs->fs_frag) {
449 1.11 fvdl errx(1, "blkfree: bad size: bno %lld bsize %d size %ld",
450 1.11 fvdl (long long)bno, fs->fs_bsize, size);
451 1.1 lukem }
452 1.1 lukem cg = dtog(fs, bno);
453 1.12 fvdl if (bno >= fs->fs_size) {
454 1.11 fvdl warnx("bad block %lld, ino %d", (long long)bno, ip->i_number);
455 1.1 lukem return;
456 1.1 lukem }
457 1.1 lukem error = bread(ip->i_fd, ip->i_fs, fsbtodb(fs, cgtod(fs, cg)),
458 1.1 lukem (int)fs->fs_cgsize, &bp);
459 1.1 lukem if (error) {
460 1.1 lukem brelse(bp);
461 1.1 lukem return;
462 1.1 lukem }
463 1.1 lukem cgp = (struct cg *)bp->b_data;
464 1.1 lukem if (!cg_chkmagic(cgp, needswap)) {
465 1.1 lukem brelse(bp);
466 1.1 lukem return;
467 1.1 lukem }
468 1.12 fvdl cgbno = dtogd(fs, bno);
469 1.1 lukem if (size == fs->fs_bsize) {
470 1.12 fvdl fragno = fragstoblks(fs, cgbno);
471 1.12 fvdl if (!ffs_isfreeblock(fs, cg_blksfree(cgp, needswap), fragno)) {
472 1.11 fvdl errx(1, "blkfree: freeing free block %lld",
473 1.11 fvdl (long long)bno);
474 1.1 lukem }
475 1.12 fvdl ffs_setblock(fs, cg_blksfree(cgp, needswap), fragno);
476 1.12 fvdl ffs_clusteracct(fs, cgp, fragno, 1);
477 1.1 lukem ufs_add32(cgp->cg_cs.cs_nbfree, 1, needswap);
478 1.1 lukem fs->fs_cstotal.cs_nbfree++;
479 1.1 lukem fs->fs_cs(fs, cg).cs_nbfree++;
480 1.1 lukem } else {
481 1.12 fvdl bbase = cgbno - fragnum(fs, cgbno);
482 1.1 lukem /*
483 1.1 lukem * decrement the counts associated with the old frags
484 1.1 lukem */
485 1.1 lukem blk = blkmap(fs, cg_blksfree(cgp, needswap), bbase);
486 1.1 lukem ffs_fragacct(fs, blk, cgp->cg_frsum, -1, needswap);
487 1.1 lukem /*
488 1.1 lukem * deallocate the fragment
489 1.1 lukem */
490 1.1 lukem frags = numfrags(fs, size);
491 1.1 lukem for (i = 0; i < frags; i++) {
492 1.12 fvdl if (isset(cg_blksfree(cgp, needswap), cgbno + i)) {
493 1.11 fvdl errx(1, "blkfree: freeing free frag: block %lld",
494 1.12 fvdl (long long)(cgbno + i));
495 1.1 lukem }
496 1.12 fvdl setbit(cg_blksfree(cgp, needswap), cgbno + i);
497 1.1 lukem }
498 1.1 lukem ufs_add32(cgp->cg_cs.cs_nffree, i, needswap);
499 1.1 lukem fs->fs_cstotal.cs_nffree += i;
500 1.1 lukem fs->fs_cs(fs, cg).cs_nffree += i;
501 1.1 lukem /*
502 1.1 lukem * add back in counts associated with the new frags
503 1.1 lukem */
504 1.1 lukem blk = blkmap(fs, cg_blksfree(cgp, needswap), bbase);
505 1.1 lukem ffs_fragacct(fs, blk, cgp->cg_frsum, 1, needswap);
506 1.1 lukem /*
507 1.1 lukem * if a complete block has been reassembled, account for it
508 1.1 lukem */
509 1.12 fvdl fragno = fragstoblks(fs, bbase);
510 1.12 fvdl if (ffs_isblock(fs, cg_blksfree(cgp, needswap), fragno)) {
511 1.1 lukem ufs_add32(cgp->cg_cs.cs_nffree, -fs->fs_frag, needswap);
512 1.1 lukem fs->fs_cstotal.cs_nffree -= fs->fs_frag;
513 1.1 lukem fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag;
514 1.12 fvdl ffs_clusteracct(fs, cgp, fragno, 1);
515 1.1 lukem ufs_add32(cgp->cg_cs.cs_nbfree, 1, needswap);
516 1.1 lukem fs->fs_cstotal.cs_nbfree++;
517 1.1 lukem fs->fs_cs(fs, cg).cs_nbfree++;
518 1.1 lukem }
519 1.1 lukem }
520 1.1 lukem fs->fs_fmod = 1;
521 1.1 lukem bdwrite(bp);
522 1.1 lukem }
523 1.1 lukem
524 1.1 lukem
525 1.1 lukem static int
526 1.1 lukem scanc(u_int size, const u_char *cp, const u_char table[], int mask)
527 1.1 lukem {
528 1.1 lukem const u_char *end = &cp[size];
529 1.1 lukem
530 1.1 lukem while (cp < end && (table[*cp] & mask) == 0)
531 1.1 lukem cp++;
532 1.1 lukem return (end - cp);
533 1.1 lukem }
534 1.1 lukem
535 1.1 lukem /*
536 1.1 lukem * Find a block of the specified size in the specified cylinder group.
537 1.1 lukem *
538 1.1 lukem * It is a panic if a request is made to find a block if none are
539 1.1 lukem * available.
540 1.1 lukem */
541 1.12 fvdl static int32_t
542 1.11 fvdl ffs_mapsearch(struct fs *fs, struct cg *cgp, daddr_t bpref, int allocsiz)
543 1.1 lukem {
544 1.12 fvdl int32_t bno;
545 1.1 lukem int start, len, loc, i;
546 1.1 lukem int blk, field, subfield, pos;
547 1.1 lukem int ostart, olen;
548 1.1 lukem const int needswap = UFS_FSNEEDSWAP(fs);
549 1.1 lukem
550 1.1 lukem /*
551 1.1 lukem * find the fragment by searching through the free block
552 1.1 lukem * map for an appropriate bit pattern
553 1.1 lukem */
554 1.1 lukem if (bpref)
555 1.1 lukem start = dtogd(fs, bpref) / NBBY;
556 1.1 lukem else
557 1.1 lukem start = ufs_rw32(cgp->cg_frotor, needswap) / NBBY;
558 1.1 lukem len = howmany(fs->fs_fpg, NBBY) - start;
559 1.1 lukem ostart = start;
560 1.1 lukem olen = len;
561 1.1 lukem loc = scanc((u_int)len,
562 1.1 lukem (const u_char *)&cg_blksfree(cgp, needswap)[start],
563 1.1 lukem (const u_char *)fragtbl[fs->fs_frag],
564 1.1 lukem (1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
565 1.1 lukem if (loc == 0) {
566 1.1 lukem len = start + 1;
567 1.1 lukem start = 0;
568 1.1 lukem loc = scanc((u_int)len,
569 1.1 lukem (const u_char *)&cg_blksfree(cgp, needswap)[0],
570 1.1 lukem (const u_char *)fragtbl[fs->fs_frag],
571 1.1 lukem (1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
572 1.1 lukem if (loc == 0) {
573 1.1 lukem errx(1,
574 1.1 lukem "ffs_alloccg: map corrupted: start %d len %d offset %d %ld",
575 1.1 lukem ostart, olen,
576 1.1 lukem ufs_rw32(cgp->cg_freeoff, needswap),
577 1.1 lukem (long)cg_blksfree(cgp, needswap) - (long)cgp);
578 1.1 lukem /* NOTREACHED */
579 1.1 lukem }
580 1.1 lukem }
581 1.1 lukem bno = (start + len - loc) * NBBY;
582 1.1 lukem cgp->cg_frotor = ufs_rw32(bno, needswap);
583 1.1 lukem /*
584 1.1 lukem * found the byte in the map
585 1.1 lukem * sift through the bits to find the selected frag
586 1.1 lukem */
587 1.1 lukem for (i = bno + NBBY; bno < i; bno += fs->fs_frag) {
588 1.1 lukem blk = blkmap(fs, cg_blksfree(cgp, needswap), bno);
589 1.1 lukem blk <<= 1;
590 1.1 lukem field = around[allocsiz];
591 1.1 lukem subfield = inside[allocsiz];
592 1.1 lukem for (pos = 0; pos <= fs->fs_frag - allocsiz; pos++) {
593 1.1 lukem if ((blk & field) == subfield)
594 1.1 lukem return (bno + pos);
595 1.1 lukem field <<= 1;
596 1.1 lukem subfield <<= 1;
597 1.1 lukem }
598 1.1 lukem }
599 1.11 fvdl errx(1, "ffs_alloccg: block not in map: bno %lld", (long long)bno);
600 1.1 lukem return (-1);
601 1.1 lukem }
602 1.1 lukem
603 1.1 lukem /*
604 1.1 lukem * Update the cluster map because of an allocation or free.
605 1.1 lukem *
606 1.1 lukem * Cnt == 1 means free; cnt == -1 means allocating.
607 1.1 lukem */
608 1.1 lukem void
609 1.12 fvdl ffs_clusteracct(struct fs *fs, struct cg *cgp, int32_t blkno, int cnt)
610 1.1 lukem {
611 1.1 lukem int32_t *sump;
612 1.1 lukem int32_t *lp;
613 1.1 lukem u_char *freemapp, *mapp;
614 1.1 lukem int i, start, end, forw, back, map, bit;
615 1.1 lukem const int needswap = UFS_FSNEEDSWAP(fs);
616 1.1 lukem
617 1.1 lukem if (fs->fs_contigsumsize <= 0)
618 1.1 lukem return;
619 1.1 lukem freemapp = cg_clustersfree(cgp, needswap);
620 1.1 lukem sump = cg_clustersum(cgp, needswap);
621 1.1 lukem /*
622 1.1 lukem * Allocate or clear the actual block.
623 1.1 lukem */
624 1.1 lukem if (cnt > 0)
625 1.1 lukem setbit(freemapp, blkno);
626 1.1 lukem else
627 1.1 lukem clrbit(freemapp, blkno);
628 1.1 lukem /*
629 1.1 lukem * Find the size of the cluster going forward.
630 1.1 lukem */
631 1.1 lukem start = blkno + 1;
632 1.1 lukem end = start + fs->fs_contigsumsize;
633 1.1 lukem if (end >= ufs_rw32(cgp->cg_nclusterblks, needswap))
634 1.1 lukem end = ufs_rw32(cgp->cg_nclusterblks, needswap);
635 1.1 lukem mapp = &freemapp[start / NBBY];
636 1.1 lukem map = *mapp++;
637 1.1 lukem bit = 1 << (start % NBBY);
638 1.1 lukem for (i = start; i < end; i++) {
639 1.1 lukem if ((map & bit) == 0)
640 1.1 lukem break;
641 1.1 lukem if ((i & (NBBY - 1)) != (NBBY - 1)) {
642 1.1 lukem bit <<= 1;
643 1.1 lukem } else {
644 1.1 lukem map = *mapp++;
645 1.1 lukem bit = 1;
646 1.1 lukem }
647 1.1 lukem }
648 1.1 lukem forw = i - start;
649 1.1 lukem /*
650 1.1 lukem * Find the size of the cluster going backward.
651 1.1 lukem */
652 1.1 lukem start = blkno - 1;
653 1.1 lukem end = start - fs->fs_contigsumsize;
654 1.1 lukem if (end < 0)
655 1.1 lukem end = -1;
656 1.1 lukem mapp = &freemapp[start / NBBY];
657 1.1 lukem map = *mapp--;
658 1.1 lukem bit = 1 << (start % NBBY);
659 1.1 lukem for (i = start; i > end; i--) {
660 1.1 lukem if ((map & bit) == 0)
661 1.1 lukem break;
662 1.1 lukem if ((i & (NBBY - 1)) != 0) {
663 1.1 lukem bit >>= 1;
664 1.1 lukem } else {
665 1.1 lukem map = *mapp--;
666 1.1 lukem bit = 1 << (NBBY - 1);
667 1.1 lukem }
668 1.1 lukem }
669 1.1 lukem back = start - i;
670 1.1 lukem /*
671 1.1 lukem * Account for old cluster and the possibly new forward and
672 1.1 lukem * back clusters.
673 1.1 lukem */
674 1.1 lukem i = back + forw + 1;
675 1.1 lukem if (i > fs->fs_contigsumsize)
676 1.1 lukem i = fs->fs_contigsumsize;
677 1.1 lukem ufs_add32(sump[i], cnt, needswap);
678 1.1 lukem if (back > 0)
679 1.1 lukem ufs_add32(sump[back], -cnt, needswap);
680 1.1 lukem if (forw > 0)
681 1.1 lukem ufs_add32(sump[forw], -cnt, needswap);
682 1.1 lukem
683 1.1 lukem /*
684 1.1 lukem * Update cluster summary information.
685 1.1 lukem */
686 1.1 lukem lp = &sump[fs->fs_contigsumsize];
687 1.1 lukem for (i = fs->fs_contigsumsize; i > 0; i--)
688 1.1 lukem if (ufs_rw32(*lp--, needswap) > 0)
689 1.1 lukem break;
690 1.1 lukem fs->fs_maxcluster[ufs_rw32(cgp->cg_cgx, needswap)] = i;
691 1.1 lukem }
692