2 * Copyright (c) 1982, 1986, 1989, 1993
3 * The Regents of the University of California. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
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.
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
33 * @(#)ffs_alloc.c 8.18 (Berkeley) 5/26/95
34 * $FreeBSD: src/sys/ufs/ffs/ffs_alloc.c,v 1.64.2.2 2001/09/21 19:15:21 dillon Exp $
35 * $DragonFly: src/sys/vfs/ufs/ffs_alloc.c,v 1.20 2006/04/28 06:13:56 dillon Exp $
38 #include "opt_quota.h"
40 #include <sys/param.h>
41 #include <sys/systm.h>
45 #include <sys/vnode.h>
46 #include <sys/mount.h>
47 #include <sys/kernel.h>
48 #include <sys/sysctl.h>
49 #include <sys/syslog.h>
51 #include <machine/inttypes.h>
55 #include "ufs_extern.h"
59 #include "ffs_extern.h"
61 typedef ufs_daddr_t allocfcn_t (struct inode *ip, int cg, ufs_daddr_t bpref,
64 static ufs_daddr_t ffs_alloccg (struct inode *, int, ufs_daddr_t, int);
66 ffs_alloccgblk (struct inode *, struct buf *, ufs_daddr_t);
68 static int ffs_checkblk (struct inode *, ufs_daddr_t, long);
70 static void ffs_clusteracct (struct fs *, struct cg *, ufs_daddr_t,
72 static ufs_daddr_t ffs_clusteralloc (struct inode *, int, ufs_daddr_t,
74 static ino_t ffs_dirpref (struct inode *);
75 static ufs_daddr_t ffs_fragextend (struct inode *, int, long, int, int);
76 static void ffs_fserr (struct fs *, uint, char *);
77 static u_long ffs_hashalloc
78 (struct inode *, int, long, int, allocfcn_t *);
79 static ino_t ffs_nodealloccg (struct inode *, int, ufs_daddr_t, int);
80 static ufs_daddr_t ffs_mapsearch (struct fs *, struct cg *, ufs_daddr_t,
84 * Allocate a block in the filesystem.
86 * The size of the requested block is given, which must be some
87 * multiple of fs_fsize and <= fs_bsize.
88 * A preference may be optionally specified. If a preference is given
89 * the following hierarchy is used to allocate a block:
90 * 1) allocate the requested block.
91 * 2) allocate a rotationally optimal block in the same cylinder.
92 * 3) allocate a block in the same cylinder group.
93 * 4) quadradically rehash into other cylinder groups, until an
94 * available block is located.
95 * If no block preference is given the following heirarchy is used
96 * to allocate a block:
97 * 1) allocate a block in the cylinder group that contains the
99 * 2) quadradically rehash into other cylinder groups, until an
100 * available block is located.
103 ffs_alloc(struct inode *ip, ufs_daddr_t lbn, ufs_daddr_t bpref, int size,
104 struct ucred *cred, ufs_daddr_t *bnp)
116 if ((uint)size > fs->fs_bsize || fragoff(fs, size) != 0) {
117 printf("dev = %s, bsize = %ld, size = %d, fs = %s\n",
118 devtoname(ip->i_dev), (long)fs->fs_bsize, size,
120 panic("ffs_alloc: bad size");
123 panic("ffs_alloc: missing credential");
124 #endif /* DIAGNOSTIC */
125 if (size == fs->fs_bsize && fs->fs_cstotal.cs_nbfree == 0)
127 if (cred->cr_uid != 0 &&
128 freespace(fs, fs->fs_minfree) - numfrags(fs, size) < 0)
131 error = ufs_chkdq(ip, (long)btodb(size), cred, 0);
135 if (bpref >= fs->fs_size)
138 cg = ino_to_cg(fs, ip->i_number);
140 cg = dtog(fs, bpref);
141 bno = (ufs_daddr_t)ffs_hashalloc(ip, cg, (long)bpref, size,
144 ip->i_blocks += btodb(size);
145 ip->i_flag |= IN_CHANGE | IN_UPDATE;
151 * Restore user's disk quota because allocation failed.
153 (void) ufs_chkdq(ip, (long)-btodb(size), cred, FORCE);
156 ffs_fserr(fs, cred->cr_uid, "filesystem full");
157 uprintf("\n%s: write failed, filesystem is full\n", fs->fs_fsmnt);
162 * Reallocate a fragment to a bigger size
164 * The number and size of the old block is given, and a preference
165 * and new size is also specified. The allocator attempts to extend
166 * the original block. Failing that, the regular block allocator is
167 * invoked to get an appropriate block.
170 ffs_realloccg(struct inode *ip, ufs_daddr_t lbprev, ufs_daddr_t bpref,
171 int osize, int nsize, struct ucred *cred, struct buf **bpp)
175 int cg, request, error;
176 ufs_daddr_t bprev, bno;
181 if ((uint)osize > fs->fs_bsize || fragoff(fs, osize) != 0 ||
182 (uint)nsize > fs->fs_bsize || fragoff(fs, nsize) != 0) {
184 "dev = %s, bsize = %ld, osize = %d, nsize = %d, fs = %s\n",
185 devtoname(ip->i_dev), (long)fs->fs_bsize, osize,
186 nsize, fs->fs_fsmnt);
187 panic("ffs_realloccg: bad size");
190 panic("ffs_realloccg: missing credential");
191 #endif /* DIAGNOSTIC */
192 if (cred->cr_uid != 0 &&
193 freespace(fs, fs->fs_minfree) - numfrags(fs, nsize - osize) < 0)
195 if ((bprev = ip->i_db[lbprev]) == 0) {
196 printf("dev = %s, bsize = %ld, bprev = %ld, fs = %s\n",
197 devtoname(ip->i_dev), (long)fs->fs_bsize, (long)bprev,
199 panic("ffs_realloccg: bad bprev");
202 * Allocate the extra space in the buffer.
204 error = bread(ITOV(ip), lblktodoff(fs, lbprev), osize, &bp);
210 if(bp->b_bio2.bio_offset == NOOFFSET) {
211 if( lbprev >= NDADDR)
212 panic("ffs_realloccg: lbprev out of range");
213 bp->b_bio2.bio_offset = fsbtodoff(fs, bprev);
217 error = ufs_chkdq(ip, (long)btodb(nsize - osize), cred, 0);
224 * Check for extension in the existing location.
226 cg = dtog(fs, bprev);
227 bno = ffs_fragextend(ip, cg, (long)bprev, osize, nsize);
229 if (bp->b_bio2.bio_offset != fsbtodoff(fs, bno))
230 panic("ffs_realloccg: bad blockno");
231 ip->i_blocks += btodb(nsize - osize);
232 ip->i_flag |= IN_CHANGE | IN_UPDATE;
234 bp->b_flags |= B_DONE;
235 bzero((char *)bp->b_data + osize, (uint)nsize - osize);
240 * Allocate a new disk location.
242 if (bpref >= fs->fs_size)
244 switch ((int)fs->fs_optim) {
247 * Allocate an exact sized fragment. Although this makes
248 * best use of space, we will waste time relocating it if
249 * the file continues to grow. If the fragmentation is
250 * less than half of the minimum free reserve, we choose
251 * to begin optimizing for time.
254 if (fs->fs_minfree <= 5 ||
255 fs->fs_cstotal.cs_nffree >
256 (off_t)fs->fs_dsize * fs->fs_minfree / (2 * 100))
258 log(LOG_NOTICE, "%s: optimization changed from SPACE to TIME\n",
260 fs->fs_optim = FS_OPTTIME;
264 * At this point we have discovered a file that is trying to
265 * grow a small fragment to a larger fragment. To save time,
266 * we allocate a full sized block, then free the unused portion.
267 * If the file continues to grow, the `ffs_fragextend' call
268 * above will be able to grow it in place without further
269 * copying. If aberrant programs cause disk fragmentation to
270 * grow within 2% of the free reserve, we choose to begin
271 * optimizing for space.
273 request = fs->fs_bsize;
274 if (fs->fs_cstotal.cs_nffree <
275 (off_t)fs->fs_dsize * (fs->fs_minfree - 2) / 100)
277 log(LOG_NOTICE, "%s: optimization changed from TIME to SPACE\n",
279 fs->fs_optim = FS_OPTSPACE;
282 printf("dev = %s, optim = %ld, fs = %s\n",
283 devtoname(ip->i_dev), (long)fs->fs_optim, fs->fs_fsmnt);
284 panic("ffs_realloccg: bad optim");
287 bno = (ufs_daddr_t)ffs_hashalloc(ip, cg, (long)bpref, request,
290 bp->b_bio2.bio_offset = fsbtodoff(fs, bno);
291 if (!DOINGSOFTDEP(ITOV(ip)))
292 ffs_blkfree(ip, bprev, (long)osize);
294 ffs_blkfree(ip, bno + numfrags(fs, nsize),
295 (long)(request - nsize));
296 ip->i_blocks += btodb(nsize - osize);
297 ip->i_flag |= IN_CHANGE | IN_UPDATE;
299 bp->b_flags |= B_DONE;
300 bzero((char *)bp->b_data + osize, (uint)nsize - osize);
306 * Restore user's disk quota because allocation failed.
308 (void) ufs_chkdq(ip, (long)-btodb(nsize - osize), cred, FORCE);
315 ffs_fserr(fs, cred->cr_uid, "filesystem full");
316 uprintf("\n%s: write failed, filesystem is full\n", fs->fs_fsmnt);
320 SYSCTL_NODE(_vfs, OID_AUTO, ffs, CTLFLAG_RW, 0, "FFS filesystem");
323 * Reallocate a sequence of blocks into a contiguous sequence of blocks.
325 * The vnode and an array of buffer pointers for a range of sequential
326 * logical blocks to be made contiguous is given. The allocator attempts
327 * to find a range of sequential blocks starting as close as possible to
328 * an fs_rotdelay offset from the end of the allocation for the logical
329 * block immediately preceeding the current range. If successful, the
330 * physical block numbers in the buffer pointers and in the inode are
331 * changed to reflect the new allocation. If unsuccessful, the allocation
332 * is left unchanged. The success in doing the reallocation is returned.
333 * Note that the error return is not reflected back to the user. Rather
334 * the previous block allocation will be used.
336 static int doasyncfree = 1;
337 SYSCTL_INT(_vfs_ffs, FFS_ASYNCFREE, doasyncfree, CTLFLAG_RW, &doasyncfree, 0, "");
339 static int doreallocblks = 1;
340 SYSCTL_INT(_vfs_ffs, FFS_REALLOCBLKS, doreallocblks, CTLFLAG_RW, &doreallocblks, 0, "");
343 static volatile int prtrealloc = 0;
347 * ffs_reallocblks(struct vnode *a_vp, struct cluster_save *a_buflist)
350 ffs_reallocblks(struct vop_reallocblks_args *ap)
355 struct buf *sbp, *ebp;
356 ufs_daddr_t *bap, *sbap, *ebap = 0;
357 struct cluster_save *buflist;
358 ufs_daddr_t start_lbn, end_lbn, soff, newblk, blkno;
362 struct indir start_ap[NIADDR + 1], end_ap[NIADDR + 1], *idp;
363 int i, len, slen, start_lvl, end_lvl, pref, ssize;
365 if (doreallocblks == 0)
370 if (fs->fs_contigsumsize <= 0)
372 buflist = ap->a_buflist;
373 len = buflist->bs_nchildren;
374 start_lbn = lblkno(fs, buflist->bs_children[0]->b_loffset);
375 end_lbn = start_lbn + len - 1;
377 for (i = 0; i < len; i++)
378 if (!ffs_checkblk(ip,
379 dofftofsb(fs, buflist->bs_children[i]->b_bio2.bio_offset), fs->fs_bsize))
380 panic("ffs_reallocblks: unallocated block 1");
381 for (i = 1; i < len; i++) {
382 if (buflist->bs_children[i]->b_loffset != lblktodoff(fs, start_lbn) + lblktodoff(fs, i))
383 panic("ffs_reallocblks: non-logical cluster");
385 boffset = buflist->bs_children[0]->b_bio2.bio_offset;
386 ssize = (int)fsbtodoff(fs, fs->fs_frag);
387 for (i = 1; i < len - 1; i++)
388 if (buflist->bs_children[i]->b_bio2.bio_offset != boffset + (i * ssize))
389 panic("ffs_reallocblks: non-physical cluster %d", i);
392 * If the latest allocation is in a new cylinder group, assume that
393 * the filesystem has decided to move and do not force it back to
394 * the previous cylinder group.
396 if (dtog(fs, dofftofsb(fs, buflist->bs_children[0]->b_bio2.bio_offset)) !=
397 dtog(fs, dofftofsb(fs, buflist->bs_children[len - 1]->b_bio2.bio_offset)))
399 if (ufs_getlbns(vp, start_lbn, start_ap, &start_lvl) ||
400 ufs_getlbns(vp, end_lbn, end_ap, &end_lvl))
403 * Get the starting offset and block map for the first block and
404 * the number of blocks that will fit into sbap starting at soff.
406 if (start_lvl == 0) {
409 slen = NDADDR - soff;
411 idp = &start_ap[start_lvl - 1];
412 if (bread(vp, lblktodoff(fs, idp->in_lbn), (int)fs->fs_bsize, &sbp)) {
416 sbap = (ufs_daddr_t *)sbp->b_data;
418 slen = fs->fs_nindir - soff;
421 * Find the preferred location for the cluster.
423 pref = ffs_blkpref(ip, start_lbn, soff, sbap);
425 * If the block range spans two block maps, get the second map.
427 if (end_lvl == 0 || (idp = &end_ap[end_lvl - 1])->in_off + 1 >= len) {
431 if (start_ap[start_lvl-1].in_lbn == idp->in_lbn)
432 panic("ffs_reallocblk: start == end");
434 ssize = len - (idp->in_off + 1);
435 if (bread(vp, lblktodoff(fs, idp->in_lbn), (int)fs->fs_bsize, &ebp))
437 ebap = (ufs_daddr_t *)ebp->b_data;
441 * Make sure we aren't spanning more then two blockmaps. ssize is
442 * our calculation of the span we have to scan in the first blockmap,
443 * while slen is our calculation of the number of entries available
444 * in the first blockmap (from soff).
447 panic("ffs_reallocblks: range spans more then two blockmaps!"
448 " start_lbn %ld len %d (%d/%d)",
449 (long)start_lbn, len, slen, ssize);
452 * Search the block map looking for an allocation of the desired size.
454 if ((newblk = (ufs_daddr_t)ffs_hashalloc(ip, dtog(fs, pref), (long)pref,
455 len, ffs_clusteralloc)) == 0)
458 * We have found a new contiguous block.
460 * First we have to replace the old block pointers with the new
461 * block pointers in the inode and indirect blocks associated
466 printf("realloc: ino %d, lbns %d-%d\n\told:", ip->i_number,
470 for (bap = &sbap[soff], i = 0; i < len; i++, blkno += fs->fs_frag) {
476 if (!ffs_checkblk(ip,
477 dofftofsb(fs, buflist->bs_children[i]->b_bio2.bio_offset), fs->fs_bsize))
478 panic("ffs_reallocblks: unallocated block 2");
479 if (dofftofsb(fs, buflist->bs_children[i]->b_bio2.bio_offset) != *bap)
480 panic("ffs_reallocblks: alloc mismatch");
484 printf(" %d,", *bap);
486 if (DOINGSOFTDEP(vp)) {
487 if (sbap == &ip->i_db[0] && i < ssize)
488 softdep_setup_allocdirect(ip, start_lbn + i,
489 blkno, *bap, fs->fs_bsize, fs->fs_bsize,
490 buflist->bs_children[i]);
492 softdep_setup_allocindir_page(ip, start_lbn + i,
493 i < ssize ? sbp : ebp, soff + i, blkno,
494 *bap, buflist->bs_children[i]);
499 * Next we must write out the modified inode and indirect blocks.
500 * For strict correctness, the writes should be synchronous since
501 * the old block values may have been written to disk. In practise
502 * they are almost never written, but if we are concerned about
503 * strict correctness, the `doasyncfree' flag should be set to zero.
505 * The test on `doasyncfree' should be changed to test a flag
506 * that shows whether the associated buffers and inodes have
507 * been written. The flag should be set when the cluster is
508 * started and cleared whenever the buffer or inode is flushed.
509 * We can then check below to see if it is set, and do the
510 * synchronous write only when it has been cleared.
512 if (sbap != &ip->i_db[0]) {
518 ip->i_flag |= IN_CHANGE | IN_UPDATE;
529 * Last, free the old blocks and assign the new blocks to the buffers.
535 for (blkno = newblk, i = 0; i < len; i++, blkno += fs->fs_frag) {
536 if (!DOINGSOFTDEP(vp))
538 dofftofsb(fs, buflist->bs_children[i]->b_bio2.bio_offset),
540 buflist->bs_children[i]->b_bio2.bio_offset = fsbtodoff(fs, blkno);
542 if (!ffs_checkblk(ip,
543 dofftofsb(fs, buflist->bs_children[i]->b_bio2.bio_offset), fs->fs_bsize))
544 panic("ffs_reallocblks: unallocated block 3");
548 printf(" %d,", blkno);
562 if (sbap != &ip->i_db[0])
568 * Allocate an inode in the filesystem.
570 * If allocating a directory, use ffs_dirpref to select the inode.
571 * If allocating in a directory, the following hierarchy is followed:
572 * 1) allocate the preferred inode.
573 * 2) allocate an inode in the same cylinder group.
574 * 3) quadradically rehash into other cylinder groups, until an
575 * available inode is located.
576 * If no inode preference is given the following heirarchy is used
577 * to allocate an inode:
578 * 1) allocate an inode in cylinder group 0.
579 * 2) quadradically rehash into other cylinder groups, until an
580 * available inode is located.
583 ffs_valloc(struct vnode *pvp, int mode, struct ucred *cred, struct vnode **vpp)
594 if (fs->fs_cstotal.cs_nifree == 0)
597 if ((mode & IFMT) == IFDIR)
598 ipref = ffs_dirpref(pip);
600 ipref = pip->i_number;
601 if (ipref >= fs->fs_ncg * fs->fs_ipg)
603 cg = ino_to_cg(fs, ipref);
605 * Track number of dirs created one after another
606 * in a same cg without intervening by files.
608 if ((mode & IFMT) == IFDIR) {
609 if (fs->fs_contigdirs[cg] < 255)
610 fs->fs_contigdirs[cg]++;
612 if (fs->fs_contigdirs[cg] > 0)
613 fs->fs_contigdirs[cg]--;
615 ino = (ino_t)ffs_hashalloc(pip, cg, (long)ipref, mode,
616 (allocfcn_t *)ffs_nodealloccg);
619 error = VFS_VGET(pvp->v_mount, ino, vpp);
621 UFS_VFREE(pvp, ino, mode);
626 printf("mode = 0%o, inum = %lu, fs = %s\n",
627 ip->i_mode, (u_long)ip->i_number, fs->fs_fsmnt);
628 panic("ffs_valloc: dup alloc");
630 if (ip->i_blocks) { /* XXX */
631 printf("free inode %s/%lu had %ld blocks\n",
632 fs->fs_fsmnt, (u_long)ino, (long)ip->i_blocks);
637 * Set up a new generation number for this inode.
639 if (ip->i_gen == 0 || ++ip->i_gen == 0)
640 ip->i_gen = random() / 2 + 1;
643 ffs_fserr(fs, cred->cr_uid, "out of inodes");
644 uprintf("\n%s: create/symlink failed, no inodes free\n", fs->fs_fsmnt);
649 * Find a cylinder group to place a directory.
651 * The policy implemented by this algorithm is to allocate a
652 * directory inode in the same cylinder group as its parent
653 * directory, but also to reserve space for its files inodes
654 * and data. Restrict the number of directories which may be
655 * allocated one after another in the same cylinder group
656 * without intervening allocation of files.
658 * If we allocate a first level directory then force allocation
659 * in another cylinder group.
662 ffs_dirpref(struct inode *pip)
665 int cg, prefcg, dirsize, cgsize;
667 int avgifree, avgbfree, avgndir, curdirsize;
668 int minifree, minbfree, maxndir;
674 avgifree = fs->fs_cstotal.cs_nifree / fs->fs_ncg;
675 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
676 avgndir = fs->fs_cstotal.cs_ndir / fs->fs_ncg;
679 * Force allocation in another cg if creating a first level dir.
681 if (ITOV(pip)->v_flag & VROOT) {
682 prefcg = arc4random() % fs->fs_ncg;
684 minndir = fs->fs_ipg;
685 for (cg = prefcg; cg < fs->fs_ncg; cg++)
686 if (fs->fs_cs(fs, cg).cs_ndir < minndir &&
687 fs->fs_cs(fs, cg).cs_nifree >= avgifree &&
688 fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
690 minndir = fs->fs_cs(fs, cg).cs_ndir;
692 for (cg = 0; cg < prefcg; cg++)
693 if (fs->fs_cs(fs, cg).cs_ndir < minndir &&
694 fs->fs_cs(fs, cg).cs_nifree >= avgifree &&
695 fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
697 minndir = fs->fs_cs(fs, cg).cs_ndir;
699 return ((ino_t)(fs->fs_ipg * mincg));
703 * Count various limits which used for
704 * optimal allocation of a directory inode.
706 maxndir = min(avgndir + fs->fs_ipg / 16, fs->fs_ipg);
707 minifree = avgifree - avgifree / 4;
710 minbfree = avgbfree - avgbfree / 4;
713 cgsize = fs->fs_fsize * fs->fs_fpg;
716 * fs_avgfilesize and fs_avgfpdir are user-settable entities and
717 * multiplying them may overflow a 32 bit integer.
719 dirsize64 = fs->fs_avgfilesize * (int64_t)fs->fs_avgfpdir;
720 if (dirsize64 > 0x7fffffff) {
723 dirsize = (int)dirsize64;
724 curdirsize = avgndir ?
725 (cgsize - avgbfree * fs->fs_bsize) / avgndir : 0;
726 if (dirsize < curdirsize)
727 dirsize = curdirsize;
728 maxcontigdirs = min((avgbfree * fs->fs_bsize) / dirsize, 255);
729 if (fs->fs_avgfpdir > 0)
730 maxcontigdirs = min(maxcontigdirs,
731 fs->fs_ipg / fs->fs_avgfpdir);
732 if (maxcontigdirs == 0)
737 * Limit number of dirs in one cg and reserve space for
738 * regular files, but only if we have no deficit in
741 prefcg = ino_to_cg(fs, pip->i_number);
742 for (cg = prefcg; cg < fs->fs_ncg; cg++)
743 if (fs->fs_cs(fs, cg).cs_ndir < maxndir &&
744 fs->fs_cs(fs, cg).cs_nifree >= minifree &&
745 fs->fs_cs(fs, cg).cs_nbfree >= minbfree) {
746 if (fs->fs_contigdirs[cg] < maxcontigdirs)
747 return ((ino_t)(fs->fs_ipg * cg));
749 for (cg = 0; cg < prefcg; cg++)
750 if (fs->fs_cs(fs, cg).cs_ndir < maxndir &&
751 fs->fs_cs(fs, cg).cs_nifree >= minifree &&
752 fs->fs_cs(fs, cg).cs_nbfree >= minbfree) {
753 if (fs->fs_contigdirs[cg] < maxcontigdirs)
754 return ((ino_t)(fs->fs_ipg * cg));
757 * This is a backstop when we have deficit in space.
759 for (cg = prefcg; cg < fs->fs_ncg; cg++)
760 if (fs->fs_cs(fs, cg).cs_nifree >= avgifree)
761 return ((ino_t)(fs->fs_ipg * cg));
762 for (cg = 0; cg < prefcg; cg++)
763 if (fs->fs_cs(fs, cg).cs_nifree >= avgifree)
765 return ((ino_t)(fs->fs_ipg * cg));
769 * Select the desired position for the next block in a file. The file is
770 * logically divided into sections. The first section is composed of the
771 * direct blocks. Each additional section contains fs_maxbpg blocks.
773 * If no blocks have been allocated in the first section, the policy is to
774 * request a block in the same cylinder group as the inode that describes
775 * the file. If no blocks have been allocated in any other section, the
776 * policy is to place the section in a cylinder group with a greater than
777 * average number of free blocks. An appropriate cylinder group is found
778 * by using a rotor that sweeps the cylinder groups. When a new group of
779 * blocks is needed, the sweep begins in the cylinder group following the
780 * cylinder group from which the previous allocation was made. The sweep
781 * continues until a cylinder group with greater than the average number
782 * of free blocks is found. If the allocation is for the first block in an
783 * indirect block, the information on the previous allocation is unavailable;
784 * here a best guess is made based upon the logical block number being
787 * If a section is already partially allocated, the policy is to
788 * contiguously allocate fs_maxcontig blocks. The end of one of these
789 * contiguous blocks and the beginning of the next is physically separated
790 * so that the disk head will be in transit between them for at least
791 * fs_rotdelay milliseconds. This is to allow time for the processor to
792 * schedule another I/O transfer.
795 ffs_blkpref(struct inode *ip, ufs_daddr_t lbn, int indx, ufs_daddr_t *bap)
799 int avgbfree, startcg;
803 if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) {
804 if (lbn < NDADDR + NINDIR(fs)) {
805 cg = ino_to_cg(fs, ip->i_number);
806 return (fs->fs_fpg * cg + fs->fs_frag);
809 * Find a cylinder with greater than average number of
810 * unused data blocks.
812 if (indx == 0 || bap[indx - 1] == 0)
814 ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg;
816 startcg = dtog(fs, bap[indx - 1]) + 1;
817 startcg %= fs->fs_ncg;
818 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
819 for (cg = startcg; cg < fs->fs_ncg; cg++)
820 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
822 return (fs->fs_fpg * cg + fs->fs_frag);
824 for (cg = 0; cg <= startcg; cg++)
825 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
827 return (fs->fs_fpg * cg + fs->fs_frag);
832 * One or more previous blocks have been laid out. If less
833 * than fs_maxcontig previous blocks are contiguous, the
834 * next block is requested contiguously, otherwise it is
835 * requested rotationally delayed by fs_rotdelay milliseconds.
837 nextblk = bap[indx - 1] + fs->fs_frag;
838 if (fs->fs_rotdelay == 0 || indx < fs->fs_maxcontig ||
839 bap[indx - fs->fs_maxcontig] +
840 blkstofrags(fs, fs->fs_maxcontig) != nextblk)
843 * Here we convert ms of delay to frags as:
844 * (frags) = (ms) * (rev/sec) * (sect/rev) /
845 * ((sect/frag) * (ms/sec))
846 * then round up to the next block.
848 nextblk += roundup(fs->fs_rotdelay * fs->fs_rps * fs->fs_nsect /
849 (NSPF(fs) * 1000), fs->fs_frag);
854 * Implement the cylinder overflow algorithm.
856 * The policy implemented by this algorithm is:
857 * 1) allocate the block in its requested cylinder group.
858 * 2) quadradically rehash on the cylinder group number.
859 * 3) brute force search for a free block.
863 ffs_hashalloc(struct inode *ip, int cg, long pref,
864 int size, /* size for data blocks, mode for inodes */
865 allocfcn_t *allocator)
868 long result; /* XXX why not same type as we return? */
873 * 1: preferred cylinder group
875 result = (*allocator)(ip, cg, pref, size);
879 * 2: quadratic rehash
881 for (i = 1; i < fs->fs_ncg; i *= 2) {
883 if (cg >= fs->fs_ncg)
885 result = (*allocator)(ip, cg, 0, size);
890 * 3: brute force search
891 * Note that we start at i == 2, since 0 was checked initially,
892 * and 1 is always checked in the quadratic rehash.
894 cg = (icg + 2) % fs->fs_ncg;
895 for (i = 2; i < fs->fs_ncg; i++) {
896 result = (*allocator)(ip, cg, 0, size);
900 if (cg == fs->fs_ncg)
907 * Determine whether a fragment can be extended.
909 * Check to see if the necessary fragments are available, and
910 * if they are, allocate them.
913 ffs_fragextend(struct inode *ip, int cg, long bprev, int osize, int nsize)
924 if (fs->fs_cs(fs, cg).cs_nffree < numfrags(fs, nsize - osize))
926 frags = numfrags(fs, nsize);
927 bbase = fragnum(fs, bprev);
928 if (bbase > fragnum(fs, (bprev + frags - 1))) {
929 /* cannot extend across a block boundary */
932 KKASSERT(blknum(fs, bprev) == blknum(fs, bprev + frags - 1));
933 error = bread(ip->i_devvp, fsbtodoff(fs, cgtod(fs, cg)),
934 (int)fs->fs_cgsize, &bp);
939 cgp = (struct cg *)bp->b_data;
940 if (!cg_chkmagic(cgp)) {
944 cgp->cg_time = time_second;
945 bno = dtogd(fs, bprev);
946 blksfree = cg_blksfree(cgp);
947 for (i = numfrags(fs, osize); i < frags; i++) {
948 if (isclr(blksfree, bno + i)) {
955 * the current fragment can be extended
956 * deduct the count on fragment being extended into
957 * increase the count on the remaining fragment (if any)
958 * allocate the extended piece
960 * ---oooooooooonnnnnnn111----
965 for (i = frags; i < fs->fs_frag - bbase; i++) {
966 if (isclr(blksfree, bno + i))
971 * Size of original free frag is [i - numfrags(fs, osize)]
972 * Size of remaining free frag is [i - frags]
974 cgp->cg_frsum[i - numfrags(fs, osize)]--;
976 cgp->cg_frsum[i - frags]++;
977 for (i = numfrags(fs, osize); i < frags; i++) {
978 clrbit(blksfree, bno + i);
979 cgp->cg_cs.cs_nffree--;
980 fs->fs_cstotal.cs_nffree--;
981 fs->fs_cs(fs, cg).cs_nffree--;
984 if (DOINGSOFTDEP(ITOV(ip)))
985 softdep_setup_blkmapdep(bp, fs, bprev);
991 * Determine whether a block can be allocated.
993 * Check to see if a block of the appropriate size is available,
994 * and if it is, allocate it.
997 ffs_alloccg(struct inode *ip, int cg, ufs_daddr_t bpref, int size)
1003 ufs_daddr_t bno, blkno;
1004 int allocsiz, error, frags;
1008 if (fs->fs_cs(fs, cg).cs_nbfree == 0 && size == fs->fs_bsize)
1010 error = bread(ip->i_devvp, fsbtodoff(fs, cgtod(fs, cg)),
1011 (int)fs->fs_cgsize, &bp);
1016 cgp = (struct cg *)bp->b_data;
1017 if (!cg_chkmagic(cgp) ||
1018 (cgp->cg_cs.cs_nbfree == 0 && size == fs->fs_bsize)) {
1022 cgp->cg_time = time_second;
1023 if (size == fs->fs_bsize) {
1024 bno = ffs_alloccgblk(ip, bp, bpref);
1029 * Check to see if any fragments of sufficient size are already
1030 * available. Fit the data into a larger fragment if necessary,
1031 * before allocating a whole new block.
1033 blksfree = cg_blksfree(cgp);
1034 frags = numfrags(fs, size);
1035 for (allocsiz = frags; allocsiz < fs->fs_frag; allocsiz++) {
1036 if (cgp->cg_frsum[allocsiz] != 0)
1039 if (allocsiz == fs->fs_frag) {
1041 * No fragments were available, allocate a whole block and
1042 * cut the requested fragment (of size frags) out of it.
1044 if (cgp->cg_cs.cs_nbfree == 0) {
1048 bno = ffs_alloccgblk(ip, bp, bpref);
1049 bpref = dtogd(fs, bno);
1050 for (i = frags; i < fs->fs_frag; i++)
1051 setbit(blksfree, bpref + i);
1054 * Calculate the number of free frags still remaining after
1055 * we have cut out the requested allocation. Indicate that
1056 * a fragment of that size is now available for future
1059 i = fs->fs_frag - frags;
1060 cgp->cg_cs.cs_nffree += i;
1061 fs->fs_cstotal.cs_nffree += i;
1062 fs->fs_cs(fs, cg).cs_nffree += i;
1070 * cg_frsum[] has told us that a free fragment of allocsiz size is
1071 * available. Find it, then clear the bitmap bits associated with
1074 bno = ffs_mapsearch(fs, cgp, bpref, allocsiz);
1079 for (i = 0; i < frags; i++)
1080 clrbit(blksfree, bno + i);
1081 cgp->cg_cs.cs_nffree -= frags;
1082 fs->fs_cstotal.cs_nffree -= frags;
1083 fs->fs_cs(fs, cg).cs_nffree -= frags;
1087 * Account for the allocation. The original searched size that we
1088 * found is no longer available. If we cut out a smaller piece then
1089 * a smaller fragment is now available.
1091 cgp->cg_frsum[allocsiz]--;
1092 if (frags != allocsiz)
1093 cgp->cg_frsum[allocsiz - frags]++;
1094 blkno = cg * fs->fs_fpg + bno;
1095 if (DOINGSOFTDEP(ITOV(ip)))
1096 softdep_setup_blkmapdep(bp, fs, blkno);
1098 return ((u_long)blkno);
1102 * Allocate a block in a cylinder group.
1104 * This algorithm implements the following policy:
1105 * 1) allocate the requested block.
1106 * 2) allocate a rotationally optimal block in the same cylinder.
1107 * 3) allocate the next available block on the block rotor for the
1108 * specified cylinder group.
1109 * Note that this routine only allocates fs_bsize blocks; these
1110 * blocks may be fragmented by the routine that allocates them.
1113 ffs_alloccgblk(struct inode *ip, struct buf *bp, ufs_daddr_t bpref)
1117 ufs_daddr_t bno, blkno;
1118 int cylno, pos, delta;
1124 cgp = (struct cg *)bp->b_data;
1125 blksfree = cg_blksfree(cgp);
1126 if (bpref == 0 || dtog(fs, bpref) != cgp->cg_cgx) {
1127 bpref = cgp->cg_rotor;
1130 bpref = blknum(fs, bpref);
1131 bpref = dtogd(fs, bpref);
1133 * if the requested block is available, use it
1135 if (ffs_isblock(fs, blksfree, fragstoblks(fs, bpref))) {
1139 if (fs->fs_nrpos <= 1 || fs->fs_cpc == 0) {
1141 * Block layout information is not available.
1142 * Leaving bpref unchanged means we take the
1143 * next available free block following the one
1144 * we just allocated. Hopefully this will at
1145 * least hit a track cache on drives of unknown
1146 * geometry (e.g. SCSI).
1151 * check for a block available on the same cylinder
1153 cylno = cbtocylno(fs, bpref);
1154 if (cg_blktot(cgp)[cylno] == 0)
1157 * check the summary information to see if a block is
1158 * available in the requested cylinder starting at the
1159 * requested rotational position and proceeding around.
1161 cylbp = cg_blks(fs, cgp, cylno);
1162 pos = cbtorpos(fs, bpref);
1163 for (i = pos; i < fs->fs_nrpos; i++)
1166 if (i == fs->fs_nrpos)
1167 for (i = 0; i < pos; i++)
1172 * found a rotational position, now find the actual
1173 * block. A panic if none is actually there.
1175 pos = cylno % fs->fs_cpc;
1176 bno = (cylno - pos) * fs->fs_spc / NSPB(fs);
1177 if (fs_postbl(fs, pos)[i] == -1) {
1178 printf("pos = %d, i = %d, fs = %s\n",
1179 pos, i, fs->fs_fsmnt);
1180 panic("ffs_alloccgblk: cyl groups corrupted");
1182 for (i = fs_postbl(fs, pos)[i];; ) {
1183 if (ffs_isblock(fs, blksfree, bno + i)) {
1184 bno = blkstofrags(fs, (bno + i));
1187 delta = fs_rotbl(fs)[i];
1189 delta + i > fragstoblks(fs, fs->fs_fpg))
1193 printf("pos = %d, i = %d, fs = %s\n", pos, i, fs->fs_fsmnt);
1194 panic("ffs_alloccgblk: can't find blk in cyl");
1198 * no blocks in the requested cylinder, so take next
1199 * available one in this cylinder group.
1201 bno = ffs_mapsearch(fs, cgp, bpref, (int)fs->fs_frag);
1204 cgp->cg_rotor = bno;
1206 blkno = fragstoblks(fs, bno);
1207 ffs_clrblock(fs, blksfree, (long)blkno);
1208 ffs_clusteracct(fs, cgp, blkno, -1);
1209 cgp->cg_cs.cs_nbfree--;
1210 fs->fs_cstotal.cs_nbfree--;
1211 fs->fs_cs(fs, cgp->cg_cgx).cs_nbfree--;
1212 cylno = cbtocylno(fs, bno);
1213 cg_blks(fs, cgp, cylno)[cbtorpos(fs, bno)]--;
1214 cg_blktot(cgp)[cylno]--;
1216 blkno = cgp->cg_cgx * fs->fs_fpg + bno;
1217 if (DOINGSOFTDEP(ITOV(ip)))
1218 softdep_setup_blkmapdep(bp, fs, blkno);
1223 * Determine whether a cluster can be allocated.
1225 * We do not currently check for optimal rotational layout if there
1226 * are multiple choices in the same cylinder group. Instead we just
1227 * take the first one that we find following bpref.
1230 ffs_clusteralloc(struct inode *ip, int cg, ufs_daddr_t bpref, int len)
1235 int i, got, run, bno, bit, map;
1241 if (fs->fs_maxcluster[cg] < len)
1243 if (bread(ip->i_devvp, fsbtodoff(fs, cgtod(fs, cg)),
1244 (int)fs->fs_cgsize, &bp)) {
1247 cgp = (struct cg *)bp->b_data;
1248 if (!cg_chkmagic(cgp))
1252 * Check to see if a cluster of the needed size (or bigger) is
1253 * available in this cylinder group.
1255 lp = &cg_clustersum(cgp)[len];
1256 for (i = len; i <= fs->fs_contigsumsize; i++)
1259 if (i > fs->fs_contigsumsize) {
1261 * This is the first time looking for a cluster in this
1262 * cylinder group. Update the cluster summary information
1263 * to reflect the true maximum sized cluster so that
1264 * future cluster allocation requests can avoid reading
1265 * the cylinder group map only to find no clusters.
1267 lp = &cg_clustersum(cgp)[len - 1];
1268 for (i = len - 1; i > 0; i--)
1271 fs->fs_maxcluster[cg] = i;
1275 * Search the cluster map to find a big enough cluster.
1276 * We take the first one that we find, even if it is larger
1277 * than we need as we prefer to get one close to the previous
1278 * block allocation. We do not search before the current
1279 * preference point as we do not want to allocate a block
1280 * that is allocated before the previous one (as we will
1281 * then have to wait for another pass of the elevator
1282 * algorithm before it will be read). We prefer to fail and
1283 * be recalled to try an allocation in the next cylinder group.
1285 if (dtog(fs, bpref) != cg)
1288 bpref = fragstoblks(fs, dtogd(fs, blknum(fs, bpref)));
1289 mapp = &cg_clustersfree(cgp)[bpref / NBBY];
1291 bit = 1 << (bpref % NBBY);
1292 for (run = 0, got = bpref; got < cgp->cg_nclusterblks; got++) {
1293 if ((map & bit) == 0) {
1300 if ((got & (NBBY - 1)) != (NBBY - 1)) {
1307 if (got >= cgp->cg_nclusterblks)
1310 * Allocate the cluster that we have found.
1312 blksfree = cg_blksfree(cgp);
1313 for (i = 1; i <= len; i++) {
1314 if (!ffs_isblock(fs, blksfree, got - run + i))
1315 panic("ffs_clusteralloc: map mismatch");
1317 bno = cg * fs->fs_fpg + blkstofrags(fs, got - run + 1);
1318 if (dtog(fs, bno) != cg)
1319 panic("ffs_clusteralloc: allocated out of group");
1320 len = blkstofrags(fs, len);
1321 for (i = 0; i < len; i += fs->fs_frag) {
1322 if ((got = ffs_alloccgblk(ip, bp, bno + i)) != bno + i)
1323 panic("ffs_clusteralloc: lost block");
1334 * Determine whether an inode can be allocated.
1336 * Check to see if an inode is available, and if it is,
1337 * allocate it using the following policy:
1338 * 1) allocate the requested inode.
1339 * 2) allocate the next available inode after the requested
1340 * inode in the specified cylinder group.
1341 * 3) the inode must not already be in the inode hash table. We
1342 * can encounter such a case because the vnode reclamation sequence
1344 * 3) the inode must not already be in the inode hash, otherwise it
1345 * may be in the process of being deallocated. This can occur
1346 * because the bitmap is updated before the inode is removed from
1347 * hash. If we were to reallocate the inode the caller could wind
1348 * up returning a vnode/inode combination which is in an indeterminate
1352 ffs_nodealloccg(struct inode *ip, int cg, ufs_daddr_t ipref, int mode)
1359 int error, len, arraysize, i;
1364 if (fs->fs_cs(fs, cg).cs_nifree == 0)
1366 error = bread(ip->i_devvp, fsbtodoff(fs, cgtod(fs, cg)),
1367 (int)fs->fs_cgsize, &bp);
1372 cgp = (struct cg *)bp->b_data;
1373 if (!cg_chkmagic(cgp) || cgp->cg_cs.cs_nifree == 0) {
1377 inosused = cg_inosused(cgp);
1381 * Quick check, reuse the most recently free inode or continue
1382 * a scan from where we left off the last time.
1384 ibase = cg * fs->fs_ipg;
1386 ipref %= fs->fs_ipg;
1387 if (isclr(inosused, ipref)) {
1388 if (ufs_ihashcheck(ip->i_dev, ibase + ipref) == 0)
1394 * Scan the inode bitmap starting at irotor, be sure to handle
1395 * the edge case by going back to the beginning of the array.
1397 * If the number of inodes is not byte-aligned, the unused bits
1398 * should be set to 1. This will be sanity checked in gotit. Note
1399 * that we have to be sure not to overlap the beginning and end
1400 * when irotor is in the middle of a byte as this will cause the
1401 * same bitmap byte to be checked twice. To solve this problem we
1402 * just convert everything to a byte index for the loop.
1404 ipref = (cgp->cg_irotor % fs->fs_ipg) >> 3; /* byte index */
1405 len = (fs->fs_ipg + 7) >> 3; /* byte size */
1409 map = inosused[ipref];
1411 for (i = 0; i < NBBY; ++i) {
1413 * If we find a free bit we have to make sure
1414 * that the inode is not in the middle of
1415 * being destroyed. The inode should not exist
1416 * in the inode hash.
1418 * Adjust the rotor to try to hit the
1419 * quick-check up above.
1421 if ((map & (1 << i)) == 0) {
1422 if (ufs_ihashcheck(ip->i_dev, ibase + (ipref << 3) + i) == 0) {
1423 ipref = (ipref << 3) + i;
1424 cgp->cg_irotor = (ipref + 1) % fs->fs_ipg;
1433 * Setup for the next byte, start at the beginning again if
1434 * we hit the end of the array.
1436 if (++ipref == arraysize)
1440 if (icheckmiss == cgp->cg_cs.cs_nifree) {
1444 printf("fs = %s\n", fs->fs_fsmnt);
1445 panic("ffs_nodealloccg: block not in map, icheckmiss/nfree %d/%d",
1446 icheckmiss, cgp->cg_cs.cs_nifree);
1450 * ipref is a bit index as of the gotit label.
1453 KKASSERT(ipref >= 0 && ipref < fs->fs_ipg);
1455 printf("Warning: inode free race avoided %d times\n",
1458 cgp->cg_time = time_second;
1459 if (DOINGSOFTDEP(ITOV(ip)))
1460 softdep_setup_inomapdep(bp, ip, ibase + ipref);
1461 setbit(inosused, ipref);
1462 cgp->cg_cs.cs_nifree--;
1463 fs->fs_cstotal.cs_nifree--;
1464 fs->fs_cs(fs, cg).cs_nifree--;
1466 if ((mode & IFMT) == IFDIR) {
1467 cgp->cg_cs.cs_ndir++;
1468 fs->fs_cstotal.cs_ndir++;
1469 fs->fs_cs(fs, cg).cs_ndir++;
1472 return (ibase + ipref);
1476 * Free a block or fragment.
1478 * The specified block or fragment is placed back in the
1479 * free map. If a fragment is deallocated, a possible
1480 * block reassembly is checked.
1483 ffs_blkfree(struct inode *ip, ufs_daddr_t bno, long size)
1489 int i, error, cg, blk, frags, bbase;
1493 VOP_FREEBLKS(ip->i_devvp, fsbtodoff(fs, bno), size);
1494 if ((uint)size > fs->fs_bsize || fragoff(fs, size) != 0 ||
1495 fragnum(fs, bno) + numfrags(fs, size) > fs->fs_frag) {
1496 printf("dev=%s, bno = %ld, bsize = %ld, size = %ld, fs = %s\n",
1497 devtoname(ip->i_dev), (long)bno, (long)fs->fs_bsize, size,
1499 panic("ffs_blkfree: bad size");
1502 if ((uint)bno >= fs->fs_size) {
1503 printf("bad block %ld, ino %lu\n",
1504 (long)bno, (u_long)ip->i_number);
1505 ffs_fserr(fs, ip->i_uid, "bad block");
1510 * Load the cylinder group
1512 error = bread(ip->i_devvp, fsbtodoff(fs, cgtod(fs, cg)),
1513 (int)fs->fs_cgsize, &bp);
1518 cgp = (struct cg *)bp->b_data;
1519 if (!cg_chkmagic(cgp)) {
1523 cgp->cg_time = time_second;
1524 bno = dtogd(fs, bno);
1525 blksfree = cg_blksfree(cgp);
1527 if (size == fs->fs_bsize) {
1529 * Free a whole block
1531 blkno = fragstoblks(fs, bno);
1532 if (!ffs_isfreeblock(fs, blksfree, blkno)) {
1533 printf("dev = %s, block = %ld, fs = %s\n",
1534 devtoname(ip->i_dev), (long)bno, fs->fs_fsmnt);
1535 panic("ffs_blkfree: freeing free block");
1537 ffs_setblock(fs, blksfree, blkno);
1538 ffs_clusteracct(fs, cgp, blkno, 1);
1539 cgp->cg_cs.cs_nbfree++;
1540 fs->fs_cstotal.cs_nbfree++;
1541 fs->fs_cs(fs, cg).cs_nbfree++;
1542 i = cbtocylno(fs, bno);
1543 cg_blks(fs, cgp, i)[cbtorpos(fs, bno)]++;
1544 cg_blktot(cgp)[i]++;
1547 * Free a fragment within a block.
1549 * bno is the starting block number of the fragment being
1552 * bbase is the starting block number for the filesystem
1553 * block containing the fragment.
1555 * blk is the current bitmap for the fragments within the
1556 * filesystem block containing the fragment.
1558 * frags is the number of fragments being freed
1560 * Call ffs_fragacct() to account for the removal of all
1561 * current fragments, then adjust the bitmap to free the
1562 * requested fragment, and finally call ffs_fragacct() again
1563 * to regenerate the accounting.
1565 bbase = bno - fragnum(fs, bno);
1566 blk = blkmap(fs, blksfree, bbase);
1567 ffs_fragacct(fs, blk, cgp->cg_frsum, -1);
1568 frags = numfrags(fs, size);
1569 for (i = 0; i < frags; i++) {
1570 if (isset(blksfree, bno + i)) {
1571 printf("dev = %s, block = %ld, fs = %s\n",
1572 devtoname(ip->i_dev), (long)(bno + i),
1574 panic("ffs_blkfree: freeing free frag");
1576 setbit(blksfree, bno + i);
1578 cgp->cg_cs.cs_nffree += i;
1579 fs->fs_cstotal.cs_nffree += i;
1580 fs->fs_cs(fs, cg).cs_nffree += i;
1583 * Add back in counts associated with the new frags
1585 blk = blkmap(fs, blksfree, bbase);
1586 ffs_fragacct(fs, blk, cgp->cg_frsum, 1);
1589 * If a complete block has been reassembled, account for it
1591 blkno = fragstoblks(fs, bbase);
1592 if (ffs_isblock(fs, blksfree, blkno)) {
1593 cgp->cg_cs.cs_nffree -= fs->fs_frag;
1594 fs->fs_cstotal.cs_nffree -= fs->fs_frag;
1595 fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag;
1596 ffs_clusteracct(fs, cgp, blkno, 1);
1597 cgp->cg_cs.cs_nbfree++;
1598 fs->fs_cstotal.cs_nbfree++;
1599 fs->fs_cs(fs, cg).cs_nbfree++;
1600 i = cbtocylno(fs, bbase);
1601 cg_blks(fs, cgp, i)[cbtorpos(fs, bbase)]++;
1602 cg_blktot(cgp)[i]++;
1611 * Verify allocation of a block or fragment. Returns true if block or
1612 * fragment is allocated, false if it is free.
1615 ffs_checkblk(struct inode *ip, ufs_daddr_t bno, long size)
1620 int i, error, frags, free;
1624 if ((uint)size > fs->fs_bsize || fragoff(fs, size) != 0) {
1625 printf("bsize = %ld, size = %ld, fs = %s\n",
1626 (long)fs->fs_bsize, size, fs->fs_fsmnt);
1627 panic("ffs_checkblk: bad size");
1629 if ((uint)bno >= fs->fs_size)
1630 panic("ffs_checkblk: bad block %d", bno);
1631 error = bread(ip->i_devvp, fsbtodoff(fs, cgtod(fs, dtog(fs, bno))),
1632 (int)fs->fs_cgsize, &bp);
1634 panic("ffs_checkblk: cg bread failed");
1635 cgp = (struct cg *)bp->b_data;
1636 if (!cg_chkmagic(cgp))
1637 panic("ffs_checkblk: cg magic mismatch");
1638 blksfree = cg_blksfree(cgp);
1639 bno = dtogd(fs, bno);
1640 if (size == fs->fs_bsize) {
1641 free = ffs_isblock(fs, blksfree, fragstoblks(fs, bno));
1643 frags = numfrags(fs, size);
1644 for (free = 0, i = 0; i < frags; i++)
1645 if (isset(blksfree, bno + i))
1647 if (free != 0 && free != frags)
1648 panic("ffs_checkblk: partially free fragment");
1653 #endif /* DIAGNOSTIC */
1659 ffs_vfree(struct vnode *pvp, ino_t ino, int mode)
1661 if (DOINGSOFTDEP(pvp)) {
1662 softdep_freefile(pvp, ino, mode);
1665 return (ffs_freefile(pvp, ino, mode));
1669 * Do the actual free operation.
1670 * The specified inode is placed back in the free map.
1673 ffs_freefile(struct vnode *pvp, ino_t ino, int mode)
1684 if ((uint)ino >= fs->fs_ipg * fs->fs_ncg)
1685 panic("ffs_vfree: range: dev = (%d,%d), ino = %"PRId64", fs = %s",
1686 major(pip->i_dev), minor(pip->i_dev), ino, fs->fs_fsmnt);
1687 cg = ino_to_cg(fs, ino);
1688 error = bread(pip->i_devvp, fsbtodoff(fs, cgtod(fs, cg)),
1689 (int)fs->fs_cgsize, &bp);
1694 cgp = (struct cg *)bp->b_data;
1695 if (!cg_chkmagic(cgp)) {
1699 cgp->cg_time = time_second;
1700 inosused = cg_inosused(cgp);
1702 if (isclr(inosused, ino)) {
1703 printf("dev = %s, ino = %lu, fs = %s\n",
1704 devtoname(pip->i_dev), (u_long)ino, fs->fs_fsmnt);
1705 if (fs->fs_ronly == 0)
1706 panic("ffs_vfree: freeing free inode");
1708 clrbit(inosused, ino);
1709 if (ino < cgp->cg_irotor)
1710 cgp->cg_irotor = ino;
1711 cgp->cg_cs.cs_nifree++;
1712 fs->fs_cstotal.cs_nifree++;
1713 fs->fs_cs(fs, cg).cs_nifree++;
1714 if ((mode & IFMT) == IFDIR) {
1715 cgp->cg_cs.cs_ndir--;
1716 fs->fs_cstotal.cs_ndir--;
1717 fs->fs_cs(fs, cg).cs_ndir--;
1725 * Find a block of the specified size in the specified cylinder group.
1727 * It is a panic if a request is made to find a block if none are
1731 ffs_mapsearch(struct fs *fs, struct cg *cgp, ufs_daddr_t bpref, int allocsiz)
1734 int start, len, loc, i;
1735 int blk, field, subfield, pos;
1739 * find the fragment by searching through the free block
1740 * map for an appropriate bit pattern.
1743 start = dtogd(fs, bpref) / NBBY;
1745 start = cgp->cg_frotor / NBBY;
1746 blksfree = cg_blksfree(cgp);
1747 len = howmany(fs->fs_fpg, NBBY) - start;
1748 loc = scanc((uint)len, (u_char *)&blksfree[start],
1749 (u_char *)fragtbl[fs->fs_frag],
1750 (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
1752 len = start + 1; /* XXX why overlap here? */
1754 loc = scanc((uint)len, (u_char *)&blksfree[0],
1755 (u_char *)fragtbl[fs->fs_frag],
1756 (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
1758 printf("start = %d, len = %d, fs = %s\n",
1759 start, len, fs->fs_fsmnt);
1760 panic("ffs_alloccg: map corrupted");
1764 bno = (start + len - loc) * NBBY;
1765 cgp->cg_frotor = bno;
1767 * found the byte in the map
1768 * sift through the bits to find the selected frag
1770 for (i = bno + NBBY; bno < i; bno += fs->fs_frag) {
1771 blk = blkmap(fs, blksfree, bno);
1773 field = around[allocsiz];
1774 subfield = inside[allocsiz];
1775 for (pos = 0; pos <= fs->fs_frag - allocsiz; pos++) {
1776 if ((blk & field) == subfield)
1782 printf("bno = %lu, fs = %s\n", (u_long)bno, fs->fs_fsmnt);
1783 panic("ffs_alloccg: block not in map");
1788 * Update the cluster map because of an allocation or free.
1790 * Cnt == 1 means free; cnt == -1 means allocating.
1793 ffs_clusteracct(struct fs *fs, struct cg *cgp, ufs_daddr_t blkno, int cnt)
1797 u_char *freemapp, *mapp;
1798 int i, start, end, forw, back, map, bit;
1800 if (fs->fs_contigsumsize <= 0)
1802 freemapp = cg_clustersfree(cgp);
1803 sump = cg_clustersum(cgp);
1805 * Allocate or clear the actual block.
1808 setbit(freemapp, blkno);
1810 clrbit(freemapp, blkno);
1812 * Find the size of the cluster going forward.
1815 end = start + fs->fs_contigsumsize;
1816 if (end >= cgp->cg_nclusterblks)
1817 end = cgp->cg_nclusterblks;
1818 mapp = &freemapp[start / NBBY];
1820 bit = 1 << (start % NBBY);
1821 for (i = start; i < end; i++) {
1822 if ((map & bit) == 0)
1824 if ((i & (NBBY - 1)) != (NBBY - 1)) {
1833 * Find the size of the cluster going backward.
1836 end = start - fs->fs_contigsumsize;
1839 mapp = &freemapp[start / NBBY];
1841 bit = 1 << (start % NBBY);
1842 for (i = start; i > end; i--) {
1843 if ((map & bit) == 0)
1845 if ((i & (NBBY - 1)) != 0) {
1849 bit = 1 << (NBBY - 1);
1854 * Account for old cluster and the possibly new forward and
1857 i = back + forw + 1;
1858 if (i > fs->fs_contigsumsize)
1859 i = fs->fs_contigsumsize;
1866 * Update cluster summary information.
1868 lp = &sump[fs->fs_contigsumsize];
1869 for (i = fs->fs_contigsumsize; i > 0; i--)
1872 fs->fs_maxcluster[cgp->cg_cgx] = i;
1876 * Fserr prints the name of a filesystem with an error diagnostic.
1878 * The form of the error message is:
1882 ffs_fserr(struct fs *fs, uint uid, char *cp)
1884 struct thread *td = curthread;
1887 if ((p = td->td_proc) != NULL) {
1888 log(LOG_ERR, "pid %d (%s), uid %d on %s: %s\n", p ? p->p_pid : -1,
1889 p ? p->p_comm : "-", uid, fs->fs_fsmnt, cp);
1891 log(LOG_ERR, "system thread %p, uid %d on %s: %s\n",
1892 td, uid, fs->fs_fsmnt, cp);