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 1.55 1999/01/06 17:04:33 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 * Last, free the old blocks and assign the new blocks to the buffers.
515 */
516#ifdef DEBUG
517 if (prtrealloc)
518 printf("\n\tnew:");
519#endif
520 for (blkno = newblk, i = 0; i < len; i++, blkno += fs->fs_frag) {
521 if (!DOINGSOFTDEP(vp))
522 ffs_blkfree(ip,
523 dbtofsb(fs, buflist->bs_children[i]->b_blkno),
524 fs->fs_bsize);
525 buflist->bs_children[i]->b_blkno = fsbtodb(fs, blkno);
526#ifdef DEBUG
527 if (!ffs_checkblk(ip,
528 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
529 panic("ffs_reallocblks: unallocated block 3");
530 if (prtrealloc)
531 printf(" %d,", blkno);
532#endif
533 }
534#ifdef DEBUG
535 if (prtrealloc) {
536 prtrealloc--;
537 printf("\n");
538 }
539#endif
540 return (0);
541
542fail:
543 if (ssize < len)
544 brelse(ebp);
545 if (sbap != &ip->i_db[0])
546 brelse(sbp);
547 return (ENOSPC);
548}
549
550/*
551 * Allocate an inode in the file system.
552 *
553 * If allocating a directory, use ffs_dirpref to select the inode.
554 * If allocating in a directory, the following hierarchy is followed:
555 * 1) allocate the preferred inode.
556 * 2) allocate an inode in the same cylinder group.
557 * 3) quadradically rehash into other cylinder groups, until an
558 * available inode is located.
559 * If no inode preference is given the following heirarchy is used
560 * to allocate an inode:
561 * 1) allocate an inode in cylinder group 0.
562 * 2) quadradically rehash into other cylinder groups, until an
563 * available inode is located.
564 */
565int
566ffs_valloc(pvp, mode, cred, vpp)
567 struct vnode *pvp;
568 int mode;
569 struct ucred *cred;
570 struct vnode **vpp;
571{
572 register struct inode *pip;
573 register struct fs *fs;
574 register struct inode *ip;
575 ino_t ino, ipref;
576 int cg, error;
577
578 *vpp = NULL;
579 pip = VTOI(pvp);
580 fs = pip->i_fs;
581 if (fs->fs_cstotal.cs_nifree == 0)
582 goto noinodes;
583
584 if ((mode & IFMT) == IFDIR)
585 ipref = ffs_dirpref(fs);
586 else
587 ipref = pip->i_number;
588 if (ipref >= fs->fs_ncg * fs->fs_ipg)
589 ipref = 0;
590 cg = ino_to_cg(fs, ipref);
591 ino = (ino_t)ffs_hashalloc(pip, cg, (long)ipref, mode,
592 (allocfcn_t *)ffs_nodealloccg);
593 if (ino == 0)
594 goto noinodes;
595 error = VFS_VGET(pvp->v_mount, ino, vpp);
596 if (error) {
597 UFS_VFREE(pvp, ino, mode);
598 return (error);
599 }
600 ip = VTOI(*vpp);
601 if (ip->i_mode) {
602 printf("mode = 0%o, inum = %lu, fs = %s\n",
603 ip->i_mode, (u_long)ip->i_number, fs->fs_fsmnt);
604 panic("ffs_valloc: dup alloc");
605 }
606 if (ip->i_blocks) { /* XXX */
607 printf("free inode %s/%lu had %ld blocks\n",
608 fs->fs_fsmnt, (u_long)ino, (long)ip->i_blocks);
609 ip->i_blocks = 0;
610 }
611 ip->i_flags = 0;
612 /*
613 * Set up a new generation number for this inode.
614 */
615 if (ip->i_gen == 0 || ++ip->i_gen == 0)
616 ip->i_gen = random() / 2 + 1;
617 return (0);
618noinodes:
619 ffs_fserr(fs, cred->cr_uid, "out of inodes");
620 uprintf("\n%s: create/symlink failed, no inodes free\n", fs->fs_fsmnt);
621 return (ENOSPC);
622}
623
624/*
625 * Find a cylinder to place a directory.
626 *
627 * The policy implemented by this algorithm is to select from
628 * among those cylinder groups with above the average number of
629 * free inodes, the one with the smallest number of directories.
630 */
631static ino_t
632ffs_dirpref(fs)
633 register struct fs *fs;
634{
635 int cg, minndir, mincg, avgifree;
636
637 avgifree = fs->fs_cstotal.cs_nifree / fs->fs_ncg;
638 minndir = fs->fs_ipg;
639 mincg = 0;
640 for (cg = 0; cg < fs->fs_ncg; cg++)
641 if (fs->fs_cs(fs, cg).cs_ndir < minndir &&
642 fs->fs_cs(fs, cg).cs_nifree >= avgifree) {
643 mincg = cg;
644 minndir = fs->fs_cs(fs, cg).cs_ndir;
645 }
646 return ((ino_t)(fs->fs_ipg * mincg));
647}
648
649/*
650 * Select the desired position for the next block in a file. The file is
651 * logically divided into sections. The first section is composed of the
652 * direct blocks. Each additional section contains fs_maxbpg blocks.
653 *
654 * If no blocks have been allocated in the first section, the policy is to
655 * request a block in the same cylinder group as the inode that describes
656 * the file. If no blocks have been allocated in any other section, the
657 * policy is to place the section in a cylinder group with a greater than
658 * average number of free blocks. An appropriate cylinder group is found
659 * by using a rotor that sweeps the cylinder groups. When a new group of
660 * blocks is needed, the sweep begins in the cylinder group following the
661 * cylinder group from which the previous allocation was made. The sweep
662 * continues until a cylinder group with greater than the average number
663 * of free blocks is found. If the allocation is for the first block in an
664 * indirect block, the information on the previous allocation is unavailable;
665 * here a best guess is made based upon the logical block number being
666 * allocated.
667 *
668 * If a section is already partially allocated, the policy is to
669 * contiguously allocate fs_maxcontig blocks. The end of one of these
670 * contiguous blocks and the beginning of the next is physically separated
671 * so that the disk head will be in transit between them for at least
672 * fs_rotdelay milliseconds. This is to allow time for the processor to
673 * schedule another I/O transfer.
674 */
675ufs_daddr_t
676ffs_blkpref(ip, lbn, indx, bap)
677 struct inode *ip;
678 ufs_daddr_t lbn;
679 int indx;
680 ufs_daddr_t *bap;
681{
682 register struct fs *fs;
683 register int cg;
684 int avgbfree, startcg;
685 ufs_daddr_t nextblk;
686
687 fs = ip->i_fs;
688 if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) {
689 if (lbn < NDADDR) {
690 cg = ino_to_cg(fs, ip->i_number);
691 return (fs->fs_fpg * cg + fs->fs_frag);
692 }
693 /*
694 * Find a cylinder with greater than average number of
695 * unused data blocks.
696 */
697 if (indx == 0 || bap[indx - 1] == 0)
698 startcg =
699 ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg;
700 else
701 startcg = dtog(fs, bap[indx - 1]) + 1;
702 startcg %= fs->fs_ncg;
703 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
704 for (cg = startcg; cg < fs->fs_ncg; cg++)
705 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
706 fs->fs_cgrotor = cg;
707 return (fs->fs_fpg * cg + fs->fs_frag);
708 }
709 for (cg = 0; cg <= startcg; cg++)
710 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
711 fs->fs_cgrotor = cg;
712 return (fs->fs_fpg * cg + fs->fs_frag);
713 }
714 return (0);
715 }
716 /*
717 * One or more previous blocks have been laid out. If less
718 * than fs_maxcontig previous blocks are contiguous, the
719 * next block is requested contiguously, otherwise it is
720 * requested rotationally delayed by fs_rotdelay milliseconds.
721 */
722 nextblk = bap[indx - 1] + fs->fs_frag;
723 if (fs->fs_rotdelay == 0 || indx < fs->fs_maxcontig ||
724 bap[indx - fs->fs_maxcontig] +
725 blkstofrags(fs, fs->fs_maxcontig) != nextblk)
726 return (nextblk);
727 /*
728 * Here we convert ms of delay to frags as:
729 * (frags) = (ms) * (rev/sec) * (sect/rev) /
730 * ((sect/frag) * (ms/sec))
731 * then round up to the next block.
732 */
733 nextblk += roundup(fs->fs_rotdelay * fs->fs_rps * fs->fs_nsect /
734 (NSPF(fs) * 1000), fs->fs_frag);
735 return (nextblk);
736}
737
738/*
739 * Implement the cylinder overflow algorithm.
740 *
741 * The policy implemented by this algorithm is:
742 * 1) allocate the block in its requested cylinder group.
743 * 2) quadradically rehash on the cylinder group number.
744 * 3) brute force search for a free block.
745 */
746/*VARARGS5*/
747static u_long
748ffs_hashalloc(ip, cg, pref, size, allocator)
749 struct inode *ip;
750 int cg;
751 long pref;
752 int size; /* size for data blocks, mode for inodes */
753 allocfcn_t *allocator;
754{
755 register struct fs *fs;
756 long result; /* XXX why not same type as we return? */
757 int i, icg = cg;
758
759 fs = ip->i_fs;
760 /*
761 * 1: preferred cylinder group
762 */
763 result = (*allocator)(ip, cg, pref, size);
764 if (result)
765 return (result);
766 /*
767 * 2: quadratic rehash
768 */
769 for (i = 1; i < fs->fs_ncg; i *= 2) {
770 cg += i;
771 if (cg >= fs->fs_ncg)
772 cg -= fs->fs_ncg;
773 result = (*allocator)(ip, cg, 0, size);
774 if (result)
775 return (result);
776 }
777 /*
778 * 3: brute force search
779 * Note that we start at i == 2, since 0 was checked initially,
780 * and 1 is always checked in the quadratic rehash.
781 */
782 cg = (icg + 2) % fs->fs_ncg;
783 for (i = 2; i < fs->fs_ncg; i++) {
784 result = (*allocator)(ip, cg, 0, size);
785 if (result)
786 return (result);
787 cg++;
788 if (cg == fs->fs_ncg)
789 cg = 0;
790 }
791 return (0);
792}
793
794/*
795 * Determine whether a fragment can be extended.
796 *
797 * Check to see if the necessary fragments are available, and
798 * if they are, allocate them.
799 */
800static ufs_daddr_t
801ffs_fragextend(ip, cg, bprev, osize, nsize)
802 struct inode *ip;
803 int cg;
804 long bprev;
805 int osize, nsize;
806{
807 register struct fs *fs;
808 register struct cg *cgp;
809 struct buf *bp;
810 long bno;
811 int frags, bbase;
812 int i, error;
813
814 fs = ip->i_fs;
815 if (fs->fs_cs(fs, cg).cs_nffree < numfrags(fs, nsize - osize))
816 return (0);
817 frags = numfrags(fs, nsize);
818 bbase = fragnum(fs, bprev);
819 if (bbase > fragnum(fs, (bprev + frags - 1))) {
820 /* cannot extend across a block boundary */
821 return (0);
822 }
823 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
824 (int)fs->fs_cgsize, NOCRED, &bp);
825 if (error) {
826 brelse(bp);
827 return (0);
828 }
829 cgp = (struct cg *)bp->b_data;
830 if (!cg_chkmagic(cgp)) {
831 brelse(bp);
832 return (0);
833 }
834 cgp->cg_time = time_second;
835 bno = dtogd(fs, bprev);
836 for (i = numfrags(fs, osize); i < frags; i++)
837 if (isclr(cg_blksfree(cgp), bno + i)) {
838 brelse(bp);
839 return (0);
840 }
841 /*
842 * the current fragment can be extended
843 * deduct the count on fragment being extended into
844 * increase the count on the remaining fragment (if any)
845 * allocate the extended piece
846 */
847 for (i = frags; i < fs->fs_frag - bbase; i++)
848 if (isclr(cg_blksfree(cgp), bno + i))
849 break;
850 cgp->cg_frsum[i - numfrags(fs, osize)]--;
851 if (i != frags)
852 cgp->cg_frsum[i - frags]++;
853 for (i = numfrags(fs, osize); i < frags; i++) {
854 clrbit(cg_blksfree(cgp), bno + i);
855 cgp->cg_cs.cs_nffree--;
856 fs->fs_cstotal.cs_nffree--;
857 fs->fs_cs(fs, cg).cs_nffree--;
858 }
859 fs->fs_fmod = 1;
860 if (DOINGSOFTDEP(ITOV(ip)))
861 softdep_setup_blkmapdep(bp, fs, bprev);
862 bdwrite(bp);
863 return (bprev);
864}
865
866/*
867 * Determine whether a block can be allocated.
868 *
869 * Check to see if a block of the appropriate size is available,
870 * and if it is, allocate it.
871 */
872static ufs_daddr_t
873ffs_alloccg(ip, cg, bpref, size)
874 struct inode *ip;
875 int cg;
876 ufs_daddr_t bpref;
877 int size;
878{
879 register struct fs *fs;
880 register struct cg *cgp;
881 struct buf *bp;
882 register int i;
883 ufs_daddr_t bno, blkno;
884 int allocsiz, error, frags;
885
886 fs = ip->i_fs;
887 if (fs->fs_cs(fs, cg).cs_nbfree == 0 && size == fs->fs_bsize)
888 return (0);
889 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
890 (int)fs->fs_cgsize, NOCRED, &bp);
891 if (error) {
892 brelse(bp);
893 return (0);
894 }
895 cgp = (struct cg *)bp->b_data;
896 if (!cg_chkmagic(cgp) ||
897 (cgp->cg_cs.cs_nbfree == 0 && size == fs->fs_bsize)) {
898 brelse(bp);
899 return (0);
900 }
901 cgp->cg_time = time_second;
902 if (size == fs->fs_bsize) {
903 bno = ffs_alloccgblk(ip, bp, bpref);
904 bdwrite(bp);
905 return (bno);
906 }
907 /*
908 * check to see if any fragments are already available
909 * allocsiz is the size which will be allocated, hacking
910 * it down to a smaller size if necessary
911 */
912 frags = numfrags(fs, size);
913 for (allocsiz = frags; allocsiz < fs->fs_frag; allocsiz++)
914 if (cgp->cg_frsum[allocsiz] != 0)
915 break;
916 if (allocsiz == fs->fs_frag) {
917 /*
918 * no fragments were available, so a block will be
919 * allocated, and hacked up
920 */
921 if (cgp->cg_cs.cs_nbfree == 0) {
922 brelse(bp);
923 return (0);
924 }
925 bno = ffs_alloccgblk(ip, bp, bpref);
926 bpref = dtogd(fs, bno);
927 for (i = frags; i < fs->fs_frag; i++)
928 setbit(cg_blksfree(cgp), bpref + i);
929 i = fs->fs_frag - frags;
930 cgp->cg_cs.cs_nffree += i;
931 fs->fs_cstotal.cs_nffree += i;
932 fs->fs_cs(fs, cg).cs_nffree += i;
933 fs->fs_fmod = 1;
934 cgp->cg_frsum[i]++;
935 bdwrite(bp);
936 return (bno);
937 }
938 bno = ffs_mapsearch(fs, cgp, bpref, allocsiz);
939 if (bno < 0) {
940 brelse(bp);
941 return (0);
942 }
943 for (i = 0; i < frags; i++)
944 clrbit(cg_blksfree(cgp), bno + i);
945 cgp->cg_cs.cs_nffree -= frags;
946 fs->fs_cstotal.cs_nffree -= frags;
947 fs->fs_cs(fs, cg).cs_nffree -= frags;
948 fs->fs_fmod = 1;
949 cgp->cg_frsum[allocsiz]--;
950 if (frags != allocsiz)
951 cgp->cg_frsum[allocsiz - frags]++;
952 blkno = cg * fs->fs_fpg + bno;
953 if (DOINGSOFTDEP(ITOV(ip)))
954 softdep_setup_blkmapdep(bp, fs, blkno);
955 bdwrite(bp);
956 return ((u_long)blkno);
957}
958
959/*
960 * Allocate a block in a cylinder group.
961 *
962 * This algorithm implements the following policy:
963 * 1) allocate the requested block.
964 * 2) allocate a rotationally optimal block in the same cylinder.
965 * 3) allocate the next available block on the block rotor for the
966 * specified cylinder group.
967 * Note that this routine only allocates fs_bsize blocks; these
968 * blocks may be fragmented by the routine that allocates them.
969 */
970static ufs_daddr_t
971ffs_alloccgblk(ip, bp, bpref)
972 struct inode *ip;
973 struct buf *bp;
974 ufs_daddr_t bpref;
975{
976 struct fs *fs;
977 struct cg *cgp;
978 ufs_daddr_t bno, blkno;
979 int cylno, pos, delta;
980 short *cylbp;
981 register int i;
982
983 fs = ip->i_fs;
984 cgp = (struct cg *)bp->b_data;
985 if (bpref == 0 || dtog(fs, bpref) != cgp->cg_cgx) {
986 bpref = cgp->cg_rotor;
987 goto norot;
988 }
989 bpref = blknum(fs, bpref);
990 bpref = dtogd(fs, bpref);
991 /*
992 * if the requested block is available, use it
993 */
994 if (ffs_isblock(fs, cg_blksfree(cgp), fragstoblks(fs, bpref))) {
995 bno = bpref;
996 goto gotit;
997 }
998 if (fs->fs_nrpos <= 1 || fs->fs_cpc == 0) {
999 /*
1000 * Block layout information is not available.
1001 * Leaving bpref unchanged means we take the
1002 * next available free block following the one
1003 * we just allocated. Hopefully this will at
1004 * least hit a track cache on drives of unknown
1005 * geometry (e.g. SCSI).
1006 */
1007 goto norot;
1008 }
1009 /*
1010 * check for a block available on the same cylinder
1011 */
1012 cylno = cbtocylno(fs, bpref);
1013 if (cg_blktot(cgp)[cylno] == 0)
1014 goto norot;
1015 /*
1016 * check the summary information to see if a block is
1017 * available in the requested cylinder starting at the
1018 * requested rotational position and proceeding around.
1019 */
1020 cylbp = cg_blks(fs, cgp, cylno);
1021 pos = cbtorpos(fs, bpref);
1022 for (i = pos; i < fs->fs_nrpos; i++)
1023 if (cylbp[i] > 0)
1024 break;
1025 if (i == fs->fs_nrpos)
1026 for (i = 0; i < pos; i++)
1027 if (cylbp[i] > 0)
1028 break;
1029 if (cylbp[i] > 0) {
1030 /*
1031 * found a rotational position, now find the actual
1032 * block. A panic if none is actually there.
1033 */
1034 pos = cylno % fs->fs_cpc;
1035 bno = (cylno - pos) * fs->fs_spc / NSPB(fs);
1036 if (fs_postbl(fs, pos)[i] == -1) {
1037 printf("pos = %d, i = %d, fs = %s\n",
1038 pos, i, fs->fs_fsmnt);
1039 panic("ffs_alloccgblk: cyl groups corrupted");
1040 }
1041 for (i = fs_postbl(fs, pos)[i];; ) {
1042 if (ffs_isblock(fs, cg_blksfree(cgp), bno + i)) {
1043 bno = blkstofrags(fs, (bno + i));
1044 goto gotit;
1045 }
1046 delta = fs_rotbl(fs)[i];
1047 if (delta <= 0 ||
1048 delta + i > fragstoblks(fs, fs->fs_fpg))
1049 break;
1050 i += delta;
1051 }
1052 printf("pos = %d, i = %d, fs = %s\n", pos, i, fs->fs_fsmnt);
1053 panic("ffs_alloccgblk: can't find blk in cyl");
1054 }
1055norot:
1056 /*
1057 * no blocks in the requested cylinder, so take next
1058 * available one in this cylinder group.
1059 */
1060 bno = ffs_mapsearch(fs, cgp, bpref, (int)fs->fs_frag);
1061 if (bno < 0)
1062 return (0);
1063 cgp->cg_rotor = bno;
1064gotit:
1065 blkno = fragstoblks(fs, bno);
1066 ffs_clrblock(fs, cg_blksfree(cgp), (long)blkno);
1067 ffs_clusteracct(fs, cgp, blkno, -1);
1068 cgp->cg_cs.cs_nbfree--;
1069 fs->fs_cstotal.cs_nbfree--;
1070 fs->fs_cs(fs, cgp->cg_cgx).cs_nbfree--;
1071 cylno = cbtocylno(fs, bno);
1072 cg_blks(fs, cgp, cylno)[cbtorpos(fs, bno)]--;
1073 cg_blktot(cgp)[cylno]--;
1074 fs->fs_fmod = 1;
1075 blkno = cgp->cg_cgx * fs->fs_fpg + bno;
1076 if (DOINGSOFTDEP(ITOV(ip)))
1077 softdep_setup_blkmapdep(bp, fs, blkno);
1078 return (blkno);
1079}
1080
1081/*
1082 * Determine whether a cluster can be allocated.
1083 *
1084 * We do not currently check for optimal rotational layout if there
1085 * are multiple choices in the same cylinder group. Instead we just
1086 * take the first one that we find following bpref.
1087 */
1088static ufs_daddr_t
1089ffs_clusteralloc(ip, cg, bpref, len)
1090 struct inode *ip;
1091 int cg;
1092 ufs_daddr_t bpref;
1093 int len;
1094{
1095 register struct fs *fs;
1096 register struct cg *cgp;
1097 struct buf *bp;
1098 int i, got, run, bno, bit, map;
1099 u_char *mapp;
1100 int32_t *lp;
1101
1102 fs = ip->i_fs;
1103 if (fs->fs_maxcluster[cg] < len)
1104 return (NULL);
1105 if (bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), (int)fs->fs_cgsize,
1106 NOCRED, &bp))
1107 goto fail;
1108 cgp = (struct cg *)bp->b_data;
1109 if (!cg_chkmagic(cgp))
1110 goto fail;
1111 /*
1112 * Check to see if a cluster of the needed size (or bigger) is
1113 * available in this cylinder group.
1114 */
1115 lp = &cg_clustersum(cgp)[len];
1116 for (i = len; i <= fs->fs_contigsumsize; i++)
1117 if (*lp++ > 0)
1118 break;
1119 if (i > fs->fs_contigsumsize) {
1120 /*
1121 * This is the first time looking for a cluster in this
1122 * cylinder group. Update the cluster summary information
1123 * to reflect the true maximum sized cluster so that
1124 * future cluster allocation requests can avoid reading
1125 * the cylinder group map only to find no clusters.
1126 */
1127 lp = &cg_clustersum(cgp)[len - 1];
1128 for (i = len - 1; i > 0; i--)
1129 if (*lp-- > 0)
1130 break;
1131 fs->fs_maxcluster[cg] = i;
1132 goto fail;
1133 }
1134 /*
1135 * Search the cluster map to find a big enough cluster.
1136 * We take the first one that we find, even if it is larger
1137 * than we need as we prefer to get one close to the previous
1138 * block allocation. We do not search before the current
1139 * preference point as we do not want to allocate a block
1140 * that is allocated before the previous one (as we will
1141 * then have to wait for another pass of the elevator
1142 * algorithm before it will be read). We prefer to fail and
1143 * be recalled to try an allocation in the next cylinder group.
1144 */
1145 if (dtog(fs, bpref) != cg)
1146 bpref = 0;
1147 else
1148 bpref = fragstoblks(fs, dtogd(fs, blknum(fs, bpref)));
1149 mapp = &cg_clustersfree(cgp)[bpref / NBBY];
1150 map = *mapp++;
1151 bit = 1 << (bpref % NBBY);
1152 for (run = 0, got = bpref; got < cgp->cg_nclusterblks; got++) {
1153 if ((map & bit) == 0) {
1154 run = 0;
1155 } else {
1156 run++;
1157 if (run == len)
1158 break;
1159 }
1160 if ((got & (NBBY - 1)) != (NBBY - 1)) {
1161 bit <<= 1;
1162 } else {
1163 map = *mapp++;
1164 bit = 1;
1165 }
1166 }
1167 if (got >= cgp->cg_nclusterblks)
1168 goto fail;
1169 /*
1170 * Allocate the cluster that we have found.
1171 */
1172 for (i = 1; i <= len; i++)
1173 if (!ffs_isblock(fs, cg_blksfree(cgp), got - run + i))
1174 panic("ffs_clusteralloc: map mismatch");
1175 bno = cg * fs->fs_fpg + blkstofrags(fs, got - run + 1);
1176 if (dtog(fs, bno) != cg)
1177 panic("ffs_clusteralloc: allocated out of group");
1178 len = blkstofrags(fs, len);
1179 for (i = 0; i < len; i += fs->fs_frag)
1180 if ((got = ffs_alloccgblk(ip, bp, bno + i)) != bno + i)
1181 panic("ffs_clusteralloc: lost block");
1182 bdwrite(bp);
1183 return (bno);
1184
1185fail:
1186 brelse(bp);
1187 return (0);
1188}
1189
1190/*
1191 * Determine whether an inode can be allocated.
1192 *
1193 * Check to see if an inode is available, and if it is,
1194 * allocate it using the following policy:
1195 * 1) allocate the requested inode.
1196 * 2) allocate the next available inode after the requested
1197 * inode in the specified cylinder group.
1198 */
1199static ino_t
1200ffs_nodealloccg(ip, cg, ipref, mode)
1201 struct inode *ip;
1202 int cg;
1203 ufs_daddr_t ipref;
1204 int mode;
1205{
1206 register struct fs *fs;
1207 register struct cg *cgp;
1208 struct buf *bp;
1209 int error, start, len, loc, map, i;
1210
1211 fs = ip->i_fs;
1212 if (fs->fs_cs(fs, cg).cs_nifree == 0)
1213 return (0);
1214 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
1215 (int)fs->fs_cgsize, NOCRED, &bp);
1216 if (error) {
1217 brelse(bp);
1218 return (0);
1219 }
1220 cgp = (struct cg *)bp->b_data;
1221 if (!cg_chkmagic(cgp) || cgp->cg_cs.cs_nifree == 0) {
1222 brelse(bp);
1223 return (0);
1224 }
1225 cgp->cg_time = time_second;
1226 if (ipref) {
1227 ipref %= fs->fs_ipg;
1228 if (isclr(cg_inosused(cgp), ipref))
1229 goto gotit;
1230 }
1231 start = cgp->cg_irotor / NBBY;
1232 len = howmany(fs->fs_ipg - cgp->cg_irotor, NBBY);
1233 loc = skpc(0xff, len, &cg_inosused(cgp)[start]);
1234 if (loc == 0) {
1235 len = start + 1;
1236 start = 0;
1237 loc = skpc(0xff, len, &cg_inosused(cgp)[0]);
1238 if (loc == 0) {
1239 printf("cg = %d, irotor = %ld, fs = %s\n",
1240 cg, (long)cgp->cg_irotor, fs->fs_fsmnt);
1241 panic("ffs_nodealloccg: map corrupted");
1242 /* NOTREACHED */
1243 }
1244 }
1245 i = start + len - loc;
1246 map = cg_inosused(cgp)[i];
1247 ipref = i * NBBY;
1248 for (i = 1; i < (1 << NBBY); i <<= 1, ipref++) {
1249 if ((map & i) == 0) {
1250 cgp->cg_irotor = ipref;
1251 goto gotit;
1252 }
1253 }
1254 printf("fs = %s\n", fs->fs_fsmnt);
1255 panic("ffs_nodealloccg: block not in map");
1256 /* NOTREACHED */
1257gotit:
1258 if (DOINGSOFTDEP(ITOV(ip)))
1259 softdep_setup_inomapdep(bp, ip, cg * fs->fs_ipg + ipref);
1260 setbit(cg_inosused(cgp), ipref);
1261 cgp->cg_cs.cs_nifree--;
1262 fs->fs_cstotal.cs_nifree--;
1263 fs->fs_cs(fs, cg).cs_nifree--;
1264 fs->fs_fmod = 1;
1265 if ((mode & IFMT) == IFDIR) {
1266 cgp->cg_cs.cs_ndir++;
1267 fs->fs_cstotal.cs_ndir++;
1268 fs->fs_cs(fs, cg).cs_ndir++;
1269 }
1270 bdwrite(bp);
1271 return (cg * fs->fs_ipg + ipref);
1272}
1273
1274/*
1275 * Free a block or fragment.
1276 *
1277 * The specified block or fragment is placed back in the
1278 * free map. If a fragment is deallocated, a possible
1279 * block reassembly is checked.
1280 */
1281void
1282ffs_blkfree(ip, bno, size)
1283 register struct inode *ip;
1284 ufs_daddr_t bno;
1285 long size;
1286{
1287 register struct fs *fs;
1288 register struct cg *cgp;
1289 struct buf *bp;
1290 ufs_daddr_t blkno;
1291 int i, error, cg, blk, frags, bbase;
1292
1293 fs = ip->i_fs;
1294 VOP_FREEBLKS(ip->i_devvp, fsbtodb(fs, bno), size);
1295 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0 ||
1296 fragnum(fs, bno) + numfrags(fs, size) > fs->fs_frag) {
1297 printf("dev=0x%lx, bno = %d, bsize = %d, size = %ld, fs = %s\n",
1298 (u_long)ip->i_dev, bno, fs->fs_bsize, size, fs->fs_fsmnt);
1299 panic("ffs_blkfree: bad size");
1300 }
1301 cg = dtog(fs, bno);
1302 if ((u_int)bno >= fs->fs_size) {
1303 printf("bad block %ld, ino %lu\n",
1304 (long)bno, (u_long)ip->i_number);
1305 ffs_fserr(fs, ip->i_uid, "bad block");
1306 return;
1307 }
1308 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
1309 (int)fs->fs_cgsize, NOCRED, &bp);
1310 if (error) {
1311 brelse(bp);
1312 return;
1313 }
1314 cgp = (struct cg *)bp->b_data;
1315 if (!cg_chkmagic(cgp)) {
1316 brelse(bp);
1317 return;
1318 }
1319 cgp->cg_time = time_second;
1320 bno = dtogd(fs, bno);
1321 if (size == fs->fs_bsize) {
1322 blkno = fragstoblks(fs, bno);
1323 if (!ffs_isfreeblock(fs, cg_blksfree(cgp), blkno)) {
1324 printf("dev = 0x%lx, block = %ld, fs = %s\n",
1325 (u_long)ip->i_dev, (long)bno, fs->fs_fsmnt);
1326 panic("ffs_blkfree: freeing free block");
1327 }
1328 ffs_setblock(fs, cg_blksfree(cgp), blkno);
1329 ffs_clusteracct(fs, cgp, blkno, 1);
1330 cgp->cg_cs.cs_nbfree++;
1331 fs->fs_cstotal.cs_nbfree++;
1332 fs->fs_cs(fs, cg).cs_nbfree++;
1333 i = cbtocylno(fs, bno);
1334 cg_blks(fs, cgp, i)[cbtorpos(fs, bno)]++;
1335 cg_blktot(cgp)[i]++;
1336 } else {
1337 bbase = bno - fragnum(fs, bno);
1338 /*
1339 * decrement the counts associated with the old frags
1340 */
1341 blk = blkmap(fs, cg_blksfree(cgp), bbase);
1342 ffs_fragacct(fs, blk, cgp->cg_frsum, -1);
1343 /*
1344 * deallocate the fragment
1345 */
1346 frags = numfrags(fs, size);
1347 for (i = 0; i < frags; i++) {
1348 if (isset(cg_blksfree(cgp), bno + i)) {
1349 printf("dev = 0x%lx, block = %ld, fs = %s\n",
1350 (u_long)ip->i_dev, (long)(bno + i),
1351 fs->fs_fsmnt);
1352 panic("ffs_blkfree: freeing free frag");
1353 }
1354 setbit(cg_blksfree(cgp), bno + i);
1355 }
1356 cgp->cg_cs.cs_nffree += i;
1357 fs->fs_cstotal.cs_nffree += i;
1358 fs->fs_cs(fs, cg).cs_nffree += i;
1359 /*
1360 * add back in counts associated with the new frags
1361 */
1362 blk = blkmap(fs, cg_blksfree(cgp), bbase);
1363 ffs_fragacct(fs, blk, cgp->cg_frsum, 1);
1364 /*
1365 * if a complete block has been reassembled, account for it
1366 */
1367 blkno = fragstoblks(fs, bbase);
1368 if (ffs_isblock(fs, cg_blksfree(cgp), blkno)) {
1369 cgp->cg_cs.cs_nffree -= fs->fs_frag;
1370 fs->fs_cstotal.cs_nffree -= fs->fs_frag;
1371 fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag;
1372 ffs_clusteracct(fs, cgp, blkno, 1);
1373 cgp->cg_cs.cs_nbfree++;
1374 fs->fs_cstotal.cs_nbfree++;
1375 fs->fs_cs(fs, cg).cs_nbfree++;
1376 i = cbtocylno(fs, bbase);
1377 cg_blks(fs, cgp, i)[cbtorpos(fs, bbase)]++;
1378 cg_blktot(cgp)[i]++;
1379 }
1380 }
1381 fs->fs_fmod = 1;
1382 bdwrite(bp);
1383}
1384
1385#ifdef DIAGNOSTIC
1386/*
1387 * Verify allocation of a block or fragment. Returns true if block or
1388 * fragment is allocated, false if it is free.
1389 */
1390static int
1391ffs_checkblk(ip, bno, size)
1392 struct inode *ip;
1393 ufs_daddr_t bno;
1394 long size;
1395{
1396 struct fs *fs;
1397 struct cg *cgp;
1398 struct buf *bp;
1399 int i, error, frags, free;
1400
1401 fs = ip->i_fs;
1402 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) {
1403 printf("bsize = %ld, size = %ld, fs = %s\n",
1404 (long)fs->fs_bsize, size, fs->fs_fsmnt);
1405 panic("ffs_checkblk: bad size");
1406 }
1407 if ((u_int)bno >= fs->fs_size)
1408 panic("ffs_checkblk: bad block %d", bno);
1409 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, dtog(fs, bno))),
1410 (int)fs->fs_cgsize, NOCRED, &bp);
1411 if (error)
1412 panic("ffs_checkblk: cg bread failed");
1413 cgp = (struct cg *)bp->b_data;
1414 if (!cg_chkmagic(cgp))
1415 panic("ffs_checkblk: cg magic mismatch");
1416 bno = dtogd(fs, bno);
1417 if (size == fs->fs_bsize) {
1418 free = ffs_isblock(fs, cg_blksfree(cgp), fragstoblks(fs, bno));
1419 } else {
1420 frags = numfrags(fs, size);
1421 for (free = 0, i = 0; i < frags; i++)
1422 if (isset(cg_blksfree(cgp), bno + i))
1423 free++;
1424 if (free != 0 && free != frags)
1425 panic("ffs_checkblk: partially free fragment");
1426 }
1427 brelse(bp);
1428 return (!free);
1429}
1430#endif /* DIAGNOSTIC */
1431
1432/*
1433 * Free an inode.
1434 */
1435int
1436ffs_vfree( pvp, ino, mode)
1437 struct vnode *pvp;
1438 ino_t ino;
1439 int mode;
1440{
1441 if (DOINGSOFTDEP(pvp)) {
1442 softdep_freefile(pvp, ino, mode);
1443 return (0);
1444 }
1445 return (ffs_freefile(pvp, ino, mode));
1446}
1447
1448/*
1449 * Do the actual free operation.
1450 * The specified inode is placed back in the free map.
1451 */
1452 int
1453 ffs_freefile( pvp, ino, mode)
1454 struct vnode *pvp;
1455 ino_t ino;
1456 int mode;
1457{
1458 register struct fs *fs;
1459 register struct cg *cgp;
1460 register struct inode *pip;
1461 struct buf *bp;
1462 int error, cg;
1463
1464 pip = VTOI(pvp);
1465 fs = pip->i_fs;
1466 if ((u_int)ino >= fs->fs_ipg * fs->fs_ncg)
1467 panic("ffs_vfree: range: dev = 0x%x, ino = %d, fs = %s",
1468 pip->i_dev, ino, fs->fs_fsmnt);
1469 cg = ino_to_cg(fs, ino);
1470 error = bread(pip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
1471 (int)fs->fs_cgsize, NOCRED, &bp);
1472 if (error) {
1473 brelse(bp);
1474 return (error);
1475 }
1476 cgp = (struct cg *)bp->b_data;
1477 if (!cg_chkmagic(cgp)) {
1478 brelse(bp);
1479 return (0);
1480 }
1481 cgp->cg_time = time_second;
1482 ino %= fs->fs_ipg;
1483 if (isclr(cg_inosused(cgp), ino)) {
1484 printf("dev = 0x%lx, ino = %lu, fs = %s\n",
1485 (u_long)pip->i_dev, (u_long)ino, fs->fs_fsmnt);
1486 if (fs->fs_ronly == 0)
1487 panic("ffs_vfree: freeing free inode");
1488 }
1489 clrbit(cg_inosused(cgp), ino);
1490 if (ino < cgp->cg_irotor)
1491 cgp->cg_irotor = ino;
1492 cgp->cg_cs.cs_nifree++;
1493 fs->fs_cstotal.cs_nifree++;
1494 fs->fs_cs(fs, cg).cs_nifree++;
1495 if ((mode & IFMT) == IFDIR) {
1496 cgp->cg_cs.cs_ndir--;
1497 fs->fs_cstotal.cs_ndir--;
1498 fs->fs_cs(fs, cg).cs_ndir--;
1499 }
1500 fs->fs_fmod = 1;
1501 bdwrite(bp);
1502 return (0);
1503}
1504
1505/*
1506 * Find a block of the specified size in the specified cylinder group.
1507 *
1508 * It is a panic if a request is made to find a block if none are
1509 * available.
1510 */
1511static ufs_daddr_t
1512ffs_mapsearch(fs, cgp, bpref, allocsiz)
1513 register struct fs *fs;
1514 register struct cg *cgp;
1515 ufs_daddr_t bpref;
1516 int allocsiz;
1517{
1518 ufs_daddr_t bno;
1519 int start, len, loc, i;
1520 int blk, field, subfield, pos;
1521
1522 /*
1523 * find the fragment by searching through the free block
1524 * map for an appropriate bit pattern
1525 */
1526 if (bpref)
1527 start = dtogd(fs, bpref) / NBBY;
1528 else
1529 start = cgp->cg_frotor / NBBY;
1530 len = howmany(fs->fs_fpg, NBBY) - start;
1531 loc = scanc((u_int)len, (u_char *)&cg_blksfree(cgp)[start],
1532 (u_char *)fragtbl[fs->fs_frag],
1533 (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
1534 if (loc == 0) {
1535 len = start + 1;
1536 start = 0;
1537 loc = scanc((u_int)len, (u_char *)&cg_blksfree(cgp)[0],
1538 (u_char *)fragtbl[fs->fs_frag],
1539 (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
1540 if (loc == 0) {
1541 printf("start = %d, len = %d, fs = %s\n",
1542 start, len, fs->fs_fsmnt);
1543 panic("ffs_alloccg: map corrupted");
1544 /* NOTREACHED */
1545 }
1546 }
1547 bno = (start + len - loc) * NBBY;
1548 cgp->cg_frotor = bno;
1549 /*
1550 * found the byte in the map
1551 * sift through the bits to find the selected frag
1552 */
1553 for (i = bno + NBBY; bno < i; bno += fs->fs_frag) {
1554 blk = blkmap(fs, cg_blksfree(cgp), bno);
1555 blk <<= 1;
1556 field = around[allocsiz];
1557 subfield = inside[allocsiz];
1558 for (pos = 0; pos <= fs->fs_frag - allocsiz; pos++) {
1559 if ((blk & field) == subfield)
1560 return (bno + pos);
1561 field <<= 1;
1562 subfield <<= 1;
1563 }
1564 }
1565 printf("bno = %lu, fs = %s\n", (u_long)bno, fs->fs_fsmnt);
1566 panic("ffs_alloccg: block not in map");
1567 return (-1);
1568}
1569
1570/*
1571 * Update the cluster map because of an allocation or free.
1572 *
1573 * Cnt == 1 means free; cnt == -1 means allocating.
1574 */
1575static void
1576ffs_clusteracct(fs, cgp, blkno, cnt)
1577 struct fs *fs;
1578 struct cg *cgp;
1579 ufs_daddr_t blkno;
1580 int cnt;
1581{
1582 int32_t *sump;
1583 int32_t *lp;
1584 u_char *freemapp, *mapp;
1585 int i, start, end, forw, back, map, bit;
1586
1587 if (fs->fs_contigsumsize <= 0)
1588 return;
1589 freemapp = cg_clustersfree(cgp);
1590 sump = cg_clustersum(cgp);
1591 /*
1592 * Allocate or clear the actual block.
1593 */
1594 if (cnt > 0)
1595 setbit(freemapp, blkno);
1596 else
1597 clrbit(freemapp, blkno);
1598 /*
1599 * Find the size of the cluster going forward.
1600 */
1601 start = blkno + 1;
1602 end = start + fs->fs_contigsumsize;
1603 if (end >= cgp->cg_nclusterblks)
1604 end = cgp->cg_nclusterblks;
1605 mapp = &freemapp[start / NBBY];
1606 map = *mapp++;
1607 bit = 1 << (start % NBBY);
1608 for (i = start; i < end; i++) {
1609 if ((map & bit) == 0)
1610 break;
1611 if ((i & (NBBY - 1)) != (NBBY - 1)) {
1612 bit <<= 1;
1613 } else {
1614 map = *mapp++;
1615 bit = 1;
1616 }
1617 }
1618 forw = i - start;
1619 /*
1620 * Find the size of the cluster going backward.
1621 */
1622 start = blkno - 1;
1623 end = start - fs->fs_contigsumsize;
1624 if (end < 0)
1625 end = -1;
1626 mapp = &freemapp[start / NBBY];
1627 map = *mapp--;
1628 bit = 1 << (start % NBBY);
1629 for (i = start; i > end; i--) {
1630 if ((map & bit) == 0)
1631 break;
1632 if ((i & (NBBY - 1)) != 0) {
1633 bit >>= 1;
1634 } else {
1635 map = *mapp--;
1636 bit = 1 << (NBBY - 1);
1637 }
1638 }
1639 back = start - i;
1640 /*
1641 * Account for old cluster and the possibly new forward and
1642 * back clusters.
1643 */
1644 i = back + forw + 1;
1645 if (i > fs->fs_contigsumsize)
1646 i = fs->fs_contigsumsize;
1647 sump[i] += cnt;
1648 if (back > 0)
1649 sump[back] -= cnt;
1650 if (forw > 0)
1651 sump[forw] -= cnt;
1652 /*
1653 * Update cluster summary information.
1654 */
1655 lp = &sump[fs->fs_contigsumsize];
1656 for (i = fs->fs_contigsumsize; i > 0; i--)
1657 if (*lp-- > 0)
1658 break;
1659 fs->fs_maxcluster[cgp->cg_cgx] = i;
1660}
1661
1662/*
1663 * Fserr prints the name of a file system with an error diagnostic.
1664 *
1665 * The form of the error message is:
1666 * fs: error message
1667 */
1668static void
1669ffs_fserr(fs, uid, cp)
1670 struct fs *fs;
1671 u_int uid;
1672 char *cp;
1673{
1674 struct proc *p = curproc; /* XXX */
1675
1676 log(LOG_ERR, "pid %d (%s), uid %d on %s: %s\n", p ? p->p_pid : -1,
1677 p ? p->p_comm : "-", uid, fs->fs_fsmnt, cp);
1678}
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 1.55 1999/01/06 17:04:33 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 * Last, free the old blocks and assign the new blocks to the buffers.
515 */
516#ifdef DEBUG
517 if (prtrealloc)
518 printf("\n\tnew:");
519#endif
520 for (blkno = newblk, i = 0; i < len; i++, blkno += fs->fs_frag) {
521 if (!DOINGSOFTDEP(vp))
522 ffs_blkfree(ip,
523 dbtofsb(fs, buflist->bs_children[i]->b_blkno),
524 fs->fs_bsize);
525 buflist->bs_children[i]->b_blkno = fsbtodb(fs, blkno);
526#ifdef DEBUG
527 if (!ffs_checkblk(ip,
528 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
529 panic("ffs_reallocblks: unallocated block 3");
530 if (prtrealloc)
531 printf(" %d,", blkno);
532#endif
533 }
534#ifdef DEBUG
535 if (prtrealloc) {
536 prtrealloc--;
537 printf("\n");
538 }
539#endif
540 return (0);
541
542fail:
543 if (ssize < len)
544 brelse(ebp);
545 if (sbap != &ip->i_db[0])
546 brelse(sbp);
547 return (ENOSPC);
548}
549
550/*
551 * Allocate an inode in the file system.
552 *
553 * If allocating a directory, use ffs_dirpref to select the inode.
554 * If allocating in a directory, the following hierarchy is followed:
555 * 1) allocate the preferred inode.
556 * 2) allocate an inode in the same cylinder group.
557 * 3) quadradically rehash into other cylinder groups, until an
558 * available inode is located.
559 * If no inode preference is given the following heirarchy is used
560 * to allocate an inode:
561 * 1) allocate an inode in cylinder group 0.
562 * 2) quadradically rehash into other cylinder groups, until an
563 * available inode is located.
564 */
565int
566ffs_valloc(pvp, mode, cred, vpp)
567 struct vnode *pvp;
568 int mode;
569 struct ucred *cred;
570 struct vnode **vpp;
571{
572 register struct inode *pip;
573 register struct fs *fs;
574 register struct inode *ip;
575 ino_t ino, ipref;
576 int cg, error;
577
578 *vpp = NULL;
579 pip = VTOI(pvp);
580 fs = pip->i_fs;
581 if (fs->fs_cstotal.cs_nifree == 0)
582 goto noinodes;
583
584 if ((mode & IFMT) == IFDIR)
585 ipref = ffs_dirpref(fs);
586 else
587 ipref = pip->i_number;
588 if (ipref >= fs->fs_ncg * fs->fs_ipg)
589 ipref = 0;
590 cg = ino_to_cg(fs, ipref);
591 ino = (ino_t)ffs_hashalloc(pip, cg, (long)ipref, mode,
592 (allocfcn_t *)ffs_nodealloccg);
593 if (ino == 0)
594 goto noinodes;
595 error = VFS_VGET(pvp->v_mount, ino, vpp);
596 if (error) {
597 UFS_VFREE(pvp, ino, mode);
598 return (error);
599 }
600 ip = VTOI(*vpp);
601 if (ip->i_mode) {
602 printf("mode = 0%o, inum = %lu, fs = %s\n",
603 ip->i_mode, (u_long)ip->i_number, fs->fs_fsmnt);
604 panic("ffs_valloc: dup alloc");
605 }
606 if (ip->i_blocks) { /* XXX */
607 printf("free inode %s/%lu had %ld blocks\n",
608 fs->fs_fsmnt, (u_long)ino, (long)ip->i_blocks);
609 ip->i_blocks = 0;
610 }
611 ip->i_flags = 0;
612 /*
613 * Set up a new generation number for this inode.
614 */
615 if (ip->i_gen == 0 || ++ip->i_gen == 0)
616 ip->i_gen = random() / 2 + 1;
617 return (0);
618noinodes:
619 ffs_fserr(fs, cred->cr_uid, "out of inodes");
620 uprintf("\n%s: create/symlink failed, no inodes free\n", fs->fs_fsmnt);
621 return (ENOSPC);
622}
623
624/*
625 * Find a cylinder to place a directory.
626 *
627 * The policy implemented by this algorithm is to select from
628 * among those cylinder groups with above the average number of
629 * free inodes, the one with the smallest number of directories.
630 */
631static ino_t
632ffs_dirpref(fs)
633 register struct fs *fs;
634{
635 int cg, minndir, mincg, avgifree;
636
637 avgifree = fs->fs_cstotal.cs_nifree / fs->fs_ncg;
638 minndir = fs->fs_ipg;
639 mincg = 0;
640 for (cg = 0; cg < fs->fs_ncg; cg++)
641 if (fs->fs_cs(fs, cg).cs_ndir < minndir &&
642 fs->fs_cs(fs, cg).cs_nifree >= avgifree) {
643 mincg = cg;
644 minndir = fs->fs_cs(fs, cg).cs_ndir;
645 }
646 return ((ino_t)(fs->fs_ipg * mincg));
647}
648
649/*
650 * Select the desired position for the next block in a file. The file is
651 * logically divided into sections. The first section is composed of the
652 * direct blocks. Each additional section contains fs_maxbpg blocks.
653 *
654 * If no blocks have been allocated in the first section, the policy is to
655 * request a block in the same cylinder group as the inode that describes
656 * the file. If no blocks have been allocated in any other section, the
657 * policy is to place the section in a cylinder group with a greater than
658 * average number of free blocks. An appropriate cylinder group is found
659 * by using a rotor that sweeps the cylinder groups. When a new group of
660 * blocks is needed, the sweep begins in the cylinder group following the
661 * cylinder group from which the previous allocation was made. The sweep
662 * continues until a cylinder group with greater than the average number
663 * of free blocks is found. If the allocation is for the first block in an
664 * indirect block, the information on the previous allocation is unavailable;
665 * here a best guess is made based upon the logical block number being
666 * allocated.
667 *
668 * If a section is already partially allocated, the policy is to
669 * contiguously allocate fs_maxcontig blocks. The end of one of these
670 * contiguous blocks and the beginning of the next is physically separated
671 * so that the disk head will be in transit between them for at least
672 * fs_rotdelay milliseconds. This is to allow time for the processor to
673 * schedule another I/O transfer.
674 */
675ufs_daddr_t
676ffs_blkpref(ip, lbn, indx, bap)
677 struct inode *ip;
678 ufs_daddr_t lbn;
679 int indx;
680 ufs_daddr_t *bap;
681{
682 register struct fs *fs;
683 register int cg;
684 int avgbfree, startcg;
685 ufs_daddr_t nextblk;
686
687 fs = ip->i_fs;
688 if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) {
689 if (lbn < NDADDR) {
690 cg = ino_to_cg(fs, ip->i_number);
691 return (fs->fs_fpg * cg + fs->fs_frag);
692 }
693 /*
694 * Find a cylinder with greater than average number of
695 * unused data blocks.
696 */
697 if (indx == 0 || bap[indx - 1] == 0)
698 startcg =
699 ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg;
700 else
701 startcg = dtog(fs, bap[indx - 1]) + 1;
702 startcg %= fs->fs_ncg;
703 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
704 for (cg = startcg; cg < fs->fs_ncg; cg++)
705 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
706 fs->fs_cgrotor = cg;
707 return (fs->fs_fpg * cg + fs->fs_frag);
708 }
709 for (cg = 0; cg <= startcg; cg++)
710 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
711 fs->fs_cgrotor = cg;
712 return (fs->fs_fpg * cg + fs->fs_frag);
713 }
714 return (0);
715 }
716 /*
717 * One or more previous blocks have been laid out. If less
718 * than fs_maxcontig previous blocks are contiguous, the
719 * next block is requested contiguously, otherwise it is
720 * requested rotationally delayed by fs_rotdelay milliseconds.
721 */
722 nextblk = bap[indx - 1] + fs->fs_frag;
723 if (fs->fs_rotdelay == 0 || indx < fs->fs_maxcontig ||
724 bap[indx - fs->fs_maxcontig] +
725 blkstofrags(fs, fs->fs_maxcontig) != nextblk)
726 return (nextblk);
727 /*
728 * Here we convert ms of delay to frags as:
729 * (frags) = (ms) * (rev/sec) * (sect/rev) /
730 * ((sect/frag) * (ms/sec))
731 * then round up to the next block.
732 */
733 nextblk += roundup(fs->fs_rotdelay * fs->fs_rps * fs->fs_nsect /
734 (NSPF(fs) * 1000), fs->fs_frag);
735 return (nextblk);
736}
737
738/*
739 * Implement the cylinder overflow algorithm.
740 *
741 * The policy implemented by this algorithm is:
742 * 1) allocate the block in its requested cylinder group.
743 * 2) quadradically rehash on the cylinder group number.
744 * 3) brute force search for a free block.
745 */
746/*VARARGS5*/
747static u_long
748ffs_hashalloc(ip, cg, pref, size, allocator)
749 struct inode *ip;
750 int cg;
751 long pref;
752 int size; /* size for data blocks, mode for inodes */
753 allocfcn_t *allocator;
754{
755 register struct fs *fs;
756 long result; /* XXX why not same type as we return? */
757 int i, icg = cg;
758
759 fs = ip->i_fs;
760 /*
761 * 1: preferred cylinder group
762 */
763 result = (*allocator)(ip, cg, pref, size);
764 if (result)
765 return (result);
766 /*
767 * 2: quadratic rehash
768 */
769 for (i = 1; i < fs->fs_ncg; i *= 2) {
770 cg += i;
771 if (cg >= fs->fs_ncg)
772 cg -= fs->fs_ncg;
773 result = (*allocator)(ip, cg, 0, size);
774 if (result)
775 return (result);
776 }
777 /*
778 * 3: brute force search
779 * Note that we start at i == 2, since 0 was checked initially,
780 * and 1 is always checked in the quadratic rehash.
781 */
782 cg = (icg + 2) % fs->fs_ncg;
783 for (i = 2; i < fs->fs_ncg; i++) {
784 result = (*allocator)(ip, cg, 0, size);
785 if (result)
786 return (result);
787 cg++;
788 if (cg == fs->fs_ncg)
789 cg = 0;
790 }
791 return (0);
792}
793
794/*
795 * Determine whether a fragment can be extended.
796 *
797 * Check to see if the necessary fragments are available, and
798 * if they are, allocate them.
799 */
800static ufs_daddr_t
801ffs_fragextend(ip, cg, bprev, osize, nsize)
802 struct inode *ip;
803 int cg;
804 long bprev;
805 int osize, nsize;
806{
807 register struct fs *fs;
808 register struct cg *cgp;
809 struct buf *bp;
810 long bno;
811 int frags, bbase;
812 int i, error;
813
814 fs = ip->i_fs;
815 if (fs->fs_cs(fs, cg).cs_nffree < numfrags(fs, nsize - osize))
816 return (0);
817 frags = numfrags(fs, nsize);
818 bbase = fragnum(fs, bprev);
819 if (bbase > fragnum(fs, (bprev + frags - 1))) {
820 /* cannot extend across a block boundary */
821 return (0);
822 }
823 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
824 (int)fs->fs_cgsize, NOCRED, &bp);
825 if (error) {
826 brelse(bp);
827 return (0);
828 }
829 cgp = (struct cg *)bp->b_data;
830 if (!cg_chkmagic(cgp)) {
831 brelse(bp);
832 return (0);
833 }
834 cgp->cg_time = time_second;
835 bno = dtogd(fs, bprev);
836 for (i = numfrags(fs, osize); i < frags; i++)
837 if (isclr(cg_blksfree(cgp), bno + i)) {
838 brelse(bp);
839 return (0);
840 }
841 /*
842 * the current fragment can be extended
843 * deduct the count on fragment being extended into
844 * increase the count on the remaining fragment (if any)
845 * allocate the extended piece
846 */
847 for (i = frags; i < fs->fs_frag - bbase; i++)
848 if (isclr(cg_blksfree(cgp), bno + i))
849 break;
850 cgp->cg_frsum[i - numfrags(fs, osize)]--;
851 if (i != frags)
852 cgp->cg_frsum[i - frags]++;
853 for (i = numfrags(fs, osize); i < frags; i++) {
854 clrbit(cg_blksfree(cgp), bno + i);
855 cgp->cg_cs.cs_nffree--;
856 fs->fs_cstotal.cs_nffree--;
857 fs->fs_cs(fs, cg).cs_nffree--;
858 }
859 fs->fs_fmod = 1;
860 if (DOINGSOFTDEP(ITOV(ip)))
861 softdep_setup_blkmapdep(bp, fs, bprev);
862 bdwrite(bp);
863 return (bprev);
864}
865
866/*
867 * Determine whether a block can be allocated.
868 *
869 * Check to see if a block of the appropriate size is available,
870 * and if it is, allocate it.
871 */
872static ufs_daddr_t
873ffs_alloccg(ip, cg, bpref, size)
874 struct inode *ip;
875 int cg;
876 ufs_daddr_t bpref;
877 int size;
878{
879 register struct fs *fs;
880 register struct cg *cgp;
881 struct buf *bp;
882 register int i;
883 ufs_daddr_t bno, blkno;
884 int allocsiz, error, frags;
885
886 fs = ip->i_fs;
887 if (fs->fs_cs(fs, cg).cs_nbfree == 0 && size == fs->fs_bsize)
888 return (0);
889 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
890 (int)fs->fs_cgsize, NOCRED, &bp);
891 if (error) {
892 brelse(bp);
893 return (0);
894 }
895 cgp = (struct cg *)bp->b_data;
896 if (!cg_chkmagic(cgp) ||
897 (cgp->cg_cs.cs_nbfree == 0 && size == fs->fs_bsize)) {
898 brelse(bp);
899 return (0);
900 }
901 cgp->cg_time = time_second;
902 if (size == fs->fs_bsize) {
903 bno = ffs_alloccgblk(ip, bp, bpref);
904 bdwrite(bp);
905 return (bno);
906 }
907 /*
908 * check to see if any fragments are already available
909 * allocsiz is the size which will be allocated, hacking
910 * it down to a smaller size if necessary
911 */
912 frags = numfrags(fs, size);
913 for (allocsiz = frags; allocsiz < fs->fs_frag; allocsiz++)
914 if (cgp->cg_frsum[allocsiz] != 0)
915 break;
916 if (allocsiz == fs->fs_frag) {
917 /*
918 * no fragments were available, so a block will be
919 * allocated, and hacked up
920 */
921 if (cgp->cg_cs.cs_nbfree == 0) {
922 brelse(bp);
923 return (0);
924 }
925 bno = ffs_alloccgblk(ip, bp, bpref);
926 bpref = dtogd(fs, bno);
927 for (i = frags; i < fs->fs_frag; i++)
928 setbit(cg_blksfree(cgp), bpref + i);
929 i = fs->fs_frag - frags;
930 cgp->cg_cs.cs_nffree += i;
931 fs->fs_cstotal.cs_nffree += i;
932 fs->fs_cs(fs, cg).cs_nffree += i;
933 fs->fs_fmod = 1;
934 cgp->cg_frsum[i]++;
935 bdwrite(bp);
936 return (bno);
937 }
938 bno = ffs_mapsearch(fs, cgp, bpref, allocsiz);
939 if (bno < 0) {
940 brelse(bp);
941 return (0);
942 }
943 for (i = 0; i < frags; i++)
944 clrbit(cg_blksfree(cgp), bno + i);
945 cgp->cg_cs.cs_nffree -= frags;
946 fs->fs_cstotal.cs_nffree -= frags;
947 fs->fs_cs(fs, cg).cs_nffree -= frags;
948 fs->fs_fmod = 1;
949 cgp->cg_frsum[allocsiz]--;
950 if (frags != allocsiz)
951 cgp->cg_frsum[allocsiz - frags]++;
952 blkno = cg * fs->fs_fpg + bno;
953 if (DOINGSOFTDEP(ITOV(ip)))
954 softdep_setup_blkmapdep(bp, fs, blkno);
955 bdwrite(bp);
956 return ((u_long)blkno);
957}
958
959/*
960 * Allocate a block in a cylinder group.
961 *
962 * This algorithm implements the following policy:
963 * 1) allocate the requested block.
964 * 2) allocate a rotationally optimal block in the same cylinder.
965 * 3) allocate the next available block on the block rotor for the
966 * specified cylinder group.
967 * Note that this routine only allocates fs_bsize blocks; these
968 * blocks may be fragmented by the routine that allocates them.
969 */
970static ufs_daddr_t
971ffs_alloccgblk(ip, bp, bpref)
972 struct inode *ip;
973 struct buf *bp;
974 ufs_daddr_t bpref;
975{
976 struct fs *fs;
977 struct cg *cgp;
978 ufs_daddr_t bno, blkno;
979 int cylno, pos, delta;
980 short *cylbp;
981 register int i;
982
983 fs = ip->i_fs;
984 cgp = (struct cg *)bp->b_data;
985 if (bpref == 0 || dtog(fs, bpref) != cgp->cg_cgx) {
986 bpref = cgp->cg_rotor;
987 goto norot;
988 }
989 bpref = blknum(fs, bpref);
990 bpref = dtogd(fs, bpref);
991 /*
992 * if the requested block is available, use it
993 */
994 if (ffs_isblock(fs, cg_blksfree(cgp), fragstoblks(fs, bpref))) {
995 bno = bpref;
996 goto gotit;
997 }
998 if (fs->fs_nrpos <= 1 || fs->fs_cpc == 0) {
999 /*
1000 * Block layout information is not available.
1001 * Leaving bpref unchanged means we take the
1002 * next available free block following the one
1003 * we just allocated. Hopefully this will at
1004 * least hit a track cache on drives of unknown
1005 * geometry (e.g. SCSI).
1006 */
1007 goto norot;
1008 }
1009 /*
1010 * check for a block available on the same cylinder
1011 */
1012 cylno = cbtocylno(fs, bpref);
1013 if (cg_blktot(cgp)[cylno] == 0)
1014 goto norot;
1015 /*
1016 * check the summary information to see if a block is
1017 * available in the requested cylinder starting at the
1018 * requested rotational position and proceeding around.
1019 */
1020 cylbp = cg_blks(fs, cgp, cylno);
1021 pos = cbtorpos(fs, bpref);
1022 for (i = pos; i < fs->fs_nrpos; i++)
1023 if (cylbp[i] > 0)
1024 break;
1025 if (i == fs->fs_nrpos)
1026 for (i = 0; i < pos; i++)
1027 if (cylbp[i] > 0)
1028 break;
1029 if (cylbp[i] > 0) {
1030 /*
1031 * found a rotational position, now find the actual
1032 * block. A panic if none is actually there.
1033 */
1034 pos = cylno % fs->fs_cpc;
1035 bno = (cylno - pos) * fs->fs_spc / NSPB(fs);
1036 if (fs_postbl(fs, pos)[i] == -1) {
1037 printf("pos = %d, i = %d, fs = %s\n",
1038 pos, i, fs->fs_fsmnt);
1039 panic("ffs_alloccgblk: cyl groups corrupted");
1040 }
1041 for (i = fs_postbl(fs, pos)[i];; ) {
1042 if (ffs_isblock(fs, cg_blksfree(cgp), bno + i)) {
1043 bno = blkstofrags(fs, (bno + i));
1044 goto gotit;
1045 }
1046 delta = fs_rotbl(fs)[i];
1047 if (delta <= 0 ||
1048 delta + i > fragstoblks(fs, fs->fs_fpg))
1049 break;
1050 i += delta;
1051 }
1052 printf("pos = %d, i = %d, fs = %s\n", pos, i, fs->fs_fsmnt);
1053 panic("ffs_alloccgblk: can't find blk in cyl");
1054 }
1055norot:
1056 /*
1057 * no blocks in the requested cylinder, so take next
1058 * available one in this cylinder group.
1059 */
1060 bno = ffs_mapsearch(fs, cgp, bpref, (int)fs->fs_frag);
1061 if (bno < 0)
1062 return (0);
1063 cgp->cg_rotor = bno;
1064gotit:
1065 blkno = fragstoblks(fs, bno);
1066 ffs_clrblock(fs, cg_blksfree(cgp), (long)blkno);
1067 ffs_clusteracct(fs, cgp, blkno, -1);
1068 cgp->cg_cs.cs_nbfree--;
1069 fs->fs_cstotal.cs_nbfree--;
1070 fs->fs_cs(fs, cgp->cg_cgx).cs_nbfree--;
1071 cylno = cbtocylno(fs, bno);
1072 cg_blks(fs, cgp, cylno)[cbtorpos(fs, bno)]--;
1073 cg_blktot(cgp)[cylno]--;
1074 fs->fs_fmod = 1;
1075 blkno = cgp->cg_cgx * fs->fs_fpg + bno;
1076 if (DOINGSOFTDEP(ITOV(ip)))
1077 softdep_setup_blkmapdep(bp, fs, blkno);
1078 return (blkno);
1079}
1080
1081/*
1082 * Determine whether a cluster can be allocated.
1083 *
1084 * We do not currently check for optimal rotational layout if there
1085 * are multiple choices in the same cylinder group. Instead we just
1086 * take the first one that we find following bpref.
1087 */
1088static ufs_daddr_t
1089ffs_clusteralloc(ip, cg, bpref, len)
1090 struct inode *ip;
1091 int cg;
1092 ufs_daddr_t bpref;
1093 int len;
1094{
1095 register struct fs *fs;
1096 register struct cg *cgp;
1097 struct buf *bp;
1098 int i, got, run, bno, bit, map;
1099 u_char *mapp;
1100 int32_t *lp;
1101
1102 fs = ip->i_fs;
1103 if (fs->fs_maxcluster[cg] < len)
1104 return (NULL);
1105 if (bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), (int)fs->fs_cgsize,
1106 NOCRED, &bp))
1107 goto fail;
1108 cgp = (struct cg *)bp->b_data;
1109 if (!cg_chkmagic(cgp))
1110 goto fail;
1111 /*
1112 * Check to see if a cluster of the needed size (or bigger) is
1113 * available in this cylinder group.
1114 */
1115 lp = &cg_clustersum(cgp)[len];
1116 for (i = len; i <= fs->fs_contigsumsize; i++)
1117 if (*lp++ > 0)
1118 break;
1119 if (i > fs->fs_contigsumsize) {
1120 /*
1121 * This is the first time looking for a cluster in this
1122 * cylinder group. Update the cluster summary information
1123 * to reflect the true maximum sized cluster so that
1124 * future cluster allocation requests can avoid reading
1125 * the cylinder group map only to find no clusters.
1126 */
1127 lp = &cg_clustersum(cgp)[len - 1];
1128 for (i = len - 1; i > 0; i--)
1129 if (*lp-- > 0)
1130 break;
1131 fs->fs_maxcluster[cg] = i;
1132 goto fail;
1133 }
1134 /*
1135 * Search the cluster map to find a big enough cluster.
1136 * We take the first one that we find, even if it is larger
1137 * than we need as we prefer to get one close to the previous
1138 * block allocation. We do not search before the current
1139 * preference point as we do not want to allocate a block
1140 * that is allocated before the previous one (as we will
1141 * then have to wait for another pass of the elevator
1142 * algorithm before it will be read). We prefer to fail and
1143 * be recalled to try an allocation in the next cylinder group.
1144 */
1145 if (dtog(fs, bpref) != cg)
1146 bpref = 0;
1147 else
1148 bpref = fragstoblks(fs, dtogd(fs, blknum(fs, bpref)));
1149 mapp = &cg_clustersfree(cgp)[bpref / NBBY];
1150 map = *mapp++;
1151 bit = 1 << (bpref % NBBY);
1152 for (run = 0, got = bpref; got < cgp->cg_nclusterblks; got++) {
1153 if ((map & bit) == 0) {
1154 run = 0;
1155 } else {
1156 run++;
1157 if (run == len)
1158 break;
1159 }
1160 if ((got & (NBBY - 1)) != (NBBY - 1)) {
1161 bit <<= 1;
1162 } else {
1163 map = *mapp++;
1164 bit = 1;
1165 }
1166 }
1167 if (got >= cgp->cg_nclusterblks)
1168 goto fail;
1169 /*
1170 * Allocate the cluster that we have found.
1171 */
1172 for (i = 1; i <= len; i++)
1173 if (!ffs_isblock(fs, cg_blksfree(cgp), got - run + i))
1174 panic("ffs_clusteralloc: map mismatch");
1175 bno = cg * fs->fs_fpg + blkstofrags(fs, got - run + 1);
1176 if (dtog(fs, bno) != cg)
1177 panic("ffs_clusteralloc: allocated out of group");
1178 len = blkstofrags(fs, len);
1179 for (i = 0; i < len; i += fs->fs_frag)
1180 if ((got = ffs_alloccgblk(ip, bp, bno + i)) != bno + i)
1181 panic("ffs_clusteralloc: lost block");
1182 bdwrite(bp);
1183 return (bno);
1184
1185fail:
1186 brelse(bp);
1187 return (0);
1188}
1189
1190/*
1191 * Determine whether an inode can be allocated.
1192 *
1193 * Check to see if an inode is available, and if it is,
1194 * allocate it using the following policy:
1195 * 1) allocate the requested inode.
1196 * 2) allocate the next available inode after the requested
1197 * inode in the specified cylinder group.
1198 */
1199static ino_t
1200ffs_nodealloccg(ip, cg, ipref, mode)
1201 struct inode *ip;
1202 int cg;
1203 ufs_daddr_t ipref;
1204 int mode;
1205{
1206 register struct fs *fs;
1207 register struct cg *cgp;
1208 struct buf *bp;
1209 int error, start, len, loc, map, i;
1210
1211 fs = ip->i_fs;
1212 if (fs->fs_cs(fs, cg).cs_nifree == 0)
1213 return (0);
1214 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
1215 (int)fs->fs_cgsize, NOCRED, &bp);
1216 if (error) {
1217 brelse(bp);
1218 return (0);
1219 }
1220 cgp = (struct cg *)bp->b_data;
1221 if (!cg_chkmagic(cgp) || cgp->cg_cs.cs_nifree == 0) {
1222 brelse(bp);
1223 return (0);
1224 }
1225 cgp->cg_time = time_second;
1226 if (ipref) {
1227 ipref %= fs->fs_ipg;
1228 if (isclr(cg_inosused(cgp), ipref))
1229 goto gotit;
1230 }
1231 start = cgp->cg_irotor / NBBY;
1232 len = howmany(fs->fs_ipg - cgp->cg_irotor, NBBY);
1233 loc = skpc(0xff, len, &cg_inosused(cgp)[start]);
1234 if (loc == 0) {
1235 len = start + 1;
1236 start = 0;
1237 loc = skpc(0xff, len, &cg_inosused(cgp)[0]);
1238 if (loc == 0) {
1239 printf("cg = %d, irotor = %ld, fs = %s\n",
1240 cg, (long)cgp->cg_irotor, fs->fs_fsmnt);
1241 panic("ffs_nodealloccg: map corrupted");
1242 /* NOTREACHED */
1243 }
1244 }
1245 i = start + len - loc;
1246 map = cg_inosused(cgp)[i];
1247 ipref = i * NBBY;
1248 for (i = 1; i < (1 << NBBY); i <<= 1, ipref++) {
1249 if ((map & i) == 0) {
1250 cgp->cg_irotor = ipref;
1251 goto gotit;
1252 }
1253 }
1254 printf("fs = %s\n", fs->fs_fsmnt);
1255 panic("ffs_nodealloccg: block not in map");
1256 /* NOTREACHED */
1257gotit:
1258 if (DOINGSOFTDEP(ITOV(ip)))
1259 softdep_setup_inomapdep(bp, ip, cg * fs->fs_ipg + ipref);
1260 setbit(cg_inosused(cgp), ipref);
1261 cgp->cg_cs.cs_nifree--;
1262 fs->fs_cstotal.cs_nifree--;
1263 fs->fs_cs(fs, cg).cs_nifree--;
1264 fs->fs_fmod = 1;
1265 if ((mode & IFMT) == IFDIR) {
1266 cgp->cg_cs.cs_ndir++;
1267 fs->fs_cstotal.cs_ndir++;
1268 fs->fs_cs(fs, cg).cs_ndir++;
1269 }
1270 bdwrite(bp);
1271 return (cg * fs->fs_ipg + ipref);
1272}
1273
1274/*
1275 * Free a block or fragment.
1276 *
1277 * The specified block or fragment is placed back in the
1278 * free map. If a fragment is deallocated, a possible
1279 * block reassembly is checked.
1280 */
1281void
1282ffs_blkfree(ip, bno, size)
1283 register struct inode *ip;
1284 ufs_daddr_t bno;
1285 long size;
1286{
1287 register struct fs *fs;
1288 register struct cg *cgp;
1289 struct buf *bp;
1290 ufs_daddr_t blkno;
1291 int i, error, cg, blk, frags, bbase;
1292
1293 fs = ip->i_fs;
1294 VOP_FREEBLKS(ip->i_devvp, fsbtodb(fs, bno), size);
1295 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0 ||
1296 fragnum(fs, bno) + numfrags(fs, size) > fs->fs_frag) {
1297 printf("dev=0x%lx, bno = %d, bsize = %d, size = %ld, fs = %s\n",
1298 (u_long)ip->i_dev, bno, fs->fs_bsize, size, fs->fs_fsmnt);
1299 panic("ffs_blkfree: bad size");
1300 }
1301 cg = dtog(fs, bno);
1302 if ((u_int)bno >= fs->fs_size) {
1303 printf("bad block %ld, ino %lu\n",
1304 (long)bno, (u_long)ip->i_number);
1305 ffs_fserr(fs, ip->i_uid, "bad block");
1306 return;
1307 }
1308 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
1309 (int)fs->fs_cgsize, NOCRED, &bp);
1310 if (error) {
1311 brelse(bp);
1312 return;
1313 }
1314 cgp = (struct cg *)bp->b_data;
1315 if (!cg_chkmagic(cgp)) {
1316 brelse(bp);
1317 return;
1318 }
1319 cgp->cg_time = time_second;
1320 bno = dtogd(fs, bno);
1321 if (size == fs->fs_bsize) {
1322 blkno = fragstoblks(fs, bno);
1323 if (!ffs_isfreeblock(fs, cg_blksfree(cgp), blkno)) {
1324 printf("dev = 0x%lx, block = %ld, fs = %s\n",
1325 (u_long)ip->i_dev, (long)bno, fs->fs_fsmnt);
1326 panic("ffs_blkfree: freeing free block");
1327 }
1328 ffs_setblock(fs, cg_blksfree(cgp), blkno);
1329 ffs_clusteracct(fs, cgp, blkno, 1);
1330 cgp->cg_cs.cs_nbfree++;
1331 fs->fs_cstotal.cs_nbfree++;
1332 fs->fs_cs(fs, cg).cs_nbfree++;
1333 i = cbtocylno(fs, bno);
1334 cg_blks(fs, cgp, i)[cbtorpos(fs, bno)]++;
1335 cg_blktot(cgp)[i]++;
1336 } else {
1337 bbase = bno - fragnum(fs, bno);
1338 /*
1339 * decrement the counts associated with the old frags
1340 */
1341 blk = blkmap(fs, cg_blksfree(cgp), bbase);
1342 ffs_fragacct(fs, blk, cgp->cg_frsum, -1);
1343 /*
1344 * deallocate the fragment
1345 */
1346 frags = numfrags(fs, size);
1347 for (i = 0; i < frags; i++) {
1348 if (isset(cg_blksfree(cgp), bno + i)) {
1349 printf("dev = 0x%lx, block = %ld, fs = %s\n",
1350 (u_long)ip->i_dev, (long)(bno + i),
1351 fs->fs_fsmnt);
1352 panic("ffs_blkfree: freeing free frag");
1353 }
1354 setbit(cg_blksfree(cgp), bno + i);
1355 }
1356 cgp->cg_cs.cs_nffree += i;
1357 fs->fs_cstotal.cs_nffree += i;
1358 fs->fs_cs(fs, cg).cs_nffree += i;
1359 /*
1360 * add back in counts associated with the new frags
1361 */
1362 blk = blkmap(fs, cg_blksfree(cgp), bbase);
1363 ffs_fragacct(fs, blk, cgp->cg_frsum, 1);
1364 /*
1365 * if a complete block has been reassembled, account for it
1366 */
1367 blkno = fragstoblks(fs, bbase);
1368 if (ffs_isblock(fs, cg_blksfree(cgp), blkno)) {
1369 cgp->cg_cs.cs_nffree -= fs->fs_frag;
1370 fs->fs_cstotal.cs_nffree -= fs->fs_frag;
1371 fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag;
1372 ffs_clusteracct(fs, cgp, blkno, 1);
1373 cgp->cg_cs.cs_nbfree++;
1374 fs->fs_cstotal.cs_nbfree++;
1375 fs->fs_cs(fs, cg).cs_nbfree++;
1376 i = cbtocylno(fs, bbase);
1377 cg_blks(fs, cgp, i)[cbtorpos(fs, bbase)]++;
1378 cg_blktot(cgp)[i]++;
1379 }
1380 }
1381 fs->fs_fmod = 1;
1382 bdwrite(bp);
1383}
1384
1385#ifdef DIAGNOSTIC
1386/*
1387 * Verify allocation of a block or fragment. Returns true if block or
1388 * fragment is allocated, false if it is free.
1389 */
1390static int
1391ffs_checkblk(ip, bno, size)
1392 struct inode *ip;
1393 ufs_daddr_t bno;
1394 long size;
1395{
1396 struct fs *fs;
1397 struct cg *cgp;
1398 struct buf *bp;
1399 int i, error, frags, free;
1400
1401 fs = ip->i_fs;
1402 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) {
1403 printf("bsize = %ld, size = %ld, fs = %s\n",
1404 (long)fs->fs_bsize, size, fs->fs_fsmnt);
1405 panic("ffs_checkblk: bad size");
1406 }
1407 if ((u_int)bno >= fs->fs_size)
1408 panic("ffs_checkblk: bad block %d", bno);
1409 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, dtog(fs, bno))),
1410 (int)fs->fs_cgsize, NOCRED, &bp);
1411 if (error)
1412 panic("ffs_checkblk: cg bread failed");
1413 cgp = (struct cg *)bp->b_data;
1414 if (!cg_chkmagic(cgp))
1415 panic("ffs_checkblk: cg magic mismatch");
1416 bno = dtogd(fs, bno);
1417 if (size == fs->fs_bsize) {
1418 free = ffs_isblock(fs, cg_blksfree(cgp), fragstoblks(fs, bno));
1419 } else {
1420 frags = numfrags(fs, size);
1421 for (free = 0, i = 0; i < frags; i++)
1422 if (isset(cg_blksfree(cgp), bno + i))
1423 free++;
1424 if (free != 0 && free != frags)
1425 panic("ffs_checkblk: partially free fragment");
1426 }
1427 brelse(bp);
1428 return (!free);
1429}
1430#endif /* DIAGNOSTIC */
1431
1432/*
1433 * Free an inode.
1434 */
1435int
1436ffs_vfree( pvp, ino, mode)
1437 struct vnode *pvp;
1438 ino_t ino;
1439 int mode;
1440{
1441 if (DOINGSOFTDEP(pvp)) {
1442 softdep_freefile(pvp, ino, mode);
1443 return (0);
1444 }
1445 return (ffs_freefile(pvp, ino, mode));
1446}
1447
1448/*
1449 * Do the actual free operation.
1450 * The specified inode is placed back in the free map.
1451 */
1452 int
1453 ffs_freefile( pvp, ino, mode)
1454 struct vnode *pvp;
1455 ino_t ino;
1456 int mode;
1457{
1458 register struct fs *fs;
1459 register struct cg *cgp;
1460 register struct inode *pip;
1461 struct buf *bp;
1462 int error, cg;
1463
1464 pip = VTOI(pvp);
1465 fs = pip->i_fs;
1466 if ((u_int)ino >= fs->fs_ipg * fs->fs_ncg)
1467 panic("ffs_vfree: range: dev = 0x%x, ino = %d, fs = %s",
1468 pip->i_dev, ino, fs->fs_fsmnt);
1469 cg = ino_to_cg(fs, ino);
1470 error = bread(pip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
1471 (int)fs->fs_cgsize, NOCRED, &bp);
1472 if (error) {
1473 brelse(bp);
1474 return (error);
1475 }
1476 cgp = (struct cg *)bp->b_data;
1477 if (!cg_chkmagic(cgp)) {
1478 brelse(bp);
1479 return (0);
1480 }
1481 cgp->cg_time = time_second;
1482 ino %= fs->fs_ipg;
1483 if (isclr(cg_inosused(cgp), ino)) {
1484 printf("dev = 0x%lx, ino = %lu, fs = %s\n",
1485 (u_long)pip->i_dev, (u_long)ino, fs->fs_fsmnt);
1486 if (fs->fs_ronly == 0)
1487 panic("ffs_vfree: freeing free inode");
1488 }
1489 clrbit(cg_inosused(cgp), ino);
1490 if (ino < cgp->cg_irotor)
1491 cgp->cg_irotor = ino;
1492 cgp->cg_cs.cs_nifree++;
1493 fs->fs_cstotal.cs_nifree++;
1494 fs->fs_cs(fs, cg).cs_nifree++;
1495 if ((mode & IFMT) == IFDIR) {
1496 cgp->cg_cs.cs_ndir--;
1497 fs->fs_cstotal.cs_ndir--;
1498 fs->fs_cs(fs, cg).cs_ndir--;
1499 }
1500 fs->fs_fmod = 1;
1501 bdwrite(bp);
1502 return (0);
1503}
1504
1505/*
1506 * Find a block of the specified size in the specified cylinder group.
1507 *
1508 * It is a panic if a request is made to find a block if none are
1509 * available.
1510 */
1511static ufs_daddr_t
1512ffs_mapsearch(fs, cgp, bpref, allocsiz)
1513 register struct fs *fs;
1514 register struct cg *cgp;
1515 ufs_daddr_t bpref;
1516 int allocsiz;
1517{
1518 ufs_daddr_t bno;
1519 int start, len, loc, i;
1520 int blk, field, subfield, pos;
1521
1522 /*
1523 * find the fragment by searching through the free block
1524 * map for an appropriate bit pattern
1525 */
1526 if (bpref)
1527 start = dtogd(fs, bpref) / NBBY;
1528 else
1529 start = cgp->cg_frotor / NBBY;
1530 len = howmany(fs->fs_fpg, NBBY) - start;
1531 loc = scanc((u_int)len, (u_char *)&cg_blksfree(cgp)[start],
1532 (u_char *)fragtbl[fs->fs_frag],
1533 (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
1534 if (loc == 0) {
1535 len = start + 1;
1536 start = 0;
1537 loc = scanc((u_int)len, (u_char *)&cg_blksfree(cgp)[0],
1538 (u_char *)fragtbl[fs->fs_frag],
1539 (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
1540 if (loc == 0) {
1541 printf("start = %d, len = %d, fs = %s\n",
1542 start, len, fs->fs_fsmnt);
1543 panic("ffs_alloccg: map corrupted");
1544 /* NOTREACHED */
1545 }
1546 }
1547 bno = (start + len - loc) * NBBY;
1548 cgp->cg_frotor = bno;
1549 /*
1550 * found the byte in the map
1551 * sift through the bits to find the selected frag
1552 */
1553 for (i = bno + NBBY; bno < i; bno += fs->fs_frag) {
1554 blk = blkmap(fs, cg_blksfree(cgp), bno);
1555 blk <<= 1;
1556 field = around[allocsiz];
1557 subfield = inside[allocsiz];
1558 for (pos = 0; pos <= fs->fs_frag - allocsiz; pos++) {
1559 if ((blk & field) == subfield)
1560 return (bno + pos);
1561 field <<= 1;
1562 subfield <<= 1;
1563 }
1564 }
1565 printf("bno = %lu, fs = %s\n", (u_long)bno, fs->fs_fsmnt);
1566 panic("ffs_alloccg: block not in map");
1567 return (-1);
1568}
1569
1570/*
1571 * Update the cluster map because of an allocation or free.
1572 *
1573 * Cnt == 1 means free; cnt == -1 means allocating.
1574 */
1575static void
1576ffs_clusteracct(fs, cgp, blkno, cnt)
1577 struct fs *fs;
1578 struct cg *cgp;
1579 ufs_daddr_t blkno;
1580 int cnt;
1581{
1582 int32_t *sump;
1583 int32_t *lp;
1584 u_char *freemapp, *mapp;
1585 int i, start, end, forw, back, map, bit;
1586
1587 if (fs->fs_contigsumsize <= 0)
1588 return;
1589 freemapp = cg_clustersfree(cgp);
1590 sump = cg_clustersum(cgp);
1591 /*
1592 * Allocate or clear the actual block.
1593 */
1594 if (cnt > 0)
1595 setbit(freemapp, blkno);
1596 else
1597 clrbit(freemapp, blkno);
1598 /*
1599 * Find the size of the cluster going forward.
1600 */
1601 start = blkno + 1;
1602 end = start + fs->fs_contigsumsize;
1603 if (end >= cgp->cg_nclusterblks)
1604 end = cgp->cg_nclusterblks;
1605 mapp = &freemapp[start / NBBY];
1606 map = *mapp++;
1607 bit = 1 << (start % NBBY);
1608 for (i = start; i < end; i++) {
1609 if ((map & bit) == 0)
1610 break;
1611 if ((i & (NBBY - 1)) != (NBBY - 1)) {
1612 bit <<= 1;
1613 } else {
1614 map = *mapp++;
1615 bit = 1;
1616 }
1617 }
1618 forw = i - start;
1619 /*
1620 * Find the size of the cluster going backward.
1621 */
1622 start = blkno - 1;
1623 end = start - fs->fs_contigsumsize;
1624 if (end < 0)
1625 end = -1;
1626 mapp = &freemapp[start / NBBY];
1627 map = *mapp--;
1628 bit = 1 << (start % NBBY);
1629 for (i = start; i > end; i--) {
1630 if ((map & bit) == 0)
1631 break;
1632 if ((i & (NBBY - 1)) != 0) {
1633 bit >>= 1;
1634 } else {
1635 map = *mapp--;
1636 bit = 1 << (NBBY - 1);
1637 }
1638 }
1639 back = start - i;
1640 /*
1641 * Account for old cluster and the possibly new forward and
1642 * back clusters.
1643 */
1644 i = back + forw + 1;
1645 if (i > fs->fs_contigsumsize)
1646 i = fs->fs_contigsumsize;
1647 sump[i] += cnt;
1648 if (back > 0)
1649 sump[back] -= cnt;
1650 if (forw > 0)
1651 sump[forw] -= cnt;
1652 /*
1653 * Update cluster summary information.
1654 */
1655 lp = &sump[fs->fs_contigsumsize];
1656 for (i = fs->fs_contigsumsize; i > 0; i--)
1657 if (*lp-- > 0)
1658 break;
1659 fs->fs_maxcluster[cgp->cg_cgx] = i;
1660}
1661
1662/*
1663 * Fserr prints the name of a file system with an error diagnostic.
1664 *
1665 * The form of the error message is:
1666 * fs: error message
1667 */
1668static void
1669ffs_fserr(fs, uid, cp)
1670 struct fs *fs;
1671 u_int uid;
1672 char *cp;
1673{
1674 struct proc *p = curproc; /* XXX */
1675
1676 log(LOG_ERR, "pid %d (%s), uid %d on %s: %s\n", p ? p->p_pid : -1,
1677 p ? p->p_comm : "-", uid, fs->fs_fsmnt, cp);
1678}