Deleted Added
full compact
ffs_alloc.c (30402) ffs_alloc.c (30418)
1/*
2 * Copyright (c) 1982, 1986, 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 3. All advertising materials mentioning features or use of this software
14 * must display the following acknowledgement:
15 * This product includes software developed by the University of
16 * California, Berkeley and its contributors.
17 * 4. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
20 *
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * SUCH DAMAGE.
32 *
33 * @(#)ffs_alloc.c 8.18 (Berkeley) 5/26/95
1/*
2 * Copyright (c) 1982, 1986, 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 3. All advertising materials mentioning features or use of this software
14 * must display the following acknowledgement:
15 * This product includes software developed by the University of
16 * California, Berkeley and its contributors.
17 * 4. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
20 *
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * SUCH DAMAGE.
32 *
33 * @(#)ffs_alloc.c 8.18 (Berkeley) 5/26/95
34 * $Id: ffs_alloc.c,v 1.37 1997/09/19 11:13:16 phk Exp $
34 * $Id: ffs_alloc.c,v 1.38 1997/10/14 14:22:23 phk Exp $
35 */
36
37#include "opt_quota.h"
38
39#include <sys/param.h>
40#include <sys/systm.h>
41#include <sys/buf.h>
42#include <sys/proc.h>
43#include <sys/vnode.h>
44#include <sys/mount.h>
45#include <sys/kernel.h>
46#include <sys/sysctl.h>
47#include <sys/syslog.h>
48
49#include <ufs/ufs/quota.h>
50#include <ufs/ufs/inode.h>
35 */
36
37#include "opt_quota.h"
38
39#include <sys/param.h>
40#include <sys/systm.h>
41#include <sys/buf.h>
42#include <sys/proc.h>
43#include <sys/vnode.h>
44#include <sys/mount.h>
45#include <sys/kernel.h>
46#include <sys/sysctl.h>
47#include <sys/syslog.h>
48
49#include <ufs/ufs/quota.h>
50#include <ufs/ufs/inode.h>
51#include <ufs/ufs/ufs_extern.h>
52
53#include <ufs/ffs/fs.h>
54#include <ufs/ffs/ffs_extern.h>
55
56typedef ufs_daddr_t allocfcn_t __P((struct inode *ip, int cg, ufs_daddr_t bpref,
57 int size));
58
59static ufs_daddr_t ffs_alloccg __P((struct inode *, int, ufs_daddr_t, int));
60static ufs_daddr_t ffs_alloccgblk __P((struct fs *, struct cg *, ufs_daddr_t));
61static void ffs_clusteracct __P((struct fs *, struct cg *, ufs_daddr_t,
62 int));
63static ufs_daddr_t ffs_clusteralloc __P((struct inode *, int, ufs_daddr_t,
64 int));
65static ino_t ffs_dirpref __P((struct fs *));
66static ufs_daddr_t ffs_fragextend __P((struct inode *, int, long, int, int));
67static void ffs_fserr __P((struct fs *, u_int, char *));
68static u_long ffs_hashalloc
69 __P((struct inode *, int, long, int, allocfcn_t *));
70static ino_t ffs_nodealloccg __P((struct inode *, int, ufs_daddr_t, int));
71static ufs_daddr_t ffs_mapsearch __P((struct fs *, struct cg *, ufs_daddr_t,
72 int));
73
74/*
75 * Allocate a block in the file system.
76 *
77 * The size of the requested block is given, which must be some
78 * multiple of fs_fsize and <= fs_bsize.
79 * A preference may be optionally specified. If a preference is given
80 * the following hierarchy is used to allocate a block:
81 * 1) allocate the requested block.
82 * 2) allocate a rotationally optimal block in the same cylinder.
83 * 3) allocate a block in the same cylinder group.
84 * 4) quadradically rehash into other cylinder groups, until an
85 * available block is located.
86 * If no block preference is given the following heirarchy is used
87 * to allocate a block:
88 * 1) allocate a block in the cylinder group that contains the
89 * inode for the file.
90 * 2) quadradically rehash into other cylinder groups, until an
91 * available block is located.
92 */
93int
94ffs_alloc(ip, lbn, bpref, size, cred, bnp)
95 register struct inode *ip;
96 ufs_daddr_t lbn, bpref;
97 int size;
98 struct ucred *cred;
99 ufs_daddr_t *bnp;
100{
101 register struct fs *fs;
102 ufs_daddr_t bno;
103 int cg;
104#ifdef QUOTA
105 int error;
106#endif
107
108 *bnp = 0;
109 fs = ip->i_fs;
110#ifdef DIAGNOSTIC
111 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) {
112 printf("dev = 0x%lx, bsize = %ld, size = %d, fs = %s\n",
113 (u_long)ip->i_dev, fs->fs_bsize, size, fs->fs_fsmnt);
114 panic("ffs_alloc: bad size");
115 }
116 if (cred == NOCRED)
117 panic("ffs_alloc: missing credential");
118#endif /* DIAGNOSTIC */
119 if (size == fs->fs_bsize && fs->fs_cstotal.cs_nbfree == 0)
120 goto nospace;
121 if (cred->cr_uid != 0 &&
122 freespace(fs, fs->fs_minfree) - numfrags(fs, size) < 0)
123 goto nospace;
124#ifdef QUOTA
125 error = chkdq(ip, (long)btodb(size), cred, 0);
126 if (error)
127 return (error);
128#endif
129 if (bpref >= fs->fs_size)
130 bpref = 0;
131 if (bpref == 0)
132 cg = ino_to_cg(fs, ip->i_number);
133 else
134 cg = dtog(fs, bpref);
135 bno = (ufs_daddr_t)ffs_hashalloc(ip, cg, (long)bpref, size,
136 ffs_alloccg);
137 if (bno > 0) {
138 ip->i_blocks += btodb(size);
139 ip->i_flag |= IN_CHANGE | IN_UPDATE;
140 *bnp = bno;
141 return (0);
142 }
143#ifdef QUOTA
144 /*
145 * Restore user's disk quota because allocation failed.
146 */
147 (void) chkdq(ip, (long)-btodb(size), cred, FORCE);
148#endif
149nospace:
150 ffs_fserr(fs, cred->cr_uid, "file system full");
151 uprintf("\n%s: write failed, file system is full\n", fs->fs_fsmnt);
152 return (ENOSPC);
153}
154
155/*
156 * Reallocate a fragment to a bigger size
157 *
158 * The number and size of the old block is given, and a preference
159 * and new size is also specified. The allocator attempts to extend
160 * the original block. Failing that, the regular block allocator is
161 * invoked to get an appropriate block.
162 */
163int
164ffs_realloccg(ip, lbprev, bpref, osize, nsize, cred, bpp)
165 register struct inode *ip;
166 ufs_daddr_t lbprev;
167 ufs_daddr_t bpref;
168 int osize, nsize;
169 struct ucred *cred;
170 struct buf **bpp;
171{
172 register struct fs *fs;
173 struct buf *bp;
174 int cg, request, error;
175 ufs_daddr_t bprev, bno;
176
177 *bpp = 0;
178 fs = ip->i_fs;
179#ifdef DIAGNOSTIC
180 if ((u_int)osize > fs->fs_bsize || fragoff(fs, osize) != 0 ||
181 (u_int)nsize > fs->fs_bsize || fragoff(fs, nsize) != 0) {
182 printf(
183 "dev = 0x%lx, bsize = %ld, osize = %d, "
184 "nsize = %d, fs = %s\n",
185 (u_long)ip->i_dev, fs->fs_bsize, osize,
186 nsize, fs->fs_fsmnt);
187 panic("ffs_realloccg: bad size");
188 }
189 if (cred == NOCRED)
190 panic("ffs_realloccg: missing credential");
191#endif /* DIAGNOSTIC */
192 if (cred->cr_uid != 0 &&
193 freespace(fs, fs->fs_minfree) - numfrags(fs, nsize - osize) < 0)
194 goto nospace;
195 if ((bprev = ip->i_db[lbprev]) == 0) {
196 printf("dev = 0x%lx, bsize = %ld, bprev = %ld, fs = %s\n",
197 (u_long) ip->i_dev, fs->fs_bsize, bprev, fs->fs_fsmnt);
198 panic("ffs_realloccg: bad bprev");
199 }
200 /*
201 * Allocate the extra space in the buffer.
202 */
203 error = bread(ITOV(ip), lbprev, osize, NOCRED, &bp);
204 if (error) {
205 brelse(bp);
206 return (error);
207 }
208
209 if( bp->b_blkno == bp->b_lblkno) {
210 if( lbprev >= NDADDR)
211 panic("ffs_realloccg: lbprev out of range");
212 bp->b_blkno = fsbtodb(fs, bprev);
213 }
214
215#ifdef QUOTA
216 error = chkdq(ip, (long)btodb(nsize - osize), cred, 0);
217 if (error) {
218 brelse(bp);
219 return (error);
220 }
221#endif
222 /*
223 * Check for extension in the existing location.
224 */
225 cg = dtog(fs, bprev);
226 bno = ffs_fragextend(ip, cg, (long)bprev, osize, nsize);
227 if (bno) {
228 if (bp->b_blkno != fsbtodb(fs, bno))
229 panic("ffs_realloccg: bad blockno");
230 ip->i_blocks += btodb(nsize - osize);
231 ip->i_flag |= IN_CHANGE | IN_UPDATE;
232 allocbuf(bp, nsize);
233 bp->b_flags |= B_DONE;
234 bzero((char *)bp->b_data + osize, (u_int)nsize - osize);
235 *bpp = bp;
236 return (0);
237 }
238 /*
239 * Allocate a new disk location.
240 */
241 if (bpref >= fs->fs_size)
242 bpref = 0;
243 switch ((int)fs->fs_optim) {
244 case FS_OPTSPACE:
245 /*
246 * Allocate an exact sized fragment. Although this makes
247 * best use of space, we will waste time relocating it if
248 * the file continues to grow. If the fragmentation is
249 * less than half of the minimum free reserve, we choose
250 * to begin optimizing for time.
251 */
252 request = nsize;
253 if (fs->fs_minfree <= 5 ||
254 fs->fs_cstotal.cs_nffree >
255 fs->fs_dsize * fs->fs_minfree / (2 * 100))
256 break;
257 log(LOG_NOTICE, "%s: optimization changed from SPACE to TIME\n",
258 fs->fs_fsmnt);
259 fs->fs_optim = FS_OPTTIME;
260 break;
261 case FS_OPTTIME:
262 /*
263 * At this point we have discovered a file that is trying to
264 * grow a small fragment to a larger fragment. To save time,
265 * we allocate a full sized block, then free the unused portion.
266 * If the file continues to grow, the `ffs_fragextend' call
267 * above will be able to grow it in place without further
268 * copying. If aberrant programs cause disk fragmentation to
269 * grow within 2% of the free reserve, we choose to begin
270 * optimizing for space.
271 */
272 request = fs->fs_bsize;
273 if (fs->fs_cstotal.cs_nffree <
274 fs->fs_dsize * (fs->fs_minfree - 2) / 100)
275 break;
276 log(LOG_NOTICE, "%s: optimization changed from TIME to SPACE\n",
277 fs->fs_fsmnt);
278 fs->fs_optim = FS_OPTSPACE;
279 break;
280 default:
281 printf("dev = 0x%lx, optim = %ld, fs = %s\n",
282 (u_long)ip->i_dev, fs->fs_optim, fs->fs_fsmnt);
283 panic("ffs_realloccg: bad optim");
284 /* NOTREACHED */
285 }
286 bno = (ufs_daddr_t)ffs_hashalloc(ip, cg, (long)bpref, request,
287 ffs_alloccg);
288 if (bno > 0) {
289 bp->b_blkno = fsbtodb(fs, bno);
290 ffs_blkfree(ip, bprev, (long)osize);
291 if (nsize < request)
292 ffs_blkfree(ip, bno + numfrags(fs, nsize),
293 (long)(request - nsize));
294 ip->i_blocks += btodb(nsize - osize);
295 ip->i_flag |= IN_CHANGE | IN_UPDATE;
296 allocbuf(bp, nsize);
297 bp->b_flags |= B_DONE;
298 bzero((char *)bp->b_data + osize, (u_int)nsize - osize);
299 *bpp = bp;
300 return (0);
301 }
302#ifdef QUOTA
303 /*
304 * Restore user's disk quota because allocation failed.
305 */
306 (void) chkdq(ip, (long)-btodb(nsize - osize), cred, FORCE);
307#endif
308 brelse(bp);
309nospace:
310 /*
311 * no space available
312 */
313 ffs_fserr(fs, cred->cr_uid, "file system full");
314 uprintf("\n%s: write failed, file system is full\n", fs->fs_fsmnt);
315 return (ENOSPC);
316}
317
318/*
319 * Reallocate a sequence of blocks into a contiguous sequence of blocks.
320 *
321 * The vnode and an array of buffer pointers for a range of sequential
322 * logical blocks to be made contiguous is given. The allocator attempts
323 * to find a range of sequential blocks starting as close as possible to
324 * an fs_rotdelay offset from the end of the allocation for the logical
325 * block immediately preceeding the current range. If successful, the
326 * physical block numbers in the buffer pointers and in the inode are
327 * changed to reflect the new allocation. If unsuccessful, the allocation
328 * is left unchanged. The success in doing the reallocation is returned.
329 * Note that the error return is not reflected back to the user. Rather
330 * the previous block allocation will be used.
331 */
332static int doasyncfree = 1;
333SYSCTL_INT(_vfs_ffs, FFS_ASYNCFREE, doasyncfree, CTLFLAG_RW, &doasyncfree, 0, "");
334
335int doreallocblks = 1;
336SYSCTL_INT(_vfs_ffs, FFS_REALLOCBLKS, doreallocblks, CTLFLAG_RW, &doreallocblks, 0, "");
337
338static int prtrealloc = 0;
339
340int
341ffs_reallocblks(ap)
342 struct vop_reallocblks_args /* {
343 struct vnode *a_vp;
344 struct cluster_save *a_buflist;
345 } */ *ap;
346{
347#if !defined (not_yes)
348 return (ENOSPC);
349#else
350 struct fs *fs;
351 struct inode *ip;
352 struct vnode *vp;
353 struct buf *sbp, *ebp;
354 ufs_daddr_t *bap, *sbap, *ebap = 0;
355 struct cluster_save *buflist;
356 ufs_daddr_t start_lbn, end_lbn, soff, newblk, blkno;
357 struct indir start_ap[NIADDR + 1], end_ap[NIADDR + 1], *idp;
358 int i, len, start_lvl, end_lvl, pref, ssize;
359 struct timeval tv;
360
361 if (doreallocblks == 0)
362 return (ENOSPC);
363 vp = ap->a_vp;
364 ip = VTOI(vp);
365 fs = ip->i_fs;
366 if (fs->fs_contigsumsize <= 0)
367 return (ENOSPC);
368 buflist = ap->a_buflist;
369 len = buflist->bs_nchildren;
370 start_lbn = buflist->bs_children[0]->b_lblkno;
371 end_lbn = start_lbn + len - 1;
372#ifdef DIAGNOSTIC
373 for (i = 0; i < len; i++)
374 if (!ffs_checkblk(ip,
375 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
376 panic("ffs_reallocblks: unallocated block 1");
377 for (i = 1; i < len; i++)
378 if (buflist->bs_children[i]->b_lblkno != start_lbn + i)
379 panic("ffs_reallocblks: non-logical cluster");
380 blkno = buflist->bs_children[0]->b_blkno;
381 ssize = fsbtodb(fs, fs->fs_frag);
382 for (i = 1; i < len - 1; i++)
383 if (buflist->bs_children[i]->b_blkno != blkno + (i * ssize))
384 panic("ffs_reallocblks: non-physical cluster %d", i);
385#endif
386 /*
387 * If the latest allocation is in a new cylinder group, assume that
388 * the filesystem has decided to move and do not force it back to
389 * the previous cylinder group.
390 */
391 if (dtog(fs, dbtofsb(fs, buflist->bs_children[0]->b_blkno)) !=
392 dtog(fs, dbtofsb(fs, buflist->bs_children[len - 1]->b_blkno)))
393 return (ENOSPC);
394 if (ufs_getlbns(vp, start_lbn, start_ap, &start_lvl) ||
395 ufs_getlbns(vp, end_lbn, end_ap, &end_lvl))
396 return (ENOSPC);
397 /*
398 * Get the starting offset and block map for the first block.
399 */
400 if (start_lvl == 0) {
401 sbap = &ip->i_db[0];
402 soff = start_lbn;
403 } else {
404 idp = &start_ap[start_lvl - 1];
405 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &sbp)) {
406 brelse(sbp);
407 return (ENOSPC);
408 }
409 sbap = (ufs_daddr_t *)sbp->b_data;
410 soff = idp->in_off;
411 }
412 /*
413 * Find the preferred location for the cluster.
414 */
415 pref = ffs_blkpref(ip, start_lbn, soff, sbap);
416 /*
417 * If the block range spans two block maps, get the second map.
418 */
419 if (end_lvl == 0 || (idp = &end_ap[end_lvl - 1])->in_off + 1 >= len) {
420 ssize = len;
421 } else {
422#ifdef DIAGNOSTIC
423 if (start_ap[start_lvl-1].in_lbn == idp->in_lbn)
424 panic("ffs_reallocblk: start == end");
425#endif
426 ssize = len - (idp->in_off + 1);
427 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &ebp))
428 goto fail;
429 ebap = (ufs_daddr_t *)ebp->b_data;
430 }
431 /*
432 * Search the block map looking for an allocation of the desired size.
433 */
434 if ((newblk = (ufs_daddr_t)ffs_hashalloc(ip, dtog(fs, pref), (long)pref,
435 len, ffs_clusteralloc)) == 0)
436 goto fail;
437 /*
438 * We have found a new contiguous block.
439 *
440 * First we have to replace the old block pointers with the new
441 * block pointers in the inode and indirect blocks associated
442 * with the file.
443 */
444#ifdef DEBUG
445 if (prtrealloc)
446 printf("realloc: ino %d, lbns %d-%d\n\told:", ip->i_number,
447 start_lbn, end_lbn);
448#endif
449 blkno = newblk;
450 for (bap = &sbap[soff], i = 0; i < len; i++, blkno += fs->fs_frag) {
451 if (i == ssize)
452 bap = ebap;
453#ifdef DIAGNOSTIC
454 if (!ffs_checkblk(ip,
455 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
456 panic("ffs_reallocblks: unallocated block 2");
457 if (dbtofsb(fs, buflist->bs_children[i]->b_blkno) != *bap)
458 panic("ffs_reallocblks: alloc mismatch");
459#endif
460#ifdef DEBUG
461 if (prtrealloc)
462 printf(" %d,", *bap);
463#endif
464 *bap++ = blkno;
465 }
466 /*
467 * Next we must write out the modified inode and indirect blocks.
468 * For strict correctness, the writes should be synchronous since
469 * the old block values may have been written to disk. In practise
470 * they are almost never written, but if we are concerned about
471 * strict correctness, the `doasyncfree' flag should be set to zero.
472 *
473 * The test on `doasyncfree' should be changed to test a flag
474 * that shows whether the associated buffers and inodes have
475 * been written. The flag should be set when the cluster is
476 * started and cleared whenever the buffer or inode is flushed.
477 * We can then check below to see if it is set, and do the
478 * synchronous write only when it has been cleared.
479 */
480 if (sbap != &ip->i_db[0]) {
481 if (doasyncfree)
482 bdwrite(sbp);
483 else
484 bwrite(sbp);
485 } else {
486 ip->i_flag |= IN_CHANGE | IN_UPDATE;
487 if (!doasyncfree) {
488 gettime(&tv);
489 VOP_UPDATE(vp, &tv, &tv, 1);
490 }
491 }
492 if (ssize < len)
493 if (doasyncfree)
494 bdwrite(ebp);
495 else
496 bwrite(ebp);
497 /*
498 * Last, free the old blocks and assign the new blocks to the buffers.
499 */
500#ifdef DEBUG
501 if (prtrealloc)
502 printf("\n\tnew:");
503#endif
504 for (blkno = newblk, i = 0; i < len; i++, blkno += fs->fs_frag) {
505 ffs_blkfree(ip, dbtofsb(fs, buflist->bs_children[i]->b_blkno),
506 fs->fs_bsize);
507 buflist->bs_children[i]->b_blkno = fsbtodb(fs, blkno);
508#ifdef DEBUG
509 if (!ffs_checkblk(ip,
510 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
511 panic("ffs_reallocblks: unallocated block 3");
512 if (prtrealloc)
513 printf(" %d,", blkno);
514#endif
515 }
516#ifdef DEBUG
517 if (prtrealloc) {
518 prtrealloc--;
519 printf("\n");
520 }
521#endif
522 return (0);
523
524fail:
525 if (ssize < len)
526 brelse(ebp);
527 if (sbap != &ip->i_db[0])
528 brelse(sbp);
529 return (ENOSPC);
530#endif
531}
532
533/*
534 * Allocate an inode in the file system.
535 *
536 * If allocating a directory, use ffs_dirpref to select the inode.
537 * If allocating in a directory, the following hierarchy is followed:
538 * 1) allocate the preferred inode.
539 * 2) allocate an inode in the same cylinder group.
540 * 3) quadradically rehash into other cylinder groups, until an
541 * available inode is located.
542 * If no inode preference is given the following heirarchy is used
543 * to allocate an inode:
544 * 1) allocate an inode in cylinder group 0.
545 * 2) quadradically rehash into other cylinder groups, until an
546 * available inode is located.
547 */
548int
549ffs_valloc(ap)
550 struct vop_valloc_args /* {
551 struct vnode *a_pvp;
552 int a_mode;
553 struct ucred *a_cred;
554 struct vnode **a_vpp;
555 } */ *ap;
556{
557 register struct vnode *pvp = ap->a_pvp;
558 register struct inode *pip;
559 register struct fs *fs;
560 register struct inode *ip;
561 mode_t mode = ap->a_mode;
562 ino_t ino, ipref;
563 int cg, error;
564
565 *ap->a_vpp = NULL;
566 pip = VTOI(pvp);
567 fs = pip->i_fs;
568 if (fs->fs_cstotal.cs_nifree == 0)
569 goto noinodes;
570
571 if ((mode & IFMT) == IFDIR)
572 ipref = ffs_dirpref(fs);
573 else
574 ipref = pip->i_number;
575 if (ipref >= fs->fs_ncg * fs->fs_ipg)
576 ipref = 0;
577 cg = ino_to_cg(fs, ipref);
578 ino = (ino_t)ffs_hashalloc(pip, cg, (long)ipref, mode,
579 (allocfcn_t *)ffs_nodealloccg);
580 if (ino == 0)
581 goto noinodes;
582 error = VFS_VGET(pvp->v_mount, ino, ap->a_vpp);
583 if (error) {
584 VOP_VFREE(pvp, ino, mode);
585 return (error);
586 }
587 ip = VTOI(*ap->a_vpp);
588 if (ip->i_mode) {
589 printf("mode = 0%o, inum = %ld, fs = %s\n",
590 ip->i_mode, ip->i_number, fs->fs_fsmnt);
591 panic("ffs_valloc: dup alloc");
592 }
593 if (ip->i_blocks) { /* XXX */
594 printf("free inode %s/%ld had %ld blocks\n",
595 fs->fs_fsmnt, ino, ip->i_blocks);
596 ip->i_blocks = 0;
597 }
598 ip->i_flags = 0;
599 /*
600 * Set up a new generation number for this inode.
601 */
602 if (ip->i_gen == 0 || ++(ip->i_gen) == 0)
603 ip->i_gen = random() / 2 + 1;
604 return (0);
605noinodes:
606 ffs_fserr(fs, ap->a_cred->cr_uid, "out of inodes");
607 uprintf("\n%s: create/symlink failed, no inodes free\n", fs->fs_fsmnt);
608 return (ENOSPC);
609}
610
611/*
612 * Find a cylinder to place a directory.
613 *
614 * The policy implemented by this algorithm is to select from
615 * among those cylinder groups with above the average number of
616 * free inodes, the one with the smallest number of directories.
617 */
618static ino_t
619ffs_dirpref(fs)
620 register struct fs *fs;
621{
622 int cg, minndir, mincg, avgifree;
623
624 avgifree = fs->fs_cstotal.cs_nifree / fs->fs_ncg;
625 minndir = fs->fs_ipg;
626 mincg = 0;
627 for (cg = 0; cg < fs->fs_ncg; cg++)
628 if (fs->fs_cs(fs, cg).cs_ndir < minndir &&
629 fs->fs_cs(fs, cg).cs_nifree >= avgifree) {
630 mincg = cg;
631 minndir = fs->fs_cs(fs, cg).cs_ndir;
632 }
633 return ((ino_t)(fs->fs_ipg * mincg));
634}
635
636/*
637 * Select the desired position for the next block in a file. The file is
638 * logically divided into sections. The first section is composed of the
639 * direct blocks. Each additional section contains fs_maxbpg blocks.
640 *
641 * If no blocks have been allocated in the first section, the policy is to
642 * request a block in the same cylinder group as the inode that describes
643 * the file. If no blocks have been allocated in any other section, the
644 * policy is to place the section in a cylinder group with a greater than
645 * average number of free blocks. An appropriate cylinder group is found
646 * by using a rotor that sweeps the cylinder groups. When a new group of
647 * blocks is needed, the sweep begins in the cylinder group following the
648 * cylinder group from which the previous allocation was made. The sweep
649 * continues until a cylinder group with greater than the average number
650 * of free blocks is found. If the allocation is for the first block in an
651 * indirect block, the information on the previous allocation is unavailable;
652 * here a best guess is made based upon the logical block number being
653 * allocated.
654 *
655 * If a section is already partially allocated, the policy is to
656 * contiguously allocate fs_maxcontig blocks. The end of one of these
657 * contiguous blocks and the beginning of the next is physically separated
658 * so that the disk head will be in transit between them for at least
659 * fs_rotdelay milliseconds. This is to allow time for the processor to
660 * schedule another I/O transfer.
661 */
662ufs_daddr_t
663ffs_blkpref(ip, lbn, indx, bap)
664 struct inode *ip;
665 ufs_daddr_t lbn;
666 int indx;
667 ufs_daddr_t *bap;
668{
669 register struct fs *fs;
670 register int cg;
671 int avgbfree, startcg;
672 ufs_daddr_t nextblk;
673
674 fs = ip->i_fs;
675 if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) {
676 if (lbn < NDADDR) {
677 cg = ino_to_cg(fs, ip->i_number);
678 return (fs->fs_fpg * cg + fs->fs_frag);
679 }
680 /*
681 * Find a cylinder with greater than average number of
682 * unused data blocks.
683 */
684 if (indx == 0 || bap[indx - 1] == 0)
685 startcg =
686 ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg;
687 else
688 startcg = dtog(fs, bap[indx - 1]) + 1;
689 startcg %= fs->fs_ncg;
690 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
691 for (cg = startcg; cg < fs->fs_ncg; cg++)
692 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
693 fs->fs_cgrotor = cg;
694 return (fs->fs_fpg * cg + fs->fs_frag);
695 }
696 for (cg = 0; cg <= startcg; cg++)
697 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
698 fs->fs_cgrotor = cg;
699 return (fs->fs_fpg * cg + fs->fs_frag);
700 }
701 return (0);
702 }
703 /*
704 * One or more previous blocks have been laid out. If less
705 * than fs_maxcontig previous blocks are contiguous, the
706 * next block is requested contiguously, otherwise it is
707 * requested rotationally delayed by fs_rotdelay milliseconds.
708 */
709 nextblk = bap[indx - 1] + fs->fs_frag;
710 if (fs->fs_rotdelay == 0 || indx < fs->fs_maxcontig ||
711 bap[indx - fs->fs_maxcontig] +
712 blkstofrags(fs, fs->fs_maxcontig) != nextblk)
713 return (nextblk);
714 /*
715 * Here we convert ms of delay to frags as:
716 * (frags) = (ms) * (rev/sec) * (sect/rev) /
717 * ((sect/frag) * (ms/sec))
718 * then round up to the next block.
719 */
720 nextblk += roundup(fs->fs_rotdelay * fs->fs_rps * fs->fs_nsect /
721 (NSPF(fs) * 1000), fs->fs_frag);
722 return (nextblk);
723}
724
725/*
726 * Implement the cylinder overflow algorithm.
727 *
728 * The policy implemented by this algorithm is:
729 * 1) allocate the block in its requested cylinder group.
730 * 2) quadradically rehash on the cylinder group number.
731 * 3) brute force search for a free block.
732 */
733/*VARARGS5*/
734static u_long
735ffs_hashalloc(ip, cg, pref, size, allocator)
736 struct inode *ip;
737 int cg;
738 long pref;
739 int size; /* size for data blocks, mode for inodes */
740 allocfcn_t *allocator;
741{
742 register struct fs *fs;
743 long result; /* XXX why not same type as we return? */
744 int i, icg = cg;
745
746 fs = ip->i_fs;
747 /*
748 * 1: preferred cylinder group
749 */
750 result = (*allocator)(ip, cg, pref, size);
751 if (result)
752 return (result);
753 /*
754 * 2: quadratic rehash
755 */
756 for (i = 1; i < fs->fs_ncg; i *= 2) {
757 cg += i;
758 if (cg >= fs->fs_ncg)
759 cg -= fs->fs_ncg;
760 result = (*allocator)(ip, cg, 0, size);
761 if (result)
762 return (result);
763 }
764 /*
765 * 3: brute force search
766 * Note that we start at i == 2, since 0 was checked initially,
767 * and 1 is always checked in the quadratic rehash.
768 */
769 cg = (icg + 2) % fs->fs_ncg;
770 for (i = 2; i < fs->fs_ncg; i++) {
771 result = (*allocator)(ip, cg, 0, size);
772 if (result)
773 return (result);
774 cg++;
775 if (cg == fs->fs_ncg)
776 cg = 0;
777 }
778 return (0);
779}
780
781/*
782 * Determine whether a fragment can be extended.
783 *
784 * Check to see if the necessary fragments are available, and
785 * if they are, allocate them.
786 */
787static ufs_daddr_t
788ffs_fragextend(ip, cg, bprev, osize, nsize)
789 struct inode *ip;
790 int cg;
791 long bprev;
792 int osize, nsize;
793{
794 register struct fs *fs;
795 register struct cg *cgp;
796 struct buf *bp;
797 long bno;
798 int frags, bbase;
799 int i, error;
800
801 fs = ip->i_fs;
802 if (fs->fs_cs(fs, cg).cs_nffree < numfrags(fs, nsize - osize))
803 return (0);
804 frags = numfrags(fs, nsize);
805 bbase = fragnum(fs, bprev);
806 if (bbase > fragnum(fs, (bprev + frags - 1))) {
807 /* cannot extend across a block boundary */
808 return (0);
809 }
810 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
811 (int)fs->fs_cgsize, NOCRED, &bp);
812 if (error) {
813 brelse(bp);
814 return (0);
815 }
816 cgp = (struct cg *)bp->b_data;
817 if (!cg_chkmagic(cgp)) {
818 brelse(bp);
819 return (0);
820 }
821 cgp->cg_time = time.tv_sec;
822 bno = dtogd(fs, bprev);
823 for (i = numfrags(fs, osize); i < frags; i++)
824 if (isclr(cg_blksfree(cgp), bno + i)) {
825 brelse(bp);
826 return (0);
827 }
828 /*
829 * the current fragment can be extended
830 * deduct the count on fragment being extended into
831 * increase the count on the remaining fragment (if any)
832 * allocate the extended piece
833 */
834 for (i = frags; i < fs->fs_frag - bbase; i++)
835 if (isclr(cg_blksfree(cgp), bno + i))
836 break;
837 cgp->cg_frsum[i - numfrags(fs, osize)]--;
838 if (i != frags)
839 cgp->cg_frsum[i - frags]++;
840 for (i = numfrags(fs, osize); i < frags; i++) {
841 clrbit(cg_blksfree(cgp), bno + i);
842 cgp->cg_cs.cs_nffree--;
843 fs->fs_cstotal.cs_nffree--;
844 fs->fs_cs(fs, cg).cs_nffree--;
845 }
846 fs->fs_fmod = 1;
847 bdwrite(bp);
848 return (bprev);
849}
850
851/*
852 * Determine whether a block can be allocated.
853 *
854 * Check to see if a block of the appropriate size is available,
855 * and if it is, allocate it.
856 */
857static ufs_daddr_t
858ffs_alloccg(ip, cg, bpref, size)
859 struct inode *ip;
860 int cg;
861 ufs_daddr_t bpref;
862 int size;
863{
864 register struct fs *fs;
865 register struct cg *cgp;
866 struct buf *bp;
867 register int i;
868 int error, bno, frags, allocsiz;
869
870 fs = ip->i_fs;
871 if (fs->fs_cs(fs, cg).cs_nbfree == 0 && size == fs->fs_bsize)
872 return (0);
873 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
874 (int)fs->fs_cgsize, NOCRED, &bp);
875 if (error) {
876 brelse(bp);
877 return (0);
878 }
879 cgp = (struct cg *)bp->b_data;
880 if (!cg_chkmagic(cgp) ||
881 (cgp->cg_cs.cs_nbfree == 0 && size == fs->fs_bsize)) {
882 brelse(bp);
883 return (0);
884 }
885 cgp->cg_time = time.tv_sec;
886 if (size == fs->fs_bsize) {
887 bno = ffs_alloccgblk(fs, cgp, bpref);
888 bdwrite(bp);
889 return (bno);
890 }
891 /*
892 * check to see if any fragments are already available
893 * allocsiz is the size which will be allocated, hacking
894 * it down to a smaller size if necessary
895 */
896 frags = numfrags(fs, size);
897 for (allocsiz = frags; allocsiz < fs->fs_frag; allocsiz++)
898 if (cgp->cg_frsum[allocsiz] != 0)
899 break;
900 if (allocsiz == fs->fs_frag) {
901 /*
902 * no fragments were available, so a block will be
903 * allocated, and hacked up
904 */
905 if (cgp->cg_cs.cs_nbfree == 0) {
906 brelse(bp);
907 return (0);
908 }
909 bno = ffs_alloccgblk(fs, cgp, bpref);
910 bpref = dtogd(fs, bno);
911 for (i = frags; i < fs->fs_frag; i++)
912 setbit(cg_blksfree(cgp), bpref + i);
913 i = fs->fs_frag - frags;
914 cgp->cg_cs.cs_nffree += i;
915 fs->fs_cstotal.cs_nffree += i;
916 fs->fs_cs(fs, cg).cs_nffree += i;
917 fs->fs_fmod = 1;
918 cgp->cg_frsum[i]++;
919 bdwrite(bp);
920 return (bno);
921 }
922 bno = ffs_mapsearch(fs, cgp, bpref, allocsiz);
923 if (bno < 0) {
924 brelse(bp);
925 return (0);
926 }
927 for (i = 0; i < frags; i++)
928 clrbit(cg_blksfree(cgp), bno + i);
929 cgp->cg_cs.cs_nffree -= frags;
930 fs->fs_cstotal.cs_nffree -= frags;
931 fs->fs_cs(fs, cg).cs_nffree -= frags;
932 fs->fs_fmod = 1;
933 cgp->cg_frsum[allocsiz]--;
934 if (frags != allocsiz)
935 cgp->cg_frsum[allocsiz - frags]++;
936 bdwrite(bp);
937 return (cg * fs->fs_fpg + bno);
938}
939
940/*
941 * Allocate a block in a cylinder group.
942 *
943 * This algorithm implements the following policy:
944 * 1) allocate the requested block.
945 * 2) allocate a rotationally optimal block in the same cylinder.
946 * 3) allocate the next available block on the block rotor for the
947 * specified cylinder group.
948 * Note that this routine only allocates fs_bsize blocks; these
949 * blocks may be fragmented by the routine that allocates them.
950 */
951static ufs_daddr_t
952ffs_alloccgblk(fs, cgp, bpref)
953 register struct fs *fs;
954 register struct cg *cgp;
955 ufs_daddr_t bpref;
956{
957 ufs_daddr_t bno, blkno;
958 int cylno, pos, delta;
959 short *cylbp;
960 register int i;
961
962 if (bpref == 0 || dtog(fs, bpref) != cgp->cg_cgx) {
963 bpref = cgp->cg_rotor;
964 goto norot;
965 }
966 bpref = blknum(fs, bpref);
967 bpref = dtogd(fs, bpref);
968 /*
969 * if the requested block is available, use it
970 */
971 if (ffs_isblock(fs, cg_blksfree(cgp), fragstoblks(fs, bpref))) {
972 bno = bpref;
973 goto gotit;
974 }
975 if (fs->fs_nrpos <= 1 || fs->fs_cpc == 0) {
976 /*
977 * Block layout information is not available.
978 * Leaving bpref unchanged means we take the
979 * next available free block following the one
980 * we just allocated. Hopefully this will at
981 * least hit a track cache on drives of unknown
982 * geometry (e.g. SCSI).
983 */
984 goto norot;
985 }
986 /*
987 * check for a block available on the same cylinder
988 */
989 cylno = cbtocylno(fs, bpref);
990 if (cg_blktot(cgp)[cylno] == 0)
991 goto norot;
992 /*
993 * check the summary information to see if a block is
994 * available in the requested cylinder starting at the
995 * requested rotational position and proceeding around.
996 */
997 cylbp = cg_blks(fs, cgp, cylno);
998 pos = cbtorpos(fs, bpref);
999 for (i = pos; i < fs->fs_nrpos; i++)
1000 if (cylbp[i] > 0)
1001 break;
1002 if (i == fs->fs_nrpos)
1003 for (i = 0; i < pos; i++)
1004 if (cylbp[i] > 0)
1005 break;
1006 if (cylbp[i] > 0) {
1007 /*
1008 * found a rotational position, now find the actual
1009 * block. A panic if none is actually there.
1010 */
1011 pos = cylno % fs->fs_cpc;
1012 bno = (cylno - pos) * fs->fs_spc / NSPB(fs);
1013 if (fs_postbl(fs, pos)[i] == -1) {
1014 printf("pos = %d, i = %d, fs = %s\n",
1015 pos, i, fs->fs_fsmnt);
1016 panic("ffs_alloccgblk: cyl groups corrupted");
1017 }
1018 for (i = fs_postbl(fs, pos)[i];; ) {
1019 if (ffs_isblock(fs, cg_blksfree(cgp), bno + i)) {
1020 bno = blkstofrags(fs, (bno + i));
1021 goto gotit;
1022 }
1023 delta = fs_rotbl(fs)[i];
1024 if (delta <= 0 ||
1025 delta + i > fragstoblks(fs, fs->fs_fpg))
1026 break;
1027 i += delta;
1028 }
1029 printf("pos = %d, i = %d, fs = %s\n", pos, i, fs->fs_fsmnt);
1030 panic("ffs_alloccgblk: can't find blk in cyl");
1031 }
1032norot:
1033 /*
1034 * no blocks in the requested cylinder, so take next
1035 * available one in this cylinder group.
1036 */
1037 bno = ffs_mapsearch(fs, cgp, bpref, (int)fs->fs_frag);
1038 if (bno < 0)
1039 return (0);
1040 cgp->cg_rotor = bno;
1041gotit:
1042 blkno = fragstoblks(fs, bno);
1043 ffs_clrblock(fs, cg_blksfree(cgp), (long)blkno);
1044 ffs_clusteracct(fs, cgp, blkno, -1);
1045 cgp->cg_cs.cs_nbfree--;
1046 fs->fs_cstotal.cs_nbfree--;
1047 fs->fs_cs(fs, cgp->cg_cgx).cs_nbfree--;
1048 cylno = cbtocylno(fs, bno);
1049 cg_blks(fs, cgp, cylno)[cbtorpos(fs, bno)]--;
1050 cg_blktot(cgp)[cylno]--;
1051 fs->fs_fmod = 1;
1052 return (cgp->cg_cgx * fs->fs_fpg + bno);
1053}
1054
1055#ifdef notyet
1056/*
1057 * Determine whether a cluster can be allocated.
1058 *
1059 * We do not currently check for optimal rotational layout if there
1060 * are multiple choices in the same cylinder group. Instead we just
1061 * take the first one that we find following bpref.
1062 */
1063static ufs_daddr_t
1064ffs_clusteralloc(ip, cg, bpref, len)
1065 struct inode *ip;
1066 int cg;
1067 ufs_daddr_t bpref;
1068 int len;
1069{
1070 register struct fs *fs;
1071 register struct cg *cgp;
1072 struct buf *bp;
1073 int i, got, run, bno, bit, map;
1074 u_char *mapp;
1075 int32_t *lp;
1076
1077 fs = ip->i_fs;
1078 if (fs->fs_maxcluster[cg] < len)
1079 return (NULL);
1080 if (bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), (int)fs->fs_cgsize,
1081 NOCRED, &bp))
1082 goto fail;
1083 cgp = (struct cg *)bp->b_data;
1084 if (!cg_chkmagic(cgp))
1085 goto fail;
1086 /*
1087 * Check to see if a cluster of the needed size (or bigger) is
1088 * available in this cylinder group.
1089 */
1090 lp = &cg_clustersum(cgp)[len];
1091 for (i = len; i <= fs->fs_contigsumsize; i++)
1092 if (*lp++ > 0)
1093 break;
1094 if (i > fs->fs_contigsumsize) {
1095 /*
1096 * This is the first time looking for a cluster in this
1097 * cylinder group. Update the cluster summary information
1098 * to reflect the true maximum sized cluster so that
1099 * future cluster allocation requests can avoid reading
1100 * the cylinder group map only to find no clusters.
1101 */
1102 lp = &cg_clustersum(cgp)[len - 1];
1103 for (i = len - 1; i > 0; i--)
1104 if (*lp-- > 0)
1105 break;
1106 fs->fs_maxcluster[cg] = i;
1107 goto fail;
1108 }
1109 /*
1110 * Search the cluster map to find a big enough cluster.
1111 * We take the first one that we find, even if it is larger
1112 * than we need as we prefer to get one close to the previous
1113 * block allocation. We do not search before the current
1114 * preference point as we do not want to allocate a block
1115 * that is allocated before the previous one (as we will
1116 * then have to wait for another pass of the elevator
1117 * algorithm before it will be read). We prefer to fail and
1118 * be recalled to try an allocation in the next cylinder group.
1119 */
1120 if (dtog(fs, bpref) != cg)
1121 bpref = 0;
1122 else
1123 bpref = fragstoblks(fs, dtogd(fs, blknum(fs, bpref)));
1124 mapp = &cg_clustersfree(cgp)[bpref / NBBY];
1125 map = *mapp++;
1126 bit = 1 << (bpref % NBBY);
1127 for (run = 0, got = bpref; got < cgp->cg_nclusterblks; got++) {
1128 if ((map & bit) == 0) {
1129 run = 0;
1130 } else {
1131 run++;
1132 if (run == len)
1133 break;
1134 }
1135 if ((got & (NBBY - 1)) != (NBBY - 1)) {
1136 bit <<= 1;
1137 } else {
1138 map = *mapp++;
1139 bit = 1;
1140 }
1141 }
1142 if (got >= cgp->cg_nclusterblks)
1143 goto fail;
1144 /*
1145 * Allocate the cluster that we have found.
1146 */
1147 for (i = 1; i <= len; i++)
1148 if (!ffs_isblock(fs, cg_blksfree(cgp), got - run + i))
1149 panic("ffs_clusteralloc: map mismatch");
1150 bno = cg * fs->fs_fpg + blkstofrags(fs, got - run + 1);
1151 if (dtog(fs, bno) != cg)
1152 panic("ffs_clusteralloc: allocated out of group");
1153 len = blkstofrags(fs, len);
1154 for (i = 0; i < len; i += fs->fs_frag)
1155 if ((got = ffs_alloccgblk(fs, cgp, bno + i)) != bno + i)
1156 panic("ffs_clusteralloc: lost block");
1157 bdwrite(bp);
1158 return (bno);
1159
1160fail:
1161 brelse(bp);
1162 return (0);
1163}
1164#endif
1165
1166/*
1167 * Determine whether an inode can be allocated.
1168 *
1169 * Check to see if an inode is available, and if it is,
1170 * allocate it using the following policy:
1171 * 1) allocate the requested inode.
1172 * 2) allocate the next available inode after the requested
1173 * inode in the specified cylinder group.
1174 */
1175static ino_t
1176ffs_nodealloccg(ip, cg, ipref, mode)
1177 struct inode *ip;
1178 int cg;
1179 ufs_daddr_t ipref;
1180 int mode;
1181{
1182 register struct fs *fs;
1183 register struct cg *cgp;
1184 struct buf *bp;
1185 int error, start, len, loc, map, i;
1186
1187 fs = ip->i_fs;
1188 if (fs->fs_cs(fs, cg).cs_nifree == 0)
1189 return (0);
1190 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
1191 (int)fs->fs_cgsize, NOCRED, &bp);
1192 if (error) {
1193 brelse(bp);
1194 return (0);
1195 }
1196 cgp = (struct cg *)bp->b_data;
1197 if (!cg_chkmagic(cgp) || cgp->cg_cs.cs_nifree == 0) {
1198 brelse(bp);
1199 return (0);
1200 }
1201 cgp->cg_time = time.tv_sec;
1202 if (ipref) {
1203 ipref %= fs->fs_ipg;
1204 if (isclr(cg_inosused(cgp), ipref))
1205 goto gotit;
1206 }
1207 start = cgp->cg_irotor / NBBY;
1208 len = howmany(fs->fs_ipg - cgp->cg_irotor, NBBY);
1209 loc = skpc(0xff, len, &cg_inosused(cgp)[start]);
1210 if (loc == 0) {
1211 len = start + 1;
1212 start = 0;
1213 loc = skpc(0xff, len, &cg_inosused(cgp)[0]);
1214 if (loc == 0) {
1215 printf("cg = %d, irotor = %ld, fs = %s\n",
1216 cg, cgp->cg_irotor, fs->fs_fsmnt);
1217 panic("ffs_nodealloccg: map corrupted");
1218 /* NOTREACHED */
1219 }
1220 }
1221 i = start + len - loc;
1222 map = cg_inosused(cgp)[i];
1223 ipref = i * NBBY;
1224 for (i = 1; i < (1 << NBBY); i <<= 1, ipref++) {
1225 if ((map & i) == 0) {
1226 cgp->cg_irotor = ipref;
1227 goto gotit;
1228 }
1229 }
1230 printf("fs = %s\n", fs->fs_fsmnt);
1231 panic("ffs_nodealloccg: block not in map");
1232 /* NOTREACHED */
1233gotit:
1234 setbit(cg_inosused(cgp), ipref);
1235 cgp->cg_cs.cs_nifree--;
1236 fs->fs_cstotal.cs_nifree--;
1237 fs->fs_cs(fs, cg).cs_nifree--;
1238 fs->fs_fmod = 1;
1239 if ((mode & IFMT) == IFDIR) {
1240 cgp->cg_cs.cs_ndir++;
1241 fs->fs_cstotal.cs_ndir++;
1242 fs->fs_cs(fs, cg).cs_ndir++;
1243 }
1244 bdwrite(bp);
1245 return (cg * fs->fs_ipg + ipref);
1246}
1247
1248/*
1249 * Free a block or fragment.
1250 *
1251 * The specified block or fragment is placed back in the
1252 * free map. If a fragment is deallocated, a possible
1253 * block reassembly is checked.
1254 */
1255void
1256ffs_blkfree(ip, bno, size)
1257 register struct inode *ip;
1258 ufs_daddr_t bno;
1259 long size;
1260{
1261 register struct fs *fs;
1262 register struct cg *cgp;
1263 struct buf *bp;
1264 ufs_daddr_t blkno;
1265 int i, error, cg, blk, frags, bbase;
1266
1267 fs = ip->i_fs;
1268 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) {
1269 printf("dev = 0x%lx, bsize = %ld, size = %ld, fs = %s\n",
1270 (u_long)ip->i_dev, fs->fs_bsize, size, fs->fs_fsmnt);
1271 panic("ffs_blkfree: bad size");
1272 }
1273 cg = dtog(fs, bno);
1274 if ((u_int)bno >= fs->fs_size) {
1275 printf("bad block %ld, ino %ld\n", bno, ip->i_number);
1276 ffs_fserr(fs, ip->i_uid, "bad block");
1277 return;
1278 }
1279 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
1280 (int)fs->fs_cgsize, NOCRED, &bp);
1281 if (error) {
1282 brelse(bp);
1283 return;
1284 }
1285 cgp = (struct cg *)bp->b_data;
1286 if (!cg_chkmagic(cgp)) {
1287 brelse(bp);
1288 return;
1289 }
1290 cgp->cg_time = time.tv_sec;
1291 bno = dtogd(fs, bno);
1292 if (size == fs->fs_bsize) {
1293 blkno = fragstoblks(fs, bno);
1294 if (ffs_isblock(fs, cg_blksfree(cgp), blkno)) {
1295 printf("dev = 0x%lx, block = %ld, fs = %s\n",
1296 (u_long) ip->i_dev, bno, fs->fs_fsmnt);
1297 panic("ffs_blkfree: freeing free block");
1298 }
1299 ffs_setblock(fs, cg_blksfree(cgp), blkno);
1300 ffs_clusteracct(fs, cgp, blkno, 1);
1301 cgp->cg_cs.cs_nbfree++;
1302 fs->fs_cstotal.cs_nbfree++;
1303 fs->fs_cs(fs, cg).cs_nbfree++;
1304 i = cbtocylno(fs, bno);
1305 cg_blks(fs, cgp, i)[cbtorpos(fs, bno)]++;
1306 cg_blktot(cgp)[i]++;
1307 } else {
1308 bbase = bno - fragnum(fs, bno);
1309 /*
1310 * decrement the counts associated with the old frags
1311 */
1312 blk = blkmap(fs, cg_blksfree(cgp), bbase);
1313 ffs_fragacct(fs, blk, cgp->cg_frsum, -1);
1314 /*
1315 * deallocate the fragment
1316 */
1317 frags = numfrags(fs, size);
1318 for (i = 0; i < frags; i++) {
1319 if (isset(cg_blksfree(cgp), bno + i)) {
1320 printf("dev = 0x%lx, block = %ld, fs = %s\n",
1321 (u_long) ip->i_dev, bno + i, fs->fs_fsmnt);
1322 panic("ffs_blkfree: freeing free frag");
1323 }
1324 setbit(cg_blksfree(cgp), bno + i);
1325 }
1326 cgp->cg_cs.cs_nffree += i;
1327 fs->fs_cstotal.cs_nffree += i;
1328 fs->fs_cs(fs, cg).cs_nffree += i;
1329 /*
1330 * add back in counts associated with the new frags
1331 */
1332 blk = blkmap(fs, cg_blksfree(cgp), bbase);
1333 ffs_fragacct(fs, blk, cgp->cg_frsum, 1);
1334 /*
1335 * if a complete block has been reassembled, account for it
1336 */
1337 blkno = fragstoblks(fs, bbase);
1338 if (ffs_isblock(fs, cg_blksfree(cgp), blkno)) {
1339 cgp->cg_cs.cs_nffree -= fs->fs_frag;
1340 fs->fs_cstotal.cs_nffree -= fs->fs_frag;
1341 fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag;
1342 ffs_clusteracct(fs, cgp, blkno, 1);
1343 cgp->cg_cs.cs_nbfree++;
1344 fs->fs_cstotal.cs_nbfree++;
1345 fs->fs_cs(fs, cg).cs_nbfree++;
1346 i = cbtocylno(fs, bbase);
1347 cg_blks(fs, cgp, i)[cbtorpos(fs, bbase)]++;
1348 cg_blktot(cgp)[i]++;
1349 }
1350 }
1351 fs->fs_fmod = 1;
1352 bdwrite(bp);
1353}
1354
1355#ifdef DIAGNOSTIC
1356/*
1357 * Verify allocation of a block or fragment. Returns true if block or
1358 * fragment is allocated, false if it is free.
1359 */
1360int
1361ffs_checkblk(ip, bno, size)
1362 struct inode *ip;
1363 ufs_daddr_t bno;
1364 long size;
1365{
1366 struct fs *fs;
1367 struct cg *cgp;
1368 struct buf *bp;
1369 int i, error, frags, free;
1370
1371 fs = ip->i_fs;
1372 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) {
1373 printf("bsize = %d, size = %d, fs = %s\n",
1374 fs->fs_bsize, size, fs->fs_fsmnt);
1375 panic("ffs_checkblk: bad size");
1376 }
1377 if ((u_int)bno >= fs->fs_size)
1378 panic("ffs_checkblk: bad block %d", bno);
1379 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, dtog(fs, bno))),
1380 (int)fs->fs_cgsize, NOCRED, &bp);
1381 if (error)
1382 panic("ffs_checkblk: cg bread failed");
1383 cgp = (struct cg *)bp->b_data;
1384 if (!cg_chkmagic(cgp))
1385 panic("ffs_checkblk: cg magic mismatch");
1386 bno = dtogd(fs, bno);
1387 if (size == fs->fs_bsize) {
1388 free = ffs_isblock(fs, cg_blksfree(cgp), fragstoblks(fs, bno));
1389 } else {
1390 frags = numfrags(fs, size);
1391 for (free = 0, i = 0; i < frags; i++)
1392 if (isset(cg_blksfree(cgp), bno + i))
1393 free++;
1394 if (free != 0 && free != frags)
1395 panic("ffs_checkblk: partially free fragment");
1396 }
1397 brelse(bp);
1398 return (!free);
1399}
1400#endif /* DIAGNOSTIC */
1401
1402/*
1403 * Free an inode.
1404 *
1405 * The specified inode is placed back in the free map.
1406 */
1407int
1408ffs_vfree(ap)
1409 struct vop_vfree_args /* {
1410 struct vnode *a_pvp;
1411 ino_t a_ino;
1412 int a_mode;
1413 } */ *ap;
1414{
1415 register struct fs *fs;
1416 register struct cg *cgp;
1417 register struct inode *pip;
1418 ino_t ino = ap->a_ino;
1419 struct buf *bp;
1420 int error, cg;
1421
1422 pip = VTOI(ap->a_pvp);
51
52#include <ufs/ffs/fs.h>
53#include <ufs/ffs/ffs_extern.h>
54
55typedef ufs_daddr_t allocfcn_t __P((struct inode *ip, int cg, ufs_daddr_t bpref,
56 int size));
57
58static ufs_daddr_t ffs_alloccg __P((struct inode *, int, ufs_daddr_t, int));
59static ufs_daddr_t ffs_alloccgblk __P((struct fs *, struct cg *, ufs_daddr_t));
60static void ffs_clusteracct __P((struct fs *, struct cg *, ufs_daddr_t,
61 int));
62static ufs_daddr_t ffs_clusteralloc __P((struct inode *, int, ufs_daddr_t,
63 int));
64static ino_t ffs_dirpref __P((struct fs *));
65static ufs_daddr_t ffs_fragextend __P((struct inode *, int, long, int, int));
66static void ffs_fserr __P((struct fs *, u_int, char *));
67static u_long ffs_hashalloc
68 __P((struct inode *, int, long, int, allocfcn_t *));
69static ino_t ffs_nodealloccg __P((struct inode *, int, ufs_daddr_t, int));
70static ufs_daddr_t ffs_mapsearch __P((struct fs *, struct cg *, ufs_daddr_t,
71 int));
72
73/*
74 * Allocate a block in the file system.
75 *
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 heirarchy is used
86 * to allocate a block:
87 * 1) allocate a block in the cylinder group that contains the
88 * inode for the file.
89 * 2) quadradically rehash into other cylinder groups, until an
90 * available block is located.
91 */
92int
93ffs_alloc(ip, lbn, bpref, size, cred, bnp)
94 register struct inode *ip;
95 ufs_daddr_t lbn, bpref;
96 int size;
97 struct ucred *cred;
98 ufs_daddr_t *bnp;
99{
100 register struct fs *fs;
101 ufs_daddr_t bno;
102 int cg;
103#ifdef QUOTA
104 int error;
105#endif
106
107 *bnp = 0;
108 fs = ip->i_fs;
109#ifdef DIAGNOSTIC
110 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) {
111 printf("dev = 0x%lx, bsize = %ld, size = %d, fs = %s\n",
112 (u_long)ip->i_dev, fs->fs_bsize, size, fs->fs_fsmnt);
113 panic("ffs_alloc: bad size");
114 }
115 if (cred == NOCRED)
116 panic("ffs_alloc: missing credential");
117#endif /* DIAGNOSTIC */
118 if (size == fs->fs_bsize && fs->fs_cstotal.cs_nbfree == 0)
119 goto nospace;
120 if (cred->cr_uid != 0 &&
121 freespace(fs, fs->fs_minfree) - numfrags(fs, size) < 0)
122 goto nospace;
123#ifdef QUOTA
124 error = chkdq(ip, (long)btodb(size), cred, 0);
125 if (error)
126 return (error);
127#endif
128 if (bpref >= fs->fs_size)
129 bpref = 0;
130 if (bpref == 0)
131 cg = ino_to_cg(fs, ip->i_number);
132 else
133 cg = dtog(fs, bpref);
134 bno = (ufs_daddr_t)ffs_hashalloc(ip, cg, (long)bpref, size,
135 ffs_alloccg);
136 if (bno > 0) {
137 ip->i_blocks += btodb(size);
138 ip->i_flag |= IN_CHANGE | IN_UPDATE;
139 *bnp = bno;
140 return (0);
141 }
142#ifdef QUOTA
143 /*
144 * Restore user's disk quota because allocation failed.
145 */
146 (void) chkdq(ip, (long)-btodb(size), cred, FORCE);
147#endif
148nospace:
149 ffs_fserr(fs, cred->cr_uid, "file system full");
150 uprintf("\n%s: write failed, file system is full\n", fs->fs_fsmnt);
151 return (ENOSPC);
152}
153
154/*
155 * Reallocate a fragment to a bigger size
156 *
157 * The number and size of the old block is given, and a preference
158 * and new size is also specified. The allocator attempts to extend
159 * the original block. Failing that, the regular block allocator is
160 * invoked to get an appropriate block.
161 */
162int
163ffs_realloccg(ip, lbprev, bpref, osize, nsize, cred, bpp)
164 register struct inode *ip;
165 ufs_daddr_t lbprev;
166 ufs_daddr_t bpref;
167 int osize, nsize;
168 struct ucred *cred;
169 struct buf **bpp;
170{
171 register struct fs *fs;
172 struct buf *bp;
173 int cg, request, error;
174 ufs_daddr_t bprev, bno;
175
176 *bpp = 0;
177 fs = ip->i_fs;
178#ifdef DIAGNOSTIC
179 if ((u_int)osize > fs->fs_bsize || fragoff(fs, osize) != 0 ||
180 (u_int)nsize > fs->fs_bsize || fragoff(fs, nsize) != 0) {
181 printf(
182 "dev = 0x%lx, bsize = %ld, osize = %d, "
183 "nsize = %d, fs = %s\n",
184 (u_long)ip->i_dev, fs->fs_bsize, osize,
185 nsize, fs->fs_fsmnt);
186 panic("ffs_realloccg: bad size");
187 }
188 if (cred == NOCRED)
189 panic("ffs_realloccg: missing credential");
190#endif /* DIAGNOSTIC */
191 if (cred->cr_uid != 0 &&
192 freespace(fs, fs->fs_minfree) - numfrags(fs, nsize - osize) < 0)
193 goto nospace;
194 if ((bprev = ip->i_db[lbprev]) == 0) {
195 printf("dev = 0x%lx, bsize = %ld, bprev = %ld, fs = %s\n",
196 (u_long) ip->i_dev, fs->fs_bsize, bprev, fs->fs_fsmnt);
197 panic("ffs_realloccg: bad bprev");
198 }
199 /*
200 * Allocate the extra space in the buffer.
201 */
202 error = bread(ITOV(ip), lbprev, osize, NOCRED, &bp);
203 if (error) {
204 brelse(bp);
205 return (error);
206 }
207
208 if( bp->b_blkno == bp->b_lblkno) {
209 if( lbprev >= NDADDR)
210 panic("ffs_realloccg: lbprev out of range");
211 bp->b_blkno = fsbtodb(fs, bprev);
212 }
213
214#ifdef QUOTA
215 error = chkdq(ip, (long)btodb(nsize - osize), cred, 0);
216 if (error) {
217 brelse(bp);
218 return (error);
219 }
220#endif
221 /*
222 * Check for extension in the existing location.
223 */
224 cg = dtog(fs, bprev);
225 bno = ffs_fragextend(ip, cg, (long)bprev, osize, nsize);
226 if (bno) {
227 if (bp->b_blkno != fsbtodb(fs, bno))
228 panic("ffs_realloccg: bad blockno");
229 ip->i_blocks += btodb(nsize - osize);
230 ip->i_flag |= IN_CHANGE | IN_UPDATE;
231 allocbuf(bp, nsize);
232 bp->b_flags |= B_DONE;
233 bzero((char *)bp->b_data + osize, (u_int)nsize - osize);
234 *bpp = bp;
235 return (0);
236 }
237 /*
238 * Allocate a new disk location.
239 */
240 if (bpref >= fs->fs_size)
241 bpref = 0;
242 switch ((int)fs->fs_optim) {
243 case FS_OPTSPACE:
244 /*
245 * Allocate an exact sized fragment. Although this makes
246 * best use of space, we will waste time relocating it if
247 * the file continues to grow. If the fragmentation is
248 * less than half of the minimum free reserve, we choose
249 * to begin optimizing for time.
250 */
251 request = nsize;
252 if (fs->fs_minfree <= 5 ||
253 fs->fs_cstotal.cs_nffree >
254 fs->fs_dsize * fs->fs_minfree / (2 * 100))
255 break;
256 log(LOG_NOTICE, "%s: optimization changed from SPACE to TIME\n",
257 fs->fs_fsmnt);
258 fs->fs_optim = FS_OPTTIME;
259 break;
260 case FS_OPTTIME:
261 /*
262 * At this point we have discovered a file that is trying to
263 * grow a small fragment to a larger fragment. To save time,
264 * we allocate a full sized block, then free the unused portion.
265 * If the file continues to grow, the `ffs_fragextend' call
266 * above will be able to grow it in place without further
267 * copying. If aberrant programs cause disk fragmentation to
268 * grow within 2% of the free reserve, we choose to begin
269 * optimizing for space.
270 */
271 request = fs->fs_bsize;
272 if (fs->fs_cstotal.cs_nffree <
273 fs->fs_dsize * (fs->fs_minfree - 2) / 100)
274 break;
275 log(LOG_NOTICE, "%s: optimization changed from TIME to SPACE\n",
276 fs->fs_fsmnt);
277 fs->fs_optim = FS_OPTSPACE;
278 break;
279 default:
280 printf("dev = 0x%lx, optim = %ld, fs = %s\n",
281 (u_long)ip->i_dev, fs->fs_optim, fs->fs_fsmnt);
282 panic("ffs_realloccg: bad optim");
283 /* NOTREACHED */
284 }
285 bno = (ufs_daddr_t)ffs_hashalloc(ip, cg, (long)bpref, request,
286 ffs_alloccg);
287 if (bno > 0) {
288 bp->b_blkno = fsbtodb(fs, bno);
289 ffs_blkfree(ip, bprev, (long)osize);
290 if (nsize < request)
291 ffs_blkfree(ip, bno + numfrags(fs, nsize),
292 (long)(request - nsize));
293 ip->i_blocks += btodb(nsize - osize);
294 ip->i_flag |= IN_CHANGE | IN_UPDATE;
295 allocbuf(bp, nsize);
296 bp->b_flags |= B_DONE;
297 bzero((char *)bp->b_data + osize, (u_int)nsize - osize);
298 *bpp = bp;
299 return (0);
300 }
301#ifdef QUOTA
302 /*
303 * Restore user's disk quota because allocation failed.
304 */
305 (void) chkdq(ip, (long)-btodb(nsize - osize), cred, FORCE);
306#endif
307 brelse(bp);
308nospace:
309 /*
310 * no space available
311 */
312 ffs_fserr(fs, cred->cr_uid, "file system full");
313 uprintf("\n%s: write failed, file system is full\n", fs->fs_fsmnt);
314 return (ENOSPC);
315}
316
317/*
318 * Reallocate a sequence of blocks into a contiguous sequence of blocks.
319 *
320 * The vnode and an array of buffer pointers for a range of sequential
321 * logical blocks to be made contiguous is given. The allocator attempts
322 * to find a range of sequential blocks starting as close as possible to
323 * an fs_rotdelay offset from the end of the allocation for the logical
324 * block immediately preceeding the current range. If successful, the
325 * physical block numbers in the buffer pointers and in the inode are
326 * changed to reflect the new allocation. If unsuccessful, the allocation
327 * is left unchanged. The success in doing the reallocation is returned.
328 * Note that the error return is not reflected back to the user. Rather
329 * the previous block allocation will be used.
330 */
331static int doasyncfree = 1;
332SYSCTL_INT(_vfs_ffs, FFS_ASYNCFREE, doasyncfree, CTLFLAG_RW, &doasyncfree, 0, "");
333
334int doreallocblks = 1;
335SYSCTL_INT(_vfs_ffs, FFS_REALLOCBLKS, doreallocblks, CTLFLAG_RW, &doreallocblks, 0, "");
336
337static int prtrealloc = 0;
338
339int
340ffs_reallocblks(ap)
341 struct vop_reallocblks_args /* {
342 struct vnode *a_vp;
343 struct cluster_save *a_buflist;
344 } */ *ap;
345{
346#if !defined (not_yes)
347 return (ENOSPC);
348#else
349 struct fs *fs;
350 struct inode *ip;
351 struct vnode *vp;
352 struct buf *sbp, *ebp;
353 ufs_daddr_t *bap, *sbap, *ebap = 0;
354 struct cluster_save *buflist;
355 ufs_daddr_t start_lbn, end_lbn, soff, newblk, blkno;
356 struct indir start_ap[NIADDR + 1], end_ap[NIADDR + 1], *idp;
357 int i, len, start_lvl, end_lvl, pref, ssize;
358 struct timeval tv;
359
360 if (doreallocblks == 0)
361 return (ENOSPC);
362 vp = ap->a_vp;
363 ip = VTOI(vp);
364 fs = ip->i_fs;
365 if (fs->fs_contigsumsize <= 0)
366 return (ENOSPC);
367 buflist = ap->a_buflist;
368 len = buflist->bs_nchildren;
369 start_lbn = buflist->bs_children[0]->b_lblkno;
370 end_lbn = start_lbn + len - 1;
371#ifdef DIAGNOSTIC
372 for (i = 0; i < len; i++)
373 if (!ffs_checkblk(ip,
374 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
375 panic("ffs_reallocblks: unallocated block 1");
376 for (i = 1; i < len; i++)
377 if (buflist->bs_children[i]->b_lblkno != start_lbn + i)
378 panic("ffs_reallocblks: non-logical cluster");
379 blkno = buflist->bs_children[0]->b_blkno;
380 ssize = fsbtodb(fs, fs->fs_frag);
381 for (i = 1; i < len - 1; i++)
382 if (buflist->bs_children[i]->b_blkno != blkno + (i * ssize))
383 panic("ffs_reallocblks: non-physical cluster %d", i);
384#endif
385 /*
386 * If the latest allocation is in a new cylinder group, assume that
387 * the filesystem has decided to move and do not force it back to
388 * the previous cylinder group.
389 */
390 if (dtog(fs, dbtofsb(fs, buflist->bs_children[0]->b_blkno)) !=
391 dtog(fs, dbtofsb(fs, buflist->bs_children[len - 1]->b_blkno)))
392 return (ENOSPC);
393 if (ufs_getlbns(vp, start_lbn, start_ap, &start_lvl) ||
394 ufs_getlbns(vp, end_lbn, end_ap, &end_lvl))
395 return (ENOSPC);
396 /*
397 * Get the starting offset and block map for the first block.
398 */
399 if (start_lvl == 0) {
400 sbap = &ip->i_db[0];
401 soff = start_lbn;
402 } else {
403 idp = &start_ap[start_lvl - 1];
404 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &sbp)) {
405 brelse(sbp);
406 return (ENOSPC);
407 }
408 sbap = (ufs_daddr_t *)sbp->b_data;
409 soff = idp->in_off;
410 }
411 /*
412 * Find the preferred location for the cluster.
413 */
414 pref = ffs_blkpref(ip, start_lbn, soff, sbap);
415 /*
416 * If the block range spans two block maps, get the second map.
417 */
418 if (end_lvl == 0 || (idp = &end_ap[end_lvl - 1])->in_off + 1 >= len) {
419 ssize = len;
420 } else {
421#ifdef DIAGNOSTIC
422 if (start_ap[start_lvl-1].in_lbn == idp->in_lbn)
423 panic("ffs_reallocblk: start == end");
424#endif
425 ssize = len - (idp->in_off + 1);
426 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &ebp))
427 goto fail;
428 ebap = (ufs_daddr_t *)ebp->b_data;
429 }
430 /*
431 * Search the block map looking for an allocation of the desired size.
432 */
433 if ((newblk = (ufs_daddr_t)ffs_hashalloc(ip, dtog(fs, pref), (long)pref,
434 len, ffs_clusteralloc)) == 0)
435 goto fail;
436 /*
437 * We have found a new contiguous block.
438 *
439 * First we have to replace the old block pointers with the new
440 * block pointers in the inode and indirect blocks associated
441 * with the file.
442 */
443#ifdef DEBUG
444 if (prtrealloc)
445 printf("realloc: ino %d, lbns %d-%d\n\told:", ip->i_number,
446 start_lbn, end_lbn);
447#endif
448 blkno = newblk;
449 for (bap = &sbap[soff], i = 0; i < len; i++, blkno += fs->fs_frag) {
450 if (i == ssize)
451 bap = ebap;
452#ifdef DIAGNOSTIC
453 if (!ffs_checkblk(ip,
454 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
455 panic("ffs_reallocblks: unallocated block 2");
456 if (dbtofsb(fs, buflist->bs_children[i]->b_blkno) != *bap)
457 panic("ffs_reallocblks: alloc mismatch");
458#endif
459#ifdef DEBUG
460 if (prtrealloc)
461 printf(" %d,", *bap);
462#endif
463 *bap++ = blkno;
464 }
465 /*
466 * Next we must write out the modified inode and indirect blocks.
467 * For strict correctness, the writes should be synchronous since
468 * the old block values may have been written to disk. In practise
469 * they are almost never written, but if we are concerned about
470 * strict correctness, the `doasyncfree' flag should be set to zero.
471 *
472 * The test on `doasyncfree' should be changed to test a flag
473 * that shows whether the associated buffers and inodes have
474 * been written. The flag should be set when the cluster is
475 * started and cleared whenever the buffer or inode is flushed.
476 * We can then check below to see if it is set, and do the
477 * synchronous write only when it has been cleared.
478 */
479 if (sbap != &ip->i_db[0]) {
480 if (doasyncfree)
481 bdwrite(sbp);
482 else
483 bwrite(sbp);
484 } else {
485 ip->i_flag |= IN_CHANGE | IN_UPDATE;
486 if (!doasyncfree) {
487 gettime(&tv);
488 VOP_UPDATE(vp, &tv, &tv, 1);
489 }
490 }
491 if (ssize < len)
492 if (doasyncfree)
493 bdwrite(ebp);
494 else
495 bwrite(ebp);
496 /*
497 * Last, free the old blocks and assign the new blocks to the buffers.
498 */
499#ifdef DEBUG
500 if (prtrealloc)
501 printf("\n\tnew:");
502#endif
503 for (blkno = newblk, i = 0; i < len; i++, blkno += fs->fs_frag) {
504 ffs_blkfree(ip, dbtofsb(fs, buflist->bs_children[i]->b_blkno),
505 fs->fs_bsize);
506 buflist->bs_children[i]->b_blkno = fsbtodb(fs, blkno);
507#ifdef DEBUG
508 if (!ffs_checkblk(ip,
509 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
510 panic("ffs_reallocblks: unallocated block 3");
511 if (prtrealloc)
512 printf(" %d,", blkno);
513#endif
514 }
515#ifdef DEBUG
516 if (prtrealloc) {
517 prtrealloc--;
518 printf("\n");
519 }
520#endif
521 return (0);
522
523fail:
524 if (ssize < len)
525 brelse(ebp);
526 if (sbap != &ip->i_db[0])
527 brelse(sbp);
528 return (ENOSPC);
529#endif
530}
531
532/*
533 * Allocate an inode in the file system.
534 *
535 * If allocating a directory, use ffs_dirpref to select the inode.
536 * If allocating in a directory, the following hierarchy is followed:
537 * 1) allocate the preferred inode.
538 * 2) allocate an inode in the same cylinder group.
539 * 3) quadradically rehash into other cylinder groups, until an
540 * available inode is located.
541 * If no inode preference is given the following heirarchy is used
542 * to allocate an inode:
543 * 1) allocate an inode in cylinder group 0.
544 * 2) quadradically rehash into other cylinder groups, until an
545 * available inode is located.
546 */
547int
548ffs_valloc(ap)
549 struct vop_valloc_args /* {
550 struct vnode *a_pvp;
551 int a_mode;
552 struct ucred *a_cred;
553 struct vnode **a_vpp;
554 } */ *ap;
555{
556 register struct vnode *pvp = ap->a_pvp;
557 register struct inode *pip;
558 register struct fs *fs;
559 register struct inode *ip;
560 mode_t mode = ap->a_mode;
561 ino_t ino, ipref;
562 int cg, error;
563
564 *ap->a_vpp = NULL;
565 pip = VTOI(pvp);
566 fs = pip->i_fs;
567 if (fs->fs_cstotal.cs_nifree == 0)
568 goto noinodes;
569
570 if ((mode & IFMT) == IFDIR)
571 ipref = ffs_dirpref(fs);
572 else
573 ipref = pip->i_number;
574 if (ipref >= fs->fs_ncg * fs->fs_ipg)
575 ipref = 0;
576 cg = ino_to_cg(fs, ipref);
577 ino = (ino_t)ffs_hashalloc(pip, cg, (long)ipref, mode,
578 (allocfcn_t *)ffs_nodealloccg);
579 if (ino == 0)
580 goto noinodes;
581 error = VFS_VGET(pvp->v_mount, ino, ap->a_vpp);
582 if (error) {
583 VOP_VFREE(pvp, ino, mode);
584 return (error);
585 }
586 ip = VTOI(*ap->a_vpp);
587 if (ip->i_mode) {
588 printf("mode = 0%o, inum = %ld, fs = %s\n",
589 ip->i_mode, ip->i_number, fs->fs_fsmnt);
590 panic("ffs_valloc: dup alloc");
591 }
592 if (ip->i_blocks) { /* XXX */
593 printf("free inode %s/%ld had %ld blocks\n",
594 fs->fs_fsmnt, ino, ip->i_blocks);
595 ip->i_blocks = 0;
596 }
597 ip->i_flags = 0;
598 /*
599 * Set up a new generation number for this inode.
600 */
601 if (ip->i_gen == 0 || ++(ip->i_gen) == 0)
602 ip->i_gen = random() / 2 + 1;
603 return (0);
604noinodes:
605 ffs_fserr(fs, ap->a_cred->cr_uid, "out of inodes");
606 uprintf("\n%s: create/symlink failed, no inodes free\n", fs->fs_fsmnt);
607 return (ENOSPC);
608}
609
610/*
611 * Find a cylinder to place a directory.
612 *
613 * The policy implemented by this algorithm is to select from
614 * among those cylinder groups with above the average number of
615 * free inodes, the one with the smallest number of directories.
616 */
617static ino_t
618ffs_dirpref(fs)
619 register struct fs *fs;
620{
621 int cg, minndir, mincg, avgifree;
622
623 avgifree = fs->fs_cstotal.cs_nifree / fs->fs_ncg;
624 minndir = fs->fs_ipg;
625 mincg = 0;
626 for (cg = 0; cg < fs->fs_ncg; cg++)
627 if (fs->fs_cs(fs, cg).cs_ndir < minndir &&
628 fs->fs_cs(fs, cg).cs_nifree >= avgifree) {
629 mincg = cg;
630 minndir = fs->fs_cs(fs, cg).cs_ndir;
631 }
632 return ((ino_t)(fs->fs_ipg * mincg));
633}
634
635/*
636 * Select the desired position for the next block in a file. The file is
637 * logically divided into sections. The first section is composed of the
638 * direct blocks. Each additional section contains fs_maxbpg blocks.
639 *
640 * If no blocks have been allocated in the first section, the policy is to
641 * request a block in the same cylinder group as the inode that describes
642 * the file. If no blocks have been allocated in any other section, the
643 * policy is to place the section in a cylinder group with a greater than
644 * average number of free blocks. An appropriate cylinder group is found
645 * by using a rotor that sweeps the cylinder groups. When a new group of
646 * blocks is needed, the sweep begins in the cylinder group following the
647 * cylinder group from which the previous allocation was made. The sweep
648 * continues until a cylinder group with greater than the average number
649 * of free blocks is found. If the allocation is for the first block in an
650 * indirect block, the information on the previous allocation is unavailable;
651 * here a best guess is made based upon the logical block number being
652 * allocated.
653 *
654 * If a section is already partially allocated, the policy is to
655 * contiguously allocate fs_maxcontig blocks. The end of one of these
656 * contiguous blocks and the beginning of the next is physically separated
657 * so that the disk head will be in transit between them for at least
658 * fs_rotdelay milliseconds. This is to allow time for the processor to
659 * schedule another I/O transfer.
660 */
661ufs_daddr_t
662ffs_blkpref(ip, lbn, indx, bap)
663 struct inode *ip;
664 ufs_daddr_t lbn;
665 int indx;
666 ufs_daddr_t *bap;
667{
668 register struct fs *fs;
669 register int cg;
670 int avgbfree, startcg;
671 ufs_daddr_t nextblk;
672
673 fs = ip->i_fs;
674 if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) {
675 if (lbn < NDADDR) {
676 cg = ino_to_cg(fs, ip->i_number);
677 return (fs->fs_fpg * cg + fs->fs_frag);
678 }
679 /*
680 * Find a cylinder with greater than average number of
681 * unused data blocks.
682 */
683 if (indx == 0 || bap[indx - 1] == 0)
684 startcg =
685 ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg;
686 else
687 startcg = dtog(fs, bap[indx - 1]) + 1;
688 startcg %= fs->fs_ncg;
689 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
690 for (cg = startcg; cg < fs->fs_ncg; cg++)
691 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
692 fs->fs_cgrotor = cg;
693 return (fs->fs_fpg * cg + fs->fs_frag);
694 }
695 for (cg = 0; cg <= startcg; cg++)
696 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
697 fs->fs_cgrotor = cg;
698 return (fs->fs_fpg * cg + fs->fs_frag);
699 }
700 return (0);
701 }
702 /*
703 * One or more previous blocks have been laid out. If less
704 * than fs_maxcontig previous blocks are contiguous, the
705 * next block is requested contiguously, otherwise it is
706 * requested rotationally delayed by fs_rotdelay milliseconds.
707 */
708 nextblk = bap[indx - 1] + fs->fs_frag;
709 if (fs->fs_rotdelay == 0 || indx < fs->fs_maxcontig ||
710 bap[indx - fs->fs_maxcontig] +
711 blkstofrags(fs, fs->fs_maxcontig) != nextblk)
712 return (nextblk);
713 /*
714 * Here we convert ms of delay to frags as:
715 * (frags) = (ms) * (rev/sec) * (sect/rev) /
716 * ((sect/frag) * (ms/sec))
717 * then round up to the next block.
718 */
719 nextblk += roundup(fs->fs_rotdelay * fs->fs_rps * fs->fs_nsect /
720 (NSPF(fs) * 1000), fs->fs_frag);
721 return (nextblk);
722}
723
724/*
725 * Implement the cylinder overflow algorithm.
726 *
727 * The policy implemented by this algorithm is:
728 * 1) allocate the block in its requested cylinder group.
729 * 2) quadradically rehash on the cylinder group number.
730 * 3) brute force search for a free block.
731 */
732/*VARARGS5*/
733static u_long
734ffs_hashalloc(ip, cg, pref, size, allocator)
735 struct inode *ip;
736 int cg;
737 long pref;
738 int size; /* size for data blocks, mode for inodes */
739 allocfcn_t *allocator;
740{
741 register struct fs *fs;
742 long result; /* XXX why not same type as we return? */
743 int i, icg = cg;
744
745 fs = ip->i_fs;
746 /*
747 * 1: preferred cylinder group
748 */
749 result = (*allocator)(ip, cg, pref, size);
750 if (result)
751 return (result);
752 /*
753 * 2: quadratic rehash
754 */
755 for (i = 1; i < fs->fs_ncg; i *= 2) {
756 cg += i;
757 if (cg >= fs->fs_ncg)
758 cg -= fs->fs_ncg;
759 result = (*allocator)(ip, cg, 0, size);
760 if (result)
761 return (result);
762 }
763 /*
764 * 3: brute force search
765 * Note that we start at i == 2, since 0 was checked initially,
766 * and 1 is always checked in the quadratic rehash.
767 */
768 cg = (icg + 2) % fs->fs_ncg;
769 for (i = 2; i < fs->fs_ncg; i++) {
770 result = (*allocator)(ip, cg, 0, size);
771 if (result)
772 return (result);
773 cg++;
774 if (cg == fs->fs_ncg)
775 cg = 0;
776 }
777 return (0);
778}
779
780/*
781 * Determine whether a fragment can be extended.
782 *
783 * Check to see if the necessary fragments are available, and
784 * if they are, allocate them.
785 */
786static ufs_daddr_t
787ffs_fragextend(ip, cg, bprev, osize, nsize)
788 struct inode *ip;
789 int cg;
790 long bprev;
791 int osize, nsize;
792{
793 register struct fs *fs;
794 register struct cg *cgp;
795 struct buf *bp;
796 long bno;
797 int frags, bbase;
798 int i, error;
799
800 fs = ip->i_fs;
801 if (fs->fs_cs(fs, cg).cs_nffree < numfrags(fs, nsize - osize))
802 return (0);
803 frags = numfrags(fs, nsize);
804 bbase = fragnum(fs, bprev);
805 if (bbase > fragnum(fs, (bprev + frags - 1))) {
806 /* cannot extend across a block boundary */
807 return (0);
808 }
809 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
810 (int)fs->fs_cgsize, NOCRED, &bp);
811 if (error) {
812 brelse(bp);
813 return (0);
814 }
815 cgp = (struct cg *)bp->b_data;
816 if (!cg_chkmagic(cgp)) {
817 brelse(bp);
818 return (0);
819 }
820 cgp->cg_time = time.tv_sec;
821 bno = dtogd(fs, bprev);
822 for (i = numfrags(fs, osize); i < frags; i++)
823 if (isclr(cg_blksfree(cgp), bno + i)) {
824 brelse(bp);
825 return (0);
826 }
827 /*
828 * the current fragment can be extended
829 * deduct the count on fragment being extended into
830 * increase the count on the remaining fragment (if any)
831 * allocate the extended piece
832 */
833 for (i = frags; i < fs->fs_frag - bbase; i++)
834 if (isclr(cg_blksfree(cgp), bno + i))
835 break;
836 cgp->cg_frsum[i - numfrags(fs, osize)]--;
837 if (i != frags)
838 cgp->cg_frsum[i - frags]++;
839 for (i = numfrags(fs, osize); i < frags; i++) {
840 clrbit(cg_blksfree(cgp), bno + i);
841 cgp->cg_cs.cs_nffree--;
842 fs->fs_cstotal.cs_nffree--;
843 fs->fs_cs(fs, cg).cs_nffree--;
844 }
845 fs->fs_fmod = 1;
846 bdwrite(bp);
847 return (bprev);
848}
849
850/*
851 * Determine whether a block can be allocated.
852 *
853 * Check to see if a block of the appropriate size is available,
854 * and if it is, allocate it.
855 */
856static ufs_daddr_t
857ffs_alloccg(ip, cg, bpref, size)
858 struct inode *ip;
859 int cg;
860 ufs_daddr_t bpref;
861 int size;
862{
863 register struct fs *fs;
864 register struct cg *cgp;
865 struct buf *bp;
866 register int i;
867 int error, bno, frags, allocsiz;
868
869 fs = ip->i_fs;
870 if (fs->fs_cs(fs, cg).cs_nbfree == 0 && size == fs->fs_bsize)
871 return (0);
872 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
873 (int)fs->fs_cgsize, NOCRED, &bp);
874 if (error) {
875 brelse(bp);
876 return (0);
877 }
878 cgp = (struct cg *)bp->b_data;
879 if (!cg_chkmagic(cgp) ||
880 (cgp->cg_cs.cs_nbfree == 0 && size == fs->fs_bsize)) {
881 brelse(bp);
882 return (0);
883 }
884 cgp->cg_time = time.tv_sec;
885 if (size == fs->fs_bsize) {
886 bno = ffs_alloccgblk(fs, cgp, bpref);
887 bdwrite(bp);
888 return (bno);
889 }
890 /*
891 * check to see if any fragments are already available
892 * allocsiz is the size which will be allocated, hacking
893 * it down to a smaller size if necessary
894 */
895 frags = numfrags(fs, size);
896 for (allocsiz = frags; allocsiz < fs->fs_frag; allocsiz++)
897 if (cgp->cg_frsum[allocsiz] != 0)
898 break;
899 if (allocsiz == fs->fs_frag) {
900 /*
901 * no fragments were available, so a block will be
902 * allocated, and hacked up
903 */
904 if (cgp->cg_cs.cs_nbfree == 0) {
905 brelse(bp);
906 return (0);
907 }
908 bno = ffs_alloccgblk(fs, cgp, bpref);
909 bpref = dtogd(fs, bno);
910 for (i = frags; i < fs->fs_frag; i++)
911 setbit(cg_blksfree(cgp), bpref + i);
912 i = fs->fs_frag - frags;
913 cgp->cg_cs.cs_nffree += i;
914 fs->fs_cstotal.cs_nffree += i;
915 fs->fs_cs(fs, cg).cs_nffree += i;
916 fs->fs_fmod = 1;
917 cgp->cg_frsum[i]++;
918 bdwrite(bp);
919 return (bno);
920 }
921 bno = ffs_mapsearch(fs, cgp, bpref, allocsiz);
922 if (bno < 0) {
923 brelse(bp);
924 return (0);
925 }
926 for (i = 0; i < frags; i++)
927 clrbit(cg_blksfree(cgp), bno + i);
928 cgp->cg_cs.cs_nffree -= frags;
929 fs->fs_cstotal.cs_nffree -= frags;
930 fs->fs_cs(fs, cg).cs_nffree -= frags;
931 fs->fs_fmod = 1;
932 cgp->cg_frsum[allocsiz]--;
933 if (frags != allocsiz)
934 cgp->cg_frsum[allocsiz - frags]++;
935 bdwrite(bp);
936 return (cg * fs->fs_fpg + bno);
937}
938
939/*
940 * Allocate a block in a cylinder group.
941 *
942 * This algorithm implements the following policy:
943 * 1) allocate the requested block.
944 * 2) allocate a rotationally optimal block in the same cylinder.
945 * 3) allocate the next available block on the block rotor for the
946 * specified cylinder group.
947 * Note that this routine only allocates fs_bsize blocks; these
948 * blocks may be fragmented by the routine that allocates them.
949 */
950static ufs_daddr_t
951ffs_alloccgblk(fs, cgp, bpref)
952 register struct fs *fs;
953 register struct cg *cgp;
954 ufs_daddr_t bpref;
955{
956 ufs_daddr_t bno, blkno;
957 int cylno, pos, delta;
958 short *cylbp;
959 register int i;
960
961 if (bpref == 0 || dtog(fs, bpref) != cgp->cg_cgx) {
962 bpref = cgp->cg_rotor;
963 goto norot;
964 }
965 bpref = blknum(fs, bpref);
966 bpref = dtogd(fs, bpref);
967 /*
968 * if the requested block is available, use it
969 */
970 if (ffs_isblock(fs, cg_blksfree(cgp), fragstoblks(fs, bpref))) {
971 bno = bpref;
972 goto gotit;
973 }
974 if (fs->fs_nrpos <= 1 || fs->fs_cpc == 0) {
975 /*
976 * Block layout information is not available.
977 * Leaving bpref unchanged means we take the
978 * next available free block following the one
979 * we just allocated. Hopefully this will at
980 * least hit a track cache on drives of unknown
981 * geometry (e.g. SCSI).
982 */
983 goto norot;
984 }
985 /*
986 * check for a block available on the same cylinder
987 */
988 cylno = cbtocylno(fs, bpref);
989 if (cg_blktot(cgp)[cylno] == 0)
990 goto norot;
991 /*
992 * check the summary information to see if a block is
993 * available in the requested cylinder starting at the
994 * requested rotational position and proceeding around.
995 */
996 cylbp = cg_blks(fs, cgp, cylno);
997 pos = cbtorpos(fs, bpref);
998 for (i = pos; i < fs->fs_nrpos; i++)
999 if (cylbp[i] > 0)
1000 break;
1001 if (i == fs->fs_nrpos)
1002 for (i = 0; i < pos; i++)
1003 if (cylbp[i] > 0)
1004 break;
1005 if (cylbp[i] > 0) {
1006 /*
1007 * found a rotational position, now find the actual
1008 * block. A panic if none is actually there.
1009 */
1010 pos = cylno % fs->fs_cpc;
1011 bno = (cylno - pos) * fs->fs_spc / NSPB(fs);
1012 if (fs_postbl(fs, pos)[i] == -1) {
1013 printf("pos = %d, i = %d, fs = %s\n",
1014 pos, i, fs->fs_fsmnt);
1015 panic("ffs_alloccgblk: cyl groups corrupted");
1016 }
1017 for (i = fs_postbl(fs, pos)[i];; ) {
1018 if (ffs_isblock(fs, cg_blksfree(cgp), bno + i)) {
1019 bno = blkstofrags(fs, (bno + i));
1020 goto gotit;
1021 }
1022 delta = fs_rotbl(fs)[i];
1023 if (delta <= 0 ||
1024 delta + i > fragstoblks(fs, fs->fs_fpg))
1025 break;
1026 i += delta;
1027 }
1028 printf("pos = %d, i = %d, fs = %s\n", pos, i, fs->fs_fsmnt);
1029 panic("ffs_alloccgblk: can't find blk in cyl");
1030 }
1031norot:
1032 /*
1033 * no blocks in the requested cylinder, so take next
1034 * available one in this cylinder group.
1035 */
1036 bno = ffs_mapsearch(fs, cgp, bpref, (int)fs->fs_frag);
1037 if (bno < 0)
1038 return (0);
1039 cgp->cg_rotor = bno;
1040gotit:
1041 blkno = fragstoblks(fs, bno);
1042 ffs_clrblock(fs, cg_blksfree(cgp), (long)blkno);
1043 ffs_clusteracct(fs, cgp, blkno, -1);
1044 cgp->cg_cs.cs_nbfree--;
1045 fs->fs_cstotal.cs_nbfree--;
1046 fs->fs_cs(fs, cgp->cg_cgx).cs_nbfree--;
1047 cylno = cbtocylno(fs, bno);
1048 cg_blks(fs, cgp, cylno)[cbtorpos(fs, bno)]--;
1049 cg_blktot(cgp)[cylno]--;
1050 fs->fs_fmod = 1;
1051 return (cgp->cg_cgx * fs->fs_fpg + bno);
1052}
1053
1054#ifdef notyet
1055/*
1056 * Determine whether a cluster can be allocated.
1057 *
1058 * We do not currently check for optimal rotational layout if there
1059 * are multiple choices in the same cylinder group. Instead we just
1060 * take the first one that we find following bpref.
1061 */
1062static ufs_daddr_t
1063ffs_clusteralloc(ip, cg, bpref, len)
1064 struct inode *ip;
1065 int cg;
1066 ufs_daddr_t bpref;
1067 int len;
1068{
1069 register struct fs *fs;
1070 register struct cg *cgp;
1071 struct buf *bp;
1072 int i, got, run, bno, bit, map;
1073 u_char *mapp;
1074 int32_t *lp;
1075
1076 fs = ip->i_fs;
1077 if (fs->fs_maxcluster[cg] < len)
1078 return (NULL);
1079 if (bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), (int)fs->fs_cgsize,
1080 NOCRED, &bp))
1081 goto fail;
1082 cgp = (struct cg *)bp->b_data;
1083 if (!cg_chkmagic(cgp))
1084 goto fail;
1085 /*
1086 * Check to see if a cluster of the needed size (or bigger) is
1087 * available in this cylinder group.
1088 */
1089 lp = &cg_clustersum(cgp)[len];
1090 for (i = len; i <= fs->fs_contigsumsize; i++)
1091 if (*lp++ > 0)
1092 break;
1093 if (i > fs->fs_contigsumsize) {
1094 /*
1095 * This is the first time looking for a cluster in this
1096 * cylinder group. Update the cluster summary information
1097 * to reflect the true maximum sized cluster so that
1098 * future cluster allocation requests can avoid reading
1099 * the cylinder group map only to find no clusters.
1100 */
1101 lp = &cg_clustersum(cgp)[len - 1];
1102 for (i = len - 1; i > 0; i--)
1103 if (*lp-- > 0)
1104 break;
1105 fs->fs_maxcluster[cg] = i;
1106 goto fail;
1107 }
1108 /*
1109 * Search the cluster map to find a big enough cluster.
1110 * We take the first one that we find, even if it is larger
1111 * than we need as we prefer to get one close to the previous
1112 * block allocation. We do not search before the current
1113 * preference point as we do not want to allocate a block
1114 * that is allocated before the previous one (as we will
1115 * then have to wait for another pass of the elevator
1116 * algorithm before it will be read). We prefer to fail and
1117 * be recalled to try an allocation in the next cylinder group.
1118 */
1119 if (dtog(fs, bpref) != cg)
1120 bpref = 0;
1121 else
1122 bpref = fragstoblks(fs, dtogd(fs, blknum(fs, bpref)));
1123 mapp = &cg_clustersfree(cgp)[bpref / NBBY];
1124 map = *mapp++;
1125 bit = 1 << (bpref % NBBY);
1126 for (run = 0, got = bpref; got < cgp->cg_nclusterblks; got++) {
1127 if ((map & bit) == 0) {
1128 run = 0;
1129 } else {
1130 run++;
1131 if (run == len)
1132 break;
1133 }
1134 if ((got & (NBBY - 1)) != (NBBY - 1)) {
1135 bit <<= 1;
1136 } else {
1137 map = *mapp++;
1138 bit = 1;
1139 }
1140 }
1141 if (got >= cgp->cg_nclusterblks)
1142 goto fail;
1143 /*
1144 * Allocate the cluster that we have found.
1145 */
1146 for (i = 1; i <= len; i++)
1147 if (!ffs_isblock(fs, cg_blksfree(cgp), got - run + i))
1148 panic("ffs_clusteralloc: map mismatch");
1149 bno = cg * fs->fs_fpg + blkstofrags(fs, got - run + 1);
1150 if (dtog(fs, bno) != cg)
1151 panic("ffs_clusteralloc: allocated out of group");
1152 len = blkstofrags(fs, len);
1153 for (i = 0; i < len; i += fs->fs_frag)
1154 if ((got = ffs_alloccgblk(fs, cgp, bno + i)) != bno + i)
1155 panic("ffs_clusteralloc: lost block");
1156 bdwrite(bp);
1157 return (bno);
1158
1159fail:
1160 brelse(bp);
1161 return (0);
1162}
1163#endif
1164
1165/*
1166 * Determine whether an inode can be allocated.
1167 *
1168 * Check to see if an inode is available, and if it is,
1169 * allocate it using the following policy:
1170 * 1) allocate the requested inode.
1171 * 2) allocate the next available inode after the requested
1172 * inode in the specified cylinder group.
1173 */
1174static ino_t
1175ffs_nodealloccg(ip, cg, ipref, mode)
1176 struct inode *ip;
1177 int cg;
1178 ufs_daddr_t ipref;
1179 int mode;
1180{
1181 register struct fs *fs;
1182 register struct cg *cgp;
1183 struct buf *bp;
1184 int error, start, len, loc, map, i;
1185
1186 fs = ip->i_fs;
1187 if (fs->fs_cs(fs, cg).cs_nifree == 0)
1188 return (0);
1189 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
1190 (int)fs->fs_cgsize, NOCRED, &bp);
1191 if (error) {
1192 brelse(bp);
1193 return (0);
1194 }
1195 cgp = (struct cg *)bp->b_data;
1196 if (!cg_chkmagic(cgp) || cgp->cg_cs.cs_nifree == 0) {
1197 brelse(bp);
1198 return (0);
1199 }
1200 cgp->cg_time = time.tv_sec;
1201 if (ipref) {
1202 ipref %= fs->fs_ipg;
1203 if (isclr(cg_inosused(cgp), ipref))
1204 goto gotit;
1205 }
1206 start = cgp->cg_irotor / NBBY;
1207 len = howmany(fs->fs_ipg - cgp->cg_irotor, NBBY);
1208 loc = skpc(0xff, len, &cg_inosused(cgp)[start]);
1209 if (loc == 0) {
1210 len = start + 1;
1211 start = 0;
1212 loc = skpc(0xff, len, &cg_inosused(cgp)[0]);
1213 if (loc == 0) {
1214 printf("cg = %d, irotor = %ld, fs = %s\n",
1215 cg, cgp->cg_irotor, fs->fs_fsmnt);
1216 panic("ffs_nodealloccg: map corrupted");
1217 /* NOTREACHED */
1218 }
1219 }
1220 i = start + len - loc;
1221 map = cg_inosused(cgp)[i];
1222 ipref = i * NBBY;
1223 for (i = 1; i < (1 << NBBY); i <<= 1, ipref++) {
1224 if ((map & i) == 0) {
1225 cgp->cg_irotor = ipref;
1226 goto gotit;
1227 }
1228 }
1229 printf("fs = %s\n", fs->fs_fsmnt);
1230 panic("ffs_nodealloccg: block not in map");
1231 /* NOTREACHED */
1232gotit:
1233 setbit(cg_inosused(cgp), ipref);
1234 cgp->cg_cs.cs_nifree--;
1235 fs->fs_cstotal.cs_nifree--;
1236 fs->fs_cs(fs, cg).cs_nifree--;
1237 fs->fs_fmod = 1;
1238 if ((mode & IFMT) == IFDIR) {
1239 cgp->cg_cs.cs_ndir++;
1240 fs->fs_cstotal.cs_ndir++;
1241 fs->fs_cs(fs, cg).cs_ndir++;
1242 }
1243 bdwrite(bp);
1244 return (cg * fs->fs_ipg + ipref);
1245}
1246
1247/*
1248 * Free a block or fragment.
1249 *
1250 * The specified block or fragment is placed back in the
1251 * free map. If a fragment is deallocated, a possible
1252 * block reassembly is checked.
1253 */
1254void
1255ffs_blkfree(ip, bno, size)
1256 register struct inode *ip;
1257 ufs_daddr_t bno;
1258 long size;
1259{
1260 register struct fs *fs;
1261 register struct cg *cgp;
1262 struct buf *bp;
1263 ufs_daddr_t blkno;
1264 int i, error, cg, blk, frags, bbase;
1265
1266 fs = ip->i_fs;
1267 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) {
1268 printf("dev = 0x%lx, bsize = %ld, size = %ld, fs = %s\n",
1269 (u_long)ip->i_dev, fs->fs_bsize, size, fs->fs_fsmnt);
1270 panic("ffs_blkfree: bad size");
1271 }
1272 cg = dtog(fs, bno);
1273 if ((u_int)bno >= fs->fs_size) {
1274 printf("bad block %ld, ino %ld\n", bno, ip->i_number);
1275 ffs_fserr(fs, ip->i_uid, "bad block");
1276 return;
1277 }
1278 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
1279 (int)fs->fs_cgsize, NOCRED, &bp);
1280 if (error) {
1281 brelse(bp);
1282 return;
1283 }
1284 cgp = (struct cg *)bp->b_data;
1285 if (!cg_chkmagic(cgp)) {
1286 brelse(bp);
1287 return;
1288 }
1289 cgp->cg_time = time.tv_sec;
1290 bno = dtogd(fs, bno);
1291 if (size == fs->fs_bsize) {
1292 blkno = fragstoblks(fs, bno);
1293 if (ffs_isblock(fs, cg_blksfree(cgp), blkno)) {
1294 printf("dev = 0x%lx, block = %ld, fs = %s\n",
1295 (u_long) ip->i_dev, bno, fs->fs_fsmnt);
1296 panic("ffs_blkfree: freeing free block");
1297 }
1298 ffs_setblock(fs, cg_blksfree(cgp), blkno);
1299 ffs_clusteracct(fs, cgp, blkno, 1);
1300 cgp->cg_cs.cs_nbfree++;
1301 fs->fs_cstotal.cs_nbfree++;
1302 fs->fs_cs(fs, cg).cs_nbfree++;
1303 i = cbtocylno(fs, bno);
1304 cg_blks(fs, cgp, i)[cbtorpos(fs, bno)]++;
1305 cg_blktot(cgp)[i]++;
1306 } else {
1307 bbase = bno - fragnum(fs, bno);
1308 /*
1309 * decrement the counts associated with the old frags
1310 */
1311 blk = blkmap(fs, cg_blksfree(cgp), bbase);
1312 ffs_fragacct(fs, blk, cgp->cg_frsum, -1);
1313 /*
1314 * deallocate the fragment
1315 */
1316 frags = numfrags(fs, size);
1317 for (i = 0; i < frags; i++) {
1318 if (isset(cg_blksfree(cgp), bno + i)) {
1319 printf("dev = 0x%lx, block = %ld, fs = %s\n",
1320 (u_long) ip->i_dev, bno + i, fs->fs_fsmnt);
1321 panic("ffs_blkfree: freeing free frag");
1322 }
1323 setbit(cg_blksfree(cgp), bno + i);
1324 }
1325 cgp->cg_cs.cs_nffree += i;
1326 fs->fs_cstotal.cs_nffree += i;
1327 fs->fs_cs(fs, cg).cs_nffree += i;
1328 /*
1329 * add back in counts associated with the new frags
1330 */
1331 blk = blkmap(fs, cg_blksfree(cgp), bbase);
1332 ffs_fragacct(fs, blk, cgp->cg_frsum, 1);
1333 /*
1334 * if a complete block has been reassembled, account for it
1335 */
1336 blkno = fragstoblks(fs, bbase);
1337 if (ffs_isblock(fs, cg_blksfree(cgp), blkno)) {
1338 cgp->cg_cs.cs_nffree -= fs->fs_frag;
1339 fs->fs_cstotal.cs_nffree -= fs->fs_frag;
1340 fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag;
1341 ffs_clusteracct(fs, cgp, blkno, 1);
1342 cgp->cg_cs.cs_nbfree++;
1343 fs->fs_cstotal.cs_nbfree++;
1344 fs->fs_cs(fs, cg).cs_nbfree++;
1345 i = cbtocylno(fs, bbase);
1346 cg_blks(fs, cgp, i)[cbtorpos(fs, bbase)]++;
1347 cg_blktot(cgp)[i]++;
1348 }
1349 }
1350 fs->fs_fmod = 1;
1351 bdwrite(bp);
1352}
1353
1354#ifdef DIAGNOSTIC
1355/*
1356 * Verify allocation of a block or fragment. Returns true if block or
1357 * fragment is allocated, false if it is free.
1358 */
1359int
1360ffs_checkblk(ip, bno, size)
1361 struct inode *ip;
1362 ufs_daddr_t bno;
1363 long size;
1364{
1365 struct fs *fs;
1366 struct cg *cgp;
1367 struct buf *bp;
1368 int i, error, frags, free;
1369
1370 fs = ip->i_fs;
1371 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) {
1372 printf("bsize = %d, size = %d, fs = %s\n",
1373 fs->fs_bsize, size, fs->fs_fsmnt);
1374 panic("ffs_checkblk: bad size");
1375 }
1376 if ((u_int)bno >= fs->fs_size)
1377 panic("ffs_checkblk: bad block %d", bno);
1378 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, dtog(fs, bno))),
1379 (int)fs->fs_cgsize, NOCRED, &bp);
1380 if (error)
1381 panic("ffs_checkblk: cg bread failed");
1382 cgp = (struct cg *)bp->b_data;
1383 if (!cg_chkmagic(cgp))
1384 panic("ffs_checkblk: cg magic mismatch");
1385 bno = dtogd(fs, bno);
1386 if (size == fs->fs_bsize) {
1387 free = ffs_isblock(fs, cg_blksfree(cgp), fragstoblks(fs, bno));
1388 } else {
1389 frags = numfrags(fs, size);
1390 for (free = 0, i = 0; i < frags; i++)
1391 if (isset(cg_blksfree(cgp), bno + i))
1392 free++;
1393 if (free != 0 && free != frags)
1394 panic("ffs_checkblk: partially free fragment");
1395 }
1396 brelse(bp);
1397 return (!free);
1398}
1399#endif /* DIAGNOSTIC */
1400
1401/*
1402 * Free an inode.
1403 *
1404 * The specified inode is placed back in the free map.
1405 */
1406int
1407ffs_vfree(ap)
1408 struct vop_vfree_args /* {
1409 struct vnode *a_pvp;
1410 ino_t a_ino;
1411 int a_mode;
1412 } */ *ap;
1413{
1414 register struct fs *fs;
1415 register struct cg *cgp;
1416 register struct inode *pip;
1417 ino_t ino = ap->a_ino;
1418 struct buf *bp;
1419 int error, cg;
1420
1421 pip = VTOI(ap->a_pvp);
1423 /* Remove the inode from its hash chain */
1424 ufs_ihashrem(pip);
1425 fs = pip->i_fs;
1426 if ((u_int)ino >= fs->fs_ipg * fs->fs_ncg)
1427 panic("ffs_vfree: range: dev = 0x%x, ino = %d, fs = %s",
1428 pip->i_dev, ino, fs->fs_fsmnt);
1429 cg = ino_to_cg(fs, ino);
1430 error = bread(pip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
1431 (int)fs->fs_cgsize, NOCRED, &bp);
1432 if (error) {
1433 brelse(bp);
1434 return (0);
1435 }
1436 cgp = (struct cg *)bp->b_data;
1437 if (!cg_chkmagic(cgp)) {
1438 brelse(bp);
1439 return (0);
1440 }
1441 cgp->cg_time = time.tv_sec;
1442 ino %= fs->fs_ipg;
1443 if (isclr(cg_inosused(cgp), ino)) {
1444 printf("dev = 0x%lx, ino = %ld, fs = %s\n",
1445 (u_long)pip->i_dev, ino, fs->fs_fsmnt);
1446 if (fs->fs_ronly == 0)
1447 panic("ffs_vfree: freeing free inode");
1448 }
1449 clrbit(cg_inosused(cgp), ino);
1450 if (ino < cgp->cg_irotor)
1451 cgp->cg_irotor = ino;
1452 cgp->cg_cs.cs_nifree++;
1453 fs->fs_cstotal.cs_nifree++;
1454 fs->fs_cs(fs, cg).cs_nifree++;
1455 if ((ap->a_mode & IFMT) == IFDIR) {
1456 cgp->cg_cs.cs_ndir--;
1457 fs->fs_cstotal.cs_ndir--;
1458 fs->fs_cs(fs, cg).cs_ndir--;
1459 }
1460 fs->fs_fmod = 1;
1461 bdwrite(bp);
1462 return (0);
1463}
1464
1465/*
1466 * Find a block of the specified size in the specified cylinder group.
1467 *
1468 * It is a panic if a request is made to find a block if none are
1469 * available.
1470 */
1471static ufs_daddr_t
1472ffs_mapsearch(fs, cgp, bpref, allocsiz)
1473 register struct fs *fs;
1474 register struct cg *cgp;
1475 ufs_daddr_t bpref;
1476 int allocsiz;
1477{
1478 ufs_daddr_t bno;
1479 int start, len, loc, i;
1480 int blk, field, subfield, pos;
1481
1482 /*
1483 * find the fragment by searching through the free block
1484 * map for an appropriate bit pattern
1485 */
1486 if (bpref)
1487 start = dtogd(fs, bpref) / NBBY;
1488 else
1489 start = cgp->cg_frotor / NBBY;
1490 len = howmany(fs->fs_fpg, NBBY) - start;
1491 loc = scanc((u_int)len, (u_char *)&cg_blksfree(cgp)[start],
1492 (u_char *)fragtbl[fs->fs_frag],
1493 (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
1494 if (loc == 0) {
1495 len = start + 1;
1496 start = 0;
1497 loc = scanc((u_int)len, (u_char *)&cg_blksfree(cgp)[0],
1498 (u_char *)fragtbl[fs->fs_frag],
1499 (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
1500 if (loc == 0) {
1501 printf("start = %d, len = %d, fs = %s\n",
1502 start, len, fs->fs_fsmnt);
1503 panic("ffs_alloccg: map corrupted");
1504 /* NOTREACHED */
1505 }
1506 }
1507 bno = (start + len - loc) * NBBY;
1508 cgp->cg_frotor = bno;
1509 /*
1510 * found the byte in the map
1511 * sift through the bits to find the selected frag
1512 */
1513 for (i = bno + NBBY; bno < i; bno += fs->fs_frag) {
1514 blk = blkmap(fs, cg_blksfree(cgp), bno);
1515 blk <<= 1;
1516 field = around[allocsiz];
1517 subfield = inside[allocsiz];
1518 for (pos = 0; pos <= fs->fs_frag - allocsiz; pos++) {
1519 if ((blk & field) == subfield)
1520 return (bno + pos);
1521 field <<= 1;
1522 subfield <<= 1;
1523 }
1524 }
1525 printf("bno = %lu, fs = %s\n", (u_long)bno, fs->fs_fsmnt);
1526 panic("ffs_alloccg: block not in map");
1527 return (-1);
1528}
1529
1530/*
1531 * Update the cluster map because of an allocation or free.
1532 *
1533 * Cnt == 1 means free; cnt == -1 means allocating.
1534 */
1535static void
1536ffs_clusteracct(fs, cgp, blkno, cnt)
1537 struct fs *fs;
1538 struct cg *cgp;
1539 ufs_daddr_t blkno;
1540 int cnt;
1541{
1542 int32_t *sump;
1543 int32_t *lp;
1544 u_char *freemapp, *mapp;
1545 int i, start, end, forw, back, map, bit;
1546
1547 if (fs->fs_contigsumsize <= 0)
1548 return;
1549 freemapp = cg_clustersfree(cgp);
1550 sump = cg_clustersum(cgp);
1551 /*
1552 * Allocate or clear the actual block.
1553 */
1554 if (cnt > 0)
1555 setbit(freemapp, blkno);
1556 else
1557 clrbit(freemapp, blkno);
1558 /*
1559 * Find the size of the cluster going forward.
1560 */
1561 start = blkno + 1;
1562 end = start + fs->fs_contigsumsize;
1563 if (end >= cgp->cg_nclusterblks)
1564 end = cgp->cg_nclusterblks;
1565 mapp = &freemapp[start / NBBY];
1566 map = *mapp++;
1567 bit = 1 << (start % NBBY);
1568 for (i = start; i < end; i++) {
1569 if ((map & bit) == 0)
1570 break;
1571 if ((i & (NBBY - 1)) != (NBBY - 1)) {
1572 bit <<= 1;
1573 } else {
1574 map = *mapp++;
1575 bit = 1;
1576 }
1577 }
1578 forw = i - start;
1579 /*
1580 * Find the size of the cluster going backward.
1581 */
1582 start = blkno - 1;
1583 end = start - fs->fs_contigsumsize;
1584 if (end < 0)
1585 end = -1;
1586 mapp = &freemapp[start / NBBY];
1587 map = *mapp--;
1588 bit = 1 << (start % NBBY);
1589 for (i = start; i > end; i--) {
1590 if ((map & bit) == 0)
1591 break;
1592 if ((i & (NBBY - 1)) != 0) {
1593 bit >>= 1;
1594 } else {
1595 map = *mapp--;
1596 bit = 1 << (NBBY - 1);
1597 }
1598 }
1599 back = start - i;
1600 /*
1601 * Account for old cluster and the possibly new forward and
1602 * back clusters.
1603 */
1604 i = back + forw + 1;
1605 if (i > fs->fs_contigsumsize)
1606 i = fs->fs_contigsumsize;
1607 sump[i] += cnt;
1608 if (back > 0)
1609 sump[back] -= cnt;
1610 if (forw > 0)
1611 sump[forw] -= cnt;
1612 /*
1613 * Update cluster summary information.
1614 */
1615 lp = &sump[fs->fs_contigsumsize];
1616 for (i = fs->fs_contigsumsize; i > 0; i--)
1617 if (*lp-- > 0)
1618 break;
1619 fs->fs_maxcluster[cgp->cg_cgx] = i;
1620}
1621
1622/*
1623 * Fserr prints the name of a file system with an error diagnostic.
1624 *
1625 * The form of the error message is:
1626 * fs: error message
1627 */
1628static void
1629ffs_fserr(fs, uid, cp)
1630 struct fs *fs;
1631 u_int uid;
1632 char *cp;
1633{
1634 struct proc *p = curproc; /* XXX */
1635
1636 log(LOG_ERR, "pid %d (%s), uid %d on %s: %s\n", p ? p->p_pid : -1,
1637 p ? p->p_comm : "-", uid, fs->fs_fsmnt, cp);
1638}
1422 fs = pip->i_fs;
1423 if ((u_int)ino >= fs->fs_ipg * fs->fs_ncg)
1424 panic("ffs_vfree: range: dev = 0x%x, ino = %d, fs = %s",
1425 pip->i_dev, ino, fs->fs_fsmnt);
1426 cg = ino_to_cg(fs, ino);
1427 error = bread(pip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
1428 (int)fs->fs_cgsize, NOCRED, &bp);
1429 if (error) {
1430 brelse(bp);
1431 return (0);
1432 }
1433 cgp = (struct cg *)bp->b_data;
1434 if (!cg_chkmagic(cgp)) {
1435 brelse(bp);
1436 return (0);
1437 }
1438 cgp->cg_time = time.tv_sec;
1439 ino %= fs->fs_ipg;
1440 if (isclr(cg_inosused(cgp), ino)) {
1441 printf("dev = 0x%lx, ino = %ld, fs = %s\n",
1442 (u_long)pip->i_dev, ino, fs->fs_fsmnt);
1443 if (fs->fs_ronly == 0)
1444 panic("ffs_vfree: freeing free inode");
1445 }
1446 clrbit(cg_inosused(cgp), ino);
1447 if (ino < cgp->cg_irotor)
1448 cgp->cg_irotor = ino;
1449 cgp->cg_cs.cs_nifree++;
1450 fs->fs_cstotal.cs_nifree++;
1451 fs->fs_cs(fs, cg).cs_nifree++;
1452 if ((ap->a_mode & IFMT) == IFDIR) {
1453 cgp->cg_cs.cs_ndir--;
1454 fs->fs_cstotal.cs_ndir--;
1455 fs->fs_cs(fs, cg).cs_ndir--;
1456 }
1457 fs->fs_fmod = 1;
1458 bdwrite(bp);
1459 return (0);
1460}
1461
1462/*
1463 * Find a block of the specified size in the specified cylinder group.
1464 *
1465 * It is a panic if a request is made to find a block if none are
1466 * available.
1467 */
1468static ufs_daddr_t
1469ffs_mapsearch(fs, cgp, bpref, allocsiz)
1470 register struct fs *fs;
1471 register struct cg *cgp;
1472 ufs_daddr_t bpref;
1473 int allocsiz;
1474{
1475 ufs_daddr_t bno;
1476 int start, len, loc, i;
1477 int blk, field, subfield, pos;
1478
1479 /*
1480 * find the fragment by searching through the free block
1481 * map for an appropriate bit pattern
1482 */
1483 if (bpref)
1484 start = dtogd(fs, bpref) / NBBY;
1485 else
1486 start = cgp->cg_frotor / NBBY;
1487 len = howmany(fs->fs_fpg, NBBY) - start;
1488 loc = scanc((u_int)len, (u_char *)&cg_blksfree(cgp)[start],
1489 (u_char *)fragtbl[fs->fs_frag],
1490 (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
1491 if (loc == 0) {
1492 len = start + 1;
1493 start = 0;
1494 loc = scanc((u_int)len, (u_char *)&cg_blksfree(cgp)[0],
1495 (u_char *)fragtbl[fs->fs_frag],
1496 (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
1497 if (loc == 0) {
1498 printf("start = %d, len = %d, fs = %s\n",
1499 start, len, fs->fs_fsmnt);
1500 panic("ffs_alloccg: map corrupted");
1501 /* NOTREACHED */
1502 }
1503 }
1504 bno = (start + len - loc) * NBBY;
1505 cgp->cg_frotor = bno;
1506 /*
1507 * found the byte in the map
1508 * sift through the bits to find the selected frag
1509 */
1510 for (i = bno + NBBY; bno < i; bno += fs->fs_frag) {
1511 blk = blkmap(fs, cg_blksfree(cgp), bno);
1512 blk <<= 1;
1513 field = around[allocsiz];
1514 subfield = inside[allocsiz];
1515 for (pos = 0; pos <= fs->fs_frag - allocsiz; pos++) {
1516 if ((blk & field) == subfield)
1517 return (bno + pos);
1518 field <<= 1;
1519 subfield <<= 1;
1520 }
1521 }
1522 printf("bno = %lu, fs = %s\n", (u_long)bno, fs->fs_fsmnt);
1523 panic("ffs_alloccg: block not in map");
1524 return (-1);
1525}
1526
1527/*
1528 * Update the cluster map because of an allocation or free.
1529 *
1530 * Cnt == 1 means free; cnt == -1 means allocating.
1531 */
1532static void
1533ffs_clusteracct(fs, cgp, blkno, cnt)
1534 struct fs *fs;
1535 struct cg *cgp;
1536 ufs_daddr_t blkno;
1537 int cnt;
1538{
1539 int32_t *sump;
1540 int32_t *lp;
1541 u_char *freemapp, *mapp;
1542 int i, start, end, forw, back, map, bit;
1543
1544 if (fs->fs_contigsumsize <= 0)
1545 return;
1546 freemapp = cg_clustersfree(cgp);
1547 sump = cg_clustersum(cgp);
1548 /*
1549 * Allocate or clear the actual block.
1550 */
1551 if (cnt > 0)
1552 setbit(freemapp, blkno);
1553 else
1554 clrbit(freemapp, blkno);
1555 /*
1556 * Find the size of the cluster going forward.
1557 */
1558 start = blkno + 1;
1559 end = start + fs->fs_contigsumsize;
1560 if (end >= cgp->cg_nclusterblks)
1561 end = cgp->cg_nclusterblks;
1562 mapp = &freemapp[start / NBBY];
1563 map = *mapp++;
1564 bit = 1 << (start % NBBY);
1565 for (i = start; i < end; i++) {
1566 if ((map & bit) == 0)
1567 break;
1568 if ((i & (NBBY - 1)) != (NBBY - 1)) {
1569 bit <<= 1;
1570 } else {
1571 map = *mapp++;
1572 bit = 1;
1573 }
1574 }
1575 forw = i - start;
1576 /*
1577 * Find the size of the cluster going backward.
1578 */
1579 start = blkno - 1;
1580 end = start - fs->fs_contigsumsize;
1581 if (end < 0)
1582 end = -1;
1583 mapp = &freemapp[start / NBBY];
1584 map = *mapp--;
1585 bit = 1 << (start % NBBY);
1586 for (i = start; i > end; i--) {
1587 if ((map & bit) == 0)
1588 break;
1589 if ((i & (NBBY - 1)) != 0) {
1590 bit >>= 1;
1591 } else {
1592 map = *mapp--;
1593 bit = 1 << (NBBY - 1);
1594 }
1595 }
1596 back = start - i;
1597 /*
1598 * Account for old cluster and the possibly new forward and
1599 * back clusters.
1600 */
1601 i = back + forw + 1;
1602 if (i > fs->fs_contigsumsize)
1603 i = fs->fs_contigsumsize;
1604 sump[i] += cnt;
1605 if (back > 0)
1606 sump[back] -= cnt;
1607 if (forw > 0)
1608 sump[forw] -= cnt;
1609 /*
1610 * Update cluster summary information.
1611 */
1612 lp = &sump[fs->fs_contigsumsize];
1613 for (i = fs->fs_contigsumsize; i > 0; i--)
1614 if (*lp-- > 0)
1615 break;
1616 fs->fs_maxcluster[cgp->cg_cgx] = i;
1617}
1618
1619/*
1620 * Fserr prints the name of a file system with an error diagnostic.
1621 *
1622 * The form of the error message is:
1623 * fs: error message
1624 */
1625static void
1626ffs_fserr(fs, uid, cp)
1627 struct fs *fs;
1628 u_int uid;
1629 char *cp;
1630{
1631 struct proc *p = curproc; /* XXX */
1632
1633 log(LOG_ERR, "pid %d (%s), uid %d on %s: %s\n", p ? p->p_pid : -1,
1634 p ? p->p_comm : "-", uid, fs->fs_fsmnt, cp);
1635}