2 * Copyright (c) 2002 Networks Associates Technology, Inc.
5 * This software was developed for the FreeBSD Project by Marshall
6 * Kirk McKusick and Network Associates Laboratories, the Security
7 * Research Division of Network Associates, Inc. under DARPA/SPAWAR
8 * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
20 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * Copyright (c) 1982, 1986, 1989, 1993
33 * The Regents of the University of California. All rights reserved.
35 * Redistribution and use in source and binary forms, with or without
36 * modification, are permitted provided that the following conditions
38 * 1. Redistributions of source code must retain the above copyright
39 * notice, this list of conditions and the following disclaimer.
40 * 2. Redistributions in binary form must reproduce the above copyright
41 * notice, this list of conditions and the following disclaimer in the
42 * documentation and/or other materials provided with the distribution.
43 * 4. Neither the name of the University nor the names of its contributors
44 * may be used to endorse or promote products derived from this software
45 * without specific prior written permission.
47 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
48 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
49 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
50 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
51 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
52 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
53 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
54 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
55 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
56 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
59 * @(#)ffs_alloc.c 8.18 (Berkeley) 5/26/95
62 #include <sys/cdefs.h>
63 __FBSDID("$FreeBSD$");
65 #include "opt_quota.h"
67 #include <sys/param.h>
68 #include <sys/capability.h>
69 #include <sys/systm.h>
73 #include <sys/fcntl.h>
75 #include <sys/filedesc.h>
78 #include <sys/vnode.h>
79 #include <sys/mount.h>
80 #include <sys/kernel.h>
81 #include <sys/syscallsubr.h>
82 #include <sys/sysctl.h>
83 #include <sys/syslog.h>
84 #include <sys/taskqueue.h>
86 #include <security/audit/audit.h>
88 #include <geom/geom.h>
90 #include <ufs/ufs/dir.h>
91 #include <ufs/ufs/extattr.h>
92 #include <ufs/ufs/quota.h>
93 #include <ufs/ufs/inode.h>
94 #include <ufs/ufs/ufs_extern.h>
95 #include <ufs/ufs/ufsmount.h>
97 #include <ufs/ffs/fs.h>
98 #include <ufs/ffs/ffs_extern.h>
99 #include <ufs/ffs/softdep.h>
101 typedef ufs2_daddr_t allocfcn_t(struct inode *ip, u_int cg, ufs2_daddr_t bpref,
102 int size, int rsize);
104 static ufs2_daddr_t ffs_alloccg(struct inode *, u_int, ufs2_daddr_t, int, int);
106 ffs_alloccgblk(struct inode *, struct buf *, ufs2_daddr_t, int);
107 static void ffs_blkfree_cg(struct ufsmount *, struct fs *,
108 struct vnode *, ufs2_daddr_t, long, ino_t,
110 static void ffs_blkfree_trim_completed(struct bio *);
111 static void ffs_blkfree_trim_task(void *ctx, int pending __unused);
113 static int ffs_checkblk(struct inode *, ufs2_daddr_t, long);
115 static ufs2_daddr_t ffs_clusteralloc(struct inode *, u_int, ufs2_daddr_t, int,
117 static ino_t ffs_dirpref(struct inode *);
118 static ufs2_daddr_t ffs_fragextend(struct inode *, u_int, ufs2_daddr_t,
120 static ufs2_daddr_t ffs_hashalloc
121 (struct inode *, u_int, ufs2_daddr_t, int, int, allocfcn_t *);
122 static ufs2_daddr_t ffs_nodealloccg(struct inode *, u_int, ufs2_daddr_t, int,
124 static ufs1_daddr_t ffs_mapsearch(struct fs *, struct cg *, ufs2_daddr_t, int);
125 static int ffs_reallocblks_ufs1(struct vop_reallocblks_args *);
126 static int ffs_reallocblks_ufs2(struct vop_reallocblks_args *);
129 * Allocate a block in the filesystem.
131 * The size of the requested block is given, which must be some
132 * multiple of fs_fsize and <= fs_bsize.
133 * A preference may be optionally specified. If a preference is given
134 * the following hierarchy is used to allocate a block:
135 * 1) allocate the requested block.
136 * 2) allocate a rotationally optimal block in the same cylinder.
137 * 3) allocate a block in the same cylinder group.
138 * 4) quadradically rehash into other cylinder groups, until an
139 * available block is located.
140 * If no block preference is given the following hierarchy is used
141 * to allocate a block:
142 * 1) allocate a block in the cylinder group that contains the
143 * inode for the file.
144 * 2) quadradically rehash into other cylinder groups, until an
145 * available block is located.
148 ffs_alloc(ip, lbn, bpref, size, flags, cred, bnp)
150 ufs2_daddr_t lbn, bpref;
156 struct ufsmount *ump;
159 static struct timeval lastfail;
169 mtx_assert(UFS_MTX(ump), MA_OWNED);
171 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) {
172 printf("dev = %s, bsize = %ld, size = %d, fs = %s\n",
173 devtoname(ip->i_dev), (long)fs->fs_bsize, size,
175 panic("ffs_alloc: bad size");
178 panic("ffs_alloc: missing credential");
179 #endif /* INVARIANTS */
184 error = chkdq(ip, btodb(size), cred, 0);
189 if (size == fs->fs_bsize && fs->fs_cstotal.cs_nbfree == 0)
191 if (priv_check_cred(cred, PRIV_VFS_BLOCKRESERVE, 0) &&
192 freespace(fs, fs->fs_minfree) - numfrags(fs, size) < 0)
194 if (bpref >= fs->fs_size)
197 cg = ino_to_cg(fs, ip->i_number);
199 cg = dtog(fs, bpref);
200 bno = ffs_hashalloc(ip, cg, bpref, size, size, ffs_alloccg);
203 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + delta);
205 ip->i_flag |= IN_CHANGE;
207 ip->i_flag |= IN_CHANGE | IN_UPDATE;
215 * Restore user's disk quota because allocation failed.
217 (void) chkdq(ip, -btodb(size), cred, FORCE);
220 if (reclaimed == 0 && (flags & IO_BUFLOCKED) == 0) {
222 softdep_request_cleanup(fs, ITOV(ip), cred, FLUSH_BLOCKS_WAIT);
226 if (reclaimed > 0 && ppsratecheck(&lastfail, &curfail, 1)) {
227 ffs_fserr(fs, ip->i_number, "filesystem full");
228 uprintf("\n%s: write failed, filesystem is full\n",
235 * Reallocate a fragment to a bigger size
237 * The number and size of the old block is given, and a preference
238 * and new size is also specified. The allocator attempts to extend
239 * the original block. Failing that, the regular block allocator is
240 * invoked to get an appropriate block.
243 ffs_realloccg(ip, lbprev, bprev, bpref, osize, nsize, flags, cred, bpp)
248 int osize, nsize, flags;
255 struct ufsmount *ump;
256 u_int cg, request, reclaimed;
259 static struct timeval lastfail;
268 gbflags = (flags & BA_UNMAPPED) != 0 ? GB_UNMAPPED : 0;
270 mtx_assert(UFS_MTX(ump), MA_OWNED);
272 if (vp->v_mount->mnt_kern_flag & MNTK_SUSPENDED)
273 panic("ffs_realloccg: allocation on suspended filesystem");
274 if ((u_int)osize > fs->fs_bsize || fragoff(fs, osize) != 0 ||
275 (u_int)nsize > fs->fs_bsize || fragoff(fs, nsize) != 0) {
277 "dev = %s, bsize = %ld, osize = %d, nsize = %d, fs = %s\n",
278 devtoname(ip->i_dev), (long)fs->fs_bsize, osize,
279 nsize, fs->fs_fsmnt);
280 panic("ffs_realloccg: bad size");
283 panic("ffs_realloccg: missing credential");
284 #endif /* INVARIANTS */
287 if (priv_check_cred(cred, PRIV_VFS_BLOCKRESERVE, 0) &&
288 freespace(fs, fs->fs_minfree) - numfrags(fs, nsize - osize) < 0) {
292 printf("dev = %s, bsize = %ld, bprev = %jd, fs = %s\n",
293 devtoname(ip->i_dev), (long)fs->fs_bsize, (intmax_t)bprev,
295 panic("ffs_realloccg: bad bprev");
299 * Allocate the extra space in the buffer.
301 error = bread_gb(vp, lbprev, osize, NOCRED, gbflags, &bp);
307 if (bp->b_blkno == bp->b_lblkno) {
308 if (lbprev >= NDADDR)
309 panic("ffs_realloccg: lbprev out of range");
310 bp->b_blkno = fsbtodb(fs, bprev);
314 error = chkdq(ip, btodb(nsize - osize), cred, 0);
321 * Check for extension in the existing location.
323 cg = dtog(fs, bprev);
325 bno = ffs_fragextend(ip, cg, bprev, osize, nsize);
327 if (bp->b_blkno != fsbtodb(fs, bno))
328 panic("ffs_realloccg: bad blockno");
329 delta = btodb(nsize - osize);
330 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + delta);
332 ip->i_flag |= IN_CHANGE;
334 ip->i_flag |= IN_CHANGE | IN_UPDATE;
336 bp->b_flags |= B_DONE;
337 vfs_bio_bzero_buf(bp, osize, nsize - osize);
338 if ((bp->b_flags & (B_MALLOC | B_VMIO)) == B_VMIO)
339 vfs_bio_set_valid(bp, osize, nsize - osize);
344 * Allocate a new disk location.
346 if (bpref >= fs->fs_size)
348 switch ((int)fs->fs_optim) {
351 * Allocate an exact sized fragment. Although this makes
352 * best use of space, we will waste time relocating it if
353 * the file continues to grow. If the fragmentation is
354 * less than half of the minimum free reserve, we choose
355 * to begin optimizing for time.
358 if (fs->fs_minfree <= 5 ||
359 fs->fs_cstotal.cs_nffree >
360 (off_t)fs->fs_dsize * fs->fs_minfree / (2 * 100))
362 log(LOG_NOTICE, "%s: optimization changed from SPACE to TIME\n",
364 fs->fs_optim = FS_OPTTIME;
368 * At this point we have discovered a file that is trying to
369 * grow a small fragment to a larger fragment. To save time,
370 * we allocate a full sized block, then free the unused portion.
371 * If the file continues to grow, the `ffs_fragextend' call
372 * above will be able to grow it in place without further
373 * copying. If aberrant programs cause disk fragmentation to
374 * grow within 2% of the free reserve, we choose to begin
375 * optimizing for space.
377 request = fs->fs_bsize;
378 if (fs->fs_cstotal.cs_nffree <
379 (off_t)fs->fs_dsize * (fs->fs_minfree - 2) / 100)
381 log(LOG_NOTICE, "%s: optimization changed from TIME to SPACE\n",
383 fs->fs_optim = FS_OPTSPACE;
386 printf("dev = %s, optim = %ld, fs = %s\n",
387 devtoname(ip->i_dev), (long)fs->fs_optim, fs->fs_fsmnt);
388 panic("ffs_realloccg: bad optim");
391 bno = ffs_hashalloc(ip, cg, bpref, request, nsize, ffs_alloccg);
393 bp->b_blkno = fsbtodb(fs, bno);
394 if (!DOINGSOFTDEP(vp))
395 ffs_blkfree(ump, fs, ip->i_devvp, bprev, (long)osize,
396 ip->i_number, vp->v_type, NULL);
397 delta = btodb(nsize - osize);
398 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + delta);
400 ip->i_flag |= IN_CHANGE;
402 ip->i_flag |= IN_CHANGE | IN_UPDATE;
404 bp->b_flags |= B_DONE;
405 vfs_bio_bzero_buf(bp, osize, nsize - osize);
406 if ((bp->b_flags & (B_MALLOC | B_VMIO)) == B_VMIO)
407 vfs_bio_set_valid(bp, osize, nsize - osize);
414 * Restore user's disk quota because allocation failed.
416 (void) chkdq(ip, -btodb(nsize - osize), cred, FORCE);
423 if (reclaimed == 0 && (flags & IO_BUFLOCKED) == 0) {
431 softdep_request_cleanup(fs, vp, cred, FLUSH_BLOCKS_WAIT);
437 if (reclaimed > 0 && ppsratecheck(&lastfail, &curfail, 1)) {
438 ffs_fserr(fs, ip->i_number, "filesystem full");
439 uprintf("\n%s: write failed, filesystem is full\n",
446 * Reallocate a sequence of blocks into a contiguous sequence of blocks.
448 * The vnode and an array of buffer pointers for a range of sequential
449 * logical blocks to be made contiguous is given. The allocator attempts
450 * to find a range of sequential blocks starting as close as possible
451 * from the end of the allocation for the logical block immediately
452 * preceding the current range. If successful, the physical block numbers
453 * in the buffer pointers and in the inode are changed to reflect the new
454 * allocation. If unsuccessful, the allocation is left unchanged. The
455 * success in doing the reallocation is returned. Note that the error
456 * return is not reflected back to the user. Rather the previous block
457 * allocation will be used.
460 SYSCTL_NODE(_vfs, OID_AUTO, ffs, CTLFLAG_RW, 0, "FFS filesystem");
462 static int doasyncfree = 1;
463 SYSCTL_INT(_vfs_ffs, OID_AUTO, doasyncfree, CTLFLAG_RW, &doasyncfree, 0, "");
465 static int doreallocblks = 1;
466 SYSCTL_INT(_vfs_ffs, OID_AUTO, doreallocblks, CTLFLAG_RW, &doreallocblks, 0, "");
469 static volatile int prtrealloc = 0;
474 struct vop_reallocblks_args /* {
476 struct cluster_save *a_buflist;
480 if (doreallocblks == 0)
483 * We can't wait in softdep prealloc as it may fsync and recurse
484 * here. Instead we simply fail to reallocate blocks if this
485 * rare condition arises.
487 if (DOINGSOFTDEP(ap->a_vp))
488 if (softdep_prealloc(ap->a_vp, MNT_NOWAIT) != 0)
490 if (VTOI(ap->a_vp)->i_ump->um_fstype == UFS1)
491 return (ffs_reallocblks_ufs1(ap));
492 return (ffs_reallocblks_ufs2(ap));
496 ffs_reallocblks_ufs1(ap)
497 struct vop_reallocblks_args /* {
499 struct cluster_save *a_buflist;
505 struct buf *sbp, *ebp;
506 ufs1_daddr_t *bap, *sbap, *ebap = 0;
507 struct cluster_save *buflist;
508 struct ufsmount *ump;
509 ufs_lbn_t start_lbn, end_lbn;
510 ufs1_daddr_t soff, newblk, blkno;
512 struct indir start_ap[NIADDR + 1], end_ap[NIADDR + 1], *idp;
513 int i, len, start_lvl, end_lvl, ssize;
520 * If we are not tracking block clusters or if we have less than 4%
521 * free blocks left, then do not attempt to cluster. Running with
522 * less than 5% free block reserve is not recommended and those that
523 * choose to do so do not expect to have good file layout.
525 if (fs->fs_contigsumsize <= 0 || freespace(fs, 4) < 0)
527 buflist = ap->a_buflist;
528 len = buflist->bs_nchildren;
529 start_lbn = buflist->bs_children[0]->b_lblkno;
530 end_lbn = start_lbn + len - 1;
532 for (i = 0; i < len; i++)
533 if (!ffs_checkblk(ip,
534 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
535 panic("ffs_reallocblks: unallocated block 1");
536 for (i = 1; i < len; i++)
537 if (buflist->bs_children[i]->b_lblkno != start_lbn + i)
538 panic("ffs_reallocblks: non-logical cluster");
539 blkno = buflist->bs_children[0]->b_blkno;
540 ssize = fsbtodb(fs, fs->fs_frag);
541 for (i = 1; i < len - 1; i++)
542 if (buflist->bs_children[i]->b_blkno != blkno + (i * ssize))
543 panic("ffs_reallocblks: non-physical cluster %d", i);
546 * If the cluster crosses the boundary for the first indirect
547 * block, leave space for the indirect block. Indirect blocks
548 * are initially laid out in a position after the last direct
549 * block. Block reallocation would usually destroy locality by
550 * moving the indirect block out of the way to make room for
551 * data blocks if we didn't compensate here. We should also do
552 * this for other indirect block boundaries, but it is only
553 * important for the first one.
555 if (start_lbn < NDADDR && end_lbn >= NDADDR)
558 * If the latest allocation is in a new cylinder group, assume that
559 * the filesystem has decided to move and do not force it back to
560 * the previous cylinder group.
562 if (dtog(fs, dbtofsb(fs, buflist->bs_children[0]->b_blkno)) !=
563 dtog(fs, dbtofsb(fs, buflist->bs_children[len - 1]->b_blkno)))
565 if (ufs_getlbns(vp, start_lbn, start_ap, &start_lvl) ||
566 ufs_getlbns(vp, end_lbn, end_ap, &end_lvl))
569 * Get the starting offset and block map for the first block.
571 if (start_lvl == 0) {
572 sbap = &ip->i_din1->di_db[0];
575 idp = &start_ap[start_lvl - 1];
576 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &sbp)) {
580 sbap = (ufs1_daddr_t *)sbp->b_data;
584 * If the block range spans two block maps, get the second map.
586 if (end_lvl == 0 || (idp = &end_ap[end_lvl - 1])->in_off + 1 >= len) {
591 start_ap[start_lvl - 1].in_lbn == idp->in_lbn)
592 panic("ffs_reallocblk: start == end");
594 ssize = len - (idp->in_off + 1);
595 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &ebp))
597 ebap = (ufs1_daddr_t *)ebp->b_data;
600 * Find the preferred location for the cluster.
603 pref = ffs_blkpref_ufs1(ip, start_lbn, soff, sbap);
605 * Search the block map looking for an allocation of the desired size.
607 if ((newblk = ffs_hashalloc(ip, dtog(fs, pref), pref,
608 len, len, ffs_clusteralloc)) == 0) {
613 * We have found a new contiguous block.
615 * First we have to replace the old block pointers with the new
616 * block pointers in the inode and indirect blocks associated
621 printf("realloc: ino %d, lbns %jd-%jd\n\told:", ip->i_number,
622 (intmax_t)start_lbn, (intmax_t)end_lbn);
625 for (bap = &sbap[soff], i = 0; i < len; i++, blkno += fs->fs_frag) {
631 if (!ffs_checkblk(ip,
632 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
633 panic("ffs_reallocblks: unallocated block 2");
634 if (dbtofsb(fs, buflist->bs_children[i]->b_blkno) != *bap)
635 panic("ffs_reallocblks: alloc mismatch");
639 printf(" %d,", *bap);
641 if (DOINGSOFTDEP(vp)) {
642 if (sbap == &ip->i_din1->di_db[0] && i < ssize)
643 softdep_setup_allocdirect(ip, start_lbn + i,
644 blkno, *bap, fs->fs_bsize, fs->fs_bsize,
645 buflist->bs_children[i]);
647 softdep_setup_allocindir_page(ip, start_lbn + i,
648 i < ssize ? sbp : ebp, soff + i, blkno,
649 *bap, buflist->bs_children[i]);
654 * Next we must write out the modified inode and indirect blocks.
655 * For strict correctness, the writes should be synchronous since
656 * the old block values may have been written to disk. In practise
657 * they are almost never written, but if we are concerned about
658 * strict correctness, the `doasyncfree' flag should be set to zero.
660 * The test on `doasyncfree' should be changed to test a flag
661 * that shows whether the associated buffers and inodes have
662 * been written. The flag should be set when the cluster is
663 * started and cleared whenever the buffer or inode is flushed.
664 * We can then check below to see if it is set, and do the
665 * synchronous write only when it has been cleared.
667 if (sbap != &ip->i_din1->di_db[0]) {
673 ip->i_flag |= IN_CHANGE | IN_UPDATE;
684 * Last, free the old blocks and assign the new blocks to the buffers.
690 for (blkno = newblk, i = 0; i < len; i++, blkno += fs->fs_frag) {
691 if (!DOINGSOFTDEP(vp))
692 ffs_blkfree(ump, fs, ip->i_devvp,
693 dbtofsb(fs, buflist->bs_children[i]->b_blkno),
694 fs->fs_bsize, ip->i_number, vp->v_type, NULL);
695 buflist->bs_children[i]->b_blkno = fsbtodb(fs, blkno);
697 if (!ffs_checkblk(ip,
698 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
699 panic("ffs_reallocblks: unallocated block 3");
703 printf(" %d,", blkno);
717 if (sbap != &ip->i_din1->di_db[0])
723 ffs_reallocblks_ufs2(ap)
724 struct vop_reallocblks_args /* {
726 struct cluster_save *a_buflist;
732 struct buf *sbp, *ebp;
733 ufs2_daddr_t *bap, *sbap, *ebap = 0;
734 struct cluster_save *buflist;
735 struct ufsmount *ump;
736 ufs_lbn_t start_lbn, end_lbn;
737 ufs2_daddr_t soff, newblk, blkno, pref;
738 struct indir start_ap[NIADDR + 1], end_ap[NIADDR + 1], *idp;
739 int i, len, start_lvl, end_lvl, ssize;
746 * If we are not tracking block clusters or if we have less than 4%
747 * free blocks left, then do not attempt to cluster. Running with
748 * less than 5% free block reserve is not recommended and those that
749 * choose to do so do not expect to have good file layout.
751 if (fs->fs_contigsumsize <= 0 || freespace(fs, 4) < 0)
753 buflist = ap->a_buflist;
754 len = buflist->bs_nchildren;
755 start_lbn = buflist->bs_children[0]->b_lblkno;
756 end_lbn = start_lbn + len - 1;
758 for (i = 0; i < len; i++)
759 if (!ffs_checkblk(ip,
760 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
761 panic("ffs_reallocblks: unallocated block 1");
762 for (i = 1; i < len; i++)
763 if (buflist->bs_children[i]->b_lblkno != start_lbn + i)
764 panic("ffs_reallocblks: non-logical cluster");
765 blkno = buflist->bs_children[0]->b_blkno;
766 ssize = fsbtodb(fs, fs->fs_frag);
767 for (i = 1; i < len - 1; i++)
768 if (buflist->bs_children[i]->b_blkno != blkno + (i * ssize))
769 panic("ffs_reallocblks: non-physical cluster %d", i);
772 * If the cluster crosses the boundary for the first indirect
773 * block, do not move anything in it. Indirect blocks are
774 * usually initially laid out in a position between the data
775 * blocks. Block reallocation would usually destroy locality by
776 * moving the indirect block out of the way to make room for
777 * data blocks if we didn't compensate here. We should also do
778 * this for other indirect block boundaries, but it is only
779 * important for the first one.
781 if (start_lbn < NDADDR && end_lbn >= NDADDR)
784 * If the latest allocation is in a new cylinder group, assume that
785 * the filesystem has decided to move and do not force it back to
786 * the previous cylinder group.
788 if (dtog(fs, dbtofsb(fs, buflist->bs_children[0]->b_blkno)) !=
789 dtog(fs, dbtofsb(fs, buflist->bs_children[len - 1]->b_blkno)))
791 if (ufs_getlbns(vp, start_lbn, start_ap, &start_lvl) ||
792 ufs_getlbns(vp, end_lbn, end_ap, &end_lvl))
795 * Get the starting offset and block map for the first block.
797 if (start_lvl == 0) {
798 sbap = &ip->i_din2->di_db[0];
801 idp = &start_ap[start_lvl - 1];
802 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &sbp)) {
806 sbap = (ufs2_daddr_t *)sbp->b_data;
810 * If the block range spans two block maps, get the second map.
812 if (end_lvl == 0 || (idp = &end_ap[end_lvl - 1])->in_off + 1 >= len) {
817 start_ap[start_lvl - 1].in_lbn == idp->in_lbn)
818 panic("ffs_reallocblk: start == end");
820 ssize = len - (idp->in_off + 1);
821 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &ebp))
823 ebap = (ufs2_daddr_t *)ebp->b_data;
826 * Find the preferred location for the cluster.
829 pref = ffs_blkpref_ufs2(ip, start_lbn, soff, sbap);
831 * Search the block map looking for an allocation of the desired size.
833 if ((newblk = ffs_hashalloc(ip, dtog(fs, pref), pref,
834 len, len, ffs_clusteralloc)) == 0) {
839 * We have found a new contiguous block.
841 * First we have to replace the old block pointers with the new
842 * block pointers in the inode and indirect blocks associated
847 printf("realloc: ino %d, lbns %jd-%jd\n\told:", ip->i_number,
848 (intmax_t)start_lbn, (intmax_t)end_lbn);
851 for (bap = &sbap[soff], i = 0; i < len; i++, blkno += fs->fs_frag) {
857 if (!ffs_checkblk(ip,
858 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
859 panic("ffs_reallocblks: unallocated block 2");
860 if (dbtofsb(fs, buflist->bs_children[i]->b_blkno) != *bap)
861 panic("ffs_reallocblks: alloc mismatch");
865 printf(" %jd,", (intmax_t)*bap);
867 if (DOINGSOFTDEP(vp)) {
868 if (sbap == &ip->i_din2->di_db[0] && i < ssize)
869 softdep_setup_allocdirect(ip, start_lbn + i,
870 blkno, *bap, fs->fs_bsize, fs->fs_bsize,
871 buflist->bs_children[i]);
873 softdep_setup_allocindir_page(ip, start_lbn + i,
874 i < ssize ? sbp : ebp, soff + i, blkno,
875 *bap, buflist->bs_children[i]);
880 * Next we must write out the modified inode and indirect blocks.
881 * For strict correctness, the writes should be synchronous since
882 * the old block values may have been written to disk. In practise
883 * they are almost never written, but if we are concerned about
884 * strict correctness, the `doasyncfree' flag should be set to zero.
886 * The test on `doasyncfree' should be changed to test a flag
887 * that shows whether the associated buffers and inodes have
888 * been written. The flag should be set when the cluster is
889 * started and cleared whenever the buffer or inode is flushed.
890 * We can then check below to see if it is set, and do the
891 * synchronous write only when it has been cleared.
893 if (sbap != &ip->i_din2->di_db[0]) {
899 ip->i_flag |= IN_CHANGE | IN_UPDATE;
910 * Last, free the old blocks and assign the new blocks to the buffers.
916 for (blkno = newblk, i = 0; i < len; i++, blkno += fs->fs_frag) {
917 if (!DOINGSOFTDEP(vp))
918 ffs_blkfree(ump, fs, ip->i_devvp,
919 dbtofsb(fs, buflist->bs_children[i]->b_blkno),
920 fs->fs_bsize, ip->i_number, vp->v_type, NULL);
921 buflist->bs_children[i]->b_blkno = fsbtodb(fs, blkno);
923 if (!ffs_checkblk(ip,
924 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
925 panic("ffs_reallocblks: unallocated block 3");
929 printf(" %jd,", (intmax_t)blkno);
943 if (sbap != &ip->i_din2->di_db[0])
949 * Allocate an inode in the filesystem.
951 * If allocating a directory, use ffs_dirpref to select the inode.
952 * If allocating in a directory, the following hierarchy is followed:
953 * 1) allocate the preferred inode.
954 * 2) allocate an inode in the same cylinder group.
955 * 3) quadradically rehash into other cylinder groups, until an
956 * available inode is located.
957 * If no inode preference is given the following hierarchy is used
958 * to allocate an inode:
959 * 1) allocate an inode in cylinder group 0.
960 * 2) quadradically rehash into other cylinder groups, until an
961 * available inode is located.
964 ffs_valloc(pvp, mode, cred, vpp)
974 struct ufsmount *ump;
977 int error, error1, reclaimed;
978 static struct timeval lastfail;
989 if (fs->fs_cstotal.cs_nifree == 0)
992 if ((mode & IFMT) == IFDIR)
993 ipref = ffs_dirpref(pip);
995 ipref = pip->i_number;
996 if (ipref >= fs->fs_ncg * fs->fs_ipg)
998 cg = ino_to_cg(fs, ipref);
1000 * Track number of dirs created one after another
1001 * in a same cg without intervening by files.
1003 if ((mode & IFMT) == IFDIR) {
1004 if (fs->fs_contigdirs[cg] < 255)
1005 fs->fs_contigdirs[cg]++;
1007 if (fs->fs_contigdirs[cg] > 0)
1008 fs->fs_contigdirs[cg]--;
1010 ino = (ino_t)ffs_hashalloc(pip, cg, ipref, mode, 0,
1011 (allocfcn_t *)ffs_nodealloccg);
1014 error = ffs_vget(pvp->v_mount, ino, LK_EXCLUSIVE, vpp);
1016 error1 = ffs_vgetf(pvp->v_mount, ino, LK_EXCLUSIVE, vpp,
1018 ffs_vfree(pvp, ino, mode);
1023 ip->i_flag |= IN_MODIFIED;
1031 printf("mode = 0%o, inum = %lu, fs = %s\n",
1032 ip->i_mode, (u_long)ip->i_number, fs->fs_fsmnt);
1033 panic("ffs_valloc: dup alloc");
1035 if (DIP(ip, i_blocks) && (fs->fs_flags & FS_UNCLEAN) == 0) { /* XXX */
1036 printf("free inode %s/%lu had %ld blocks\n",
1037 fs->fs_fsmnt, (u_long)ino, (long)DIP(ip, i_blocks));
1038 DIP_SET(ip, i_blocks, 0);
1041 DIP_SET(ip, i_flags, 0);
1043 * Set up a new generation number for this inode.
1045 if (ip->i_gen == 0 || ++ip->i_gen == 0)
1046 ip->i_gen = arc4random() / 2 + 1;
1047 DIP_SET(ip, i_gen, ip->i_gen);
1048 if (fs->fs_magic == FS_UFS2_MAGIC) {
1050 ip->i_din2->di_birthtime = ts.tv_sec;
1051 ip->i_din2->di_birthnsec = ts.tv_nsec;
1053 ufs_prepare_reclaim(*vpp);
1055 (*vpp)->v_vflag = 0;
1056 (*vpp)->v_type = VNON;
1057 if (fs->fs_magic == FS_UFS2_MAGIC)
1058 (*vpp)->v_op = &ffs_vnodeops2;
1060 (*vpp)->v_op = &ffs_vnodeops1;
1063 if (reclaimed == 0) {
1065 softdep_request_cleanup(fs, pvp, cred, FLUSH_INODES_WAIT);
1069 if (ppsratecheck(&lastfail, &curfail, 1)) {
1070 ffs_fserr(fs, pip->i_number, "out of inodes");
1071 uprintf("\n%s: create/symlink failed, no inodes free\n",
1078 * Find a cylinder group to place a directory.
1080 * The policy implemented by this algorithm is to allocate a
1081 * directory inode in the same cylinder group as its parent
1082 * directory, but also to reserve space for its files inodes
1083 * and data. Restrict the number of directories which may be
1084 * allocated one after another in the same cylinder group
1085 * without intervening allocation of files.
1087 * If we allocate a first level directory then force allocation
1088 * in another cylinder group.
1095 int cg, prefcg, dirsize, cgsize;
1096 u_int avgifree, avgbfree, avgndir, curdirsize;
1097 u_int minifree, minbfree, maxndir;
1098 u_int mincg, minndir;
1099 u_int maxcontigdirs;
1101 mtx_assert(UFS_MTX(pip->i_ump), MA_OWNED);
1104 avgifree = fs->fs_cstotal.cs_nifree / fs->fs_ncg;
1105 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
1106 avgndir = fs->fs_cstotal.cs_ndir / fs->fs_ncg;
1109 * Force allocation in another cg if creating a first level dir.
1111 ASSERT_VOP_LOCKED(ITOV(pip), "ffs_dirpref");
1112 if (ITOV(pip)->v_vflag & VV_ROOT) {
1113 prefcg = arc4random() % fs->fs_ncg;
1115 minndir = fs->fs_ipg;
1116 for (cg = prefcg; cg < fs->fs_ncg; cg++)
1117 if (fs->fs_cs(fs, cg).cs_ndir < minndir &&
1118 fs->fs_cs(fs, cg).cs_nifree >= avgifree &&
1119 fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
1121 minndir = fs->fs_cs(fs, cg).cs_ndir;
1123 for (cg = 0; cg < prefcg; cg++)
1124 if (fs->fs_cs(fs, cg).cs_ndir < minndir &&
1125 fs->fs_cs(fs, cg).cs_nifree >= avgifree &&
1126 fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
1128 minndir = fs->fs_cs(fs, cg).cs_ndir;
1130 return ((ino_t)(fs->fs_ipg * mincg));
1134 * Count various limits which used for
1135 * optimal allocation of a directory inode.
1137 maxndir = min(avgndir + fs->fs_ipg / 16, fs->fs_ipg);
1138 minifree = avgifree - avgifree / 4;
1141 minbfree = avgbfree - avgbfree / 4;
1144 cgsize = fs->fs_fsize * fs->fs_fpg;
1145 dirsize = fs->fs_avgfilesize * fs->fs_avgfpdir;
1146 curdirsize = avgndir ? (cgsize - avgbfree * fs->fs_bsize) / avgndir : 0;
1147 if (dirsize < curdirsize)
1148 dirsize = curdirsize;
1150 maxcontigdirs = 0; /* dirsize overflowed */
1152 maxcontigdirs = min((avgbfree * fs->fs_bsize) / dirsize, 255);
1153 if (fs->fs_avgfpdir > 0)
1154 maxcontigdirs = min(maxcontigdirs,
1155 fs->fs_ipg / fs->fs_avgfpdir);
1156 if (maxcontigdirs == 0)
1160 * Limit number of dirs in one cg and reserve space for
1161 * regular files, but only if we have no deficit in
1164 * We are trying to find a suitable cylinder group nearby
1165 * our preferred cylinder group to place a new directory.
1166 * We scan from our preferred cylinder group forward looking
1167 * for a cylinder group that meets our criterion. If we get
1168 * to the final cylinder group and do not find anything,
1169 * we start scanning backwards from our preferred cylinder
1170 * group. The ideal would be to alternate looking forward
1171 * and backward, but that is just too complex to code for
1172 * the gain it would get. The most likely place where the
1173 * backward scan would take effect is when we start near
1174 * the end of the filesystem and do not find anything from
1175 * where we are to the end. In that case, scanning backward
1176 * will likely find us a suitable cylinder group much closer
1177 * to our desired location than if we were to start scanning
1178 * forward from the beginning of the filesystem.
1180 prefcg = ino_to_cg(fs, pip->i_number);
1181 for (cg = prefcg; cg < fs->fs_ncg; cg++)
1182 if (fs->fs_cs(fs, cg).cs_ndir < maxndir &&
1183 fs->fs_cs(fs, cg).cs_nifree >= minifree &&
1184 fs->fs_cs(fs, cg).cs_nbfree >= minbfree) {
1185 if (fs->fs_contigdirs[cg] < maxcontigdirs)
1186 return ((ino_t)(fs->fs_ipg * cg));
1188 for (cg = 0; cg < prefcg; cg++)
1189 if (fs->fs_cs(fs, cg).cs_ndir < maxndir &&
1190 fs->fs_cs(fs, cg).cs_nifree >= minifree &&
1191 fs->fs_cs(fs, cg).cs_nbfree >= minbfree) {
1192 if (fs->fs_contigdirs[cg] < maxcontigdirs)
1193 return ((ino_t)(fs->fs_ipg * cg));
1196 * This is a backstop when we have deficit in space.
1198 for (cg = prefcg; cg < fs->fs_ncg; cg++)
1199 if (fs->fs_cs(fs, cg).cs_nifree >= avgifree)
1200 return ((ino_t)(fs->fs_ipg * cg));
1201 for (cg = 0; cg < prefcg; cg++)
1202 if (fs->fs_cs(fs, cg).cs_nifree >= avgifree)
1204 return ((ino_t)(fs->fs_ipg * cg));
1208 * Select the desired position for the next block in a file. The file is
1209 * logically divided into sections. The first section is composed of the
1210 * direct blocks. Each additional section contains fs_maxbpg blocks.
1212 * If no blocks have been allocated in the first section, the policy is to
1213 * request a block in the same cylinder group as the inode that describes
1214 * the file. The first indirect is allocated immediately following the last
1215 * direct block and the data blocks for the first indirect immediately
1218 * If no blocks have been allocated in any other section, the indirect
1219 * block(s) are allocated in the same cylinder group as its inode in an
1220 * area reserved immediately following the inode blocks. The policy for
1221 * the data blocks is to place them in a cylinder group with a greater than
1222 * average number of free blocks. An appropriate cylinder group is found
1223 * by using a rotor that sweeps the cylinder groups. When a new group of
1224 * blocks is needed, the sweep begins in the cylinder group following the
1225 * cylinder group from which the previous allocation was made. The sweep
1226 * continues until a cylinder group with greater than the average number
1227 * of free blocks is found. If the allocation is for the first block in an
1228 * indirect block, the information on the previous allocation is unavailable;
1229 * here a best guess is made based upon the logical block number being
1232 * If a section is already partially allocated, the policy is to
1233 * contiguously allocate fs_maxcontig blocks. The end of one of these
1234 * contiguous blocks and the beginning of the next is laid out
1235 * contiguously if possible.
1238 ffs_blkpref_ufs1(ip, lbn, indx, bap)
1246 u_int avgbfree, startcg;
1249 KASSERT(indx <= 0 || bap != NULL, ("need non-NULL bap"));
1250 mtx_assert(UFS_MTX(ip->i_ump), MA_OWNED);
1253 * Allocation of indirect blocks is indicated by passing negative
1254 * values in indx: -1 for single indirect, -2 for double indirect,
1255 * -3 for triple indirect. As noted below, we attempt to allocate
1256 * the first indirect inline with the file data. For all later
1257 * indirect blocks, the data is often allocated in other cylinder
1258 * groups. However to speed random file access and to speed up
1259 * fsck, the filesystem reserves the first fs_metaspace blocks
1260 * (typically half of fs_minfree) of the data area of each cylinder
1261 * group to hold these later indirect blocks.
1263 inocg = ino_to_cg(fs, ip->i_number);
1266 * Our preference for indirect blocks is the zone at the
1267 * beginning of the inode's cylinder group data area that
1268 * we try to reserve for indirect blocks.
1270 pref = cgmeta(fs, inocg);
1272 * If we are allocating the first indirect block, try to
1273 * place it immediately following the last direct block.
1275 if (indx == -1 && lbn < NDADDR + NINDIR(fs) &&
1276 ip->i_din1->di_db[NDADDR - 1] != 0)
1277 pref = ip->i_din1->di_db[NDADDR - 1] + fs->fs_frag;
1281 * If we are allocating the first data block in the first indirect
1282 * block and the indirect has been allocated in the data block area,
1283 * try to place it immediately following the indirect block.
1285 if (lbn == NDADDR) {
1286 pref = ip->i_din1->di_ib[0];
1287 if (pref != 0 && pref >= cgdata(fs, inocg) &&
1288 pref < cgbase(fs, inocg + 1))
1289 return (pref + fs->fs_frag);
1292 * If we are at the beginning of a file, or we have already allocated
1293 * the maximum number of blocks per cylinder group, or we do not
1294 * have a block allocated immediately preceeding us, then we need
1295 * to decide where to start allocating new blocks.
1297 if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) {
1299 * If we are allocating a directory data block, we want
1300 * to place it in the metadata area.
1302 if ((ip->i_mode & IFMT) == IFDIR)
1303 return (cgmeta(fs, inocg));
1305 * Until we fill all the direct and all the first indirect's
1306 * blocks, we try to allocate in the data area of the inode's
1309 if (lbn < NDADDR + NINDIR(fs))
1310 return (cgdata(fs, inocg));
1312 * Find a cylinder with greater than average number of
1313 * unused data blocks.
1315 if (indx == 0 || bap[indx - 1] == 0)
1316 startcg = inocg + lbn / fs->fs_maxbpg;
1318 startcg = dtog(fs, bap[indx - 1]) + 1;
1319 startcg %= fs->fs_ncg;
1320 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
1321 for (cg = startcg; cg < fs->fs_ncg; cg++)
1322 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
1323 fs->fs_cgrotor = cg;
1324 return (cgdata(fs, cg));
1326 for (cg = 0; cg <= startcg; cg++)
1327 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
1328 fs->fs_cgrotor = cg;
1329 return (cgdata(fs, cg));
1334 * Otherwise, we just always try to lay things out contiguously.
1336 return (bap[indx - 1] + fs->fs_frag);
1340 * Same as above, but for UFS2
1343 ffs_blkpref_ufs2(ip, lbn, indx, bap)
1351 u_int avgbfree, startcg;
1354 KASSERT(indx <= 0 || bap != NULL, ("need non-NULL bap"));
1355 mtx_assert(UFS_MTX(ip->i_ump), MA_OWNED);
1358 * Allocation of indirect blocks is indicated by passing negative
1359 * values in indx: -1 for single indirect, -2 for double indirect,
1360 * -3 for triple indirect. As noted below, we attempt to allocate
1361 * the first indirect inline with the file data. For all later
1362 * indirect blocks, the data is often allocated in other cylinder
1363 * groups. However to speed random file access and to speed up
1364 * fsck, the filesystem reserves the first fs_metaspace blocks
1365 * (typically half of fs_minfree) of the data area of each cylinder
1366 * group to hold these later indirect blocks.
1368 inocg = ino_to_cg(fs, ip->i_number);
1371 * Our preference for indirect blocks is the zone at the
1372 * beginning of the inode's cylinder group data area that
1373 * we try to reserve for indirect blocks.
1375 pref = cgmeta(fs, inocg);
1377 * If we are allocating the first indirect block, try to
1378 * place it immediately following the last direct block.
1380 if (indx == -1 && lbn < NDADDR + NINDIR(fs) &&
1381 ip->i_din2->di_db[NDADDR - 1] != 0)
1382 pref = ip->i_din2->di_db[NDADDR - 1] + fs->fs_frag;
1386 * If we are allocating the first data block in the first indirect
1387 * block and the indirect has been allocated in the data block area,
1388 * try to place it immediately following the indirect block.
1390 if (lbn == NDADDR) {
1391 pref = ip->i_din2->di_ib[0];
1392 if (pref != 0 && pref >= cgdata(fs, inocg) &&
1393 pref < cgbase(fs, inocg + 1))
1394 return (pref + fs->fs_frag);
1397 * If we are at the beginning of a file, or we have already allocated
1398 * the maximum number of blocks per cylinder group, or we do not
1399 * have a block allocated immediately preceeding us, then we need
1400 * to decide where to start allocating new blocks.
1402 if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) {
1404 * If we are allocating a directory data block, we want
1405 * to place it in the metadata area.
1407 if ((ip->i_mode & IFMT) == IFDIR)
1408 return (cgmeta(fs, inocg));
1410 * Until we fill all the direct and all the first indirect's
1411 * blocks, we try to allocate in the data area of the inode's
1414 if (lbn < NDADDR + NINDIR(fs))
1415 return (cgdata(fs, inocg));
1417 * Find a cylinder with greater than average number of
1418 * unused data blocks.
1420 if (indx == 0 || bap[indx - 1] == 0)
1421 startcg = inocg + lbn / fs->fs_maxbpg;
1423 startcg = dtog(fs, bap[indx - 1]) + 1;
1424 startcg %= fs->fs_ncg;
1425 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
1426 for (cg = startcg; cg < fs->fs_ncg; cg++)
1427 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
1428 fs->fs_cgrotor = cg;
1429 return (cgdata(fs, cg));
1431 for (cg = 0; cg <= startcg; cg++)
1432 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
1433 fs->fs_cgrotor = cg;
1434 return (cgdata(fs, cg));
1439 * Otherwise, we just always try to lay things out contiguously.
1441 return (bap[indx - 1] + fs->fs_frag);
1445 * Implement the cylinder overflow algorithm.
1447 * The policy implemented by this algorithm is:
1448 * 1) allocate the block in its requested cylinder group.
1449 * 2) quadradically rehash on the cylinder group number.
1450 * 3) brute force search for a free block.
1452 * Must be called with the UFS lock held. Will release the lock on success
1453 * and return with it held on failure.
1457 ffs_hashalloc(ip, cg, pref, size, rsize, allocator)
1461 int size; /* Search size for data blocks, mode for inodes */
1462 int rsize; /* Real allocated size. */
1463 allocfcn_t *allocator;
1466 ufs2_daddr_t result;
1469 mtx_assert(UFS_MTX(ip->i_ump), MA_OWNED);
1471 if (ITOV(ip)->v_mount->mnt_kern_flag & MNTK_SUSPENDED)
1472 panic("ffs_hashalloc: allocation on suspended filesystem");
1476 * 1: preferred cylinder group
1478 result = (*allocator)(ip, cg, pref, size, rsize);
1482 * 2: quadratic rehash
1484 for (i = 1; i < fs->fs_ncg; i *= 2) {
1486 if (cg >= fs->fs_ncg)
1488 result = (*allocator)(ip, cg, 0, size, rsize);
1493 * 3: brute force search
1494 * Note that we start at i == 2, since 0 was checked initially,
1495 * and 1 is always checked in the quadratic rehash.
1497 cg = (icg + 2) % fs->fs_ncg;
1498 for (i = 2; i < fs->fs_ncg; i++) {
1499 result = (*allocator)(ip, cg, 0, size, rsize);
1503 if (cg == fs->fs_ncg)
1510 * Determine whether a fragment can be extended.
1512 * Check to see if the necessary fragments are available, and
1513 * if they are, allocate them.
1516 ffs_fragextend(ip, cg, bprev, osize, nsize)
1525 struct ufsmount *ump;
1534 if (fs->fs_cs(fs, cg).cs_nffree < numfrags(fs, nsize - osize))
1536 frags = numfrags(fs, nsize);
1537 bbase = fragnum(fs, bprev);
1538 if (bbase > fragnum(fs, (bprev + frags - 1))) {
1539 /* cannot extend across a block boundary */
1543 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
1544 (int)fs->fs_cgsize, NOCRED, &bp);
1547 cgp = (struct cg *)bp->b_data;
1548 if (!cg_chkmagic(cgp))
1550 bp->b_xflags |= BX_BKGRDWRITE;
1551 cgp->cg_old_time = cgp->cg_time = time_second;
1552 bno = dtogd(fs, bprev);
1553 blksfree = cg_blksfree(cgp);
1554 for (i = numfrags(fs, osize); i < frags; i++)
1555 if (isclr(blksfree, bno + i))
1558 * the current fragment can be extended
1559 * deduct the count on fragment being extended into
1560 * increase the count on the remaining fragment (if any)
1561 * allocate the extended piece
1563 for (i = frags; i < fs->fs_frag - bbase; i++)
1564 if (isclr(blksfree, bno + i))
1566 cgp->cg_frsum[i - numfrags(fs, osize)]--;
1568 cgp->cg_frsum[i - frags]++;
1569 for (i = numfrags(fs, osize), nffree = 0; i < frags; i++) {
1570 clrbit(blksfree, bno + i);
1571 cgp->cg_cs.cs_nffree--;
1575 fs->fs_cstotal.cs_nffree -= nffree;
1576 fs->fs_cs(fs, cg).cs_nffree -= nffree;
1578 ACTIVECLEAR(fs, cg);
1580 if (DOINGSOFTDEP(ITOV(ip)))
1581 softdep_setup_blkmapdep(bp, UFSTOVFS(ump), bprev,
1582 frags, numfrags(fs, osize));
1594 * Determine whether a block can be allocated.
1596 * Check to see if a block of the appropriate size is available,
1597 * and if it is, allocate it.
1600 ffs_alloccg(ip, cg, bpref, size, rsize)
1610 struct ufsmount *ump;
1613 int i, allocsiz, error, frags;
1618 if (fs->fs_cs(fs, cg).cs_nbfree == 0 && size == fs->fs_bsize)
1621 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
1622 (int)fs->fs_cgsize, NOCRED, &bp);
1625 cgp = (struct cg *)bp->b_data;
1626 if (!cg_chkmagic(cgp) ||
1627 (cgp->cg_cs.cs_nbfree == 0 && size == fs->fs_bsize))
1629 bp->b_xflags |= BX_BKGRDWRITE;
1630 cgp->cg_old_time = cgp->cg_time = time_second;
1631 if (size == fs->fs_bsize) {
1633 blkno = ffs_alloccgblk(ip, bp, bpref, rsize);
1634 ACTIVECLEAR(fs, cg);
1640 * check to see if any fragments are already available
1641 * allocsiz is the size which will be allocated, hacking
1642 * it down to a smaller size if necessary
1644 blksfree = cg_blksfree(cgp);
1645 frags = numfrags(fs, size);
1646 for (allocsiz = frags; allocsiz < fs->fs_frag; allocsiz++)
1647 if (cgp->cg_frsum[allocsiz] != 0)
1649 if (allocsiz == fs->fs_frag) {
1651 * no fragments were available, so a block will be
1652 * allocated, and hacked up
1654 if (cgp->cg_cs.cs_nbfree == 0)
1657 blkno = ffs_alloccgblk(ip, bp, bpref, rsize);
1658 ACTIVECLEAR(fs, cg);
1663 KASSERT(size == rsize,
1664 ("ffs_alloccg: size(%d) != rsize(%d)", size, rsize));
1665 bno = ffs_mapsearch(fs, cgp, bpref, allocsiz);
1668 for (i = 0; i < frags; i++)
1669 clrbit(blksfree, bno + i);
1670 cgp->cg_cs.cs_nffree -= frags;
1671 cgp->cg_frsum[allocsiz]--;
1672 if (frags != allocsiz)
1673 cgp->cg_frsum[allocsiz - frags]++;
1675 fs->fs_cstotal.cs_nffree -= frags;
1676 fs->fs_cs(fs, cg).cs_nffree -= frags;
1678 blkno = cgbase(fs, cg) + bno;
1679 ACTIVECLEAR(fs, cg);
1681 if (DOINGSOFTDEP(ITOV(ip)))
1682 softdep_setup_blkmapdep(bp, UFSTOVFS(ump), blkno, frags, 0);
1693 * Allocate a block in a cylinder group.
1695 * This algorithm implements the following policy:
1696 * 1) allocate the requested block.
1697 * 2) allocate a rotationally optimal block in the same cylinder.
1698 * 3) allocate the next available block on the block rotor for the
1699 * specified cylinder group.
1700 * Note that this routine only allocates fs_bsize blocks; these
1701 * blocks may be fragmented by the routine that allocates them.
1704 ffs_alloccgblk(ip, bp, bpref, size)
1712 struct ufsmount *ump;
1720 mtx_assert(UFS_MTX(ump), MA_OWNED);
1721 cgp = (struct cg *)bp->b_data;
1722 blksfree = cg_blksfree(cgp);
1724 bpref = cgbase(fs, cgp->cg_cgx) + cgp->cg_rotor + fs->fs_frag;
1725 } else if ((cgbpref = dtog(fs, bpref)) != cgp->cg_cgx) {
1726 /* map bpref to correct zone in this cg */
1727 if (bpref < cgdata(fs, cgbpref))
1728 bpref = cgmeta(fs, cgp->cg_cgx);
1730 bpref = cgdata(fs, cgp->cg_cgx);
1733 * if the requested block is available, use it
1735 bno = dtogd(fs, blknum(fs, bpref));
1736 if (ffs_isblock(fs, blksfree, fragstoblks(fs, bno)))
1739 * Take the next available block in this cylinder group.
1741 bno = ffs_mapsearch(fs, cgp, bpref, (int)fs->fs_frag);
1744 /* Update cg_rotor only if allocated from the data zone */
1745 if (bno >= dtogd(fs, cgdata(fs, cgp->cg_cgx)))
1746 cgp->cg_rotor = bno;
1748 blkno = fragstoblks(fs, bno);
1749 ffs_clrblock(fs, blksfree, (long)blkno);
1750 ffs_clusteracct(fs, cgp, blkno, -1);
1751 cgp->cg_cs.cs_nbfree--;
1752 fs->fs_cstotal.cs_nbfree--;
1753 fs->fs_cs(fs, cgp->cg_cgx).cs_nbfree--;
1755 blkno = cgbase(fs, cgp->cg_cgx) + bno;
1757 * If the caller didn't want the whole block free the frags here.
1759 size = numfrags(fs, size);
1760 if (size != fs->fs_frag) {
1761 bno = dtogd(fs, blkno);
1762 for (i = size; i < fs->fs_frag; i++)
1763 setbit(blksfree, bno + i);
1764 i = fs->fs_frag - size;
1765 cgp->cg_cs.cs_nffree += i;
1766 fs->fs_cstotal.cs_nffree += i;
1767 fs->fs_cs(fs, cgp->cg_cgx).cs_nffree += i;
1773 if (DOINGSOFTDEP(ITOV(ip)))
1774 softdep_setup_blkmapdep(bp, UFSTOVFS(ump), blkno,
1781 * Determine whether a cluster can be allocated.
1783 * We do not currently check for optimal rotational layout if there
1784 * are multiple choices in the same cylinder group. Instead we just
1785 * take the first one that we find following bpref.
1788 ffs_clusteralloc(ip, cg, bpref, len, unused)
1798 struct ufsmount *ump;
1799 int i, run, bit, map, got;
1807 if (fs->fs_maxcluster[cg] < len)
1810 if (bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), (int)fs->fs_cgsize,
1813 cgp = (struct cg *)bp->b_data;
1814 if (!cg_chkmagic(cgp))
1816 bp->b_xflags |= BX_BKGRDWRITE;
1818 * Check to see if a cluster of the needed size (or bigger) is
1819 * available in this cylinder group.
1821 lp = &cg_clustersum(cgp)[len];
1822 for (i = len; i <= fs->fs_contigsumsize; i++)
1825 if (i > fs->fs_contigsumsize) {
1827 * This is the first time looking for a cluster in this
1828 * cylinder group. Update the cluster summary information
1829 * to reflect the true maximum sized cluster so that
1830 * future cluster allocation requests can avoid reading
1831 * the cylinder group map only to find no clusters.
1833 lp = &cg_clustersum(cgp)[len - 1];
1834 for (i = len - 1; i > 0; i--)
1838 fs->fs_maxcluster[cg] = i;
1842 * Search the cluster map to find a big enough cluster.
1843 * We take the first one that we find, even if it is larger
1844 * than we need as we prefer to get one close to the previous
1845 * block allocation. We do not search before the current
1846 * preference point as we do not want to allocate a block
1847 * that is allocated before the previous one (as we will
1848 * then have to wait for another pass of the elevator
1849 * algorithm before it will be read). We prefer to fail and
1850 * be recalled to try an allocation in the next cylinder group.
1852 if (dtog(fs, bpref) != cg)
1853 bpref = cgdata(fs, cg);
1855 bpref = blknum(fs, bpref);
1856 bpref = fragstoblks(fs, dtogd(fs, bpref));
1857 mapp = &cg_clustersfree(cgp)[bpref / NBBY];
1859 bit = 1 << (bpref % NBBY);
1860 for (run = 0, got = bpref; got < cgp->cg_nclusterblks; got++) {
1861 if ((map & bit) == 0) {
1868 if ((got & (NBBY - 1)) != (NBBY - 1)) {
1875 if (got >= cgp->cg_nclusterblks)
1878 * Allocate the cluster that we have found.
1880 blksfree = cg_blksfree(cgp);
1881 for (i = 1; i <= len; i++)
1882 if (!ffs_isblock(fs, blksfree, got - run + i))
1883 panic("ffs_clusteralloc: map mismatch");
1884 bno = cgbase(fs, cg) + blkstofrags(fs, got - run + 1);
1885 if (dtog(fs, bno) != cg)
1886 panic("ffs_clusteralloc: allocated out of group");
1887 len = blkstofrags(fs, len);
1889 for (i = 0; i < len; i += fs->fs_frag)
1890 if (ffs_alloccgblk(ip, bp, bno + i, fs->fs_bsize) != bno + i)
1891 panic("ffs_clusteralloc: lost block");
1892 ACTIVECLEAR(fs, cg);
1904 static inline struct buf *
1905 getinobuf(struct inode *ip, u_int cg, u_int32_t cginoblk, int gbflags)
1910 return (getblk(ip->i_devvp, fsbtodb(fs, ino_to_fsba(fs,
1911 cg * fs->fs_ipg + cginoblk)), (int)fs->fs_bsize, 0, 0,
1916 * Determine whether an inode can be allocated.
1918 * Check to see if an inode is available, and if it is,
1919 * allocate it using the following policy:
1920 * 1) allocate the requested inode.
1921 * 2) allocate the next available inode after the requested
1922 * inode in the specified cylinder group.
1925 ffs_nodealloccg(ip, cg, ipref, mode, unused)
1934 struct buf *bp, *ibp;
1935 struct ufsmount *ump;
1937 struct ufs2_dinode *dp2;
1938 int error, start, len, loc, map, i;
1939 u_int32_t old_initediblk;
1944 if (fs->fs_cs(fs, cg).cs_nifree == 0)
1947 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
1948 (int)fs->fs_cgsize, NOCRED, &bp);
1954 cgp = (struct cg *)bp->b_data;
1956 if (!cg_chkmagic(cgp) || cgp->cg_cs.cs_nifree == 0) {
1961 bp->b_xflags |= BX_BKGRDWRITE;
1962 inosused = cg_inosused(cgp);
1964 ipref %= fs->fs_ipg;
1965 if (isclr(inosused, ipref))
1968 start = cgp->cg_irotor / NBBY;
1969 len = howmany(fs->fs_ipg - cgp->cg_irotor, NBBY);
1970 loc = skpc(0xff, len, &inosused[start]);
1974 loc = skpc(0xff, len, &inosused[0]);
1976 printf("cg = %d, irotor = %ld, fs = %s\n",
1977 cg, (long)cgp->cg_irotor, fs->fs_fsmnt);
1978 panic("ffs_nodealloccg: map corrupted");
1982 i = start + len - loc;
1983 map = inosused[i] ^ 0xff;
1985 printf("fs = %s\n", fs->fs_fsmnt);
1986 panic("ffs_nodealloccg: block not in map");
1988 ipref = i * NBBY + ffs(map) - 1;
1991 * Check to see if we need to initialize more inodes.
1993 if (fs->fs_magic == FS_UFS2_MAGIC &&
1994 ipref + INOPB(fs) > cgp->cg_initediblk &&
1995 cgp->cg_initediblk < cgp->cg_niblk) {
1996 old_initediblk = cgp->cg_initediblk;
1999 * Free the cylinder group lock before writing the
2000 * initialized inode block. Entering the
2001 * babarrierwrite() with the cylinder group lock
2002 * causes lock order violation between the lock and
2005 * Another thread can decide to initialize the same
2006 * inode block, but whichever thread first gets the
2007 * cylinder group lock after writing the newly
2008 * allocated inode block will update it and the other
2009 * will realize that it has lost and leave the
2010 * cylinder group unchanged.
2012 ibp = getinobuf(ip, cg, old_initediblk, GB_LOCK_NOWAIT);
2016 * The inode block buffer is already owned by
2017 * another thread, which must initialize it.
2018 * Wait on the buffer to allow another thread
2019 * to finish the updates, with dropped cg
2020 * buffer lock, then retry.
2022 ibp = getinobuf(ip, cg, old_initediblk, 0);
2027 bzero(ibp->b_data, (int)fs->fs_bsize);
2028 dp2 = (struct ufs2_dinode *)(ibp->b_data);
2029 for (i = 0; i < INOPB(fs); i++) {
2030 dp2->di_gen = arc4random() / 2 + 1;
2034 * Rather than adding a soft updates dependency to ensure
2035 * that the new inode block is written before it is claimed
2036 * by the cylinder group map, we just do a barrier write
2037 * here. The barrier write will ensure that the inode block
2038 * gets written before the updated cylinder group map can be
2039 * written. The barrier write should only slow down bulk
2040 * loading of newly created filesystems.
2042 babarrierwrite(ibp);
2045 * After the inode block is written, try to update the
2046 * cg initediblk pointer. If another thread beat us
2047 * to it, then leave it unchanged as the other thread
2048 * has already set it correctly.
2050 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
2051 (int)fs->fs_cgsize, NOCRED, &bp);
2053 ACTIVECLEAR(fs, cg);
2059 cgp = (struct cg *)bp->b_data;
2060 if (cgp->cg_initediblk == old_initediblk)
2061 cgp->cg_initediblk += INOPB(fs);
2064 cgp->cg_old_time = cgp->cg_time = time_second;
2065 cgp->cg_irotor = ipref;
2067 ACTIVECLEAR(fs, cg);
2068 setbit(inosused, ipref);
2069 cgp->cg_cs.cs_nifree--;
2070 fs->fs_cstotal.cs_nifree--;
2071 fs->fs_cs(fs, cg).cs_nifree--;
2073 if ((mode & IFMT) == IFDIR) {
2074 cgp->cg_cs.cs_ndir++;
2075 fs->fs_cstotal.cs_ndir++;
2076 fs->fs_cs(fs, cg).cs_ndir++;
2079 if (DOINGSOFTDEP(ITOV(ip)))
2080 softdep_setup_inomapdep(bp, ip, cg * fs->fs_ipg + ipref, mode);
2082 return ((ino_t)(cg * fs->fs_ipg + ipref));
2086 * Free a block or fragment.
2088 * The specified block or fragment is placed back in the
2089 * free map. If a fragment is deallocated, a possible
2090 * block reassembly is checked.
2093 ffs_blkfree_cg(ump, fs, devvp, bno, size, inum, dephd)
2094 struct ufsmount *ump;
2096 struct vnode *devvp;
2100 struct workhead *dephd;
2105 ufs1_daddr_t fragno, cgbno;
2106 ufs2_daddr_t cgblkno;
2107 int i, blk, frags, bbase;
2113 if (devvp->v_type == VREG) {
2114 /* devvp is a snapshot */
2115 dev = VTOI(devvp)->i_devvp->v_rdev;
2116 cgblkno = fragstoblks(fs, cgtod(fs, cg));
2118 /* devvp is a normal disk device */
2119 dev = devvp->v_rdev;
2120 cgblkno = fsbtodb(fs, cgtod(fs, cg));
2121 ASSERT_VOP_LOCKED(devvp, "ffs_blkfree_cg");
2124 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0 ||
2125 fragnum(fs, bno) + numfrags(fs, size) > fs->fs_frag) {
2126 printf("dev=%s, bno = %jd, bsize = %ld, size = %ld, fs = %s\n",
2127 devtoname(dev), (intmax_t)bno, (long)fs->fs_bsize,
2128 size, fs->fs_fsmnt);
2129 panic("ffs_blkfree_cg: bad size");
2132 if ((u_int)bno >= fs->fs_size) {
2133 printf("bad block %jd, ino %lu\n", (intmax_t)bno,
2135 ffs_fserr(fs, inum, "bad block");
2138 if (bread(devvp, cgblkno, (int)fs->fs_cgsize, NOCRED, &bp)) {
2142 cgp = (struct cg *)bp->b_data;
2143 if (!cg_chkmagic(cgp)) {
2147 bp->b_xflags |= BX_BKGRDWRITE;
2148 cgp->cg_old_time = cgp->cg_time = time_second;
2149 cgbno = dtogd(fs, bno);
2150 blksfree = cg_blksfree(cgp);
2152 if (size == fs->fs_bsize) {
2153 fragno = fragstoblks(fs, cgbno);
2154 if (!ffs_isfreeblock(fs, blksfree, fragno)) {
2155 if (devvp->v_type == VREG) {
2157 /* devvp is a snapshot */
2161 printf("dev = %s, block = %jd, fs = %s\n",
2162 devtoname(dev), (intmax_t)bno, fs->fs_fsmnt);
2163 panic("ffs_blkfree_cg: freeing free block");
2165 ffs_setblock(fs, blksfree, fragno);
2166 ffs_clusteracct(fs, cgp, fragno, 1);
2167 cgp->cg_cs.cs_nbfree++;
2168 fs->fs_cstotal.cs_nbfree++;
2169 fs->fs_cs(fs, cg).cs_nbfree++;
2171 bbase = cgbno - fragnum(fs, cgbno);
2173 * decrement the counts associated with the old frags
2175 blk = blkmap(fs, blksfree, bbase);
2176 ffs_fragacct(fs, blk, cgp->cg_frsum, -1);
2178 * deallocate the fragment
2180 frags = numfrags(fs, size);
2181 for (i = 0; i < frags; i++) {
2182 if (isset(blksfree, cgbno + i)) {
2183 printf("dev = %s, block = %jd, fs = %s\n",
2184 devtoname(dev), (intmax_t)(bno + i),
2186 panic("ffs_blkfree_cg: freeing free frag");
2188 setbit(blksfree, cgbno + i);
2190 cgp->cg_cs.cs_nffree += i;
2191 fs->fs_cstotal.cs_nffree += i;
2192 fs->fs_cs(fs, cg).cs_nffree += i;
2194 * add back in counts associated with the new frags
2196 blk = blkmap(fs, blksfree, bbase);
2197 ffs_fragacct(fs, blk, cgp->cg_frsum, 1);
2199 * if a complete block has been reassembled, account for it
2201 fragno = fragstoblks(fs, bbase);
2202 if (ffs_isblock(fs, blksfree, fragno)) {
2203 cgp->cg_cs.cs_nffree -= fs->fs_frag;
2204 fs->fs_cstotal.cs_nffree -= fs->fs_frag;
2205 fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag;
2206 ffs_clusteracct(fs, cgp, fragno, 1);
2207 cgp->cg_cs.cs_nbfree++;
2208 fs->fs_cstotal.cs_nbfree++;
2209 fs->fs_cs(fs, cg).cs_nbfree++;
2213 ACTIVECLEAR(fs, cg);
2216 if (MOUNTEDSOFTDEP(mp) && devvp->v_type != VREG)
2217 softdep_setup_blkfree(UFSTOVFS(ump), bp, bno,
2218 numfrags(fs, size), dephd);
2222 TASKQUEUE_DEFINE_THREAD(ffs_trim);
2224 struct ffs_blkfree_trim_params {
2226 struct ufsmount *ump;
2227 struct vnode *devvp;
2231 struct workhead *pdephd;
2232 struct workhead dephd;
2236 ffs_blkfree_trim_task(ctx, pending)
2240 struct ffs_blkfree_trim_params *tp;
2243 ffs_blkfree_cg(tp->ump, tp->ump->um_fs, tp->devvp, tp->bno, tp->size,
2244 tp->inum, tp->pdephd);
2245 vn_finished_secondary_write(UFSTOVFS(tp->ump));
2250 ffs_blkfree_trim_completed(bip)
2253 struct ffs_blkfree_trim_params *tp;
2255 tp = bip->bio_caller2;
2257 TASK_INIT(&tp->task, 0, ffs_blkfree_trim_task, tp);
2258 taskqueue_enqueue(taskqueue_ffs_trim, &tp->task);
2262 ffs_blkfree(ump, fs, devvp, bno, size, inum, vtype, dephd)
2263 struct ufsmount *ump;
2265 struct vnode *devvp;
2270 struct workhead *dephd;
2274 struct ffs_blkfree_trim_params *tp;
2277 * Check to see if a snapshot wants to claim the block.
2278 * Check that devvp is a normal disk device, not a snapshot,
2279 * it has a snapshot(s) associated with it, and one of the
2280 * snapshots wants to claim the block.
2282 if (devvp->v_type != VREG &&
2283 (devvp->v_vflag & VV_COPYONWRITE) &&
2284 ffs_snapblkfree(fs, devvp, bno, size, inum, vtype, dephd)) {
2288 * Nothing to delay if TRIM is disabled, or the operation is
2289 * performed on the snapshot.
2291 if (!ump->um_candelete || devvp->v_type == VREG) {
2292 ffs_blkfree_cg(ump, fs, devvp, bno, size, inum, dephd);
2297 * Postpone the set of the free bit in the cg bitmap until the
2298 * BIO_DELETE is completed. Otherwise, due to disk queue
2299 * reordering, TRIM might be issued after we reuse the block
2300 * and write some new data into it.
2302 tp = malloc(sizeof(struct ffs_blkfree_trim_params), M_TEMP, M_WAITOK);
2308 if (dephd != NULL) {
2309 LIST_INIT(&tp->dephd);
2310 LIST_SWAP(dephd, &tp->dephd, worklist, wk_list);
2311 tp->pdephd = &tp->dephd;
2315 bip = g_alloc_bio();
2316 bip->bio_cmd = BIO_DELETE;
2317 bip->bio_offset = dbtob(fsbtodb(fs, bno));
2318 bip->bio_done = ffs_blkfree_trim_completed;
2319 bip->bio_length = size;
2320 bip->bio_caller2 = tp;
2323 vn_start_secondary_write(NULL, &mp, 0);
2324 g_io_request(bip, (struct g_consumer *)devvp->v_bufobj.bo_private);
2329 * Verify allocation of a block or fragment. Returns true if block or
2330 * fragment is allocated, false if it is free.
2333 ffs_checkblk(ip, bno, size)
2342 int i, error, frags, free;
2346 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) {
2347 printf("bsize = %ld, size = %ld, fs = %s\n",
2348 (long)fs->fs_bsize, size, fs->fs_fsmnt);
2349 panic("ffs_checkblk: bad size");
2351 if ((u_int)bno >= fs->fs_size)
2352 panic("ffs_checkblk: bad block %jd", (intmax_t)bno);
2353 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, dtog(fs, bno))),
2354 (int)fs->fs_cgsize, NOCRED, &bp);
2356 panic("ffs_checkblk: cg bread failed");
2357 cgp = (struct cg *)bp->b_data;
2358 if (!cg_chkmagic(cgp))
2359 panic("ffs_checkblk: cg magic mismatch");
2360 bp->b_xflags |= BX_BKGRDWRITE;
2361 blksfree = cg_blksfree(cgp);
2362 cgbno = dtogd(fs, bno);
2363 if (size == fs->fs_bsize) {
2364 free = ffs_isblock(fs, blksfree, fragstoblks(fs, cgbno));
2366 frags = numfrags(fs, size);
2367 for (free = 0, i = 0; i < frags; i++)
2368 if (isset(blksfree, cgbno + i))
2370 if (free != 0 && free != frags)
2371 panic("ffs_checkblk: partially free fragment");
2376 #endif /* INVARIANTS */
2382 ffs_vfree(pvp, ino, mode)
2389 if (DOINGSOFTDEP(pvp)) {
2390 softdep_freefile(pvp, ino, mode);
2394 return (ffs_freefile(ip->i_ump, ip->i_fs, ip->i_devvp, ino, mode,
2399 * Do the actual free operation.
2400 * The specified inode is placed back in the free map.
2403 ffs_freefile(ump, fs, devvp, ino, mode, wkhd)
2404 struct ufsmount *ump;
2406 struct vnode *devvp;
2409 struct workhead *wkhd;
2419 cg = ino_to_cg(fs, ino);
2420 if (devvp->v_type == VREG) {
2421 /* devvp is a snapshot */
2422 dev = VTOI(devvp)->i_devvp->v_rdev;
2423 cgbno = fragstoblks(fs, cgtod(fs, cg));
2425 /* devvp is a normal disk device */
2426 dev = devvp->v_rdev;
2427 cgbno = fsbtodb(fs, cgtod(fs, cg));
2429 if (ino >= fs->fs_ipg * fs->fs_ncg)
2430 panic("ffs_freefile: range: dev = %s, ino = %lu, fs = %s",
2431 devtoname(dev), (u_long)ino, fs->fs_fsmnt);
2432 if ((error = bread(devvp, cgbno, (int)fs->fs_cgsize, NOCRED, &bp))) {
2436 cgp = (struct cg *)bp->b_data;
2437 if (!cg_chkmagic(cgp)) {
2441 bp->b_xflags |= BX_BKGRDWRITE;
2442 cgp->cg_old_time = cgp->cg_time = time_second;
2443 inosused = cg_inosused(cgp);
2445 if (isclr(inosused, ino)) {
2446 printf("dev = %s, ino = %u, fs = %s\n", devtoname(dev),
2447 ino + cg * fs->fs_ipg, fs->fs_fsmnt);
2448 if (fs->fs_ronly == 0)
2449 panic("ffs_freefile: freeing free inode");
2451 clrbit(inosused, ino);
2452 if (ino < cgp->cg_irotor)
2453 cgp->cg_irotor = ino;
2454 cgp->cg_cs.cs_nifree++;
2456 fs->fs_cstotal.cs_nifree++;
2457 fs->fs_cs(fs, cg).cs_nifree++;
2458 if ((mode & IFMT) == IFDIR) {
2459 cgp->cg_cs.cs_ndir--;
2460 fs->fs_cstotal.cs_ndir--;
2461 fs->fs_cs(fs, cg).cs_ndir--;
2464 ACTIVECLEAR(fs, cg);
2466 if (MOUNTEDSOFTDEP(UFSTOVFS(ump)) && devvp->v_type != VREG)
2467 softdep_setup_inofree(UFSTOVFS(ump), bp,
2468 ino + cg * fs->fs_ipg, wkhd);
2474 * Check to see if a file is free.
2477 ffs_checkfreefile(fs, devvp, ino)
2479 struct vnode *devvp;
2489 cg = ino_to_cg(fs, ino);
2490 if (devvp->v_type == VREG) {
2491 /* devvp is a snapshot */
2492 cgbno = fragstoblks(fs, cgtod(fs, cg));
2494 /* devvp is a normal disk device */
2495 cgbno = fsbtodb(fs, cgtod(fs, cg));
2497 if (ino >= fs->fs_ipg * fs->fs_ncg)
2499 if (bread(devvp, cgbno, (int)fs->fs_cgsize, NOCRED, &bp)) {
2503 cgp = (struct cg *)bp->b_data;
2504 if (!cg_chkmagic(cgp)) {
2508 inosused = cg_inosused(cgp);
2510 ret = isclr(inosused, ino);
2516 * Find a block of the specified size in the specified cylinder group.
2518 * It is a panic if a request is made to find a block if none are
2522 ffs_mapsearch(fs, cgp, bpref, allocsiz)
2529 int start, len, loc, i;
2530 int blk, field, subfield, pos;
2534 * find the fragment by searching through the free block
2535 * map for an appropriate bit pattern
2538 start = dtogd(fs, bpref) / NBBY;
2540 start = cgp->cg_frotor / NBBY;
2541 blksfree = cg_blksfree(cgp);
2542 len = howmany(fs->fs_fpg, NBBY) - start;
2543 loc = scanc((u_int)len, (u_char *)&blksfree[start],
2544 fragtbl[fs->fs_frag],
2545 (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
2549 loc = scanc((u_int)len, (u_char *)&blksfree[0],
2550 fragtbl[fs->fs_frag],
2551 (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
2553 printf("start = %d, len = %d, fs = %s\n",
2554 start, len, fs->fs_fsmnt);
2555 panic("ffs_alloccg: map corrupted");
2559 bno = (start + len - loc) * NBBY;
2560 cgp->cg_frotor = bno;
2562 * found the byte in the map
2563 * sift through the bits to find the selected frag
2565 for (i = bno + NBBY; bno < i; bno += fs->fs_frag) {
2566 blk = blkmap(fs, blksfree, bno);
2568 field = around[allocsiz];
2569 subfield = inside[allocsiz];
2570 for (pos = 0; pos <= fs->fs_frag - allocsiz; pos++) {
2571 if ((blk & field) == subfield)
2577 printf("bno = %lu, fs = %s\n", (u_long)bno, fs->fs_fsmnt);
2578 panic("ffs_alloccg: block not in map");
2583 * Fserr prints the name of a filesystem with an error diagnostic.
2585 * The form of the error message is:
2589 ffs_fserr(fs, inum, cp)
2594 struct thread *td = curthread; /* XXX */
2595 struct proc *p = td->td_proc;
2597 log(LOG_ERR, "pid %d (%s), uid %d inumber %d on %s: %s\n",
2598 p->p_pid, p->p_comm, td->td_ucred->cr_uid, inum, fs->fs_fsmnt, cp);
2602 * This function provides the capability for the fsck program to
2603 * update an active filesystem. Fourteen operations are provided:
2605 * adjrefcnt(inode, amt) - adjusts the reference count on the
2606 * specified inode by the specified amount. Under normal
2607 * operation the count should always go down. Decrementing
2608 * the count to zero will cause the inode to be freed.
2609 * adjblkcnt(inode, amt) - adjust the number of blocks used by the
2610 * inode by the specified amount.
2611 * adjndir, adjbfree, adjifree, adjffree, adjnumclusters(amt) -
2612 * adjust the superblock summary.
2613 * freedirs(inode, count) - directory inodes [inode..inode + count - 1]
2614 * are marked as free. Inodes should never have to be marked
2616 * freefiles(inode, count) - file inodes [inode..inode + count - 1]
2617 * are marked as free. Inodes should never have to be marked
2619 * freeblks(blockno, size) - blocks [blockno..blockno + size - 1]
2620 * are marked as free. Blocks should never have to be marked
2622 * setflags(flags, set/clear) - the fs_flags field has the specified
2623 * flags set (second parameter +1) or cleared (second parameter -1).
2624 * setcwd(dirinode) - set the current directory to dirinode in the
2625 * filesystem associated with the snapshot.
2626 * setdotdot(oldvalue, newvalue) - Verify that the inode number for ".."
2627 * in the current directory is oldvalue then change it to newvalue.
2628 * unlink(nameptr, oldvalue) - Verify that the inode number associated
2629 * with nameptr in the current directory is oldvalue then unlink it.
2631 * The following functions may only be used on a quiescent filesystem
2632 * by the soft updates journal. They are not safe to be run on an active
2635 * setinode(inode, dip) - the specified disk inode is replaced with the
2636 * contents pointed to by dip.
2637 * setbufoutput(fd, flags) - output associated with the specified file
2638 * descriptor (which must reference the character device supporting
2639 * the filesystem) switches from using physio to running through the
2640 * buffer cache when flags is set to 1. The descriptor reverts to
2641 * physio for output when flags is set to zero.
2644 static int sysctl_ffs_fsck(SYSCTL_HANDLER_ARGS);
2646 SYSCTL_PROC(_vfs_ffs, FFS_ADJ_REFCNT, adjrefcnt, CTLFLAG_WR|CTLTYPE_STRUCT,
2647 0, 0, sysctl_ffs_fsck, "S,fsck", "Adjust Inode Reference Count");
2649 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_BLKCNT, adjblkcnt, CTLFLAG_WR,
2650 sysctl_ffs_fsck, "Adjust Inode Used Blocks Count");
2652 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NDIR, adjndir, CTLFLAG_WR,
2653 sysctl_ffs_fsck, "Adjust number of directories");
2655 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NBFREE, adjnbfree, CTLFLAG_WR,
2656 sysctl_ffs_fsck, "Adjust number of free blocks");
2658 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NIFREE, adjnifree, CTLFLAG_WR,
2659 sysctl_ffs_fsck, "Adjust number of free inodes");
2661 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NFFREE, adjnffree, CTLFLAG_WR,
2662 sysctl_ffs_fsck, "Adjust number of free frags");
2664 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NUMCLUSTERS, adjnumclusters, CTLFLAG_WR,
2665 sysctl_ffs_fsck, "Adjust number of free clusters");
2667 static SYSCTL_NODE(_vfs_ffs, FFS_DIR_FREE, freedirs, CTLFLAG_WR,
2668 sysctl_ffs_fsck, "Free Range of Directory Inodes");
2670 static SYSCTL_NODE(_vfs_ffs, FFS_FILE_FREE, freefiles, CTLFLAG_WR,
2671 sysctl_ffs_fsck, "Free Range of File Inodes");
2673 static SYSCTL_NODE(_vfs_ffs, FFS_BLK_FREE, freeblks, CTLFLAG_WR,
2674 sysctl_ffs_fsck, "Free Range of Blocks");
2676 static SYSCTL_NODE(_vfs_ffs, FFS_SET_FLAGS, setflags, CTLFLAG_WR,
2677 sysctl_ffs_fsck, "Change Filesystem Flags");
2679 static SYSCTL_NODE(_vfs_ffs, FFS_SET_CWD, setcwd, CTLFLAG_WR,
2680 sysctl_ffs_fsck, "Set Current Working Directory");
2682 static SYSCTL_NODE(_vfs_ffs, FFS_SET_DOTDOT, setdotdot, CTLFLAG_WR,
2683 sysctl_ffs_fsck, "Change Value of .. Entry");
2685 static SYSCTL_NODE(_vfs_ffs, FFS_UNLINK, unlink, CTLFLAG_WR,
2686 sysctl_ffs_fsck, "Unlink a Duplicate Name");
2688 static SYSCTL_NODE(_vfs_ffs, FFS_SET_INODE, setinode, CTLFLAG_WR,
2689 sysctl_ffs_fsck, "Update an On-Disk Inode");
2691 static SYSCTL_NODE(_vfs_ffs, FFS_SET_BUFOUTPUT, setbufoutput, CTLFLAG_WR,
2692 sysctl_ffs_fsck, "Set Buffered Writing for Descriptor");
2696 static int fsckcmds = 0;
2697 SYSCTL_INT(_debug, OID_AUTO, fsckcmds, CTLFLAG_RW, &fsckcmds, 0, "");
2700 static int buffered_write(struct file *, struct uio *, struct ucred *,
2701 int, struct thread *);
2704 sysctl_ffs_fsck(SYSCTL_HANDLER_ARGS)
2706 struct thread *td = curthread;
2707 struct fsck_cmd cmd;
2708 struct ufsmount *ump;
2709 struct vnode *vp, *vpold, *dvp, *fdvp;
2710 struct inode *ip, *dp;
2714 long blkcnt, blksize;
2715 struct filedesc *fdp;
2716 struct file *fp, *vfp;
2717 int vfslocked, filetype, error;
2718 static struct fileops *origops, bufferedops;
2720 if (req->newlen > sizeof cmd)
2722 if ((error = SYSCTL_IN(req, &cmd, sizeof cmd)) != 0)
2724 if (cmd.version != FFS_CMD_VERSION)
2725 return (ERPCMISMATCH);
2726 if ((error = getvnode(td->td_proc->p_fd, cmd.handle, CAP_FSCK,
2730 if (vp->v_type != VREG && vp->v_type != VDIR) {
2734 vn_start_write(vp, &mp, V_WAIT);
2735 if (mp == 0 || strncmp(mp->mnt_stat.f_fstypename, "ufs", MFSNAMELEN)) {
2736 vn_finished_write(mp);
2741 if ((mp->mnt_flag & MNT_RDONLY) &&
2742 ump->um_fsckpid != td->td_proc->p_pid) {
2743 vn_finished_write(mp);
2750 switch (oidp->oid_number) {
2755 printf("%s: %s flags\n", mp->mnt_stat.f_mntonname,
2756 cmd.size > 0 ? "set" : "clear");
2759 fs->fs_flags |= (long)cmd.value;
2761 fs->fs_flags &= ~(long)cmd.value;
2764 case FFS_ADJ_REFCNT:
2767 printf("%s: adjust inode %jd link count by %jd\n",
2768 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value,
2769 (intmax_t)cmd.size);
2772 if ((error = ffs_vget(mp, (ino_t)cmd.value, LK_EXCLUSIVE, &vp)))
2775 ip->i_nlink += cmd.size;
2776 DIP_SET(ip, i_nlink, ip->i_nlink);
2777 ip->i_effnlink += cmd.size;
2778 ip->i_flag |= IN_CHANGE | IN_MODIFIED;
2779 error = ffs_update(vp, 1);
2780 if (DOINGSOFTDEP(vp))
2781 softdep_change_linkcnt(ip);
2785 case FFS_ADJ_BLKCNT:
2788 printf("%s: adjust inode %jd block count by %jd\n",
2789 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value,
2790 (intmax_t)cmd.size);
2793 if ((error = ffs_vget(mp, (ino_t)cmd.value, LK_EXCLUSIVE, &vp)))
2796 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + cmd.size);
2797 ip->i_flag |= IN_CHANGE | IN_MODIFIED;
2798 error = ffs_update(vp, 1);
2810 printf("%s: free %s inode %d\n",
2811 mp->mnt_stat.f_mntonname,
2812 filetype == IFDIR ? "directory" : "file",
2815 printf("%s: free %s inodes %d-%d\n",
2816 mp->mnt_stat.f_mntonname,
2817 filetype == IFDIR ? "directory" : "file",
2819 (ino_t)(cmd.value + cmd.size - 1));
2822 while (cmd.size > 0) {
2823 if ((error = ffs_freefile(ump, fs, ump->um_devvp,
2824 cmd.value, filetype, NULL)))
2835 printf("%s: free block %jd\n",
2836 mp->mnt_stat.f_mntonname,
2837 (intmax_t)cmd.value);
2839 printf("%s: free blocks %jd-%jd\n",
2840 mp->mnt_stat.f_mntonname,
2841 (intmax_t)cmd.value,
2842 (intmax_t)cmd.value + cmd.size - 1);
2847 blksize = fs->fs_frag - (blkno % fs->fs_frag);
2848 while (blkcnt > 0) {
2849 if (blksize > blkcnt)
2851 ffs_blkfree(ump, fs, ump->um_devvp, blkno,
2852 blksize * fs->fs_fsize, ROOTINO, VDIR, NULL);
2855 blksize = fs->fs_frag;
2860 * Adjust superblock summaries. fsck(8) is expected to
2861 * submit deltas when necessary.
2866 printf("%s: adjust number of directories by %jd\n",
2867 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value);
2870 fs->fs_cstotal.cs_ndir += cmd.value;
2873 case FFS_ADJ_NBFREE:
2876 printf("%s: adjust number of free blocks by %+jd\n",
2877 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value);
2880 fs->fs_cstotal.cs_nbfree += cmd.value;
2883 case FFS_ADJ_NIFREE:
2886 printf("%s: adjust number of free inodes by %+jd\n",
2887 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value);
2890 fs->fs_cstotal.cs_nifree += cmd.value;
2893 case FFS_ADJ_NFFREE:
2896 printf("%s: adjust number of free frags by %+jd\n",
2897 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value);
2900 fs->fs_cstotal.cs_nffree += cmd.value;
2903 case FFS_ADJ_NUMCLUSTERS:
2906 printf("%s: adjust number of free clusters by %+jd\n",
2907 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value);
2910 fs->fs_cstotal.cs_numclusters += cmd.value;
2916 printf("%s: set current directory to inode %jd\n",
2917 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value);
2920 if ((error = ffs_vget(mp, (ino_t)cmd.value, LK_SHARED, &vp)))
2922 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
2923 AUDIT_ARG_VNODE1(vp);
2924 if ((error = change_dir(vp, td)) != 0) {
2926 VFS_UNLOCK_GIANT(vfslocked);
2930 VFS_UNLOCK_GIANT(vfslocked);
2931 fdp = td->td_proc->p_fd;
2932 FILEDESC_XLOCK(fdp);
2933 vpold = fdp->fd_cdir;
2935 FILEDESC_XUNLOCK(fdp);
2936 vfslocked = VFS_LOCK_GIANT(vpold->v_mount);
2938 VFS_UNLOCK_GIANT(vfslocked);
2941 case FFS_SET_DOTDOT:
2944 printf("%s: change .. in cwd from %jd to %jd\n",
2945 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value,
2946 (intmax_t)cmd.size);
2950 * First we have to get and lock the parent directory
2951 * to which ".." points.
2953 error = ffs_vget(mp, (ino_t)cmd.value, LK_EXCLUSIVE, &fdvp);
2957 * Now we get and lock the child directory containing "..".
2959 FILEDESC_SLOCK(td->td_proc->p_fd);
2960 dvp = td->td_proc->p_fd->fd_cdir;
2961 FILEDESC_SUNLOCK(td->td_proc->p_fd);
2962 if ((error = vget(dvp, LK_EXCLUSIVE, td)) != 0) {
2967 dp->i_offset = 12; /* XXX mastertemplate.dot_reclen */
2968 error = ufs_dirrewrite(dp, VTOI(fdvp), (ino_t)cmd.size,
2981 if (copyinstr((char *)(intptr_t)cmd.value, buf,32,NULL))
2982 strncpy(buf, "Name_too_long", 32);
2983 printf("%s: unlink %s (inode %jd)\n",
2984 mp->mnt_stat.f_mntonname, buf, (intmax_t)cmd.size);
2988 * kern_unlinkat will do its own start/finish writes and
2989 * they do not nest, so drop ours here. Setting mp == NULL
2990 * indicates that vn_finished_write is not needed down below.
2992 vn_finished_write(mp);
2994 error = kern_unlinkat(td, AT_FDCWD, (char *)(intptr_t)cmd.value,
2995 UIO_USERSPACE, (ino_t)cmd.size);
2999 if (ump->um_fsckpid != td->td_proc->p_pid) {
3005 printf("%s: update inode %jd\n",
3006 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value);
3009 if ((error = ffs_vget(mp, (ino_t)cmd.value, LK_EXCLUSIVE, &vp)))
3011 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
3012 AUDIT_ARG_VNODE1(vp);
3014 if (ip->i_ump->um_fstype == UFS1)
3015 error = copyin((void *)(intptr_t)cmd.size, ip->i_din1,
3016 sizeof(struct ufs1_dinode));
3018 error = copyin((void *)(intptr_t)cmd.size, ip->i_din2,
3019 sizeof(struct ufs2_dinode));
3022 VFS_UNLOCK_GIANT(vfslocked);
3025 ip->i_flag |= IN_CHANGE | IN_MODIFIED;
3026 error = ffs_update(vp, 1);
3028 VFS_UNLOCK_GIANT(vfslocked);
3031 case FFS_SET_BUFOUTPUT:
3032 if (ump->um_fsckpid != td->td_proc->p_pid) {
3036 if (VTOI(vp)->i_ump != ump) {
3042 printf("%s: %s buffered output for descriptor %jd\n",
3043 mp->mnt_stat.f_mntonname,
3044 cmd.size == 1 ? "enable" : "disable",
3045 (intmax_t)cmd.value);
3048 if ((error = getvnode(td->td_proc->p_fd, cmd.value,
3049 CAP_FSCK, &vfp)) != 0)
3051 if (vfp->f_vnode->v_type != VCHR) {
3056 if (origops == NULL) {
3057 origops = vfp->f_ops;
3058 bcopy((void *)origops, (void *)&bufferedops,
3059 sizeof(bufferedops));
3060 bufferedops.fo_write = buffered_write;
3063 atomic_store_rel_ptr((volatile uintptr_t *)&vfp->f_ops,
3064 (uintptr_t)&bufferedops);
3066 atomic_store_rel_ptr((volatile uintptr_t *)&vfp->f_ops,
3067 (uintptr_t)origops);
3074 printf("Invalid request %d from fsck\n",
3083 vn_finished_write(mp);
3088 * Function to switch a descriptor to use the buffer cache to stage
3089 * its I/O. This is needed so that writes to the filesystem device
3090 * will give snapshots a chance to copy modified blocks for which it
3091 * needs to retain copies.
3094 buffered_write(fp, uio, active_cred, flags, td)
3097 struct ucred *active_cred;
3101 struct vnode *devvp, *vp;
3105 struct filedesc *fdp;
3106 int error, vfslocked;
3110 * The devvp is associated with the /dev filesystem. To discover
3111 * the filesystem with which the device is associated, we depend
3112 * on the application setting the current directory to a location
3113 * within the filesystem being written. Yes, this is an ugly hack.
3115 devvp = fp->f_vnode;
3116 if (!vn_isdisk(devvp, NULL))
3118 fdp = td->td_proc->p_fd;
3119 FILEDESC_SLOCK(fdp);
3122 FILEDESC_SUNLOCK(fdp);
3123 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
3124 vn_lock(vp, LK_SHARED | LK_RETRY);
3126 * Check that the current directory vnode indeed belongs to
3127 * UFS before trying to dereference UFS-specific v_data fields.
3129 if (vp->v_op != &ffs_vnodeops1 && vp->v_op != &ffs_vnodeops2) {
3131 VFS_UNLOCK_GIANT(vfslocked);
3135 if (ip->i_devvp != devvp) {
3137 VFS_UNLOCK_GIANT(vfslocked);
3142 VFS_UNLOCK_GIANT(vfslocked);
3143 foffset_lock_uio(fp, uio, flags);
3144 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
3147 printf("%s: buffered write for block %jd\n",
3148 fs->fs_fsmnt, (intmax_t)btodb(uio->uio_offset));
3152 * All I/O must be contained within a filesystem block, start on
3153 * a fragment boundary, and be a multiple of fragments in length.
3155 if (uio->uio_resid > fs->fs_bsize - (uio->uio_offset % fs->fs_bsize) ||
3156 fragoff(fs, uio->uio_offset) != 0 ||
3157 fragoff(fs, uio->uio_resid) != 0) {
3161 lbn = numfrags(fs, uio->uio_offset);
3162 bp = getblk(devvp, lbn, uio->uio_resid, 0, 0, 0);
3163 bp->b_flags |= B_RELBUF;
3164 if ((error = uiomove((char *)bp->b_data, uio->uio_resid, uio)) != 0) {
3170 VOP_UNLOCK(devvp, 0);
3171 foffset_unlock_uio(fp, uio, flags | FOF_NEXTOFF);