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 1.56 1999/01/07 16:14:16 bde 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>
51#include <ufs/ufs/ufs_extern.h>
52#include <ufs/ufs/ufsmount.h>
53
54#include <ufs/ffs/fs.h>
55#include <ufs/ffs/ffs_extern.h>
56
57typedef ufs_daddr_t allocfcn_t __P((struct inode *ip, int cg, ufs_daddr_t bpref,
58 int size));
59
60static ufs_daddr_t ffs_alloccg __P((struct inode *, int, ufs_daddr_t, int));
61static ufs_daddr_t
62 ffs_alloccgblk __P((struct inode *, struct buf *, ufs_daddr_t));
63#ifdef DIAGNOSTIC
64static int ffs_checkblk __P((struct inode *, ufs_daddr_t, long));
65#endif
66static void ffs_clusteracct __P((struct fs *, struct cg *, ufs_daddr_t,
67 int));
68static ufs_daddr_t ffs_clusteralloc __P((struct inode *, int, ufs_daddr_t,
69 int));
70static ino_t ffs_dirpref __P((struct fs *));
71static ufs_daddr_t ffs_fragextend __P((struct inode *, int, long, int, int));
72static void ffs_fserr __P((struct fs *, u_int, char *));
73static u_long ffs_hashalloc
74 __P((struct inode *, int, long, int, allocfcn_t *));
75static ino_t ffs_nodealloccg __P((struct inode *, int, ufs_daddr_t, int));
76static ufs_daddr_t ffs_mapsearch __P((struct fs *, struct cg *, ufs_daddr_t,
77 int));
78
79/*
80 * Allocate a block in the file system.
81 *
82 * The size of the requested block is given, which must be some
83 * multiple of fs_fsize and <= fs_bsize.
84 * A preference may be optionally specified. If a preference is given
85 * the following hierarchy is used to allocate a block:
86 * 1) allocate the requested block.
87 * 2) allocate a rotationally optimal block in the same cylinder.
88 * 3) allocate a block in the same cylinder group.
89 * 4) quadradically rehash into other cylinder groups, until an
90 * available block is located.
91 * If no block preference is given the following heirarchy is used
92 * to allocate a block:
93 * 1) allocate a block in the cylinder group that contains the
94 * inode for the file.
95 * 2) quadradically rehash into other cylinder groups, until an
96 * available block is located.
97 */
98int
99ffs_alloc(ip, lbn, bpref, size, cred, bnp)
100 register struct inode *ip;
101 ufs_daddr_t lbn, bpref;
102 int size;
103 struct ucred *cred;
104 ufs_daddr_t *bnp;
105{
106 register struct fs *fs;
107 ufs_daddr_t bno;
108 int cg;
109#ifdef QUOTA
110 int error;
111#endif
112
113 *bnp = 0;
114 fs = ip->i_fs;
115#ifdef DIAGNOSTIC
116 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) {
117 printf("dev = 0x%lx, bsize = %ld, size = %d, fs = %s\n",
118 (u_long)ip->i_dev, (long)fs->fs_bsize, size, fs->fs_fsmnt);
119 panic("ffs_alloc: bad size");
120 }
121 if (cred == NOCRED)
122 panic("ffs_alloc: missing credential");
123#endif /* DIAGNOSTIC */
124 if (size == fs->fs_bsize && fs->fs_cstotal.cs_nbfree == 0)
125 goto nospace;
126 if (cred->cr_uid != 0 &&
127 freespace(fs, fs->fs_minfree) - numfrags(fs, size) < 0)
128 goto nospace;
129#ifdef QUOTA
130 error = chkdq(ip, (long)btodb(size), cred, 0);
131 if (error)
132 return (error);
133#endif
134 if (bpref >= fs->fs_size)
135 bpref = 0;
136 if (bpref == 0)
137 cg = ino_to_cg(fs, ip->i_number);
138 else
139 cg = dtog(fs, bpref);
140 bno = (ufs_daddr_t)ffs_hashalloc(ip, cg, (long)bpref, size,
141 ffs_alloccg);
142 if (bno > 0) {
143 ip->i_blocks += btodb(size);
144 ip->i_flag |= IN_CHANGE | IN_UPDATE;
145 *bnp = bno;
146 return (0);
147 }
148#ifdef QUOTA
149 /*
150 * Restore user's disk quota because allocation failed.
151 */
152 (void) chkdq(ip, (long)-btodb(size), cred, FORCE);
153#endif
154nospace:
155 ffs_fserr(fs, cred->cr_uid, "file system full");
156 uprintf("\n%s: write failed, file system is full\n", fs->fs_fsmnt);
157 return (ENOSPC);
158}
159
160/*
161 * Reallocate a fragment to a bigger size
162 *
163 * The number and size of the old block is given, and a preference
164 * and new size is also specified. The allocator attempts to extend
165 * the original block. Failing that, the regular block allocator is
166 * invoked to get an appropriate block.
167 */
168int
169ffs_realloccg(ip, lbprev, bpref, osize, nsize, cred, bpp)
170 register struct inode *ip;
171 ufs_daddr_t lbprev;
172 ufs_daddr_t bpref;
173 int osize, nsize;
174 struct ucred *cred;
175 struct buf **bpp;
176{
177 register struct fs *fs;
178 struct buf *bp;
179 int cg, request, error;
180 ufs_daddr_t bprev, bno;
181
182 *bpp = 0;
183 fs = ip->i_fs;
184#ifdef DIAGNOSTIC
185 if ((u_int)osize > fs->fs_bsize || fragoff(fs, osize) != 0 ||
186 (u_int)nsize > fs->fs_bsize || fragoff(fs, nsize) != 0) {
187 printf(
188 "dev = 0x%lx, bsize = %ld, osize = %d, nsize = %d, fs = %s\n",
189 (u_long)ip->i_dev, (long)fs->fs_bsize, osize,
190 nsize, fs->fs_fsmnt);
191 panic("ffs_realloccg: bad size");
192 }
193 if (cred == NOCRED)
194 panic("ffs_realloccg: missing credential");
195#endif /* DIAGNOSTIC */
196 if (cred->cr_uid != 0 &&
197 freespace(fs, fs->fs_minfree) - numfrags(fs, nsize - osize) < 0)
198 goto nospace;
199 if ((bprev = ip->i_db[lbprev]) == 0) {
200 printf("dev = 0x%lx, bsize = %ld, bprev = %ld, fs = %s\n",
201 (u_long)ip->i_dev, (long)fs->fs_bsize, (long)bprev,
202 fs->fs_fsmnt);
203 panic("ffs_realloccg: bad bprev");
204 }
205 /*
206 * Allocate the extra space in the buffer.
207 */
208 error = bread(ITOV(ip), lbprev, osize, NOCRED, &bp);
209 if (error) {
210 brelse(bp);
211 return (error);
212 }
213
214 if( bp->b_blkno == bp->b_lblkno) {
215 if( lbprev >= NDADDR)
216 panic("ffs_realloccg: lbprev out of range");
217 bp->b_blkno = fsbtodb(fs, bprev);
218 }
219
220#ifdef QUOTA
221 error = chkdq(ip, (long)btodb(nsize - osize), cred, 0);
222 if (error) {
223 brelse(bp);
224 return (error);
225 }
226#endif
227 /*
228 * Check for extension in the existing location.
229 */
230 cg = dtog(fs, bprev);
231 bno = ffs_fragextend(ip, cg, (long)bprev, osize, nsize);
232 if (bno) {
233 if (bp->b_blkno != fsbtodb(fs, bno))
234 panic("ffs_realloccg: bad blockno");
235 ip->i_blocks += btodb(nsize - osize);
236 ip->i_flag |= IN_CHANGE | IN_UPDATE;
237 allocbuf(bp, nsize);
238 bp->b_flags |= B_DONE;
239 bzero((char *)bp->b_data + osize, (u_int)nsize - osize);
240 *bpp = bp;
241 return (0);
242 }
243 /*
244 * Allocate a new disk location.
245 */
246 if (bpref >= fs->fs_size)
247 bpref = 0;
248 switch ((int)fs->fs_optim) {
249 case FS_OPTSPACE:
250 /*
251 * Allocate an exact sized fragment. Although this makes
252 * best use of space, we will waste time relocating it if
253 * the file continues to grow. If the fragmentation is
254 * less than half of the minimum free reserve, we choose
255 * to begin optimizing for time.
256 */
257 request = nsize;
258 if (fs->fs_minfree <= 5 ||
259 fs->fs_cstotal.cs_nffree >
260 fs->fs_dsize * fs->fs_minfree / (2 * 100))
261 break;
262 log(LOG_NOTICE, "%s: optimization changed from SPACE to TIME\n",
263 fs->fs_fsmnt);
264 fs->fs_optim = FS_OPTTIME;
265 break;
266 case FS_OPTTIME:
267 /*
268 * At this point we have discovered a file that is trying to
269 * grow a small fragment to a larger fragment. To save time,
270 * we allocate a full sized block, then free the unused portion.
271 * If the file continues to grow, the `ffs_fragextend' call
272 * above will be able to grow it in place without further
273 * copying. If aberrant programs cause disk fragmentation to
274 * grow within 2% of the free reserve, we choose to begin
275 * optimizing for space.
276 */
277 request = fs->fs_bsize;
278 if (fs->fs_cstotal.cs_nffree <
279 fs->fs_dsize * (fs->fs_minfree - 2) / 100)
280 break;
281 log(LOG_NOTICE, "%s: optimization changed from TIME to SPACE\n",
282 fs->fs_fsmnt);
283 fs->fs_optim = FS_OPTSPACE;
284 break;
285 default:
286 printf("dev = 0x%lx, optim = %ld, fs = %s\n",
287 (u_long)ip->i_dev, (long)fs->fs_optim, fs->fs_fsmnt);
288 panic("ffs_realloccg: bad optim");
289 /* NOTREACHED */
290 }
291 bno = (ufs_daddr_t)ffs_hashalloc(ip, cg, (long)bpref, request,
292 ffs_alloccg);
293 if (bno > 0) {
294 bp->b_blkno = fsbtodb(fs, bno);
295 if (!DOINGSOFTDEP(ITOV(ip)))
296 ffs_blkfree(ip, bprev, (long)osize);
297 if (nsize < request)
298 ffs_blkfree(ip, bno + numfrags(fs, nsize),
299 (long)(request - nsize));
300 ip->i_blocks += btodb(nsize - osize);
301 ip->i_flag |= IN_CHANGE | IN_UPDATE;
302 allocbuf(bp, nsize);
303 bp->b_flags |= B_DONE;
304 bzero((char *)bp->b_data + osize, (u_int)nsize - osize);
305 *bpp = bp;
306 return (0);
307 }
308#ifdef QUOTA
309 /*
310 * Restore user's disk quota because allocation failed.
311 */
312 (void) chkdq(ip, (long)-btodb(nsize - osize), cred, FORCE);
313#endif
314 brelse(bp);
315nospace:
316 /*
317 * no space available
318 */
319 ffs_fserr(fs, cred->cr_uid, "file system full");
320 uprintf("\n%s: write failed, file system is full\n", fs->fs_fsmnt);
321 return (ENOSPC);
322}
323
324SYSCTL_NODE(_vfs, OID_AUTO, ffs, CTLFLAG_RW, 0, "FFS filesystem");
325
326/*
327 * Reallocate a sequence of blocks into a contiguous sequence of blocks.
328 *
329 * The vnode and an array of buffer pointers for a range of sequential
330 * logical blocks to be made contiguous is given. The allocator attempts
331 * to find a range of sequential blocks starting as close as possible to
332 * an fs_rotdelay offset from the end of the allocation for the logical
333 * block immediately preceeding the current range. If successful, the
334 * physical block numbers in the buffer pointers and in the inode are
335 * changed to reflect the new allocation. If unsuccessful, the allocation
336 * is left unchanged. The success in doing the reallocation is returned.
337 * Note that the error return is not reflected back to the user. Rather
338 * the previous block allocation will be used.
339 */
340static int doasyncfree = 1;
341SYSCTL_INT(_vfs_ffs, FFS_ASYNCFREE, doasyncfree, CTLFLAG_RW, &doasyncfree, 0, "");
342
343static int doreallocblks = 1;
344SYSCTL_INT(_vfs_ffs, FFS_REALLOCBLKS, doreallocblks, CTLFLAG_RW, &doreallocblks, 0, "");
345
346#ifdef DEBUG
347static volatile int prtrealloc = 0;
348#endif
349
350int
351ffs_reallocblks(ap)
352 struct vop_reallocblks_args /* {
353 struct vnode *a_vp;
354 struct cluster_save *a_buflist;
355 } */ *ap;
356{
357 struct fs *fs;
358 struct inode *ip;
359 struct vnode *vp;
360 struct buf *sbp, *ebp;
361 ufs_daddr_t *bap, *sbap, *ebap = 0;
362 struct cluster_save *buflist;
363 ufs_daddr_t start_lbn, end_lbn, soff, newblk, blkno;
364 struct indir start_ap[NIADDR + 1], end_ap[NIADDR + 1], *idp;
365 int i, len, start_lvl, end_lvl, pref, ssize;
366
367 if (doreallocblks == 0)
368 return (ENOSPC);
369 vp = ap->a_vp;
370 ip = VTOI(vp);
371 fs = ip->i_fs;
372 if (fs->fs_contigsumsize <= 0)
373 return (ENOSPC);
374 buflist = ap->a_buflist;
375 len = buflist->bs_nchildren;
376 start_lbn = buflist->bs_children[0]->b_lblkno;
377 end_lbn = start_lbn + len - 1;
378#ifdef DIAGNOSTIC
379 for (i = 0; i < len; i++)
380 if (!ffs_checkblk(ip,
381 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
382 panic("ffs_reallocblks: unallocated block 1");
383 for (i = 1; i < len; i++)
384 if (buflist->bs_children[i]->b_lblkno != start_lbn + i)
385 panic("ffs_reallocblks: non-logical cluster");
386 blkno = buflist->bs_children[0]->b_blkno;
387 ssize = fsbtodb(fs, fs->fs_frag);
388 for (i = 1; i < len - 1; i++)
389 if (buflist->bs_children[i]->b_blkno != blkno + (i * ssize))
390 panic("ffs_reallocblks: non-physical cluster %d", i);
391#endif
392 /*
393 * If the latest allocation is in a new cylinder group, assume that
394 * the filesystem has decided to move and do not force it back to
395 * the previous cylinder group.
396 */
397 if (dtog(fs, dbtofsb(fs, buflist->bs_children[0]->b_blkno)) !=
398 dtog(fs, dbtofsb(fs, buflist->bs_children[len - 1]->b_blkno)))
399 return (ENOSPC);
400 if (ufs_getlbns(vp, start_lbn, start_ap, &start_lvl) ||
401 ufs_getlbns(vp, end_lbn, end_ap, &end_lvl))
402 return (ENOSPC);
403 /*
404 * Get the starting offset and block map for the first block.
405 */
406 if (start_lvl == 0) {
407 sbap = &ip->i_db[0];
408 soff = start_lbn;
409 } else {
410 idp = &start_ap[start_lvl - 1];
411 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &sbp)) {
412 brelse(sbp);
413 return (ENOSPC);
414 }
415 sbap = (ufs_daddr_t *)sbp->b_data;
416 soff = idp->in_off;
417 }
418 /*
419 * Find the preferred location for the cluster.
420 */
421 pref = ffs_blkpref(ip, start_lbn, soff, sbap);
422 /*
423 * If the block range spans two block maps, get the second map.
424 */
425 if (end_lvl == 0 || (idp = &end_ap[end_lvl - 1])->in_off + 1 >= len) {
426 ssize = len;
427 } else {
428#ifdef DIAGNOSTIC
429 if (start_ap[start_lvl-1].in_lbn == idp->in_lbn)
430 panic("ffs_reallocblk: start == end");
431#endif
432 ssize = len - (idp->in_off + 1);
433 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &ebp))
434 goto fail;
435 ebap = (ufs_daddr_t *)ebp->b_data;
436 }
437 /*
438 * Search the block map looking for an allocation of the desired size.
439 */
440 if ((newblk = (ufs_daddr_t)ffs_hashalloc(ip, dtog(fs, pref), (long)pref,
441 len, ffs_clusteralloc)) == 0)
442 goto fail;
443 /*
444 * We have found a new contiguous block.
445 *
446 * First we have to replace the old block pointers with the new
447 * block pointers in the inode and indirect blocks associated
448 * with the file.
449 */
450#ifdef DEBUG
451 if (prtrealloc)
452 printf("realloc: ino %d, lbns %d-%d\n\told:", ip->i_number,
453 start_lbn, end_lbn);
454#endif
455 blkno = newblk;
456 for (bap = &sbap[soff], i = 0; i < len; i++, blkno += fs->fs_frag) {
457 if (i == ssize) {
458 bap = ebap;
459 soff = -i;
460 }
461#ifdef DIAGNOSTIC
462 if (!ffs_checkblk(ip,
463 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
464 panic("ffs_reallocblks: unallocated block 2");
465 if (dbtofsb(fs, buflist->bs_children[i]->b_blkno) != *bap)
466 panic("ffs_reallocblks: alloc mismatch");
467#endif
468#ifdef DEBUG
469 if (prtrealloc)
470 printf(" %d,", *bap);
471#endif
472 if (DOINGSOFTDEP(vp)) {
473 if (sbap == &ip->i_db[0] && i < ssize)
474 softdep_setup_allocdirect(ip, start_lbn + i,
475 blkno, *bap, fs->fs_bsize, fs->fs_bsize,
476 buflist->bs_children[i]);
477 else
478 softdep_setup_allocindir_page(ip, start_lbn + i,
479 i < ssize ? sbp : ebp, soff + i, blkno,
480 *bap, buflist->bs_children[i]);
481 }
482 *bap++ = blkno;
483 }
484 /*
485 * Next we must write out the modified inode and indirect blocks.
486 * For strict correctness, the writes should be synchronous since
487 * the old block values may have been written to disk. In practise
488 * they are almost never written, but if we are concerned about
489 * strict correctness, the `doasyncfree' flag should be set to zero.
490 *
491 * The test on `doasyncfree' should be changed to test a flag
492 * that shows whether the associated buffers and inodes have
493 * been written. The flag should be set when the cluster is
494 * started and cleared whenever the buffer or inode is flushed.
495 * We can then check below to see if it is set, and do the
496 * synchronous write only when it has been cleared.
497 */
498 if (sbap != &ip->i_db[0]) {
499 if (doasyncfree)
500 bdwrite(sbp);
501 else
502 bwrite(sbp);
503 } else {
504 ip->i_flag |= IN_CHANGE | IN_UPDATE;
505 if (!doasyncfree)
506 UFS_UPDATE(vp, 1);
507 }
508 if (ssize < len) {
509 if (doasyncfree)
510 bdwrite(ebp);
511 else
512 bwrite(ebp);
513 }
514 /*
515 * Last, free the old blocks and assign the new blocks to the buffers.
516 */
517#ifdef DEBUG
518 if (prtrealloc)
519 printf("\n\tnew:");
520#endif
521 for (blkno = newblk, i = 0; i < len; i++, blkno += fs->fs_frag) {
522 if (!DOINGSOFTDEP(vp))
523 ffs_blkfree(ip,
524 dbtofsb(fs, buflist->bs_children[i]->b_blkno),
525 fs->fs_bsize);
526 buflist->bs_children[i]->b_blkno = fsbtodb(fs, blkno);
527#ifdef DEBUG
528 if (!ffs_checkblk(ip,
529 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
530 panic("ffs_reallocblks: unallocated block 3");
531 if (prtrealloc)
532 printf(" %d,", blkno);
533#endif
534 }
535#ifdef DEBUG
536 if (prtrealloc) {
537 prtrealloc--;
538 printf("\n");
539 }
540#endif
541 return (0);
542
543fail:
544 if (ssize < len)
545 brelse(ebp);
546 if (sbap != &ip->i_db[0])
547 brelse(sbp);
548 return (ENOSPC);
549}
550
551/*
552 * Allocate an inode in the file system.
553 *
554 * If allocating a directory, use ffs_dirpref to select the inode.
555 * If allocating in a directory, the following hierarchy is followed:
556 * 1) allocate the preferred inode.
557 * 2) allocate an inode in the same cylinder group.
558 * 3) quadradically rehash into other cylinder groups, until an
559 * available inode is located.
560 * If no inode preference is given the following heirarchy is used
561 * to allocate an inode:
562 * 1) allocate an inode in cylinder group 0.
563 * 2) quadradically rehash into other cylinder groups, until an
564 * available inode is located.
565 */
566int
567ffs_valloc(pvp, mode, cred, vpp)
568 struct vnode *pvp;
569 int mode;
570 struct ucred *cred;
571 struct vnode **vpp;
572{
573 register struct inode *pip;
574 register struct fs *fs;
575 register struct inode *ip;
576 ino_t ino, ipref;
577 int cg, error;
578
579 *vpp = NULL;
580 pip = VTOI(pvp);
581 fs = pip->i_fs;
582 if (fs->fs_cstotal.cs_nifree == 0)
583 goto noinodes;
584
585 if ((mode & IFMT) == IFDIR)
586 ipref = ffs_dirpref(fs);
587 else
588 ipref = pip->i_number;
589 if (ipref >= fs->fs_ncg * fs->fs_ipg)
590 ipref = 0;
591 cg = ino_to_cg(fs, ipref);
592 ino = (ino_t)ffs_hashalloc(pip, cg, (long)ipref, mode,
593 (allocfcn_t *)ffs_nodealloccg);
594 if (ino == 0)
595 goto noinodes;
596 error = VFS_VGET(pvp->v_mount, ino, vpp);
597 if (error) {
598 UFS_VFREE(pvp, ino, mode);
599 return (error);
600 }
601 ip = VTOI(*vpp);
602 if (ip->i_mode) {
603 printf("mode = 0%o, inum = %lu, fs = %s\n",
604 ip->i_mode, (u_long)ip->i_number, fs->fs_fsmnt);
605 panic("ffs_valloc: dup alloc");
606 }
607 if (ip->i_blocks) { /* XXX */
608 printf("free inode %s/%lu had %ld blocks\n",
609 fs->fs_fsmnt, (u_long)ino, (long)ip->i_blocks);
610 ip->i_blocks = 0;
611 }
612 ip->i_flags = 0;
613 /*
614 * Set up a new generation number for this inode.
615 */
616 if (ip->i_gen == 0 || ++ip->i_gen == 0)
617 ip->i_gen = random() / 2 + 1;
618 return (0);
619noinodes:
620 ffs_fserr(fs, cred->cr_uid, "out of inodes");
621 uprintf("\n%s: create/symlink failed, no inodes free\n", fs->fs_fsmnt);
622 return (ENOSPC);
623}
624
625/*
626 * Find a cylinder to place a directory.
627 *
628 * The policy implemented by this algorithm is to select from
629 * among those cylinder groups with above the average number of
630 * free inodes, the one with the smallest number of directories.
631 */
632static ino_t
633ffs_dirpref(fs)
634 register struct fs *fs;
635{
636 int cg, minndir, mincg, avgifree;
637
638 avgifree = fs->fs_cstotal.cs_nifree / fs->fs_ncg;
639 minndir = fs->fs_ipg;
640 mincg = 0;
641 for (cg = 0; cg < fs->fs_ncg; cg++)
642 if (fs->fs_cs(fs, cg).cs_ndir < minndir &&
643 fs->fs_cs(fs, cg).cs_nifree >= avgifree) {
644 mincg = cg;
645 minndir = fs->fs_cs(fs, cg).cs_ndir;
646 }
647 return ((ino_t)(fs->fs_ipg * mincg));
648}
649
650/*
651 * Select the desired position for the next block in a file. The file is
652 * logically divided into sections. The first section is composed of the
653 * direct blocks. Each additional section contains fs_maxbpg blocks.
654 *
655 * If no blocks have been allocated in the first section, the policy is to
656 * request a block in the same cylinder group as the inode that describes
657 * the file. If no blocks have been allocated in any other section, the
658 * policy is to place the section in a cylinder group with a greater than
659 * average number of free blocks. An appropriate cylinder group is found
660 * by using a rotor that sweeps the cylinder groups. When a new group of
661 * blocks is needed, the sweep begins in the cylinder group following the
662 * cylinder group from which the previous allocation was made. The sweep
663 * continues until a cylinder group with greater than the average number
664 * of free blocks is found. If the allocation is for the first block in an
665 * indirect block, the information on the previous allocation is unavailable;
666 * here a best guess is made based upon the logical block number being
667 * allocated.
668 *
669 * If a section is already partially allocated, the policy is to
670 * contiguously allocate fs_maxcontig blocks. The end of one of these
671 * contiguous blocks and the beginning of the next is physically separated
672 * so that the disk head will be in transit between them for at least
673 * fs_rotdelay milliseconds. This is to allow time for the processor to
674 * schedule another I/O transfer.
675 */
676ufs_daddr_t
677ffs_blkpref(ip, lbn, indx, bap)
678 struct inode *ip;
679 ufs_daddr_t lbn;
680 int indx;
681 ufs_daddr_t *bap;
682{
683 register struct fs *fs;
684 register int cg;
685 int avgbfree, startcg;
686 ufs_daddr_t nextblk;
687
688 fs = ip->i_fs;
689 if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) {
690 if (lbn < NDADDR) {
691 cg = ino_to_cg(fs, ip->i_number);
692 return (fs->fs_fpg * cg + fs->fs_frag);
693 }
694 /*
695 * Find a cylinder with greater than average number of
696 * unused data blocks.
697 */
698 if (indx == 0 || bap[indx - 1] == 0)
699 startcg =
700 ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg;
701 else
702 startcg = dtog(fs, bap[indx - 1]) + 1;
703 startcg %= fs->fs_ncg;
704 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
705 for (cg = startcg; cg < fs->fs_ncg; cg++)
706 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
707 fs->fs_cgrotor = cg;
708 return (fs->fs_fpg * cg + fs->fs_frag);
709 }
710 for (cg = 0; cg <= startcg; cg++)
711 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
712 fs->fs_cgrotor = cg;
713 return (fs->fs_fpg * cg + fs->fs_frag);
714 }
715 return (0);
716 }
717 /*
718 * One or more previous blocks have been laid out. If less
719 * than fs_maxcontig previous blocks are contiguous, the
720 * next block is requested contiguously, otherwise it is
721 * requested rotationally delayed by fs_rotdelay milliseconds.
722 */
723 nextblk = bap[indx - 1] + fs->fs_frag;
724 if (fs->fs_rotdelay == 0 || indx < fs->fs_maxcontig ||
725 bap[indx - fs->fs_maxcontig] +
726 blkstofrags(fs, fs->fs_maxcontig) != nextblk)
727 return (nextblk);
728 /*
729 * Here we convert ms of delay to frags as:
730 * (frags) = (ms) * (rev/sec) * (sect/rev) /
731 * ((sect/frag) * (ms/sec))
732 * then round up to the next block.
733 */
734 nextblk += roundup(fs->fs_rotdelay * fs->fs_rps * fs->fs_nsect /
735 (NSPF(fs) * 1000), fs->fs_frag);
736 return (nextblk);
737}
738
739/*
740 * Implement the cylinder overflow algorithm.
741 *
742 * The policy implemented by this algorithm is:
743 * 1) allocate the block in its requested cylinder group.
744 * 2) quadradically rehash on the cylinder group number.
745 * 3) brute force search for a free block.
746 */
747/*VARARGS5*/
748static u_long
749ffs_hashalloc(ip, cg, pref, size, allocator)
750 struct inode *ip;
751 int cg;
752 long pref;
753 int size; /* size for data blocks, mode for inodes */
754 allocfcn_t *allocator;
755{
756 register struct fs *fs;
757 long result; /* XXX why not same type as we return? */
758 int i, icg = cg;
759
760 fs = ip->i_fs;
761 /*
762 * 1: preferred cylinder group
763 */
764 result = (*allocator)(ip, cg, pref, size);
765 if (result)
766 return (result);
767 /*
768 * 2: quadratic rehash
769 */
770 for (i = 1; i < fs->fs_ncg; i *= 2) {
771 cg += i;
772 if (cg >= fs->fs_ncg)
773 cg -= fs->fs_ncg;
774 result = (*allocator)(ip, cg, 0, size);
775 if (result)
776 return (result);
777 }
778 /*
779 * 3: brute force search
780 * Note that we start at i == 2, since 0 was checked initially,
781 * and 1 is always checked in the quadratic rehash.
782 */
783 cg = (icg + 2) % fs->fs_ncg;
784 for (i = 2; i < fs->fs_ncg; i++) {
785 result = (*allocator)(ip, cg, 0, size);
786 if (result)
787 return (result);
788 cg++;
789 if (cg == fs->fs_ncg)
790 cg = 0;
791 }
792 return (0);
793}
794
795/*
796 * Determine whether a fragment can be extended.
797 *
798 * Check to see if the necessary fragments are available, and
799 * if they are, allocate them.
800 */
801static ufs_daddr_t
802ffs_fragextend(ip, cg, bprev, osize, nsize)
803 struct inode *ip;
804 int cg;
805 long bprev;
806 int osize, nsize;
807{
808 register struct fs *fs;
809 register struct cg *cgp;
810 struct buf *bp;
811 long bno;
812 int frags, bbase;
813 int i, error;
814
815 fs = ip->i_fs;
816 if (fs->fs_cs(fs, cg).cs_nffree < numfrags(fs, nsize - osize))
817 return (0);
818 frags = numfrags(fs, nsize);
819 bbase = fragnum(fs, bprev);
820 if (bbase > fragnum(fs, (bprev + frags - 1))) {
821 /* cannot extend across a block boundary */
822 return (0);
823 }
824 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
825 (int)fs->fs_cgsize, NOCRED, &bp);
826 if (error) {
827 brelse(bp);
828 return (0);
829 }
830 cgp = (struct cg *)bp->b_data;
831 if (!cg_chkmagic(cgp)) {
832 brelse(bp);
833 return (0);
834 }
835 cgp->cg_time = time_second;
836 bno = dtogd(fs, bprev);
837 for (i = numfrags(fs, osize); i < frags; i++)
838 if (isclr(cg_blksfree(cgp), bno + i)) {
839 brelse(bp);
840 return (0);
841 }
842 /*
843 * the current fragment can be extended
844 * deduct the count on fragment being extended into
845 * increase the count on the remaining fragment (if any)
846 * allocate the extended piece
847 */
848 for (i = frags; i < fs->fs_frag - bbase; i++)
849 if (isclr(cg_blksfree(cgp), bno + i))
850 break;
851 cgp->cg_frsum[i - numfrags(fs, osize)]--;
852 if (i != frags)
853 cgp->cg_frsum[i - frags]++;
854 for (i = numfrags(fs, osize); i < frags; i++) {
855 clrbit(cg_blksfree(cgp), bno + i);
856 cgp->cg_cs.cs_nffree--;
857 fs->fs_cstotal.cs_nffree--;
858 fs->fs_cs(fs, cg).cs_nffree--;
859 }
860 fs->fs_fmod = 1;
861 if (DOINGSOFTDEP(ITOV(ip)))
862 softdep_setup_blkmapdep(bp, fs, bprev);
863 bdwrite(bp);
864 return (bprev);
865}
866
867/*
868 * Determine whether a block can be allocated.
869 *
870 * Check to see if a block of the appropriate size is available,
871 * and if it is, allocate it.
872 */
873static ufs_daddr_t
874ffs_alloccg(ip, cg, bpref, size)
875 struct inode *ip;
876 int cg;
877 ufs_daddr_t bpref;
878 int size;
879{
880 register struct fs *fs;
881 register struct cg *cgp;
882 struct buf *bp;
883 register int i;
884 ufs_daddr_t bno, blkno;
885 int allocsiz, error, frags;
886
887 fs = ip->i_fs;
888 if (fs->fs_cs(fs, cg).cs_nbfree == 0 && size == fs->fs_bsize)
889 return (0);
890 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
891 (int)fs->fs_cgsize, NOCRED, &bp);
892 if (error) {
893 brelse(bp);
894 return (0);
895 }
896 cgp = (struct cg *)bp->b_data;
897 if (!cg_chkmagic(cgp) ||
898 (cgp->cg_cs.cs_nbfree == 0 && size == fs->fs_bsize)) {
899 brelse(bp);
900 return (0);
901 }
902 cgp->cg_time = time_second;
903 if (size == fs->fs_bsize) {
904 bno = ffs_alloccgblk(ip, bp, bpref);
905 bdwrite(bp);
906 return (bno);
907 }
908 /*
909 * check to see if any fragments are already available
910 * allocsiz is the size which will be allocated, hacking
911 * it down to a smaller size if necessary
912 */
913 frags = numfrags(fs, size);
914 for (allocsiz = frags; allocsiz < fs->fs_frag; allocsiz++)
915 if (cgp->cg_frsum[allocsiz] != 0)
916 break;
917 if (allocsiz == fs->fs_frag) {
918 /*
919 * no fragments were available, so a block will be
920 * allocated, and hacked up
921 */
922 if (cgp->cg_cs.cs_nbfree == 0) {
923 brelse(bp);
924 return (0);
925 }
926 bno = ffs_alloccgblk(ip, bp, bpref);
927 bpref = dtogd(fs, bno);
928 for (i = frags; i < fs->fs_frag; i++)
929 setbit(cg_blksfree(cgp), bpref + i);
930 i = fs->fs_frag - frags;
931 cgp->cg_cs.cs_nffree += i;
932 fs->fs_cstotal.cs_nffree += i;
933 fs->fs_cs(fs, cg).cs_nffree += i;
934 fs->fs_fmod = 1;
935 cgp->cg_frsum[i]++;
936 bdwrite(bp);
937 return (bno);
938 }
939 bno = ffs_mapsearch(fs, cgp, bpref, allocsiz);
940 if (bno < 0) {
941 brelse(bp);
942 return (0);
943 }
944 for (i = 0; i < frags; i++)
945 clrbit(cg_blksfree(cgp), bno + i);
946 cgp->cg_cs.cs_nffree -= frags;
947 fs->fs_cstotal.cs_nffree -= frags;
948 fs->fs_cs(fs, cg).cs_nffree -= frags;
949 fs->fs_fmod = 1;
950 cgp->cg_frsum[allocsiz]--;
951 if (frags != allocsiz)
952 cgp->cg_frsum[allocsiz - frags]++;
953 blkno = cg * fs->fs_fpg + bno;
954 if (DOINGSOFTDEP(ITOV(ip)))
955 softdep_setup_blkmapdep(bp, fs, blkno);
956 bdwrite(bp);
957 return ((u_long)blkno);
958}
959
960/*
961 * Allocate a block in a cylinder group.
962 *
963 * This algorithm implements the following policy:
964 * 1) allocate the requested block.
965 * 2) allocate a rotationally optimal block in the same cylinder.
966 * 3) allocate the next available block on the block rotor for the
967 * specified cylinder group.
968 * Note that this routine only allocates fs_bsize blocks; these
969 * blocks may be fragmented by the routine that allocates them.
970 */
971static ufs_daddr_t
972ffs_alloccgblk(ip, bp, bpref)
973 struct inode *ip;
974 struct buf *bp;
975 ufs_daddr_t bpref;
976{
977 struct fs *fs;
978 struct cg *cgp;
979 ufs_daddr_t bno, blkno;
980 int cylno, pos, delta;
981 short *cylbp;
982 register int i;
983
984 fs = ip->i_fs;
985 cgp = (struct cg *)bp->b_data;
986 if (bpref == 0 || dtog(fs, bpref) != cgp->cg_cgx) {
987 bpref = cgp->cg_rotor;
988 goto norot;
989 }
990 bpref = blknum(fs, bpref);
991 bpref = dtogd(fs, bpref);
992 /*
993 * if the requested block is available, use it
994 */
995 if (ffs_isblock(fs, cg_blksfree(cgp), fragstoblks(fs, bpref))) {
996 bno = bpref;
997 goto gotit;
998 }
999 if (fs->fs_nrpos <= 1 || fs->fs_cpc == 0) {
1000 /*
1001 * Block layout information is not available.
1002 * Leaving bpref unchanged means we take the
1003 * next available free block following the one
1004 * we just allocated. Hopefully this will at
1005 * least hit a track cache on drives of unknown
1006 * geometry (e.g. SCSI).
1007 */
1008 goto norot;
1009 }
1010 /*
1011 * check for a block available on the same cylinder
1012 */
1013 cylno = cbtocylno(fs, bpref);
1014 if (cg_blktot(cgp)[cylno] == 0)
1015 goto norot;
1016 /*
1017 * check the summary information to see if a block is
1018 * available in the requested cylinder starting at the
1019 * requested rotational position and proceeding around.
1020 */
1021 cylbp = cg_blks(fs, cgp, cylno);
1022 pos = cbtorpos(fs, bpref);
1023 for (i = pos; i < fs->fs_nrpos; i++)
1024 if (cylbp[i] > 0)
1025 break;
1026 if (i == fs->fs_nrpos)
1027 for (i = 0; i < pos; i++)
1028 if (cylbp[i] > 0)
1029 break;
1030 if (cylbp[i] > 0) {
1031 /*
1032 * found a rotational position, now find the actual
1033 * block. A panic if none is actually there.
1034 */
1035 pos = cylno % fs->fs_cpc;
1036 bno = (cylno - pos) * fs->fs_spc / NSPB(fs);
1037 if (fs_postbl(fs, pos)[i] == -1) {
1038 printf("pos = %d, i = %d, fs = %s\n",
1039 pos, i, fs->fs_fsmnt);
1040 panic("ffs_alloccgblk: cyl groups corrupted");
1041 }
1042 for (i = fs_postbl(fs, pos)[i];; ) {
1043 if (ffs_isblock(fs, cg_blksfree(cgp), bno + i)) {
1044 bno = blkstofrags(fs, (bno + i));
1045 goto gotit;
1046 }
1047 delta = fs_rotbl(fs)[i];
1048 if (delta <= 0 ||
1049 delta + i > fragstoblks(fs, fs->fs_fpg))
1050 break;
1051 i += delta;
1052 }
1053 printf("pos = %d, i = %d, fs = %s\n", pos, i, fs->fs_fsmnt);
1054 panic("ffs_alloccgblk: can't find blk in cyl");
1055 }
1056norot:
1057 /*
1058 * no blocks in the requested cylinder, so take next
1059 * available one in this cylinder group.
1060 */
1061 bno = ffs_mapsearch(fs, cgp, bpref, (int)fs->fs_frag);
1062 if (bno < 0)
1063 return (0);
1064 cgp->cg_rotor = bno;
1065gotit:
1066 blkno = fragstoblks(fs, bno);
1067 ffs_clrblock(fs, cg_blksfree(cgp), (long)blkno);
1068 ffs_clusteracct(fs, cgp, blkno, -1);
1069 cgp->cg_cs.cs_nbfree--;
1070 fs->fs_cstotal.cs_nbfree--;
1071 fs->fs_cs(fs, cgp->cg_cgx).cs_nbfree--;
1072 cylno = cbtocylno(fs, bno);
1073 cg_blks(fs, cgp, cylno)[cbtorpos(fs, bno)]--;
1074 cg_blktot(cgp)[cylno]--;
1075 fs->fs_fmod = 1;
1076 blkno = cgp->cg_cgx * fs->fs_fpg + bno;
1077 if (DOINGSOFTDEP(ITOV(ip)))
1078 softdep_setup_blkmapdep(bp, fs, blkno);
1079 return (blkno);
1080}
1081
1082/*
1083 * Determine whether a cluster can be allocated.
1084 *
1085 * We do not currently check for optimal rotational layout if there
1086 * are multiple choices in the same cylinder group. Instead we just
1087 * take the first one that we find following bpref.
1088 */
1089static ufs_daddr_t
1090ffs_clusteralloc(ip, cg, bpref, len)
1091 struct inode *ip;
1092 int cg;
1093 ufs_daddr_t bpref;
1094 int len;
1095{
1096 register struct fs *fs;
1097 register struct cg *cgp;
1098 struct buf *bp;
1099 int i, got, run, bno, bit, map;
1100 u_char *mapp;
1101 int32_t *lp;
1102
1103 fs = ip->i_fs;
1104 if (fs->fs_maxcluster[cg] < len)
1105 return (NULL);
1106 if (bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), (int)fs->fs_cgsize,
1107 NOCRED, &bp))
1108 goto fail;
1109 cgp = (struct cg *)bp->b_data;
1110 if (!cg_chkmagic(cgp))
1111 goto fail;
1112 /*
1113 * Check to see if a cluster of the needed size (or bigger) is
1114 * available in this cylinder group.
1115 */
1116 lp = &cg_clustersum(cgp)[len];
1117 for (i = len; i <= fs->fs_contigsumsize; i++)
1118 if (*lp++ > 0)
1119 break;
1120 if (i > fs->fs_contigsumsize) {
1121 /*
1122 * This is the first time looking for a cluster in this
1123 * cylinder group. Update the cluster summary information
1124 * to reflect the true maximum sized cluster so that
1125 * future cluster allocation requests can avoid reading
1126 * the cylinder group map only to find no clusters.
1127 */
1128 lp = &cg_clustersum(cgp)[len - 1];
1129 for (i = len - 1; i > 0; i--)
1130 if (*lp-- > 0)
1131 break;
1132 fs->fs_maxcluster[cg] = i;
1133 goto fail;
1134 }
1135 /*
1136 * Search the cluster map to find a big enough cluster.
1137 * We take the first one that we find, even if it is larger
1138 * than we need as we prefer to get one close to the previous
1139 * block allocation. We do not search before the current
1140 * preference point as we do not want to allocate a block
1141 * that is allocated before the previous one (as we will
1142 * then have to wait for another pass of the elevator
1143 * algorithm before it will be read). We prefer to fail and
1144 * be recalled to try an allocation in the next cylinder group.
1145 */
1146 if (dtog(fs, bpref) != cg)
1147 bpref = 0;
1148 else
1149 bpref = fragstoblks(fs, dtogd(fs, blknum(fs, bpref)));
1150 mapp = &cg_clustersfree(cgp)[bpref / NBBY];
1151 map = *mapp++;
1152 bit = 1 << (bpref % NBBY);
1153 for (run = 0, got = bpref; got < cgp->cg_nclusterblks; got++) {
1154 if ((map & bit) == 0) {
1155 run = 0;
1156 } else {
1157 run++;
1158 if (run == len)
1159 break;
1160 }
1161 if ((got & (NBBY - 1)) != (NBBY - 1)) {
1162 bit <<= 1;
1163 } else {
1164 map = *mapp++;
1165 bit = 1;
1166 }
1167 }
1168 if (got >= cgp->cg_nclusterblks)
1169 goto fail;
1170 /*
1171 * Allocate the cluster that we have found.
1172 */
1173 for (i = 1; i <= len; i++)
1174 if (!ffs_isblock(fs, cg_blksfree(cgp), got - run + i))
1175 panic("ffs_clusteralloc: map mismatch");
1176 bno = cg * fs->fs_fpg + blkstofrags(fs, got - run + 1);
1177 if (dtog(fs, bno) != cg)
1178 panic("ffs_clusteralloc: allocated out of group");
1179 len = blkstofrags(fs, len);
1180 for (i = 0; i < len; i += fs->fs_frag)
1181 if ((got = ffs_alloccgblk(ip, bp, bno + i)) != bno + i)
1182 panic("ffs_clusteralloc: lost block");
1183 bdwrite(bp);
1184 return (bno);
1185
1186fail:
1187 brelse(bp);
1188 return (0);
1189}
1190
1191/*
1192 * Determine whether an inode can be allocated.
1193 *
1194 * Check to see if an inode is available, and if it is,
1195 * allocate it using the following policy:
1196 * 1) allocate the requested inode.
1197 * 2) allocate the next available inode after the requested
1198 * inode in the specified cylinder group.
1199 */
1200static ino_t
1201ffs_nodealloccg(ip, cg, ipref, mode)
1202 struct inode *ip;
1203 int cg;
1204 ufs_daddr_t ipref;
1205 int mode;
1206{
1207 register struct fs *fs;
1208 register struct cg *cgp;
1209 struct buf *bp;
1210 int error, start, len, loc, map, i;
1211
1212 fs = ip->i_fs;
1213 if (fs->fs_cs(fs, cg).cs_nifree == 0)
1214 return (0);
1215 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
1216 (int)fs->fs_cgsize, NOCRED, &bp);
1217 if (error) {
1218 brelse(bp);
1219 return (0);
1220 }
1221 cgp = (struct cg *)bp->b_data;
1222 if (!cg_chkmagic(cgp) || cgp->cg_cs.cs_nifree == 0) {
1223 brelse(bp);
1224 return (0);
1225 }
1226 cgp->cg_time = time_second;
1227 if (ipref) {
1228 ipref %= fs->fs_ipg;
1229 if (isclr(cg_inosused(cgp), ipref))
1230 goto gotit;
1231 }
1232 start = cgp->cg_irotor / NBBY;
1233 len = howmany(fs->fs_ipg - cgp->cg_irotor, NBBY);
1234 loc = skpc(0xff, len, &cg_inosused(cgp)[start]);
1235 if (loc == 0) {
1236 len = start + 1;
1237 start = 0;
1238 loc = skpc(0xff, len, &cg_inosused(cgp)[0]);
1239 if (loc == 0) {
1240 printf("cg = %d, irotor = %ld, fs = %s\n",
1241 cg, (long)cgp->cg_irotor, fs->fs_fsmnt);
1242 panic("ffs_nodealloccg: map corrupted");
1243 /* NOTREACHED */
1244 }
1245 }
1246 i = start + len - loc;
1247 map = cg_inosused(cgp)[i];
1248 ipref = i * NBBY;
1249 for (i = 1; i < (1 << NBBY); i <<= 1, ipref++) {
1250 if ((map & i) == 0) {
1251 cgp->cg_irotor = ipref;
1252 goto gotit;
1253 }
1254 }
1255 printf("fs = %s\n", fs->fs_fsmnt);
1256 panic("ffs_nodealloccg: block not in map");
1257 /* NOTREACHED */
1258gotit:
1259 if (DOINGSOFTDEP(ITOV(ip)))
1260 softdep_setup_inomapdep(bp, ip, cg * fs->fs_ipg + ipref);
1261 setbit(cg_inosused(cgp), ipref);
1262 cgp->cg_cs.cs_nifree--;
1263 fs->fs_cstotal.cs_nifree--;
1264 fs->fs_cs(fs, cg).cs_nifree--;
1265 fs->fs_fmod = 1;
1266 if ((mode & IFMT) == IFDIR) {
1267 cgp->cg_cs.cs_ndir++;
1268 fs->fs_cstotal.cs_ndir++;
1269 fs->fs_cs(fs, cg).cs_ndir++;
1270 }
1271 bdwrite(bp);
1272 return (cg * fs->fs_ipg + ipref);
1273}
1274
1275/*
1276 * Free a block or fragment.
1277 *
1278 * The specified block or fragment is placed back in the
1279 * free map. If a fragment is deallocated, a possible
1280 * block reassembly is checked.
1281 */
1282void
1283ffs_blkfree(ip, bno, size)
1284 register struct inode *ip;
1285 ufs_daddr_t bno;
1286 long size;
1287{
1288 register struct fs *fs;
1289 register struct cg *cgp;
1290 struct buf *bp;
1291 ufs_daddr_t blkno;
1292 int i, error, cg, blk, frags, bbase;
1293
1294 fs = ip->i_fs;
1295 VOP_FREEBLKS(ip->i_devvp, fsbtodb(fs, bno), size);
1296 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0 ||
1297 fragnum(fs, bno) + numfrags(fs, size) > fs->fs_frag) {
1298 printf("dev=0x%lx, bno = %d, bsize = %d, size = %ld, fs = %s\n",
1299 (u_long)ip->i_dev, bno, fs->fs_bsize, size, fs->fs_fsmnt);
1300 panic("ffs_blkfree: bad size");
1301 }
1302 cg = dtog(fs, bno);
1303 if ((u_int)bno >= fs->fs_size) {
1304 printf("bad block %ld, ino %lu\n",
1305 (long)bno, (u_long)ip->i_number);
1306 ffs_fserr(fs, ip->i_uid, "bad block");
1307 return;
1308 }
1309 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
1310 (int)fs->fs_cgsize, NOCRED, &bp);
1311 if (error) {
1312 brelse(bp);
1313 return;
1314 }
1315 cgp = (struct cg *)bp->b_data;
1316 if (!cg_chkmagic(cgp)) {
1317 brelse(bp);
1318 return;
1319 }
1320 cgp->cg_time = time_second;
1321 bno = dtogd(fs, bno);
1322 if (size == fs->fs_bsize) {
1323 blkno = fragstoblks(fs, bno);
1324 if (!ffs_isfreeblock(fs, cg_blksfree(cgp), blkno)) {
1325 printf("dev = 0x%lx, block = %ld, fs = %s\n",
1326 (u_long)ip->i_dev, (long)bno, fs->fs_fsmnt);
1327 panic("ffs_blkfree: freeing free block");
1328 }
1329 ffs_setblock(fs, cg_blksfree(cgp), blkno);
1330 ffs_clusteracct(fs, cgp, blkno, 1);
1331 cgp->cg_cs.cs_nbfree++;
1332 fs->fs_cstotal.cs_nbfree++;
1333 fs->fs_cs(fs, cg).cs_nbfree++;
1334 i = cbtocylno(fs, bno);
1335 cg_blks(fs, cgp, i)[cbtorpos(fs, bno)]++;
1336 cg_blktot(cgp)[i]++;
1337 } else {
1338 bbase = bno - fragnum(fs, bno);
1339 /*
1340 * decrement the counts associated with the old frags
1341 */
1342 blk = blkmap(fs, cg_blksfree(cgp), bbase);
1343 ffs_fragacct(fs, blk, cgp->cg_frsum, -1);
1344 /*
1345 * deallocate the fragment
1346 */
1347 frags = numfrags(fs, size);
1348 for (i = 0; i < frags; i++) {
1349 if (isset(cg_blksfree(cgp), bno + i)) {
1350 printf("dev = 0x%lx, block = %ld, fs = %s\n",
1351 (u_long)ip->i_dev, (long)(bno + i),
1352 fs->fs_fsmnt);
1353 panic("ffs_blkfree: freeing free frag");
1354 }
1355 setbit(cg_blksfree(cgp), bno + i);
1356 }
1357 cgp->cg_cs.cs_nffree += i;
1358 fs->fs_cstotal.cs_nffree += i;
1359 fs->fs_cs(fs, cg).cs_nffree += i;
1360 /*
1361 * add back in counts associated with the new frags
1362 */
1363 blk = blkmap(fs, cg_blksfree(cgp), bbase);
1364 ffs_fragacct(fs, blk, cgp->cg_frsum, 1);
1365 /*
1366 * if a complete block has been reassembled, account for it
1367 */
1368 blkno = fragstoblks(fs, bbase);
1369 if (ffs_isblock(fs, cg_blksfree(cgp), blkno)) {
1370 cgp->cg_cs.cs_nffree -= fs->fs_frag;
1371 fs->fs_cstotal.cs_nffree -= fs->fs_frag;
1372 fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag;
1373 ffs_clusteracct(fs, cgp, blkno, 1);
1374 cgp->cg_cs.cs_nbfree++;
1375 fs->fs_cstotal.cs_nbfree++;
1376 fs->fs_cs(fs, cg).cs_nbfree++;
1377 i = cbtocylno(fs, bbase);
1378 cg_blks(fs, cgp, i)[cbtorpos(fs, bbase)]++;
1379 cg_blktot(cgp)[i]++;
1380 }
1381 }
1382 fs->fs_fmod = 1;
1383 bdwrite(bp);
1384}
1385
1386#ifdef DIAGNOSTIC
1387/*
1388 * Verify allocation of a block or fragment. Returns true if block or
1389 * fragment is allocated, false if it is free.
1390 */
1391static int
1392ffs_checkblk(ip, bno, size)
1393 struct inode *ip;
1394 ufs_daddr_t bno;
1395 long size;
1396{
1397 struct fs *fs;
1398 struct cg *cgp;
1399 struct buf *bp;
1400 int i, error, frags, free;
1401
1402 fs = ip->i_fs;
1403 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) {
1404 printf("bsize = %ld, size = %ld, fs = %s\n",
1405 (long)fs->fs_bsize, size, fs->fs_fsmnt);
1406 panic("ffs_checkblk: bad size");
1407 }
1408 if ((u_int)bno >= fs->fs_size)
1409 panic("ffs_checkblk: bad block %d", bno);
1410 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, dtog(fs, bno))),
1411 (int)fs->fs_cgsize, NOCRED, &bp);
1412 if (error)
1413 panic("ffs_checkblk: cg bread failed");
1414 cgp = (struct cg *)bp->b_data;
1415 if (!cg_chkmagic(cgp))
1416 panic("ffs_checkblk: cg magic mismatch");
1417 bno = dtogd(fs, bno);
1418 if (size == fs->fs_bsize) {
1419 free = ffs_isblock(fs, cg_blksfree(cgp), fragstoblks(fs, bno));
1420 } else {
1421 frags = numfrags(fs, size);
1422 for (free = 0, i = 0; i < frags; i++)
1423 if (isset(cg_blksfree(cgp), bno + i))
1424 free++;
1425 if (free != 0 && free != frags)
1426 panic("ffs_checkblk: partially free fragment");
1427 }
1428 brelse(bp);
1429 return (!free);
1430}
1431#endif /* DIAGNOSTIC */
1432
1433/*
1434 * Free an inode.
1435 */
1436int
1437ffs_vfree( pvp, ino, mode)
1438 struct vnode *pvp;
1439 ino_t ino;
1440 int mode;
1441{
1442 if (DOINGSOFTDEP(pvp)) {
1443 softdep_freefile(pvp, ino, mode);
1444 return (0);
1445 }
1446 return (ffs_freefile(pvp, ino, mode));
1447}
1448
1449/*
1450 * Do the actual free operation.
1451 * The specified inode is placed back in the free map.
1452 */
1453 int
1454 ffs_freefile( pvp, ino, mode)
1455 struct vnode *pvp;
1456 ino_t ino;
1457 int mode;
1458{
1459 register struct fs *fs;
1460 register struct cg *cgp;
1461 register struct inode *pip;
1462 struct buf *bp;
1463 int error, cg;
1464
1465 pip = VTOI(pvp);
1466 fs = pip->i_fs;
1467 if ((u_int)ino >= fs->fs_ipg * fs->fs_ncg)
1468 panic("ffs_vfree: range: dev = 0x%x, ino = %d, fs = %s",
1469 pip->i_dev, ino, fs->fs_fsmnt);
1470 cg = ino_to_cg(fs, ino);
1471 error = bread(pip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
1472 (int)fs->fs_cgsize, NOCRED, &bp);
1473 if (error) {
1474 brelse(bp);
1475 return (error);
1476 }
1477 cgp = (struct cg *)bp->b_data;
1478 if (!cg_chkmagic(cgp)) {
1479 brelse(bp);
1480 return (0);
1481 }
1482 cgp->cg_time = time_second;
1483 ino %= fs->fs_ipg;
1484 if (isclr(cg_inosused(cgp), ino)) {
1485 printf("dev = 0x%lx, ino = %lu, fs = %s\n",
1486 (u_long)pip->i_dev, (u_long)ino, fs->fs_fsmnt);
1487 if (fs->fs_ronly == 0)
1488 panic("ffs_vfree: freeing free inode");
1489 }
1490 clrbit(cg_inosused(cgp), ino);
1491 if (ino < cgp->cg_irotor)
1492 cgp->cg_irotor = ino;
1493 cgp->cg_cs.cs_nifree++;
1494 fs->fs_cstotal.cs_nifree++;
1495 fs->fs_cs(fs, cg).cs_nifree++;
1496 if ((mode & IFMT) == IFDIR) {
1497 cgp->cg_cs.cs_ndir--;
1498 fs->fs_cstotal.cs_ndir--;
1499 fs->fs_cs(fs, cg).cs_ndir--;
1500 }
1501 fs->fs_fmod = 1;
1502 bdwrite(bp);
1503 return (0);
1504}
1505
1506/*
1507 * Find a block of the specified size in the specified cylinder group.
1508 *
1509 * It is a panic if a request is made to find a block if none are
1510 * available.
1511 */
1512static ufs_daddr_t
1513ffs_mapsearch(fs, cgp, bpref, allocsiz)
1514 register struct fs *fs;
1515 register struct cg *cgp;
1516 ufs_daddr_t bpref;
1517 int allocsiz;
1518{
1519 ufs_daddr_t bno;
1520 int start, len, loc, i;
1521 int blk, field, subfield, pos;
1522
1523 /*
1524 * find the fragment by searching through the free block
1525 * map for an appropriate bit pattern
1526 */
1527 if (bpref)
1528 start = dtogd(fs, bpref) / NBBY;
1529 else
1530 start = cgp->cg_frotor / NBBY;
1531 len = howmany(fs->fs_fpg, NBBY) - start;
1532 loc = scanc((u_int)len, (u_char *)&cg_blksfree(cgp)[start],
1533 (u_char *)fragtbl[fs->fs_frag],
1534 (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
1535 if (loc == 0) {
1536 len = start + 1;
1537 start = 0;
1538 loc = scanc((u_int)len, (u_char *)&cg_blksfree(cgp)[0],
1539 (u_char *)fragtbl[fs->fs_frag],
1540 (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
1541 if (loc == 0) {
1542 printf("start = %d, len = %d, fs = %s\n",
1543 start, len, fs->fs_fsmnt);
1544 panic("ffs_alloccg: map corrupted");
1545 /* NOTREACHED */
1546 }
1547 }
1548 bno = (start + len - loc) * NBBY;
1549 cgp->cg_frotor = bno;
1550 /*
1551 * found the byte in the map
1552 * sift through the bits to find the selected frag
1553 */
1554 for (i = bno + NBBY; bno < i; bno += fs->fs_frag) {
1555 blk = blkmap(fs, cg_blksfree(cgp), bno);
1556 blk <<= 1;
1557 field = around[allocsiz];
1558 subfield = inside[allocsiz];
1559 for (pos = 0; pos <= fs->fs_frag - allocsiz; pos++) {
1560 if ((blk & field) == subfield)
1561 return (bno + pos);
1562 field <<= 1;
1563 subfield <<= 1;
1564 }
1565 }
1566 printf("bno = %lu, fs = %s\n", (u_long)bno, fs->fs_fsmnt);
1567 panic("ffs_alloccg: block not in map");
1568 return (-1);
1569}
1570
1571/*
1572 * Update the cluster map because of an allocation or free.
1573 *
1574 * Cnt == 1 means free; cnt == -1 means allocating.
1575 */
1576static void
1577ffs_clusteracct(fs, cgp, blkno, cnt)
1578 struct fs *fs;
1579 struct cg *cgp;
1580 ufs_daddr_t blkno;
1581 int cnt;
1582{
1583 int32_t *sump;
1584 int32_t *lp;
1585 u_char *freemapp, *mapp;
1586 int i, start, end, forw, back, map, bit;
1587
1588 if (fs->fs_contigsumsize <= 0)
1589 return;
1590 freemapp = cg_clustersfree(cgp);
1591 sump = cg_clustersum(cgp);
1592 /*
1593 * Allocate or clear the actual block.
1594 */
1595 if (cnt > 0)
1596 setbit(freemapp, blkno);
1597 else
1598 clrbit(freemapp, blkno);
1599 /*
1600 * Find the size of the cluster going forward.
1601 */
1602 start = blkno + 1;
1603 end = start + fs->fs_contigsumsize;
1604 if (end >= cgp->cg_nclusterblks)
1605 end = cgp->cg_nclusterblks;
1606 mapp = &freemapp[start / NBBY];
1607 map = *mapp++;
1608 bit = 1 << (start % NBBY);
1609 for (i = start; i < end; i++) {
1610 if ((map & bit) == 0)
1611 break;
1612 if ((i & (NBBY - 1)) != (NBBY - 1)) {
1613 bit <<= 1;
1614 } else {
1615 map = *mapp++;
1616 bit = 1;
1617 }
1618 }
1619 forw = i - start;
1620 /*
1621 * Find the size of the cluster going backward.
1622 */
1623 start = blkno - 1;
1624 end = start - fs->fs_contigsumsize;
1625 if (end < 0)
1626 end = -1;
1627 mapp = &freemapp[start / NBBY];
1628 map = *mapp--;
1629 bit = 1 << (start % NBBY);
1630 for (i = start; i > end; i--) {
1631 if ((map & bit) == 0)
1632 break;
1633 if ((i & (NBBY - 1)) != 0) {
1634 bit >>= 1;
1635 } else {
1636 map = *mapp--;
1637 bit = 1 << (NBBY - 1);
1638 }
1639 }
1640 back = start - i;
1641 /*
1642 * Account for old cluster and the possibly new forward and
1643 * back clusters.
1644 */
1645 i = back + forw + 1;
1646 if (i > fs->fs_contigsumsize)
1647 i = fs->fs_contigsumsize;
1648 sump[i] += cnt;
1649 if (back > 0)
1650 sump[back] -= cnt;
1651 if (forw > 0)
1652 sump[forw] -= cnt;
1653 /*
1654 * Update cluster summary information.
1655 */
1656 lp = &sump[fs->fs_contigsumsize];
1657 for (i = fs->fs_contigsumsize; i > 0; i--)
1658 if (*lp-- > 0)
1659 break;
1660 fs->fs_maxcluster[cgp->cg_cgx] = i;
1661}
1662
1663/*
1664 * Fserr prints the name of a file system with an error diagnostic.
1665 *
1666 * The form of the error message is:
1667 * fs: error message
1668 */
1669static void
1670ffs_fserr(fs, uid, cp)
1671 struct fs *fs;
1672 u_int uid;
1673 char *cp;
1674{
1675 struct proc *p = curproc; /* XXX */
1676
1677 log(LOG_ERR, "pid %d (%s), uid %d on %s: %s\n", p ? p->p_pid : -1,
1678 p ? p->p_comm : "-", uid, fs->fs_fsmnt, cp);
1679}
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 1.56 1999/01/07 16:14:16 bde 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>
51#include <ufs/ufs/ufs_extern.h>
52#include <ufs/ufs/ufsmount.h>
53
54#include <ufs/ffs/fs.h>
55#include <ufs/ffs/ffs_extern.h>
56
57typedef ufs_daddr_t allocfcn_t __P((struct inode *ip, int cg, ufs_daddr_t bpref,
58 int size));
59
60static ufs_daddr_t ffs_alloccg __P((struct inode *, int, ufs_daddr_t, int));
61static ufs_daddr_t
62 ffs_alloccgblk __P((struct inode *, struct buf *, ufs_daddr_t));
63#ifdef DIAGNOSTIC
64static int ffs_checkblk __P((struct inode *, ufs_daddr_t, long));
65#endif
66static void ffs_clusteracct __P((struct fs *, struct cg *, ufs_daddr_t,
67 int));
68static ufs_daddr_t ffs_clusteralloc __P((struct inode *, int, ufs_daddr_t,
69 int));
70static ino_t ffs_dirpref __P((struct fs *));
71static ufs_daddr_t ffs_fragextend __P((struct inode *, int, long, int, int));
72static void ffs_fserr __P((struct fs *, u_int, char *));
73static u_long ffs_hashalloc
74 __P((struct inode *, int, long, int, allocfcn_t *));
75static ino_t ffs_nodealloccg __P((struct inode *, int, ufs_daddr_t, int));
76static ufs_daddr_t ffs_mapsearch __P((struct fs *, struct cg *, ufs_daddr_t,
77 int));
78
79/*
80 * Allocate a block in the file system.
81 *
82 * The size of the requested block is given, which must be some
83 * multiple of fs_fsize and <= fs_bsize.
84 * A preference may be optionally specified. If a preference is given
85 * the following hierarchy is used to allocate a block:
86 * 1) allocate the requested block.
87 * 2) allocate a rotationally optimal block in the same cylinder.
88 * 3) allocate a block in the same cylinder group.
89 * 4) quadradically rehash into other cylinder groups, until an
90 * available block is located.
91 * If no block preference is given the following heirarchy is used
92 * to allocate a block:
93 * 1) allocate a block in the cylinder group that contains the
94 * inode for the file.
95 * 2) quadradically rehash into other cylinder groups, until an
96 * available block is located.
97 */
98int
99ffs_alloc(ip, lbn, bpref, size, cred, bnp)
100 register struct inode *ip;
101 ufs_daddr_t lbn, bpref;
102 int size;
103 struct ucred *cred;
104 ufs_daddr_t *bnp;
105{
106 register struct fs *fs;
107 ufs_daddr_t bno;
108 int cg;
109#ifdef QUOTA
110 int error;
111#endif
112
113 *bnp = 0;
114 fs = ip->i_fs;
115#ifdef DIAGNOSTIC
116 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) {
117 printf("dev = 0x%lx, bsize = %ld, size = %d, fs = %s\n",
118 (u_long)ip->i_dev, (long)fs->fs_bsize, size, fs->fs_fsmnt);
119 panic("ffs_alloc: bad size");
120 }
121 if (cred == NOCRED)
122 panic("ffs_alloc: missing credential");
123#endif /* DIAGNOSTIC */
124 if (size == fs->fs_bsize && fs->fs_cstotal.cs_nbfree == 0)
125 goto nospace;
126 if (cred->cr_uid != 0 &&
127 freespace(fs, fs->fs_minfree) - numfrags(fs, size) < 0)
128 goto nospace;
129#ifdef QUOTA
130 error = chkdq(ip, (long)btodb(size), cred, 0);
131 if (error)
132 return (error);
133#endif
134 if (bpref >= fs->fs_size)
135 bpref = 0;
136 if (bpref == 0)
137 cg = ino_to_cg(fs, ip->i_number);
138 else
139 cg = dtog(fs, bpref);
140 bno = (ufs_daddr_t)ffs_hashalloc(ip, cg, (long)bpref, size,
141 ffs_alloccg);
142 if (bno > 0) {
143 ip->i_blocks += btodb(size);
144 ip->i_flag |= IN_CHANGE | IN_UPDATE;
145 *bnp = bno;
146 return (0);
147 }
148#ifdef QUOTA
149 /*
150 * Restore user's disk quota because allocation failed.
151 */
152 (void) chkdq(ip, (long)-btodb(size), cred, FORCE);
153#endif
154nospace:
155 ffs_fserr(fs, cred->cr_uid, "file system full");
156 uprintf("\n%s: write failed, file system is full\n", fs->fs_fsmnt);
157 return (ENOSPC);
158}
159
160/*
161 * Reallocate a fragment to a bigger size
162 *
163 * The number and size of the old block is given, and a preference
164 * and new size is also specified. The allocator attempts to extend
165 * the original block. Failing that, the regular block allocator is
166 * invoked to get an appropriate block.
167 */
168int
169ffs_realloccg(ip, lbprev, bpref, osize, nsize, cred, bpp)
170 register struct inode *ip;
171 ufs_daddr_t lbprev;
172 ufs_daddr_t bpref;
173 int osize, nsize;
174 struct ucred *cred;
175 struct buf **bpp;
176{
177 register struct fs *fs;
178 struct buf *bp;
179 int cg, request, error;
180 ufs_daddr_t bprev, bno;
181
182 *bpp = 0;
183 fs = ip->i_fs;
184#ifdef DIAGNOSTIC
185 if ((u_int)osize > fs->fs_bsize || fragoff(fs, osize) != 0 ||
186 (u_int)nsize > fs->fs_bsize || fragoff(fs, nsize) != 0) {
187 printf(
188 "dev = 0x%lx, bsize = %ld, osize = %d, nsize = %d, fs = %s\n",
189 (u_long)ip->i_dev, (long)fs->fs_bsize, osize,
190 nsize, fs->fs_fsmnt);
191 panic("ffs_realloccg: bad size");
192 }
193 if (cred == NOCRED)
194 panic("ffs_realloccg: missing credential");
195#endif /* DIAGNOSTIC */
196 if (cred->cr_uid != 0 &&
197 freespace(fs, fs->fs_minfree) - numfrags(fs, nsize - osize) < 0)
198 goto nospace;
199 if ((bprev = ip->i_db[lbprev]) == 0) {
200 printf("dev = 0x%lx, bsize = %ld, bprev = %ld, fs = %s\n",
201 (u_long)ip->i_dev, (long)fs->fs_bsize, (long)bprev,
202 fs->fs_fsmnt);
203 panic("ffs_realloccg: bad bprev");
204 }
205 /*
206 * Allocate the extra space in the buffer.
207 */
208 error = bread(ITOV(ip), lbprev, osize, NOCRED, &bp);
209 if (error) {
210 brelse(bp);
211 return (error);
212 }
213
214 if( bp->b_blkno == bp->b_lblkno) {
215 if( lbprev >= NDADDR)
216 panic("ffs_realloccg: lbprev out of range");
217 bp->b_blkno = fsbtodb(fs, bprev);
218 }
219
220#ifdef QUOTA
221 error = chkdq(ip, (long)btodb(nsize - osize), cred, 0);
222 if (error) {
223 brelse(bp);
224 return (error);
225 }
226#endif
227 /*
228 * Check for extension in the existing location.
229 */
230 cg = dtog(fs, bprev);
231 bno = ffs_fragextend(ip, cg, (long)bprev, osize, nsize);
232 if (bno) {
233 if (bp->b_blkno != fsbtodb(fs, bno))
234 panic("ffs_realloccg: bad blockno");
235 ip->i_blocks += btodb(nsize - osize);
236 ip->i_flag |= IN_CHANGE | IN_UPDATE;
237 allocbuf(bp, nsize);
238 bp->b_flags |= B_DONE;
239 bzero((char *)bp->b_data + osize, (u_int)nsize - osize);
240 *bpp = bp;
241 return (0);
242 }
243 /*
244 * Allocate a new disk location.
245 */
246 if (bpref >= fs->fs_size)
247 bpref = 0;
248 switch ((int)fs->fs_optim) {
249 case FS_OPTSPACE:
250 /*
251 * Allocate an exact sized fragment. Although this makes
252 * best use of space, we will waste time relocating it if
253 * the file continues to grow. If the fragmentation is
254 * less than half of the minimum free reserve, we choose
255 * to begin optimizing for time.
256 */
257 request = nsize;
258 if (fs->fs_minfree <= 5 ||
259 fs->fs_cstotal.cs_nffree >
260 fs->fs_dsize * fs->fs_minfree / (2 * 100))
261 break;
262 log(LOG_NOTICE, "%s: optimization changed from SPACE to TIME\n",
263 fs->fs_fsmnt);
264 fs->fs_optim = FS_OPTTIME;
265 break;
266 case FS_OPTTIME:
267 /*
268 * At this point we have discovered a file that is trying to
269 * grow a small fragment to a larger fragment. To save time,
270 * we allocate a full sized block, then free the unused portion.
271 * If the file continues to grow, the `ffs_fragextend' call
272 * above will be able to grow it in place without further
273 * copying. If aberrant programs cause disk fragmentation to
274 * grow within 2% of the free reserve, we choose to begin
275 * optimizing for space.
276 */
277 request = fs->fs_bsize;
278 if (fs->fs_cstotal.cs_nffree <
279 fs->fs_dsize * (fs->fs_minfree - 2) / 100)
280 break;
281 log(LOG_NOTICE, "%s: optimization changed from TIME to SPACE\n",
282 fs->fs_fsmnt);
283 fs->fs_optim = FS_OPTSPACE;
284 break;
285 default:
286 printf("dev = 0x%lx, optim = %ld, fs = %s\n",
287 (u_long)ip->i_dev, (long)fs->fs_optim, fs->fs_fsmnt);
288 panic("ffs_realloccg: bad optim");
289 /* NOTREACHED */
290 }
291 bno = (ufs_daddr_t)ffs_hashalloc(ip, cg, (long)bpref, request,
292 ffs_alloccg);
293 if (bno > 0) {
294 bp->b_blkno = fsbtodb(fs, bno);
295 if (!DOINGSOFTDEP(ITOV(ip)))
296 ffs_blkfree(ip, bprev, (long)osize);
297 if (nsize < request)
298 ffs_blkfree(ip, bno + numfrags(fs, nsize),
299 (long)(request - nsize));
300 ip->i_blocks += btodb(nsize - osize);
301 ip->i_flag |= IN_CHANGE | IN_UPDATE;
302 allocbuf(bp, nsize);
303 bp->b_flags |= B_DONE;
304 bzero((char *)bp->b_data + osize, (u_int)nsize - osize);
305 *bpp = bp;
306 return (0);
307 }
308#ifdef QUOTA
309 /*
310 * Restore user's disk quota because allocation failed.
311 */
312 (void) chkdq(ip, (long)-btodb(nsize - osize), cred, FORCE);
313#endif
314 brelse(bp);
315nospace:
316 /*
317 * no space available
318 */
319 ffs_fserr(fs, cred->cr_uid, "file system full");
320 uprintf("\n%s: write failed, file system is full\n", fs->fs_fsmnt);
321 return (ENOSPC);
322}
323
324SYSCTL_NODE(_vfs, OID_AUTO, ffs, CTLFLAG_RW, 0, "FFS filesystem");
325
326/*
327 * Reallocate a sequence of blocks into a contiguous sequence of blocks.
328 *
329 * The vnode and an array of buffer pointers for a range of sequential
330 * logical blocks to be made contiguous is given. The allocator attempts
331 * to find a range of sequential blocks starting as close as possible to
332 * an fs_rotdelay offset from the end of the allocation for the logical
333 * block immediately preceeding the current range. If successful, the
334 * physical block numbers in the buffer pointers and in the inode are
335 * changed to reflect the new allocation. If unsuccessful, the allocation
336 * is left unchanged. The success in doing the reallocation is returned.
337 * Note that the error return is not reflected back to the user. Rather
338 * the previous block allocation will be used.
339 */
340static int doasyncfree = 1;
341SYSCTL_INT(_vfs_ffs, FFS_ASYNCFREE, doasyncfree, CTLFLAG_RW, &doasyncfree, 0, "");
342
343static int doreallocblks = 1;
344SYSCTL_INT(_vfs_ffs, FFS_REALLOCBLKS, doreallocblks, CTLFLAG_RW, &doreallocblks, 0, "");
345
346#ifdef DEBUG
347static volatile int prtrealloc = 0;
348#endif
349
350int
351ffs_reallocblks(ap)
352 struct vop_reallocblks_args /* {
353 struct vnode *a_vp;
354 struct cluster_save *a_buflist;
355 } */ *ap;
356{
357 struct fs *fs;
358 struct inode *ip;
359 struct vnode *vp;
360 struct buf *sbp, *ebp;
361 ufs_daddr_t *bap, *sbap, *ebap = 0;
362 struct cluster_save *buflist;
363 ufs_daddr_t start_lbn, end_lbn, soff, newblk, blkno;
364 struct indir start_ap[NIADDR + 1], end_ap[NIADDR + 1], *idp;
365 int i, len, start_lvl, end_lvl, pref, ssize;
366
367 if (doreallocblks == 0)
368 return (ENOSPC);
369 vp = ap->a_vp;
370 ip = VTOI(vp);
371 fs = ip->i_fs;
372 if (fs->fs_contigsumsize <= 0)
373 return (ENOSPC);
374 buflist = ap->a_buflist;
375 len = buflist->bs_nchildren;
376 start_lbn = buflist->bs_children[0]->b_lblkno;
377 end_lbn = start_lbn + len - 1;
378#ifdef DIAGNOSTIC
379 for (i = 0; i < len; i++)
380 if (!ffs_checkblk(ip,
381 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
382 panic("ffs_reallocblks: unallocated block 1");
383 for (i = 1; i < len; i++)
384 if (buflist->bs_children[i]->b_lblkno != start_lbn + i)
385 panic("ffs_reallocblks: non-logical cluster");
386 blkno = buflist->bs_children[0]->b_blkno;
387 ssize = fsbtodb(fs, fs->fs_frag);
388 for (i = 1; i < len - 1; i++)
389 if (buflist->bs_children[i]->b_blkno != blkno + (i * ssize))
390 panic("ffs_reallocblks: non-physical cluster %d", i);
391#endif
392 /*
393 * If the latest allocation is in a new cylinder group, assume that
394 * the filesystem has decided to move and do not force it back to
395 * the previous cylinder group.
396 */
397 if (dtog(fs, dbtofsb(fs, buflist->bs_children[0]->b_blkno)) !=
398 dtog(fs, dbtofsb(fs, buflist->bs_children[len - 1]->b_blkno)))
399 return (ENOSPC);
400 if (ufs_getlbns(vp, start_lbn, start_ap, &start_lvl) ||
401 ufs_getlbns(vp, end_lbn, end_ap, &end_lvl))
402 return (ENOSPC);
403 /*
404 * Get the starting offset and block map for the first block.
405 */
406 if (start_lvl == 0) {
407 sbap = &ip->i_db[0];
408 soff = start_lbn;
409 } else {
410 idp = &start_ap[start_lvl - 1];
411 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &sbp)) {
412 brelse(sbp);
413 return (ENOSPC);
414 }
415 sbap = (ufs_daddr_t *)sbp->b_data;
416 soff = idp->in_off;
417 }
418 /*
419 * Find the preferred location for the cluster.
420 */
421 pref = ffs_blkpref(ip, start_lbn, soff, sbap);
422 /*
423 * If the block range spans two block maps, get the second map.
424 */
425 if (end_lvl == 0 || (idp = &end_ap[end_lvl - 1])->in_off + 1 >= len) {
426 ssize = len;
427 } else {
428#ifdef DIAGNOSTIC
429 if (start_ap[start_lvl-1].in_lbn == idp->in_lbn)
430 panic("ffs_reallocblk: start == end");
431#endif
432 ssize = len - (idp->in_off + 1);
433 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &ebp))
434 goto fail;
435 ebap = (ufs_daddr_t *)ebp->b_data;
436 }
437 /*
438 * Search the block map looking for an allocation of the desired size.
439 */
440 if ((newblk = (ufs_daddr_t)ffs_hashalloc(ip, dtog(fs, pref), (long)pref,
441 len, ffs_clusteralloc)) == 0)
442 goto fail;
443 /*
444 * We have found a new contiguous block.
445 *
446 * First we have to replace the old block pointers with the new
447 * block pointers in the inode and indirect blocks associated
448 * with the file.
449 */
450#ifdef DEBUG
451 if (prtrealloc)
452 printf("realloc: ino %d, lbns %d-%d\n\told:", ip->i_number,
453 start_lbn, end_lbn);
454#endif
455 blkno = newblk;
456 for (bap = &sbap[soff], i = 0; i < len; i++, blkno += fs->fs_frag) {
457 if (i == ssize) {
458 bap = ebap;
459 soff = -i;
460 }
461#ifdef DIAGNOSTIC
462 if (!ffs_checkblk(ip,
463 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
464 panic("ffs_reallocblks: unallocated block 2");
465 if (dbtofsb(fs, buflist->bs_children[i]->b_blkno) != *bap)
466 panic("ffs_reallocblks: alloc mismatch");
467#endif
468#ifdef DEBUG
469 if (prtrealloc)
470 printf(" %d,", *bap);
471#endif
472 if (DOINGSOFTDEP(vp)) {
473 if (sbap == &ip->i_db[0] && i < ssize)
474 softdep_setup_allocdirect(ip, start_lbn + i,
475 blkno, *bap, fs->fs_bsize, fs->fs_bsize,
476 buflist->bs_children[i]);
477 else
478 softdep_setup_allocindir_page(ip, start_lbn + i,
479 i < ssize ? sbp : ebp, soff + i, blkno,
480 *bap, buflist->bs_children[i]);
481 }
482 *bap++ = blkno;
483 }
484 /*
485 * Next we must write out the modified inode and indirect blocks.
486 * For strict correctness, the writes should be synchronous since
487 * the old block values may have been written to disk. In practise
488 * they are almost never written, but if we are concerned about
489 * strict correctness, the `doasyncfree' flag should be set to zero.
490 *
491 * The test on `doasyncfree' should be changed to test a flag
492 * that shows whether the associated buffers and inodes have
493 * been written. The flag should be set when the cluster is
494 * started and cleared whenever the buffer or inode is flushed.
495 * We can then check below to see if it is set, and do the
496 * synchronous write only when it has been cleared.
497 */
498 if (sbap != &ip->i_db[0]) {
499 if (doasyncfree)
500 bdwrite(sbp);
501 else
502 bwrite(sbp);
503 } else {
504 ip->i_flag |= IN_CHANGE | IN_UPDATE;
505 if (!doasyncfree)
506 UFS_UPDATE(vp, 1);
507 }
508 if (ssize < len) {
509 if (doasyncfree)
510 bdwrite(ebp);
511 else
512 bwrite(ebp);
513 }
514 /*
515 * Last, free the old blocks and assign the new blocks to the buffers.
516 */
517#ifdef DEBUG
518 if (prtrealloc)
519 printf("\n\tnew:");
520#endif
521 for (blkno = newblk, i = 0; i < len; i++, blkno += fs->fs_frag) {
522 if (!DOINGSOFTDEP(vp))
523 ffs_blkfree(ip,
524 dbtofsb(fs, buflist->bs_children[i]->b_blkno),
525 fs->fs_bsize);
526 buflist->bs_children[i]->b_blkno = fsbtodb(fs, blkno);
527#ifdef DEBUG
528 if (!ffs_checkblk(ip,
529 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
530 panic("ffs_reallocblks: unallocated block 3");
531 if (prtrealloc)
532 printf(" %d,", blkno);
533#endif
534 }
535#ifdef DEBUG
536 if (prtrealloc) {
537 prtrealloc--;
538 printf("\n");
539 }
540#endif
541 return (0);
542
543fail:
544 if (ssize < len)
545 brelse(ebp);
546 if (sbap != &ip->i_db[0])
547 brelse(sbp);
548 return (ENOSPC);
549}
550
551/*
552 * Allocate an inode in the file system.
553 *
554 * If allocating a directory, use ffs_dirpref to select the inode.
555 * If allocating in a directory, the following hierarchy is followed:
556 * 1) allocate the preferred inode.
557 * 2) allocate an inode in the same cylinder group.
558 * 3) quadradically rehash into other cylinder groups, until an
559 * available inode is located.
560 * If no inode preference is given the following heirarchy is used
561 * to allocate an inode:
562 * 1) allocate an inode in cylinder group 0.
563 * 2) quadradically rehash into other cylinder groups, until an
564 * available inode is located.
565 */
566int
567ffs_valloc(pvp, mode, cred, vpp)
568 struct vnode *pvp;
569 int mode;
570 struct ucred *cred;
571 struct vnode **vpp;
572{
573 register struct inode *pip;
574 register struct fs *fs;
575 register struct inode *ip;
576 ino_t ino, ipref;
577 int cg, error;
578
579 *vpp = NULL;
580 pip = VTOI(pvp);
581 fs = pip->i_fs;
582 if (fs->fs_cstotal.cs_nifree == 0)
583 goto noinodes;
584
585 if ((mode & IFMT) == IFDIR)
586 ipref = ffs_dirpref(fs);
587 else
588 ipref = pip->i_number;
589 if (ipref >= fs->fs_ncg * fs->fs_ipg)
590 ipref = 0;
591 cg = ino_to_cg(fs, ipref);
592 ino = (ino_t)ffs_hashalloc(pip, cg, (long)ipref, mode,
593 (allocfcn_t *)ffs_nodealloccg);
594 if (ino == 0)
595 goto noinodes;
596 error = VFS_VGET(pvp->v_mount, ino, vpp);
597 if (error) {
598 UFS_VFREE(pvp, ino, mode);
599 return (error);
600 }
601 ip = VTOI(*vpp);
602 if (ip->i_mode) {
603 printf("mode = 0%o, inum = %lu, fs = %s\n",
604 ip->i_mode, (u_long)ip->i_number, fs->fs_fsmnt);
605 panic("ffs_valloc: dup alloc");
606 }
607 if (ip->i_blocks) { /* XXX */
608 printf("free inode %s/%lu had %ld blocks\n",
609 fs->fs_fsmnt, (u_long)ino, (long)ip->i_blocks);
610 ip->i_blocks = 0;
611 }
612 ip->i_flags = 0;
613 /*
614 * Set up a new generation number for this inode.
615 */
616 if (ip->i_gen == 0 || ++ip->i_gen == 0)
617 ip->i_gen = random() / 2 + 1;
618 return (0);
619noinodes:
620 ffs_fserr(fs, cred->cr_uid, "out of inodes");
621 uprintf("\n%s: create/symlink failed, no inodes free\n", fs->fs_fsmnt);
622 return (ENOSPC);
623}
624
625/*
626 * Find a cylinder to place a directory.
627 *
628 * The policy implemented by this algorithm is to select from
629 * among those cylinder groups with above the average number of
630 * free inodes, the one with the smallest number of directories.
631 */
632static ino_t
633ffs_dirpref(fs)
634 register struct fs *fs;
635{
636 int cg, minndir, mincg, avgifree;
637
638 avgifree = fs->fs_cstotal.cs_nifree / fs->fs_ncg;
639 minndir = fs->fs_ipg;
640 mincg = 0;
641 for (cg = 0; cg < fs->fs_ncg; cg++)
642 if (fs->fs_cs(fs, cg).cs_ndir < minndir &&
643 fs->fs_cs(fs, cg).cs_nifree >= avgifree) {
644 mincg = cg;
645 minndir = fs->fs_cs(fs, cg).cs_ndir;
646 }
647 return ((ino_t)(fs->fs_ipg * mincg));
648}
649
650/*
651 * Select the desired position for the next block in a file. The file is
652 * logically divided into sections. The first section is composed of the
653 * direct blocks. Each additional section contains fs_maxbpg blocks.
654 *
655 * If no blocks have been allocated in the first section, the policy is to
656 * request a block in the same cylinder group as the inode that describes
657 * the file. If no blocks have been allocated in any other section, the
658 * policy is to place the section in a cylinder group with a greater than
659 * average number of free blocks. An appropriate cylinder group is found
660 * by using a rotor that sweeps the cylinder groups. When a new group of
661 * blocks is needed, the sweep begins in the cylinder group following the
662 * cylinder group from which the previous allocation was made. The sweep
663 * continues until a cylinder group with greater than the average number
664 * of free blocks is found. If the allocation is for the first block in an
665 * indirect block, the information on the previous allocation is unavailable;
666 * here a best guess is made based upon the logical block number being
667 * allocated.
668 *
669 * If a section is already partially allocated, the policy is to
670 * contiguously allocate fs_maxcontig blocks. The end of one of these
671 * contiguous blocks and the beginning of the next is physically separated
672 * so that the disk head will be in transit between them for at least
673 * fs_rotdelay milliseconds. This is to allow time for the processor to
674 * schedule another I/O transfer.
675 */
676ufs_daddr_t
677ffs_blkpref(ip, lbn, indx, bap)
678 struct inode *ip;
679 ufs_daddr_t lbn;
680 int indx;
681 ufs_daddr_t *bap;
682{
683 register struct fs *fs;
684 register int cg;
685 int avgbfree, startcg;
686 ufs_daddr_t nextblk;
687
688 fs = ip->i_fs;
689 if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) {
690 if (lbn < NDADDR) {
691 cg = ino_to_cg(fs, ip->i_number);
692 return (fs->fs_fpg * cg + fs->fs_frag);
693 }
694 /*
695 * Find a cylinder with greater than average number of
696 * unused data blocks.
697 */
698 if (indx == 0 || bap[indx - 1] == 0)
699 startcg =
700 ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg;
701 else
702 startcg = dtog(fs, bap[indx - 1]) + 1;
703 startcg %= fs->fs_ncg;
704 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
705 for (cg = startcg; cg < fs->fs_ncg; cg++)
706 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
707 fs->fs_cgrotor = cg;
708 return (fs->fs_fpg * cg + fs->fs_frag);
709 }
710 for (cg = 0; cg <= startcg; cg++)
711 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
712 fs->fs_cgrotor = cg;
713 return (fs->fs_fpg * cg + fs->fs_frag);
714 }
715 return (0);
716 }
717 /*
718 * One or more previous blocks have been laid out. If less
719 * than fs_maxcontig previous blocks are contiguous, the
720 * next block is requested contiguously, otherwise it is
721 * requested rotationally delayed by fs_rotdelay milliseconds.
722 */
723 nextblk = bap[indx - 1] + fs->fs_frag;
724 if (fs->fs_rotdelay == 0 || indx < fs->fs_maxcontig ||
725 bap[indx - fs->fs_maxcontig] +
726 blkstofrags(fs, fs->fs_maxcontig) != nextblk)
727 return (nextblk);
728 /*
729 * Here we convert ms of delay to frags as:
730 * (frags) = (ms) * (rev/sec) * (sect/rev) /
731 * ((sect/frag) * (ms/sec))
732 * then round up to the next block.
733 */
734 nextblk += roundup(fs->fs_rotdelay * fs->fs_rps * fs->fs_nsect /
735 (NSPF(fs) * 1000), fs->fs_frag);
736 return (nextblk);
737}
738
739/*
740 * Implement the cylinder overflow algorithm.
741 *
742 * The policy implemented by this algorithm is:
743 * 1) allocate the block in its requested cylinder group.
744 * 2) quadradically rehash on the cylinder group number.
745 * 3) brute force search for a free block.
746 */
747/*VARARGS5*/
748static u_long
749ffs_hashalloc(ip, cg, pref, size, allocator)
750 struct inode *ip;
751 int cg;
752 long pref;
753 int size; /* size for data blocks, mode for inodes */
754 allocfcn_t *allocator;
755{
756 register struct fs *fs;
757 long result; /* XXX why not same type as we return? */
758 int i, icg = cg;
759
760 fs = ip->i_fs;
761 /*
762 * 1: preferred cylinder group
763 */
764 result = (*allocator)(ip, cg, pref, size);
765 if (result)
766 return (result);
767 /*
768 * 2: quadratic rehash
769 */
770 for (i = 1; i < fs->fs_ncg; i *= 2) {
771 cg += i;
772 if (cg >= fs->fs_ncg)
773 cg -= fs->fs_ncg;
774 result = (*allocator)(ip, cg, 0, size);
775 if (result)
776 return (result);
777 }
778 /*
779 * 3: brute force search
780 * Note that we start at i == 2, since 0 was checked initially,
781 * and 1 is always checked in the quadratic rehash.
782 */
783 cg = (icg + 2) % fs->fs_ncg;
784 for (i = 2; i < fs->fs_ncg; i++) {
785 result = (*allocator)(ip, cg, 0, size);
786 if (result)
787 return (result);
788 cg++;
789 if (cg == fs->fs_ncg)
790 cg = 0;
791 }
792 return (0);
793}
794
795/*
796 * Determine whether a fragment can be extended.
797 *
798 * Check to see if the necessary fragments are available, and
799 * if they are, allocate them.
800 */
801static ufs_daddr_t
802ffs_fragextend(ip, cg, bprev, osize, nsize)
803 struct inode *ip;
804 int cg;
805 long bprev;
806 int osize, nsize;
807{
808 register struct fs *fs;
809 register struct cg *cgp;
810 struct buf *bp;
811 long bno;
812 int frags, bbase;
813 int i, error;
814
815 fs = ip->i_fs;
816 if (fs->fs_cs(fs, cg).cs_nffree < numfrags(fs, nsize - osize))
817 return (0);
818 frags = numfrags(fs, nsize);
819 bbase = fragnum(fs, bprev);
820 if (bbase > fragnum(fs, (bprev + frags - 1))) {
821 /* cannot extend across a block boundary */
822 return (0);
823 }
824 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
825 (int)fs->fs_cgsize, NOCRED, &bp);
826 if (error) {
827 brelse(bp);
828 return (0);
829 }
830 cgp = (struct cg *)bp->b_data;
831 if (!cg_chkmagic(cgp)) {
832 brelse(bp);
833 return (0);
834 }
835 cgp->cg_time = time_second;
836 bno = dtogd(fs, bprev);
837 for (i = numfrags(fs, osize); i < frags; i++)
838 if (isclr(cg_blksfree(cgp), bno + i)) {
839 brelse(bp);
840 return (0);
841 }
842 /*
843 * the current fragment can be extended
844 * deduct the count on fragment being extended into
845 * increase the count on the remaining fragment (if any)
846 * allocate the extended piece
847 */
848 for (i = frags; i < fs->fs_frag - bbase; i++)
849 if (isclr(cg_blksfree(cgp), bno + i))
850 break;
851 cgp->cg_frsum[i - numfrags(fs, osize)]--;
852 if (i != frags)
853 cgp->cg_frsum[i - frags]++;
854 for (i = numfrags(fs, osize); i < frags; i++) {
855 clrbit(cg_blksfree(cgp), bno + i);
856 cgp->cg_cs.cs_nffree--;
857 fs->fs_cstotal.cs_nffree--;
858 fs->fs_cs(fs, cg).cs_nffree--;
859 }
860 fs->fs_fmod = 1;
861 if (DOINGSOFTDEP(ITOV(ip)))
862 softdep_setup_blkmapdep(bp, fs, bprev);
863 bdwrite(bp);
864 return (bprev);
865}
866
867/*
868 * Determine whether a block can be allocated.
869 *
870 * Check to see if a block of the appropriate size is available,
871 * and if it is, allocate it.
872 */
873static ufs_daddr_t
874ffs_alloccg(ip, cg, bpref, size)
875 struct inode *ip;
876 int cg;
877 ufs_daddr_t bpref;
878 int size;
879{
880 register struct fs *fs;
881 register struct cg *cgp;
882 struct buf *bp;
883 register int i;
884 ufs_daddr_t bno, blkno;
885 int allocsiz, error, frags;
886
887 fs = ip->i_fs;
888 if (fs->fs_cs(fs, cg).cs_nbfree == 0 && size == fs->fs_bsize)
889 return (0);
890 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
891 (int)fs->fs_cgsize, NOCRED, &bp);
892 if (error) {
893 brelse(bp);
894 return (0);
895 }
896 cgp = (struct cg *)bp->b_data;
897 if (!cg_chkmagic(cgp) ||
898 (cgp->cg_cs.cs_nbfree == 0 && size == fs->fs_bsize)) {
899 brelse(bp);
900 return (0);
901 }
902 cgp->cg_time = time_second;
903 if (size == fs->fs_bsize) {
904 bno = ffs_alloccgblk(ip, bp, bpref);
905 bdwrite(bp);
906 return (bno);
907 }
908 /*
909 * check to see if any fragments are already available
910 * allocsiz is the size which will be allocated, hacking
911 * it down to a smaller size if necessary
912 */
913 frags = numfrags(fs, size);
914 for (allocsiz = frags; allocsiz < fs->fs_frag; allocsiz++)
915 if (cgp->cg_frsum[allocsiz] != 0)
916 break;
917 if (allocsiz == fs->fs_frag) {
918 /*
919 * no fragments were available, so a block will be
920 * allocated, and hacked up
921 */
922 if (cgp->cg_cs.cs_nbfree == 0) {
923 brelse(bp);
924 return (0);
925 }
926 bno = ffs_alloccgblk(ip, bp, bpref);
927 bpref = dtogd(fs, bno);
928 for (i = frags; i < fs->fs_frag; i++)
929 setbit(cg_blksfree(cgp), bpref + i);
930 i = fs->fs_frag - frags;
931 cgp->cg_cs.cs_nffree += i;
932 fs->fs_cstotal.cs_nffree += i;
933 fs->fs_cs(fs, cg).cs_nffree += i;
934 fs->fs_fmod = 1;
935 cgp->cg_frsum[i]++;
936 bdwrite(bp);
937 return (bno);
938 }
939 bno = ffs_mapsearch(fs, cgp, bpref, allocsiz);
940 if (bno < 0) {
941 brelse(bp);
942 return (0);
943 }
944 for (i = 0; i < frags; i++)
945 clrbit(cg_blksfree(cgp), bno + i);
946 cgp->cg_cs.cs_nffree -= frags;
947 fs->fs_cstotal.cs_nffree -= frags;
948 fs->fs_cs(fs, cg).cs_nffree -= frags;
949 fs->fs_fmod = 1;
950 cgp->cg_frsum[allocsiz]--;
951 if (frags != allocsiz)
952 cgp->cg_frsum[allocsiz - frags]++;
953 blkno = cg * fs->fs_fpg + bno;
954 if (DOINGSOFTDEP(ITOV(ip)))
955 softdep_setup_blkmapdep(bp, fs, blkno);
956 bdwrite(bp);
957 return ((u_long)blkno);
958}
959
960/*
961 * Allocate a block in a cylinder group.
962 *
963 * This algorithm implements the following policy:
964 * 1) allocate the requested block.
965 * 2) allocate a rotationally optimal block in the same cylinder.
966 * 3) allocate the next available block on the block rotor for the
967 * specified cylinder group.
968 * Note that this routine only allocates fs_bsize blocks; these
969 * blocks may be fragmented by the routine that allocates them.
970 */
971static ufs_daddr_t
972ffs_alloccgblk(ip, bp, bpref)
973 struct inode *ip;
974 struct buf *bp;
975 ufs_daddr_t bpref;
976{
977 struct fs *fs;
978 struct cg *cgp;
979 ufs_daddr_t bno, blkno;
980 int cylno, pos, delta;
981 short *cylbp;
982 register int i;
983
984 fs = ip->i_fs;
985 cgp = (struct cg *)bp->b_data;
986 if (bpref == 0 || dtog(fs, bpref) != cgp->cg_cgx) {
987 bpref = cgp->cg_rotor;
988 goto norot;
989 }
990 bpref = blknum(fs, bpref);
991 bpref = dtogd(fs, bpref);
992 /*
993 * if the requested block is available, use it
994 */
995 if (ffs_isblock(fs, cg_blksfree(cgp), fragstoblks(fs, bpref))) {
996 bno = bpref;
997 goto gotit;
998 }
999 if (fs->fs_nrpos <= 1 || fs->fs_cpc == 0) {
1000 /*
1001 * Block layout information is not available.
1002 * Leaving bpref unchanged means we take the
1003 * next available free block following the one
1004 * we just allocated. Hopefully this will at
1005 * least hit a track cache on drives of unknown
1006 * geometry (e.g. SCSI).
1007 */
1008 goto norot;
1009 }
1010 /*
1011 * check for a block available on the same cylinder
1012 */
1013 cylno = cbtocylno(fs, bpref);
1014 if (cg_blktot(cgp)[cylno] == 0)
1015 goto norot;
1016 /*
1017 * check the summary information to see if a block is
1018 * available in the requested cylinder starting at the
1019 * requested rotational position and proceeding around.
1020 */
1021 cylbp = cg_blks(fs, cgp, cylno);
1022 pos = cbtorpos(fs, bpref);
1023 for (i = pos; i < fs->fs_nrpos; i++)
1024 if (cylbp[i] > 0)
1025 break;
1026 if (i == fs->fs_nrpos)
1027 for (i = 0; i < pos; i++)
1028 if (cylbp[i] > 0)
1029 break;
1030 if (cylbp[i] > 0) {
1031 /*
1032 * found a rotational position, now find the actual
1033 * block. A panic if none is actually there.
1034 */
1035 pos = cylno % fs->fs_cpc;
1036 bno = (cylno - pos) * fs->fs_spc / NSPB(fs);
1037 if (fs_postbl(fs, pos)[i] == -1) {
1038 printf("pos = %d, i = %d, fs = %s\n",
1039 pos, i, fs->fs_fsmnt);
1040 panic("ffs_alloccgblk: cyl groups corrupted");
1041 }
1042 for (i = fs_postbl(fs, pos)[i];; ) {
1043 if (ffs_isblock(fs, cg_blksfree(cgp), bno + i)) {
1044 bno = blkstofrags(fs, (bno + i));
1045 goto gotit;
1046 }
1047 delta = fs_rotbl(fs)[i];
1048 if (delta <= 0 ||
1049 delta + i > fragstoblks(fs, fs->fs_fpg))
1050 break;
1051 i += delta;
1052 }
1053 printf("pos = %d, i = %d, fs = %s\n", pos, i, fs->fs_fsmnt);
1054 panic("ffs_alloccgblk: can't find blk in cyl");
1055 }
1056norot:
1057 /*
1058 * no blocks in the requested cylinder, so take next
1059 * available one in this cylinder group.
1060 */
1061 bno = ffs_mapsearch(fs, cgp, bpref, (int)fs->fs_frag);
1062 if (bno < 0)
1063 return (0);
1064 cgp->cg_rotor = bno;
1065gotit:
1066 blkno = fragstoblks(fs, bno);
1067 ffs_clrblock(fs, cg_blksfree(cgp), (long)blkno);
1068 ffs_clusteracct(fs, cgp, blkno, -1);
1069 cgp->cg_cs.cs_nbfree--;
1070 fs->fs_cstotal.cs_nbfree--;
1071 fs->fs_cs(fs, cgp->cg_cgx).cs_nbfree--;
1072 cylno = cbtocylno(fs, bno);
1073 cg_blks(fs, cgp, cylno)[cbtorpos(fs, bno)]--;
1074 cg_blktot(cgp)[cylno]--;
1075 fs->fs_fmod = 1;
1076 blkno = cgp->cg_cgx * fs->fs_fpg + bno;
1077 if (DOINGSOFTDEP(ITOV(ip)))
1078 softdep_setup_blkmapdep(bp, fs, blkno);
1079 return (blkno);
1080}
1081
1082/*
1083 * Determine whether a cluster can be allocated.
1084 *
1085 * We do not currently check for optimal rotational layout if there
1086 * are multiple choices in the same cylinder group. Instead we just
1087 * take the first one that we find following bpref.
1088 */
1089static ufs_daddr_t
1090ffs_clusteralloc(ip, cg, bpref, len)
1091 struct inode *ip;
1092 int cg;
1093 ufs_daddr_t bpref;
1094 int len;
1095{
1096 register struct fs *fs;
1097 register struct cg *cgp;
1098 struct buf *bp;
1099 int i, got, run, bno, bit, map;
1100 u_char *mapp;
1101 int32_t *lp;
1102
1103 fs = ip->i_fs;
1104 if (fs->fs_maxcluster[cg] < len)
1105 return (NULL);
1106 if (bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), (int)fs->fs_cgsize,
1107 NOCRED, &bp))
1108 goto fail;
1109 cgp = (struct cg *)bp->b_data;
1110 if (!cg_chkmagic(cgp))
1111 goto fail;
1112 /*
1113 * Check to see if a cluster of the needed size (or bigger) is
1114 * available in this cylinder group.
1115 */
1116 lp = &cg_clustersum(cgp)[len];
1117 for (i = len; i <= fs->fs_contigsumsize; i++)
1118 if (*lp++ > 0)
1119 break;
1120 if (i > fs->fs_contigsumsize) {
1121 /*
1122 * This is the first time looking for a cluster in this
1123 * cylinder group. Update the cluster summary information
1124 * to reflect the true maximum sized cluster so that
1125 * future cluster allocation requests can avoid reading
1126 * the cylinder group map only to find no clusters.
1127 */
1128 lp = &cg_clustersum(cgp)[len - 1];
1129 for (i = len - 1; i > 0; i--)
1130 if (*lp-- > 0)
1131 break;
1132 fs->fs_maxcluster[cg] = i;
1133 goto fail;
1134 }
1135 /*
1136 * Search the cluster map to find a big enough cluster.
1137 * We take the first one that we find, even if it is larger
1138 * than we need as we prefer to get one close to the previous
1139 * block allocation. We do not search before the current
1140 * preference point as we do not want to allocate a block
1141 * that is allocated before the previous one (as we will
1142 * then have to wait for another pass of the elevator
1143 * algorithm before it will be read). We prefer to fail and
1144 * be recalled to try an allocation in the next cylinder group.
1145 */
1146 if (dtog(fs, bpref) != cg)
1147 bpref = 0;
1148 else
1149 bpref = fragstoblks(fs, dtogd(fs, blknum(fs, bpref)));
1150 mapp = &cg_clustersfree(cgp)[bpref / NBBY];
1151 map = *mapp++;
1152 bit = 1 << (bpref % NBBY);
1153 for (run = 0, got = bpref; got < cgp->cg_nclusterblks; got++) {
1154 if ((map & bit) == 0) {
1155 run = 0;
1156 } else {
1157 run++;
1158 if (run == len)
1159 break;
1160 }
1161 if ((got & (NBBY - 1)) != (NBBY - 1)) {
1162 bit <<= 1;
1163 } else {
1164 map = *mapp++;
1165 bit = 1;
1166 }
1167 }
1168 if (got >= cgp->cg_nclusterblks)
1169 goto fail;
1170 /*
1171 * Allocate the cluster that we have found.
1172 */
1173 for (i = 1; i <= len; i++)
1174 if (!ffs_isblock(fs, cg_blksfree(cgp), got - run + i))
1175 panic("ffs_clusteralloc: map mismatch");
1176 bno = cg * fs->fs_fpg + blkstofrags(fs, got - run + 1);
1177 if (dtog(fs, bno) != cg)
1178 panic("ffs_clusteralloc: allocated out of group");
1179 len = blkstofrags(fs, len);
1180 for (i = 0; i < len; i += fs->fs_frag)
1181 if ((got = ffs_alloccgblk(ip, bp, bno + i)) != bno + i)
1182 panic("ffs_clusteralloc: lost block");
1183 bdwrite(bp);
1184 return (bno);
1185
1186fail:
1187 brelse(bp);
1188 return (0);
1189}
1190
1191/*
1192 * Determine whether an inode can be allocated.
1193 *
1194 * Check to see if an inode is available, and if it is,
1195 * allocate it using the following policy:
1196 * 1) allocate the requested inode.
1197 * 2) allocate the next available inode after the requested
1198 * inode in the specified cylinder group.
1199 */
1200static ino_t
1201ffs_nodealloccg(ip, cg, ipref, mode)
1202 struct inode *ip;
1203 int cg;
1204 ufs_daddr_t ipref;
1205 int mode;
1206{
1207 register struct fs *fs;
1208 register struct cg *cgp;
1209 struct buf *bp;
1210 int error, start, len, loc, map, i;
1211
1212 fs = ip->i_fs;
1213 if (fs->fs_cs(fs, cg).cs_nifree == 0)
1214 return (0);
1215 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
1216 (int)fs->fs_cgsize, NOCRED, &bp);
1217 if (error) {
1218 brelse(bp);
1219 return (0);
1220 }
1221 cgp = (struct cg *)bp->b_data;
1222 if (!cg_chkmagic(cgp) || cgp->cg_cs.cs_nifree == 0) {
1223 brelse(bp);
1224 return (0);
1225 }
1226 cgp->cg_time = time_second;
1227 if (ipref) {
1228 ipref %= fs->fs_ipg;
1229 if (isclr(cg_inosused(cgp), ipref))
1230 goto gotit;
1231 }
1232 start = cgp->cg_irotor / NBBY;
1233 len = howmany(fs->fs_ipg - cgp->cg_irotor, NBBY);
1234 loc = skpc(0xff, len, &cg_inosused(cgp)[start]);
1235 if (loc == 0) {
1236 len = start + 1;
1237 start = 0;
1238 loc = skpc(0xff, len, &cg_inosused(cgp)[0]);
1239 if (loc == 0) {
1240 printf("cg = %d, irotor = %ld, fs = %s\n",
1241 cg, (long)cgp->cg_irotor, fs->fs_fsmnt);
1242 panic("ffs_nodealloccg: map corrupted");
1243 /* NOTREACHED */
1244 }
1245 }
1246 i = start + len - loc;
1247 map = cg_inosused(cgp)[i];
1248 ipref = i * NBBY;
1249 for (i = 1; i < (1 << NBBY); i <<= 1, ipref++) {
1250 if ((map & i) == 0) {
1251 cgp->cg_irotor = ipref;
1252 goto gotit;
1253 }
1254 }
1255 printf("fs = %s\n", fs->fs_fsmnt);
1256 panic("ffs_nodealloccg: block not in map");
1257 /* NOTREACHED */
1258gotit:
1259 if (DOINGSOFTDEP(ITOV(ip)))
1260 softdep_setup_inomapdep(bp, ip, cg * fs->fs_ipg + ipref);
1261 setbit(cg_inosused(cgp), ipref);
1262 cgp->cg_cs.cs_nifree--;
1263 fs->fs_cstotal.cs_nifree--;
1264 fs->fs_cs(fs, cg).cs_nifree--;
1265 fs->fs_fmod = 1;
1266 if ((mode & IFMT) == IFDIR) {
1267 cgp->cg_cs.cs_ndir++;
1268 fs->fs_cstotal.cs_ndir++;
1269 fs->fs_cs(fs, cg).cs_ndir++;
1270 }
1271 bdwrite(bp);
1272 return (cg * fs->fs_ipg + ipref);
1273}
1274
1275/*
1276 * Free a block or fragment.
1277 *
1278 * The specified block or fragment is placed back in the
1279 * free map. If a fragment is deallocated, a possible
1280 * block reassembly is checked.
1281 */
1282void
1283ffs_blkfree(ip, bno, size)
1284 register struct inode *ip;
1285 ufs_daddr_t bno;
1286 long size;
1287{
1288 register struct fs *fs;
1289 register struct cg *cgp;
1290 struct buf *bp;
1291 ufs_daddr_t blkno;
1292 int i, error, cg, blk, frags, bbase;
1293
1294 fs = ip->i_fs;
1295 VOP_FREEBLKS(ip->i_devvp, fsbtodb(fs, bno), size);
1296 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0 ||
1297 fragnum(fs, bno) + numfrags(fs, size) > fs->fs_frag) {
1298 printf("dev=0x%lx, bno = %d, bsize = %d, size = %ld, fs = %s\n",
1299 (u_long)ip->i_dev, bno, fs->fs_bsize, size, fs->fs_fsmnt);
1300 panic("ffs_blkfree: bad size");
1301 }
1302 cg = dtog(fs, bno);
1303 if ((u_int)bno >= fs->fs_size) {
1304 printf("bad block %ld, ino %lu\n",
1305 (long)bno, (u_long)ip->i_number);
1306 ffs_fserr(fs, ip->i_uid, "bad block");
1307 return;
1308 }
1309 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
1310 (int)fs->fs_cgsize, NOCRED, &bp);
1311 if (error) {
1312 brelse(bp);
1313 return;
1314 }
1315 cgp = (struct cg *)bp->b_data;
1316 if (!cg_chkmagic(cgp)) {
1317 brelse(bp);
1318 return;
1319 }
1320 cgp->cg_time = time_second;
1321 bno = dtogd(fs, bno);
1322 if (size == fs->fs_bsize) {
1323 blkno = fragstoblks(fs, bno);
1324 if (!ffs_isfreeblock(fs, cg_blksfree(cgp), blkno)) {
1325 printf("dev = 0x%lx, block = %ld, fs = %s\n",
1326 (u_long)ip->i_dev, (long)bno, fs->fs_fsmnt);
1327 panic("ffs_blkfree: freeing free block");
1328 }
1329 ffs_setblock(fs, cg_blksfree(cgp), blkno);
1330 ffs_clusteracct(fs, cgp, blkno, 1);
1331 cgp->cg_cs.cs_nbfree++;
1332 fs->fs_cstotal.cs_nbfree++;
1333 fs->fs_cs(fs, cg).cs_nbfree++;
1334 i = cbtocylno(fs, bno);
1335 cg_blks(fs, cgp, i)[cbtorpos(fs, bno)]++;
1336 cg_blktot(cgp)[i]++;
1337 } else {
1338 bbase = bno - fragnum(fs, bno);
1339 /*
1340 * decrement the counts associated with the old frags
1341 */
1342 blk = blkmap(fs, cg_blksfree(cgp), bbase);
1343 ffs_fragacct(fs, blk, cgp->cg_frsum, -1);
1344 /*
1345 * deallocate the fragment
1346 */
1347 frags = numfrags(fs, size);
1348 for (i = 0; i < frags; i++) {
1349 if (isset(cg_blksfree(cgp), bno + i)) {
1350 printf("dev = 0x%lx, block = %ld, fs = %s\n",
1351 (u_long)ip->i_dev, (long)(bno + i),
1352 fs->fs_fsmnt);
1353 panic("ffs_blkfree: freeing free frag");
1354 }
1355 setbit(cg_blksfree(cgp), bno + i);
1356 }
1357 cgp->cg_cs.cs_nffree += i;
1358 fs->fs_cstotal.cs_nffree += i;
1359 fs->fs_cs(fs, cg).cs_nffree += i;
1360 /*
1361 * add back in counts associated with the new frags
1362 */
1363 blk = blkmap(fs, cg_blksfree(cgp), bbase);
1364 ffs_fragacct(fs, blk, cgp->cg_frsum, 1);
1365 /*
1366 * if a complete block has been reassembled, account for it
1367 */
1368 blkno = fragstoblks(fs, bbase);
1369 if (ffs_isblock(fs, cg_blksfree(cgp), blkno)) {
1370 cgp->cg_cs.cs_nffree -= fs->fs_frag;
1371 fs->fs_cstotal.cs_nffree -= fs->fs_frag;
1372 fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag;
1373 ffs_clusteracct(fs, cgp, blkno, 1);
1374 cgp->cg_cs.cs_nbfree++;
1375 fs->fs_cstotal.cs_nbfree++;
1376 fs->fs_cs(fs, cg).cs_nbfree++;
1377 i = cbtocylno(fs, bbase);
1378 cg_blks(fs, cgp, i)[cbtorpos(fs, bbase)]++;
1379 cg_blktot(cgp)[i]++;
1380 }
1381 }
1382 fs->fs_fmod = 1;
1383 bdwrite(bp);
1384}
1385
1386#ifdef DIAGNOSTIC
1387/*
1388 * Verify allocation of a block or fragment. Returns true if block or
1389 * fragment is allocated, false if it is free.
1390 */
1391static int
1392ffs_checkblk(ip, bno, size)
1393 struct inode *ip;
1394 ufs_daddr_t bno;
1395 long size;
1396{
1397 struct fs *fs;
1398 struct cg *cgp;
1399 struct buf *bp;
1400 int i, error, frags, free;
1401
1402 fs = ip->i_fs;
1403 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) {
1404 printf("bsize = %ld, size = %ld, fs = %s\n",
1405 (long)fs->fs_bsize, size, fs->fs_fsmnt);
1406 panic("ffs_checkblk: bad size");
1407 }
1408 if ((u_int)bno >= fs->fs_size)
1409 panic("ffs_checkblk: bad block %d", bno);
1410 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, dtog(fs, bno))),
1411 (int)fs->fs_cgsize, NOCRED, &bp);
1412 if (error)
1413 panic("ffs_checkblk: cg bread failed");
1414 cgp = (struct cg *)bp->b_data;
1415 if (!cg_chkmagic(cgp))
1416 panic("ffs_checkblk: cg magic mismatch");
1417 bno = dtogd(fs, bno);
1418 if (size == fs->fs_bsize) {
1419 free = ffs_isblock(fs, cg_blksfree(cgp), fragstoblks(fs, bno));
1420 } else {
1421 frags = numfrags(fs, size);
1422 for (free = 0, i = 0; i < frags; i++)
1423 if (isset(cg_blksfree(cgp), bno + i))
1424 free++;
1425 if (free != 0 && free != frags)
1426 panic("ffs_checkblk: partially free fragment");
1427 }
1428 brelse(bp);
1429 return (!free);
1430}
1431#endif /* DIAGNOSTIC */
1432
1433/*
1434 * Free an inode.
1435 */
1436int
1437ffs_vfree( pvp, ino, mode)
1438 struct vnode *pvp;
1439 ino_t ino;
1440 int mode;
1441{
1442 if (DOINGSOFTDEP(pvp)) {
1443 softdep_freefile(pvp, ino, mode);
1444 return (0);
1445 }
1446 return (ffs_freefile(pvp, ino, mode));
1447}
1448
1449/*
1450 * Do the actual free operation.
1451 * The specified inode is placed back in the free map.
1452 */
1453 int
1454 ffs_freefile( pvp, ino, mode)
1455 struct vnode *pvp;
1456 ino_t ino;
1457 int mode;
1458{
1459 register struct fs *fs;
1460 register struct cg *cgp;
1461 register struct inode *pip;
1462 struct buf *bp;
1463 int error, cg;
1464
1465 pip = VTOI(pvp);
1466 fs = pip->i_fs;
1467 if ((u_int)ino >= fs->fs_ipg * fs->fs_ncg)
1468 panic("ffs_vfree: range: dev = 0x%x, ino = %d, fs = %s",
1469 pip->i_dev, ino, fs->fs_fsmnt);
1470 cg = ino_to_cg(fs, ino);
1471 error = bread(pip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
1472 (int)fs->fs_cgsize, NOCRED, &bp);
1473 if (error) {
1474 brelse(bp);
1475 return (error);
1476 }
1477 cgp = (struct cg *)bp->b_data;
1478 if (!cg_chkmagic(cgp)) {
1479 brelse(bp);
1480 return (0);
1481 }
1482 cgp->cg_time = time_second;
1483 ino %= fs->fs_ipg;
1484 if (isclr(cg_inosused(cgp), ino)) {
1485 printf("dev = 0x%lx, ino = %lu, fs = %s\n",
1486 (u_long)pip->i_dev, (u_long)ino, fs->fs_fsmnt);
1487 if (fs->fs_ronly == 0)
1488 panic("ffs_vfree: freeing free inode");
1489 }
1490 clrbit(cg_inosused(cgp), ino);
1491 if (ino < cgp->cg_irotor)
1492 cgp->cg_irotor = ino;
1493 cgp->cg_cs.cs_nifree++;
1494 fs->fs_cstotal.cs_nifree++;
1495 fs->fs_cs(fs, cg).cs_nifree++;
1496 if ((mode & IFMT) == IFDIR) {
1497 cgp->cg_cs.cs_ndir--;
1498 fs->fs_cstotal.cs_ndir--;
1499 fs->fs_cs(fs, cg).cs_ndir--;
1500 }
1501 fs->fs_fmod = 1;
1502 bdwrite(bp);
1503 return (0);
1504}
1505
1506/*
1507 * Find a block of the specified size in the specified cylinder group.
1508 *
1509 * It is a panic if a request is made to find a block if none are
1510 * available.
1511 */
1512static ufs_daddr_t
1513ffs_mapsearch(fs, cgp, bpref, allocsiz)
1514 register struct fs *fs;
1515 register struct cg *cgp;
1516 ufs_daddr_t bpref;
1517 int allocsiz;
1518{
1519 ufs_daddr_t bno;
1520 int start, len, loc, i;
1521 int blk, field, subfield, pos;
1522
1523 /*
1524 * find the fragment by searching through the free block
1525 * map for an appropriate bit pattern
1526 */
1527 if (bpref)
1528 start = dtogd(fs, bpref) / NBBY;
1529 else
1530 start = cgp->cg_frotor / NBBY;
1531 len = howmany(fs->fs_fpg, NBBY) - start;
1532 loc = scanc((u_int)len, (u_char *)&cg_blksfree(cgp)[start],
1533 (u_char *)fragtbl[fs->fs_frag],
1534 (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
1535 if (loc == 0) {
1536 len = start + 1;
1537 start = 0;
1538 loc = scanc((u_int)len, (u_char *)&cg_blksfree(cgp)[0],
1539 (u_char *)fragtbl[fs->fs_frag],
1540 (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
1541 if (loc == 0) {
1542 printf("start = %d, len = %d, fs = %s\n",
1543 start, len, fs->fs_fsmnt);
1544 panic("ffs_alloccg: map corrupted");
1545 /* NOTREACHED */
1546 }
1547 }
1548 bno = (start + len - loc) * NBBY;
1549 cgp->cg_frotor = bno;
1550 /*
1551 * found the byte in the map
1552 * sift through the bits to find the selected frag
1553 */
1554 for (i = bno + NBBY; bno < i; bno += fs->fs_frag) {
1555 blk = blkmap(fs, cg_blksfree(cgp), bno);
1556 blk <<= 1;
1557 field = around[allocsiz];
1558 subfield = inside[allocsiz];
1559 for (pos = 0; pos <= fs->fs_frag - allocsiz; pos++) {
1560 if ((blk & field) == subfield)
1561 return (bno + pos);
1562 field <<= 1;
1563 subfield <<= 1;
1564 }
1565 }
1566 printf("bno = %lu, fs = %s\n", (u_long)bno, fs->fs_fsmnt);
1567 panic("ffs_alloccg: block not in map");
1568 return (-1);
1569}
1570
1571/*
1572 * Update the cluster map because of an allocation or free.
1573 *
1574 * Cnt == 1 means free; cnt == -1 means allocating.
1575 */
1576static void
1577ffs_clusteracct(fs, cgp, blkno, cnt)
1578 struct fs *fs;
1579 struct cg *cgp;
1580 ufs_daddr_t blkno;
1581 int cnt;
1582{
1583 int32_t *sump;
1584 int32_t *lp;
1585 u_char *freemapp, *mapp;
1586 int i, start, end, forw, back, map, bit;
1587
1588 if (fs->fs_contigsumsize <= 0)
1589 return;
1590 freemapp = cg_clustersfree(cgp);
1591 sump = cg_clustersum(cgp);
1592 /*
1593 * Allocate or clear the actual block.
1594 */
1595 if (cnt > 0)
1596 setbit(freemapp, blkno);
1597 else
1598 clrbit(freemapp, blkno);
1599 /*
1600 * Find the size of the cluster going forward.
1601 */
1602 start = blkno + 1;
1603 end = start + fs->fs_contigsumsize;
1604 if (end >= cgp->cg_nclusterblks)
1605 end = cgp->cg_nclusterblks;
1606 mapp = &freemapp[start / NBBY];
1607 map = *mapp++;
1608 bit = 1 << (start % NBBY);
1609 for (i = start; i < end; i++) {
1610 if ((map & bit) == 0)
1611 break;
1612 if ((i & (NBBY - 1)) != (NBBY - 1)) {
1613 bit <<= 1;
1614 } else {
1615 map = *mapp++;
1616 bit = 1;
1617 }
1618 }
1619 forw = i - start;
1620 /*
1621 * Find the size of the cluster going backward.
1622 */
1623 start = blkno - 1;
1624 end = start - fs->fs_contigsumsize;
1625 if (end < 0)
1626 end = -1;
1627 mapp = &freemapp[start / NBBY];
1628 map = *mapp--;
1629 bit = 1 << (start % NBBY);
1630 for (i = start; i > end; i--) {
1631 if ((map & bit) == 0)
1632 break;
1633 if ((i & (NBBY - 1)) != 0) {
1634 bit >>= 1;
1635 } else {
1636 map = *mapp--;
1637 bit = 1 << (NBBY - 1);
1638 }
1639 }
1640 back = start - i;
1641 /*
1642 * Account for old cluster and the possibly new forward and
1643 * back clusters.
1644 */
1645 i = back + forw + 1;
1646 if (i > fs->fs_contigsumsize)
1647 i = fs->fs_contigsumsize;
1648 sump[i] += cnt;
1649 if (back > 0)
1650 sump[back] -= cnt;
1651 if (forw > 0)
1652 sump[forw] -= cnt;
1653 /*
1654 * Update cluster summary information.
1655 */
1656 lp = &sump[fs->fs_contigsumsize];
1657 for (i = fs->fs_contigsumsize; i > 0; i--)
1658 if (*lp-- > 0)
1659 break;
1660 fs->fs_maxcluster[cgp->cg_cgx] = i;
1661}
1662
1663/*
1664 * Fserr prints the name of a file system with an error diagnostic.
1665 *
1666 * The form of the error message is:
1667 * fs: error message
1668 */
1669static void
1670ffs_fserr(fs, uid, cp)
1671 struct fs *fs;
1672 u_int uid;
1673 char *cp;
1674{
1675 struct proc *p = curproc; /* XXX */
1676
1677 log(LOG_ERR, "pid %d (%s), uid %d on %s: %s\n", p ? p->p_pid : -1,
1678 p ? p->p_comm : "-", uid, fs->fs_fsmnt, cp);
1679}