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/capsicum.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);
116 static ino_t ffs_dirpref(struct inode *);
117 static ufs2_daddr_t ffs_fragextend(struct inode *, u_int, ufs2_daddr_t,
119 static ufs2_daddr_t ffs_hashalloc
120 (struct inode *, u_int, ufs2_daddr_t, int, int, allocfcn_t *);
121 static ufs2_daddr_t ffs_nodealloccg(struct inode *, u_int, ufs2_daddr_t, int,
123 static ufs1_daddr_t ffs_mapsearch(struct fs *, struct cg *, ufs2_daddr_t, int);
124 static int ffs_reallocblks_ufs1(struct vop_reallocblks_args *);
125 static int ffs_reallocblks_ufs2(struct vop_reallocblks_args *);
128 * Allocate a block in the filesystem.
130 * The size of the requested block is given, which must be some
131 * multiple of fs_fsize and <= fs_bsize.
132 * A preference may be optionally specified. If a preference is given
133 * the following hierarchy is used to allocate a block:
134 * 1) allocate the requested block.
135 * 2) allocate a rotationally optimal block in the same cylinder.
136 * 3) allocate a block in the same cylinder group.
137 * 4) quadradically rehash into other cylinder groups, until an
138 * available block is located.
139 * If no block preference is given the following hierarchy is used
140 * to allocate a block:
141 * 1) allocate a block in the cylinder group that contains the
142 * inode for the file.
143 * 2) quadradically rehash into other cylinder groups, until an
144 * available block is located.
147 ffs_alloc(ip, lbn, bpref, size, flags, cred, bnp)
149 ufs2_daddr_t lbn, bpref;
155 struct ufsmount *ump;
158 static struct timeval lastfail;
168 mtx_assert(UFS_MTX(ump), MA_OWNED);
170 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) {
171 printf("dev = %s, bsize = %ld, size = %d, fs = %s\n",
172 devtoname(ip->i_dev), (long)fs->fs_bsize, size,
174 panic("ffs_alloc: bad size");
177 panic("ffs_alloc: missing credential");
178 #endif /* INVARIANTS */
183 error = chkdq(ip, btodb(size), cred, 0);
188 if (size == fs->fs_bsize && fs->fs_cstotal.cs_nbfree == 0)
190 if (priv_check_cred(cred, PRIV_VFS_BLOCKRESERVE, 0) &&
191 freespace(fs, fs->fs_minfree) - numfrags(fs, size) < 0)
193 if (bpref >= fs->fs_size)
196 cg = ino_to_cg(fs, ip->i_number);
198 cg = dtog(fs, bpref);
199 bno = ffs_hashalloc(ip, cg, bpref, size, size, ffs_alloccg);
202 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + delta);
204 ip->i_flag |= IN_CHANGE;
206 ip->i_flag |= IN_CHANGE | IN_UPDATE;
214 * Restore user's disk quota because allocation failed.
216 (void) chkdq(ip, -btodb(size), cred, FORCE);
219 if (reclaimed == 0 && (flags & IO_BUFLOCKED) == 0) {
221 softdep_request_cleanup(fs, ITOV(ip), cred, FLUSH_BLOCKS_WAIT);
225 if (reclaimed > 0 && ppsratecheck(&lastfail, &curfail, 1)) {
226 ffs_fserr(fs, ip->i_number, "filesystem full");
227 uprintf("\n%s: write failed, filesystem is full\n",
234 * Reallocate a fragment to a bigger size
236 * The number and size of the old block is given, and a preference
237 * and new size is also specified. The allocator attempts to extend
238 * the original block. Failing that, the regular block allocator is
239 * invoked to get an appropriate block.
242 ffs_realloccg(ip, lbprev, bprev, bpref, osize, nsize, flags, cred, bpp)
247 int osize, nsize, flags;
254 struct ufsmount *ump;
255 u_int cg, request, reclaimed;
258 static struct timeval lastfail;
267 gbflags = (flags & BA_UNMAPPED) != 0 ? GB_UNMAPPED : 0;
269 mtx_assert(UFS_MTX(ump), MA_OWNED);
271 if (vp->v_mount->mnt_kern_flag & MNTK_SUSPENDED)
272 panic("ffs_realloccg: allocation on suspended filesystem");
273 if ((u_int)osize > fs->fs_bsize || fragoff(fs, osize) != 0 ||
274 (u_int)nsize > fs->fs_bsize || fragoff(fs, nsize) != 0) {
276 "dev = %s, bsize = %ld, osize = %d, nsize = %d, fs = %s\n",
277 devtoname(ip->i_dev), (long)fs->fs_bsize, osize,
278 nsize, fs->fs_fsmnt);
279 panic("ffs_realloccg: bad size");
282 panic("ffs_realloccg: missing credential");
283 #endif /* INVARIANTS */
286 if (priv_check_cred(cred, PRIV_VFS_BLOCKRESERVE, 0) &&
287 freespace(fs, fs->fs_minfree) - numfrags(fs, nsize - osize) < 0) {
291 printf("dev = %s, bsize = %ld, bprev = %jd, fs = %s\n",
292 devtoname(ip->i_dev), (long)fs->fs_bsize, (intmax_t)bprev,
294 panic("ffs_realloccg: bad bprev");
298 * Allocate the extra space in the buffer.
300 error = bread_gb(vp, lbprev, osize, NOCRED, gbflags, &bp);
306 if (bp->b_blkno == bp->b_lblkno) {
307 if (lbprev >= NDADDR)
308 panic("ffs_realloccg: lbprev out of range");
309 bp->b_blkno = fsbtodb(fs, bprev);
313 error = chkdq(ip, btodb(nsize - osize), cred, 0);
320 * Check for extension in the existing location.
322 cg = dtog(fs, bprev);
324 bno = ffs_fragextend(ip, cg, bprev, osize, nsize);
326 if (bp->b_blkno != fsbtodb(fs, bno))
327 panic("ffs_realloccg: bad blockno");
328 delta = btodb(nsize - osize);
329 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + delta);
331 ip->i_flag |= IN_CHANGE;
333 ip->i_flag |= IN_CHANGE | IN_UPDATE;
335 bp->b_flags |= B_DONE;
336 vfs_bio_bzero_buf(bp, osize, nsize - osize);
337 if ((bp->b_flags & (B_MALLOC | B_VMIO)) == B_VMIO)
338 vfs_bio_set_valid(bp, osize, nsize - osize);
343 * Allocate a new disk location.
345 if (bpref >= fs->fs_size)
347 switch ((int)fs->fs_optim) {
350 * Allocate an exact sized fragment. Although this makes
351 * best use of space, we will waste time relocating it if
352 * the file continues to grow. If the fragmentation is
353 * less than half of the minimum free reserve, we choose
354 * to begin optimizing for time.
357 if (fs->fs_minfree <= 5 ||
358 fs->fs_cstotal.cs_nffree >
359 (off_t)fs->fs_dsize * fs->fs_minfree / (2 * 100))
361 log(LOG_NOTICE, "%s: optimization changed from SPACE to TIME\n",
363 fs->fs_optim = FS_OPTTIME;
367 * At this point we have discovered a file that is trying to
368 * grow a small fragment to a larger fragment. To save time,
369 * we allocate a full sized block, then free the unused portion.
370 * If the file continues to grow, the `ffs_fragextend' call
371 * above will be able to grow it in place without further
372 * copying. If aberrant programs cause disk fragmentation to
373 * grow within 2% of the free reserve, we choose to begin
374 * optimizing for space.
376 request = fs->fs_bsize;
377 if (fs->fs_cstotal.cs_nffree <
378 (off_t)fs->fs_dsize * (fs->fs_minfree - 2) / 100)
380 log(LOG_NOTICE, "%s: optimization changed from TIME to SPACE\n",
382 fs->fs_optim = FS_OPTSPACE;
385 printf("dev = %s, optim = %ld, fs = %s\n",
386 devtoname(ip->i_dev), (long)fs->fs_optim, fs->fs_fsmnt);
387 panic("ffs_realloccg: bad optim");
390 bno = ffs_hashalloc(ip, cg, bpref, request, nsize, ffs_alloccg);
392 bp->b_blkno = fsbtodb(fs, bno);
393 if (!DOINGSOFTDEP(vp))
394 ffs_blkfree(ump, fs, ip->i_devvp, bprev, (long)osize,
395 ip->i_number, vp->v_type, NULL);
396 delta = btodb(nsize - osize);
397 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + delta);
399 ip->i_flag |= IN_CHANGE;
401 ip->i_flag |= IN_CHANGE | IN_UPDATE;
403 bp->b_flags |= B_DONE;
404 vfs_bio_bzero_buf(bp, osize, nsize - osize);
405 if ((bp->b_flags & (B_MALLOC | B_VMIO)) == B_VMIO)
406 vfs_bio_set_valid(bp, osize, nsize - osize);
413 * Restore user's disk quota because allocation failed.
415 (void) chkdq(ip, -btodb(nsize - osize), cred, FORCE);
422 if (reclaimed == 0 && (flags & IO_BUFLOCKED) == 0) {
430 softdep_request_cleanup(fs, vp, cred, FLUSH_BLOCKS_WAIT);
436 if (reclaimed > 0 && ppsratecheck(&lastfail, &curfail, 1)) {
437 ffs_fserr(fs, ip->i_number, "filesystem full");
438 uprintf("\n%s: write failed, filesystem is full\n",
445 * Reallocate a sequence of blocks into a contiguous sequence of blocks.
447 * The vnode and an array of buffer pointers for a range of sequential
448 * logical blocks to be made contiguous is given. The allocator attempts
449 * to find a range of sequential blocks starting as close as possible
450 * from the end of the allocation for the logical block immediately
451 * preceding the current range. If successful, the physical block numbers
452 * in the buffer pointers and in the inode are changed to reflect the new
453 * allocation. If unsuccessful, the allocation is left unchanged. The
454 * success in doing the reallocation is returned. Note that the error
455 * return is not reflected back to the user. Rather the previous block
456 * allocation will be used.
459 SYSCTL_NODE(_vfs, OID_AUTO, ffs, CTLFLAG_RW, 0, "FFS filesystem");
461 static int doasyncfree = 1;
462 SYSCTL_INT(_vfs_ffs, OID_AUTO, doasyncfree, CTLFLAG_RW, &doasyncfree, 0,
463 "do not force synchronous writes when blocks are reallocated");
465 static int doreallocblks = 1;
466 SYSCTL_INT(_vfs_ffs, OID_AUTO, doreallocblks, CTLFLAG_RW, &doreallocblks, 0,
467 "enable block reallocation");
469 static int maxclustersearch = 10;
470 SYSCTL_INT(_vfs_ffs, OID_AUTO, maxclustersearch, CTLFLAG_RW, &maxclustersearch,
471 0, "max number of cylinder group to search for contigous blocks");
474 static volatile int prtrealloc = 0;
479 struct vop_reallocblks_args /* {
481 struct cluster_save *a_buflist;
485 if (doreallocblks == 0)
488 * We can't wait in softdep prealloc as it may fsync and recurse
489 * here. Instead we simply fail to reallocate blocks if this
490 * rare condition arises.
492 if (DOINGSOFTDEP(ap->a_vp))
493 if (softdep_prealloc(ap->a_vp, MNT_NOWAIT) != 0)
495 if (VTOI(ap->a_vp)->i_ump->um_fstype == UFS1)
496 return (ffs_reallocblks_ufs1(ap));
497 return (ffs_reallocblks_ufs2(ap));
501 ffs_reallocblks_ufs1(ap)
502 struct vop_reallocblks_args /* {
504 struct cluster_save *a_buflist;
510 struct buf *sbp, *ebp;
511 ufs1_daddr_t *bap, *sbap, *ebap = 0;
512 struct cluster_save *buflist;
513 struct ufsmount *ump;
514 ufs_lbn_t start_lbn, end_lbn;
515 ufs1_daddr_t soff, newblk, blkno;
517 struct indir start_ap[NIADDR + 1], end_ap[NIADDR + 1], *idp;
518 int i, cg, len, start_lvl, end_lvl, ssize;
525 * If we are not tracking block clusters or if we have less than 4%
526 * free blocks left, then do not attempt to cluster. Running with
527 * less than 5% free block reserve is not recommended and those that
528 * choose to do so do not expect to have good file layout.
530 if (fs->fs_contigsumsize <= 0 || freespace(fs, 4) < 0)
532 buflist = ap->a_buflist;
533 len = buflist->bs_nchildren;
534 start_lbn = buflist->bs_children[0]->b_lblkno;
535 end_lbn = start_lbn + len - 1;
537 for (i = 0; i < len; i++)
538 if (!ffs_checkblk(ip,
539 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
540 panic("ffs_reallocblks: unallocated block 1");
541 for (i = 1; i < len; i++)
542 if (buflist->bs_children[i]->b_lblkno != start_lbn + i)
543 panic("ffs_reallocblks: non-logical cluster");
544 blkno = buflist->bs_children[0]->b_blkno;
545 ssize = fsbtodb(fs, fs->fs_frag);
546 for (i = 1; i < len - 1; i++)
547 if (buflist->bs_children[i]->b_blkno != blkno + (i * ssize))
548 panic("ffs_reallocblks: non-physical cluster %d", i);
551 * If the cluster crosses the boundary for the first indirect
552 * block, leave space for the indirect block. Indirect blocks
553 * are initially laid out in a position after the last direct
554 * block. Block reallocation would usually destroy locality by
555 * moving the indirect block out of the way to make room for
556 * data blocks if we didn't compensate here. We should also do
557 * this for other indirect block boundaries, but it is only
558 * important for the first one.
560 if (start_lbn < NDADDR && end_lbn >= NDADDR)
563 * If the latest allocation is in a new cylinder group, assume that
564 * the filesystem has decided to move and do not force it back to
565 * the previous cylinder group.
567 if (dtog(fs, dbtofsb(fs, buflist->bs_children[0]->b_blkno)) !=
568 dtog(fs, dbtofsb(fs, buflist->bs_children[len - 1]->b_blkno)))
570 if (ufs_getlbns(vp, start_lbn, start_ap, &start_lvl) ||
571 ufs_getlbns(vp, end_lbn, end_ap, &end_lvl))
574 * Get the starting offset and block map for the first block.
576 if (start_lvl == 0) {
577 sbap = &ip->i_din1->di_db[0];
580 idp = &start_ap[start_lvl - 1];
581 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &sbp)) {
585 sbap = (ufs1_daddr_t *)sbp->b_data;
589 * If the block range spans two block maps, get the second map.
591 if (end_lvl == 0 || (idp = &end_ap[end_lvl - 1])->in_off + 1 >= len) {
596 start_ap[start_lvl - 1].in_lbn == idp->in_lbn)
597 panic("ffs_reallocblk: start == end");
599 ssize = len - (idp->in_off + 1);
600 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &ebp))
602 ebap = (ufs1_daddr_t *)ebp->b_data;
605 * Find the preferred location for the cluster. If we have not
606 * previously failed at this endeavor, then follow our standard
607 * preference calculation. If we have failed at it, then pick up
608 * where we last ended our search.
611 if (ip->i_nextclustercg == -1)
612 pref = ffs_blkpref_ufs1(ip, start_lbn, soff, sbap);
614 pref = cgdata(fs, ip->i_nextclustercg);
616 * Search the block map looking for an allocation of the desired size.
617 * To avoid wasting too much time, we limit the number of cylinder
618 * groups that we will search.
621 for (i = min(maxclustersearch, fs->fs_ncg); i > 0; i--) {
622 if ((newblk = ffs_clusteralloc(ip, cg, pref, len)) != 0)
625 if (cg >= fs->fs_ncg)
629 * If we have failed in our search, record where we gave up for
630 * next time. Otherwise, fall back to our usual search citerion.
633 ip->i_nextclustercg = cg;
637 ip->i_nextclustercg = -1;
639 * We have found a new contiguous block.
641 * First we have to replace the old block pointers with the new
642 * block pointers in the inode and indirect blocks associated
647 printf("realloc: ino %ju, lbns %jd-%jd\n\told:",
648 (uintmax_t)ip->i_number,
649 (intmax_t)start_lbn, (intmax_t)end_lbn);
652 for (bap = &sbap[soff], i = 0; i < len; i++, blkno += fs->fs_frag) {
658 if (!ffs_checkblk(ip,
659 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
660 panic("ffs_reallocblks: unallocated block 2");
661 if (dbtofsb(fs, buflist->bs_children[i]->b_blkno) != *bap)
662 panic("ffs_reallocblks: alloc mismatch");
666 printf(" %d,", *bap);
668 if (DOINGSOFTDEP(vp)) {
669 if (sbap == &ip->i_din1->di_db[0] && i < ssize)
670 softdep_setup_allocdirect(ip, start_lbn + i,
671 blkno, *bap, fs->fs_bsize, fs->fs_bsize,
672 buflist->bs_children[i]);
674 softdep_setup_allocindir_page(ip, start_lbn + i,
675 i < ssize ? sbp : ebp, soff + i, blkno,
676 *bap, buflist->bs_children[i]);
681 * Next we must write out the modified inode and indirect blocks.
682 * For strict correctness, the writes should be synchronous since
683 * the old block values may have been written to disk. In practise
684 * they are almost never written, but if we are concerned about
685 * strict correctness, the `doasyncfree' flag should be set to zero.
687 * The test on `doasyncfree' should be changed to test a flag
688 * that shows whether the associated buffers and inodes have
689 * been written. The flag should be set when the cluster is
690 * started and cleared whenever the buffer or inode is flushed.
691 * We can then check below to see if it is set, and do the
692 * synchronous write only when it has been cleared.
694 if (sbap != &ip->i_din1->di_db[0]) {
700 ip->i_flag |= IN_CHANGE | IN_UPDATE;
711 * Last, free the old blocks and assign the new blocks to the buffers.
717 for (blkno = newblk, i = 0; i < len; i++, blkno += fs->fs_frag) {
718 if (!DOINGSOFTDEP(vp))
719 ffs_blkfree(ump, fs, ip->i_devvp,
720 dbtofsb(fs, buflist->bs_children[i]->b_blkno),
721 fs->fs_bsize, ip->i_number, vp->v_type, NULL);
722 buflist->bs_children[i]->b_blkno = fsbtodb(fs, blkno);
724 if (!ffs_checkblk(ip,
725 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
726 panic("ffs_reallocblks: unallocated block 3");
730 printf(" %d,", blkno);
744 if (sbap != &ip->i_din1->di_db[0])
750 ffs_reallocblks_ufs2(ap)
751 struct vop_reallocblks_args /* {
753 struct cluster_save *a_buflist;
759 struct buf *sbp, *ebp;
760 ufs2_daddr_t *bap, *sbap, *ebap = 0;
761 struct cluster_save *buflist;
762 struct ufsmount *ump;
763 ufs_lbn_t start_lbn, end_lbn;
764 ufs2_daddr_t soff, newblk, blkno, pref;
765 struct indir start_ap[NIADDR + 1], end_ap[NIADDR + 1], *idp;
766 int i, cg, len, start_lvl, end_lvl, ssize;
773 * If we are not tracking block clusters or if we have less than 4%
774 * free blocks left, then do not attempt to cluster. Running with
775 * less than 5% free block reserve is not recommended and those that
776 * choose to do so do not expect to have good file layout.
778 if (fs->fs_contigsumsize <= 0 || freespace(fs, 4) < 0)
780 buflist = ap->a_buflist;
781 len = buflist->bs_nchildren;
782 start_lbn = buflist->bs_children[0]->b_lblkno;
783 end_lbn = start_lbn + len - 1;
785 for (i = 0; i < len; i++)
786 if (!ffs_checkblk(ip,
787 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
788 panic("ffs_reallocblks: unallocated block 1");
789 for (i = 1; i < len; i++)
790 if (buflist->bs_children[i]->b_lblkno != start_lbn + i)
791 panic("ffs_reallocblks: non-logical cluster");
792 blkno = buflist->bs_children[0]->b_blkno;
793 ssize = fsbtodb(fs, fs->fs_frag);
794 for (i = 1; i < len - 1; i++)
795 if (buflist->bs_children[i]->b_blkno != blkno + (i * ssize))
796 panic("ffs_reallocblks: non-physical cluster %d", i);
799 * If the cluster crosses the boundary for the first indirect
800 * block, do not move anything in it. Indirect blocks are
801 * usually initially laid out in a position between the data
802 * blocks. Block reallocation would usually destroy locality by
803 * moving the indirect block out of the way to make room for
804 * data blocks if we didn't compensate here. We should also do
805 * this for other indirect block boundaries, but it is only
806 * important for the first one.
808 if (start_lbn < NDADDR && end_lbn >= NDADDR)
811 * If the latest allocation is in a new cylinder group, assume that
812 * the filesystem has decided to move and do not force it back to
813 * the previous cylinder group.
815 if (dtog(fs, dbtofsb(fs, buflist->bs_children[0]->b_blkno)) !=
816 dtog(fs, dbtofsb(fs, buflist->bs_children[len - 1]->b_blkno)))
818 if (ufs_getlbns(vp, start_lbn, start_ap, &start_lvl) ||
819 ufs_getlbns(vp, end_lbn, end_ap, &end_lvl))
822 * Get the starting offset and block map for the first block.
824 if (start_lvl == 0) {
825 sbap = &ip->i_din2->di_db[0];
828 idp = &start_ap[start_lvl - 1];
829 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &sbp)) {
833 sbap = (ufs2_daddr_t *)sbp->b_data;
837 * If the block range spans two block maps, get the second map.
839 if (end_lvl == 0 || (idp = &end_ap[end_lvl - 1])->in_off + 1 >= len) {
844 start_ap[start_lvl - 1].in_lbn == idp->in_lbn)
845 panic("ffs_reallocblk: start == end");
847 ssize = len - (idp->in_off + 1);
848 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &ebp))
850 ebap = (ufs2_daddr_t *)ebp->b_data;
853 * Find the preferred location for the cluster. If we have not
854 * previously failed at this endeavor, then follow our standard
855 * preference calculation. If we have failed at it, then pick up
856 * where we last ended our search.
859 if (ip->i_nextclustercg == -1)
860 pref = ffs_blkpref_ufs2(ip, start_lbn, soff, sbap);
862 pref = cgdata(fs, ip->i_nextclustercg);
864 * Search the block map looking for an allocation of the desired size.
865 * To avoid wasting too much time, we limit the number of cylinder
866 * groups that we will search.
869 for (i = min(maxclustersearch, fs->fs_ncg); i > 0; i--) {
870 if ((newblk = ffs_clusteralloc(ip, cg, pref, len)) != 0)
873 if (cg >= fs->fs_ncg)
877 * If we have failed in our search, record where we gave up for
878 * next time. Otherwise, fall back to our usual search citerion.
881 ip->i_nextclustercg = cg;
885 ip->i_nextclustercg = -1;
887 * We have found a new contiguous block.
889 * First we have to replace the old block pointers with the new
890 * block pointers in the inode and indirect blocks associated
895 printf("realloc: ino %d, lbns %jd-%jd\n\told:", ip->i_number,
896 (intmax_t)start_lbn, (intmax_t)end_lbn);
899 for (bap = &sbap[soff], i = 0; i < len; i++, blkno += fs->fs_frag) {
905 if (!ffs_checkblk(ip,
906 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
907 panic("ffs_reallocblks: unallocated block 2");
908 if (dbtofsb(fs, buflist->bs_children[i]->b_blkno) != *bap)
909 panic("ffs_reallocblks: alloc mismatch");
913 printf(" %jd,", (intmax_t)*bap);
915 if (DOINGSOFTDEP(vp)) {
916 if (sbap == &ip->i_din2->di_db[0] && i < ssize)
917 softdep_setup_allocdirect(ip, start_lbn + i,
918 blkno, *bap, fs->fs_bsize, fs->fs_bsize,
919 buflist->bs_children[i]);
921 softdep_setup_allocindir_page(ip, start_lbn + i,
922 i < ssize ? sbp : ebp, soff + i, blkno,
923 *bap, buflist->bs_children[i]);
928 * Next we must write out the modified inode and indirect blocks.
929 * For strict correctness, the writes should be synchronous since
930 * the old block values may have been written to disk. In practise
931 * they are almost never written, but if we are concerned about
932 * strict correctness, the `doasyncfree' flag should be set to zero.
934 * The test on `doasyncfree' should be changed to test a flag
935 * that shows whether the associated buffers and inodes have
936 * been written. The flag should be set when the cluster is
937 * started and cleared whenever the buffer or inode is flushed.
938 * We can then check below to see if it is set, and do the
939 * synchronous write only when it has been cleared.
941 if (sbap != &ip->i_din2->di_db[0]) {
947 ip->i_flag |= IN_CHANGE | IN_UPDATE;
958 * Last, free the old blocks and assign the new blocks to the buffers.
964 for (blkno = newblk, i = 0; i < len; i++, blkno += fs->fs_frag) {
965 if (!DOINGSOFTDEP(vp))
966 ffs_blkfree(ump, fs, ip->i_devvp,
967 dbtofsb(fs, buflist->bs_children[i]->b_blkno),
968 fs->fs_bsize, ip->i_number, vp->v_type, NULL);
969 buflist->bs_children[i]->b_blkno = fsbtodb(fs, blkno);
971 if (!ffs_checkblk(ip,
972 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
973 panic("ffs_reallocblks: unallocated block 3");
977 printf(" %jd,", (intmax_t)blkno);
991 if (sbap != &ip->i_din2->di_db[0])
997 * Allocate an inode in the filesystem.
999 * If allocating a directory, use ffs_dirpref to select the inode.
1000 * If allocating in a directory, the following hierarchy is followed:
1001 * 1) allocate the preferred inode.
1002 * 2) allocate an inode in the same cylinder group.
1003 * 3) quadradically rehash into other cylinder groups, until an
1004 * available inode is located.
1005 * If no inode preference is given the following hierarchy is used
1006 * to allocate an inode:
1007 * 1) allocate an inode in cylinder group 0.
1008 * 2) quadradically rehash into other cylinder groups, until an
1009 * available inode is located.
1012 ffs_valloc(pvp, mode, cred, vpp)
1022 struct ufsmount *ump;
1025 int error, error1, reclaimed;
1026 static struct timeval lastfail;
1037 if (fs->fs_cstotal.cs_nifree == 0)
1040 if ((mode & IFMT) == IFDIR)
1041 ipref = ffs_dirpref(pip);
1043 ipref = pip->i_number;
1044 if (ipref >= fs->fs_ncg * fs->fs_ipg)
1046 cg = ino_to_cg(fs, ipref);
1048 * Track number of dirs created one after another
1049 * in a same cg without intervening by files.
1051 if ((mode & IFMT) == IFDIR) {
1052 if (fs->fs_contigdirs[cg] < 255)
1053 fs->fs_contigdirs[cg]++;
1055 if (fs->fs_contigdirs[cg] > 0)
1056 fs->fs_contigdirs[cg]--;
1058 ino = (ino_t)ffs_hashalloc(pip, cg, ipref, mode, 0,
1059 (allocfcn_t *)ffs_nodealloccg);
1062 error = ffs_vget(pvp->v_mount, ino, LK_EXCLUSIVE, vpp);
1064 error1 = ffs_vgetf(pvp->v_mount, ino, LK_EXCLUSIVE, vpp,
1066 ffs_vfree(pvp, ino, mode);
1071 ip->i_flag |= IN_MODIFIED;
1079 printf("mode = 0%o, inum = %lu, fs = %s\n",
1080 ip->i_mode, (u_long)ip->i_number, fs->fs_fsmnt);
1081 panic("ffs_valloc: dup alloc");
1083 if (DIP(ip, i_blocks) && (fs->fs_flags & FS_UNCLEAN) == 0) { /* XXX */
1084 printf("free inode %s/%lu had %ld blocks\n",
1085 fs->fs_fsmnt, (u_long)ino, (long)DIP(ip, i_blocks));
1086 DIP_SET(ip, i_blocks, 0);
1089 DIP_SET(ip, i_flags, 0);
1091 * Set up a new generation number for this inode.
1093 if (ip->i_gen == 0 || ++ip->i_gen == 0)
1094 ip->i_gen = arc4random() / 2 + 1;
1095 DIP_SET(ip, i_gen, ip->i_gen);
1096 if (fs->fs_magic == FS_UFS2_MAGIC) {
1098 ip->i_din2->di_birthtime = ts.tv_sec;
1099 ip->i_din2->di_birthnsec = ts.tv_nsec;
1101 ufs_prepare_reclaim(*vpp);
1103 (*vpp)->v_vflag = 0;
1104 (*vpp)->v_type = VNON;
1105 if (fs->fs_magic == FS_UFS2_MAGIC)
1106 (*vpp)->v_op = &ffs_vnodeops2;
1108 (*vpp)->v_op = &ffs_vnodeops1;
1111 if (reclaimed == 0) {
1113 softdep_request_cleanup(fs, pvp, cred, FLUSH_INODES_WAIT);
1117 if (ppsratecheck(&lastfail, &curfail, 1)) {
1118 ffs_fserr(fs, pip->i_number, "out of inodes");
1119 uprintf("\n%s: create/symlink failed, no inodes free\n",
1126 * Find a cylinder group to place a directory.
1128 * The policy implemented by this algorithm is to allocate a
1129 * directory inode in the same cylinder group as its parent
1130 * directory, but also to reserve space for its files inodes
1131 * and data. Restrict the number of directories which may be
1132 * allocated one after another in the same cylinder group
1133 * without intervening allocation of files.
1135 * If we allocate a first level directory then force allocation
1136 * in another cylinder group.
1143 int cg, prefcg, dirsize, cgsize;
1144 u_int avgifree, avgbfree, avgndir, curdirsize;
1145 u_int minifree, minbfree, maxndir;
1146 u_int mincg, minndir;
1147 u_int maxcontigdirs;
1149 mtx_assert(UFS_MTX(pip->i_ump), MA_OWNED);
1152 avgifree = fs->fs_cstotal.cs_nifree / fs->fs_ncg;
1153 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
1154 avgndir = fs->fs_cstotal.cs_ndir / fs->fs_ncg;
1157 * Force allocation in another cg if creating a first level dir.
1159 ASSERT_VOP_LOCKED(ITOV(pip), "ffs_dirpref");
1160 if (ITOV(pip)->v_vflag & VV_ROOT) {
1161 prefcg = arc4random() % fs->fs_ncg;
1163 minndir = fs->fs_ipg;
1164 for (cg = prefcg; cg < fs->fs_ncg; cg++)
1165 if (fs->fs_cs(fs, cg).cs_ndir < minndir &&
1166 fs->fs_cs(fs, cg).cs_nifree >= avgifree &&
1167 fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
1169 minndir = fs->fs_cs(fs, cg).cs_ndir;
1171 for (cg = 0; cg < prefcg; cg++)
1172 if (fs->fs_cs(fs, cg).cs_ndir < minndir &&
1173 fs->fs_cs(fs, cg).cs_nifree >= avgifree &&
1174 fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
1176 minndir = fs->fs_cs(fs, cg).cs_ndir;
1178 return ((ino_t)(fs->fs_ipg * mincg));
1182 * Count various limits which used for
1183 * optimal allocation of a directory inode.
1185 maxndir = min(avgndir + fs->fs_ipg / 16, fs->fs_ipg);
1186 minifree = avgifree - avgifree / 4;
1189 minbfree = avgbfree - avgbfree / 4;
1192 cgsize = fs->fs_fsize * fs->fs_fpg;
1193 dirsize = fs->fs_avgfilesize * fs->fs_avgfpdir;
1194 curdirsize = avgndir ? (cgsize - avgbfree * fs->fs_bsize) / avgndir : 0;
1195 if (dirsize < curdirsize)
1196 dirsize = curdirsize;
1198 maxcontigdirs = 0; /* dirsize overflowed */
1200 maxcontigdirs = min((avgbfree * fs->fs_bsize) / dirsize, 255);
1201 if (fs->fs_avgfpdir > 0)
1202 maxcontigdirs = min(maxcontigdirs,
1203 fs->fs_ipg / fs->fs_avgfpdir);
1204 if (maxcontigdirs == 0)
1208 * Limit number of dirs in one cg and reserve space for
1209 * regular files, but only if we have no deficit in
1212 * We are trying to find a suitable cylinder group nearby
1213 * our preferred cylinder group to place a new directory.
1214 * We scan from our preferred cylinder group forward looking
1215 * for a cylinder group that meets our criterion. If we get
1216 * to the final cylinder group and do not find anything,
1217 * we start scanning backwards from our preferred cylinder
1218 * group. The ideal would be to alternate looking forward
1219 * and backward, but that is just too complex to code for
1220 * the gain it would get. The most likely place where the
1221 * backward scan would take effect is when we start near
1222 * the end of the filesystem and do not find anything from
1223 * where we are to the end. In that case, scanning backward
1224 * will likely find us a suitable cylinder group much closer
1225 * to our desired location than if we were to start scanning
1226 * forward from the beginning of the filesystem.
1228 prefcg = ino_to_cg(fs, pip->i_number);
1229 for (cg = prefcg; cg < fs->fs_ncg; cg++)
1230 if (fs->fs_cs(fs, cg).cs_ndir < maxndir &&
1231 fs->fs_cs(fs, cg).cs_nifree >= minifree &&
1232 fs->fs_cs(fs, cg).cs_nbfree >= minbfree) {
1233 if (fs->fs_contigdirs[cg] < maxcontigdirs)
1234 return ((ino_t)(fs->fs_ipg * cg));
1236 for (cg = 0; cg < prefcg; cg++)
1237 if (fs->fs_cs(fs, cg).cs_ndir < maxndir &&
1238 fs->fs_cs(fs, cg).cs_nifree >= minifree &&
1239 fs->fs_cs(fs, cg).cs_nbfree >= minbfree) {
1240 if (fs->fs_contigdirs[cg] < maxcontigdirs)
1241 return ((ino_t)(fs->fs_ipg * cg));
1244 * This is a backstop when we have deficit in space.
1246 for (cg = prefcg; cg < fs->fs_ncg; cg++)
1247 if (fs->fs_cs(fs, cg).cs_nifree >= avgifree)
1248 return ((ino_t)(fs->fs_ipg * cg));
1249 for (cg = 0; cg < prefcg; cg++)
1250 if (fs->fs_cs(fs, cg).cs_nifree >= avgifree)
1252 return ((ino_t)(fs->fs_ipg * cg));
1256 * Select the desired position for the next block in a file. The file is
1257 * logically divided into sections. The first section is composed of the
1258 * direct blocks and the next fs_maxbpg blocks. Each additional section
1259 * contains fs_maxbpg blocks.
1261 * If no blocks have been allocated in the first section, the policy is to
1262 * request a block in the same cylinder group as the inode that describes
1263 * the file. The first indirect is allocated immediately following the last
1264 * direct block and the data blocks for the first indirect immediately
1267 * If no blocks have been allocated in any other section, the indirect
1268 * block(s) are allocated in the same cylinder group as its inode in an
1269 * area reserved immediately following the inode blocks. The policy for
1270 * the data blocks is to place them in a cylinder group with a greater than
1271 * average number of free blocks. An appropriate cylinder group is found
1272 * by using a rotor that sweeps the cylinder groups. When a new group of
1273 * blocks is needed, the sweep begins in the cylinder group following the
1274 * cylinder group from which the previous allocation was made. The sweep
1275 * continues until a cylinder group with greater than the average number
1276 * of free blocks is found. If the allocation is for the first block in an
1277 * indirect block or the previous block is a hole, then the information on
1278 * the previous allocation is unavailable; here a best guess is made based
1279 * on the logical block number being allocated.
1281 * If a section is already partially allocated, the policy is to
1282 * allocate blocks contiguously within the section if possible.
1285 ffs_blkpref_ufs1(ip, lbn, indx, bap)
1293 u_int avgbfree, startcg;
1296 KASSERT(indx <= 0 || bap != NULL, ("need non-NULL bap"));
1297 mtx_assert(UFS_MTX(ip->i_ump), MA_OWNED);
1300 * Allocation of indirect blocks is indicated by passing negative
1301 * values in indx: -1 for single indirect, -2 for double indirect,
1302 * -3 for triple indirect. As noted below, we attempt to allocate
1303 * the first indirect inline with the file data. For all later
1304 * indirect blocks, the data is often allocated in other cylinder
1305 * groups. However to speed random file access and to speed up
1306 * fsck, the filesystem reserves the first fs_metaspace blocks
1307 * (typically half of fs_minfree) of the data area of each cylinder
1308 * group to hold these later indirect blocks.
1310 inocg = ino_to_cg(fs, ip->i_number);
1313 * Our preference for indirect blocks is the zone at the
1314 * beginning of the inode's cylinder group data area that
1315 * we try to reserve for indirect blocks.
1317 pref = cgmeta(fs, inocg);
1319 * If we are allocating the first indirect block, try to
1320 * place it immediately following the last direct block.
1322 if (indx == -1 && lbn < NDADDR + NINDIR(fs) &&
1323 ip->i_din1->di_db[NDADDR - 1] != 0)
1324 pref = ip->i_din1->di_db[NDADDR - 1] + fs->fs_frag;
1328 * If we are allocating the first data block in the first indirect
1329 * block and the indirect has been allocated in the data block area,
1330 * try to place it immediately following the indirect block.
1332 if (lbn == NDADDR) {
1333 pref = ip->i_din1->di_ib[0];
1334 if (pref != 0 && pref >= cgdata(fs, inocg) &&
1335 pref < cgbase(fs, inocg + 1))
1336 return (pref + fs->fs_frag);
1339 * If we are at the beginning of a file, or we have already allocated
1340 * the maximum number of blocks per cylinder group, or we do not
1341 * have a block allocated immediately preceeding us, then we need
1342 * to decide where to start allocating new blocks.
1344 if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) {
1346 * If we are allocating a directory data block, we want
1347 * to place it in the metadata area.
1349 if ((ip->i_mode & IFMT) == IFDIR)
1350 return (cgmeta(fs, inocg));
1352 * Until we fill all the direct and all the first indirect's
1353 * blocks, we try to allocate in the data area of the inode's
1356 if (lbn < NDADDR + NINDIR(fs))
1357 return (cgdata(fs, inocg));
1359 * Find a cylinder with greater than average number of
1360 * unused data blocks.
1362 if (indx == 0 || bap[indx - 1] == 0)
1363 startcg = inocg + lbn / fs->fs_maxbpg;
1365 startcg = dtog(fs, bap[indx - 1]) + 1;
1366 startcg %= fs->fs_ncg;
1367 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
1368 for (cg = startcg; cg < fs->fs_ncg; cg++)
1369 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
1370 fs->fs_cgrotor = cg;
1371 return (cgdata(fs, cg));
1373 for (cg = 0; cg <= startcg; cg++)
1374 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
1375 fs->fs_cgrotor = cg;
1376 return (cgdata(fs, cg));
1381 * Otherwise, we just always try to lay things out contiguously.
1383 return (bap[indx - 1] + fs->fs_frag);
1387 * Same as above, but for UFS2
1390 ffs_blkpref_ufs2(ip, lbn, indx, bap)
1398 u_int avgbfree, startcg;
1401 KASSERT(indx <= 0 || bap != NULL, ("need non-NULL bap"));
1402 mtx_assert(UFS_MTX(ip->i_ump), MA_OWNED);
1405 * Allocation of indirect blocks is indicated by passing negative
1406 * values in indx: -1 for single indirect, -2 for double indirect,
1407 * -3 for triple indirect. As noted below, we attempt to allocate
1408 * the first indirect inline with the file data. For all later
1409 * indirect blocks, the data is often allocated in other cylinder
1410 * groups. However to speed random file access and to speed up
1411 * fsck, the filesystem reserves the first fs_metaspace blocks
1412 * (typically half of fs_minfree) of the data area of each cylinder
1413 * group to hold these later indirect blocks.
1415 inocg = ino_to_cg(fs, ip->i_number);
1418 * Our preference for indirect blocks is the zone at the
1419 * beginning of the inode's cylinder group data area that
1420 * we try to reserve for indirect blocks.
1422 pref = cgmeta(fs, inocg);
1424 * If we are allocating the first indirect block, try to
1425 * place it immediately following the last direct block.
1427 if (indx == -1 && lbn < NDADDR + NINDIR(fs) &&
1428 ip->i_din2->di_db[NDADDR - 1] != 0)
1429 pref = ip->i_din2->di_db[NDADDR - 1] + fs->fs_frag;
1433 * If we are allocating the first data block in the first indirect
1434 * block and the indirect has been allocated in the data block area,
1435 * try to place it immediately following the indirect block.
1437 if (lbn == NDADDR) {
1438 pref = ip->i_din2->di_ib[0];
1439 if (pref != 0 && pref >= cgdata(fs, inocg) &&
1440 pref < cgbase(fs, inocg + 1))
1441 return (pref + fs->fs_frag);
1444 * If we are at the beginning of a file, or we have already allocated
1445 * the maximum number of blocks per cylinder group, or we do not
1446 * have a block allocated immediately preceeding us, then we need
1447 * to decide where to start allocating new blocks.
1449 if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) {
1451 * If we are allocating a directory data block, we want
1452 * to place it in the metadata area.
1454 if ((ip->i_mode & IFMT) == IFDIR)
1455 return (cgmeta(fs, inocg));
1457 * Until we fill all the direct and all the first indirect's
1458 * blocks, we try to allocate in the data area of the inode's
1461 if (lbn < NDADDR + NINDIR(fs))
1462 return (cgdata(fs, inocg));
1464 * Find a cylinder with greater than average number of
1465 * unused data blocks.
1467 if (indx == 0 || bap[indx - 1] == 0)
1468 startcg = inocg + lbn / fs->fs_maxbpg;
1470 startcg = dtog(fs, bap[indx - 1]) + 1;
1471 startcg %= fs->fs_ncg;
1472 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
1473 for (cg = startcg; cg < fs->fs_ncg; cg++)
1474 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
1475 fs->fs_cgrotor = cg;
1476 return (cgdata(fs, cg));
1478 for (cg = 0; cg <= startcg; cg++)
1479 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
1480 fs->fs_cgrotor = cg;
1481 return (cgdata(fs, cg));
1486 * Otherwise, we just always try to lay things out contiguously.
1488 return (bap[indx - 1] + fs->fs_frag);
1492 * Implement the cylinder overflow algorithm.
1494 * The policy implemented by this algorithm is:
1495 * 1) allocate the block in its requested cylinder group.
1496 * 2) quadradically rehash on the cylinder group number.
1497 * 3) brute force search for a free block.
1499 * Must be called with the UFS lock held. Will release the lock on success
1500 * and return with it held on failure.
1504 ffs_hashalloc(ip, cg, pref, size, rsize, allocator)
1508 int size; /* Search size for data blocks, mode for inodes */
1509 int rsize; /* Real allocated size. */
1510 allocfcn_t *allocator;
1513 ufs2_daddr_t result;
1516 mtx_assert(UFS_MTX(ip->i_ump), MA_OWNED);
1518 if (ITOV(ip)->v_mount->mnt_kern_flag & MNTK_SUSPENDED)
1519 panic("ffs_hashalloc: allocation on suspended filesystem");
1523 * 1: preferred cylinder group
1525 result = (*allocator)(ip, cg, pref, size, rsize);
1529 * 2: quadratic rehash
1531 for (i = 1; i < fs->fs_ncg; i *= 2) {
1533 if (cg >= fs->fs_ncg)
1535 result = (*allocator)(ip, cg, 0, size, rsize);
1540 * 3: brute force search
1541 * Note that we start at i == 2, since 0 was checked initially,
1542 * and 1 is always checked in the quadratic rehash.
1544 cg = (icg + 2) % fs->fs_ncg;
1545 for (i = 2; i < fs->fs_ncg; i++) {
1546 result = (*allocator)(ip, cg, 0, size, rsize);
1550 if (cg == fs->fs_ncg)
1557 * Determine whether a fragment can be extended.
1559 * Check to see if the necessary fragments are available, and
1560 * if they are, allocate them.
1563 ffs_fragextend(ip, cg, bprev, osize, nsize)
1572 struct ufsmount *ump;
1581 if (fs->fs_cs(fs, cg).cs_nffree < numfrags(fs, nsize - osize))
1583 frags = numfrags(fs, nsize);
1584 bbase = fragnum(fs, bprev);
1585 if (bbase > fragnum(fs, (bprev + frags - 1))) {
1586 /* cannot extend across a block boundary */
1590 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
1591 (int)fs->fs_cgsize, NOCRED, &bp);
1594 cgp = (struct cg *)bp->b_data;
1595 if (!cg_chkmagic(cgp))
1597 bp->b_xflags |= BX_BKGRDWRITE;
1598 cgp->cg_old_time = cgp->cg_time = time_second;
1599 bno = dtogd(fs, bprev);
1600 blksfree = cg_blksfree(cgp);
1601 for (i = numfrags(fs, osize); i < frags; i++)
1602 if (isclr(blksfree, bno + i))
1605 * the current fragment can be extended
1606 * deduct the count on fragment being extended into
1607 * increase the count on the remaining fragment (if any)
1608 * allocate the extended piece
1610 for (i = frags; i < fs->fs_frag - bbase; i++)
1611 if (isclr(blksfree, bno + i))
1613 cgp->cg_frsum[i - numfrags(fs, osize)]--;
1615 cgp->cg_frsum[i - frags]++;
1616 for (i = numfrags(fs, osize), nffree = 0; i < frags; i++) {
1617 clrbit(blksfree, bno + i);
1618 cgp->cg_cs.cs_nffree--;
1622 fs->fs_cstotal.cs_nffree -= nffree;
1623 fs->fs_cs(fs, cg).cs_nffree -= nffree;
1625 ACTIVECLEAR(fs, cg);
1627 if (DOINGSOFTDEP(ITOV(ip)))
1628 softdep_setup_blkmapdep(bp, UFSTOVFS(ump), bprev,
1629 frags, numfrags(fs, osize));
1641 * Determine whether a block can be allocated.
1643 * Check to see if a block of the appropriate size is available,
1644 * and if it is, allocate it.
1647 ffs_alloccg(ip, cg, bpref, size, rsize)
1657 struct ufsmount *ump;
1660 int i, allocsiz, error, frags;
1665 if (fs->fs_cs(fs, cg).cs_nbfree == 0 && size == fs->fs_bsize)
1668 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
1669 (int)fs->fs_cgsize, NOCRED, &bp);
1672 cgp = (struct cg *)bp->b_data;
1673 if (!cg_chkmagic(cgp) ||
1674 (cgp->cg_cs.cs_nbfree == 0 && size == fs->fs_bsize))
1676 bp->b_xflags |= BX_BKGRDWRITE;
1677 cgp->cg_old_time = cgp->cg_time = time_second;
1678 if (size == fs->fs_bsize) {
1680 blkno = ffs_alloccgblk(ip, bp, bpref, rsize);
1681 ACTIVECLEAR(fs, cg);
1687 * check to see if any fragments are already available
1688 * allocsiz is the size which will be allocated, hacking
1689 * it down to a smaller size if necessary
1691 blksfree = cg_blksfree(cgp);
1692 frags = numfrags(fs, size);
1693 for (allocsiz = frags; allocsiz < fs->fs_frag; allocsiz++)
1694 if (cgp->cg_frsum[allocsiz] != 0)
1696 if (allocsiz == fs->fs_frag) {
1698 * no fragments were available, so a block will be
1699 * allocated, and hacked up
1701 if (cgp->cg_cs.cs_nbfree == 0)
1704 blkno = ffs_alloccgblk(ip, bp, bpref, rsize);
1705 ACTIVECLEAR(fs, cg);
1710 KASSERT(size == rsize,
1711 ("ffs_alloccg: size(%d) != rsize(%d)", size, rsize));
1712 bno = ffs_mapsearch(fs, cgp, bpref, allocsiz);
1715 for (i = 0; i < frags; i++)
1716 clrbit(blksfree, bno + i);
1717 cgp->cg_cs.cs_nffree -= frags;
1718 cgp->cg_frsum[allocsiz]--;
1719 if (frags != allocsiz)
1720 cgp->cg_frsum[allocsiz - frags]++;
1722 fs->fs_cstotal.cs_nffree -= frags;
1723 fs->fs_cs(fs, cg).cs_nffree -= frags;
1725 blkno = cgbase(fs, cg) + bno;
1726 ACTIVECLEAR(fs, cg);
1728 if (DOINGSOFTDEP(ITOV(ip)))
1729 softdep_setup_blkmapdep(bp, UFSTOVFS(ump), blkno, frags, 0);
1740 * Allocate a block in a cylinder group.
1742 * This algorithm implements the following policy:
1743 * 1) allocate the requested block.
1744 * 2) allocate a rotationally optimal block in the same cylinder.
1745 * 3) allocate the next available block on the block rotor for the
1746 * specified cylinder group.
1747 * Note that this routine only allocates fs_bsize blocks; these
1748 * blocks may be fragmented by the routine that allocates them.
1751 ffs_alloccgblk(ip, bp, bpref, size)
1759 struct ufsmount *ump;
1767 mtx_assert(UFS_MTX(ump), MA_OWNED);
1768 cgp = (struct cg *)bp->b_data;
1769 blksfree = cg_blksfree(cgp);
1771 bpref = cgbase(fs, cgp->cg_cgx) + cgp->cg_rotor + fs->fs_frag;
1772 } else if ((cgbpref = dtog(fs, bpref)) != cgp->cg_cgx) {
1773 /* map bpref to correct zone in this cg */
1774 if (bpref < cgdata(fs, cgbpref))
1775 bpref = cgmeta(fs, cgp->cg_cgx);
1777 bpref = cgdata(fs, cgp->cg_cgx);
1780 * if the requested block is available, use it
1782 bno = dtogd(fs, blknum(fs, bpref));
1783 if (ffs_isblock(fs, blksfree, fragstoblks(fs, bno)))
1786 * Take the next available block in this cylinder group.
1788 bno = ffs_mapsearch(fs, cgp, bpref, (int)fs->fs_frag);
1791 /* Update cg_rotor only if allocated from the data zone */
1792 if (bno >= dtogd(fs, cgdata(fs, cgp->cg_cgx)))
1793 cgp->cg_rotor = bno;
1795 blkno = fragstoblks(fs, bno);
1796 ffs_clrblock(fs, blksfree, (long)blkno);
1797 ffs_clusteracct(fs, cgp, blkno, -1);
1798 cgp->cg_cs.cs_nbfree--;
1799 fs->fs_cstotal.cs_nbfree--;
1800 fs->fs_cs(fs, cgp->cg_cgx).cs_nbfree--;
1802 blkno = cgbase(fs, cgp->cg_cgx) + bno;
1804 * If the caller didn't want the whole block free the frags here.
1806 size = numfrags(fs, size);
1807 if (size != fs->fs_frag) {
1808 bno = dtogd(fs, blkno);
1809 for (i = size; i < fs->fs_frag; i++)
1810 setbit(blksfree, bno + i);
1811 i = fs->fs_frag - size;
1812 cgp->cg_cs.cs_nffree += i;
1813 fs->fs_cstotal.cs_nffree += i;
1814 fs->fs_cs(fs, cgp->cg_cgx).cs_nffree += i;
1820 if (DOINGSOFTDEP(ITOV(ip)))
1821 softdep_setup_blkmapdep(bp, UFSTOVFS(ump), blkno,
1828 * Determine whether a cluster can be allocated.
1830 * We do not currently check for optimal rotational layout if there
1831 * are multiple choices in the same cylinder group. Instead we just
1832 * take the first one that we find following bpref.
1835 ffs_clusteralloc(ip, cg, bpref, len)
1844 struct ufsmount *ump;
1845 int i, run, bit, map, got;
1853 if (fs->fs_maxcluster[cg] < len)
1856 if (bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), (int)fs->fs_cgsize,
1859 cgp = (struct cg *)bp->b_data;
1860 if (!cg_chkmagic(cgp))
1862 bp->b_xflags |= BX_BKGRDWRITE;
1864 * Check to see if a cluster of the needed size (or bigger) is
1865 * available in this cylinder group.
1867 lp = &cg_clustersum(cgp)[len];
1868 for (i = len; i <= fs->fs_contigsumsize; i++)
1871 if (i > fs->fs_contigsumsize) {
1873 * This is the first time looking for a cluster in this
1874 * cylinder group. Update the cluster summary information
1875 * to reflect the true maximum sized cluster so that
1876 * future cluster allocation requests can avoid reading
1877 * the cylinder group map only to find no clusters.
1879 lp = &cg_clustersum(cgp)[len - 1];
1880 for (i = len - 1; i > 0; i--)
1884 fs->fs_maxcluster[cg] = i;
1888 * Search the cluster map to find a big enough cluster.
1889 * We take the first one that we find, even if it is larger
1890 * than we need as we prefer to get one close to the previous
1891 * block allocation. We do not search before the current
1892 * preference point as we do not want to allocate a block
1893 * that is allocated before the previous one (as we will
1894 * then have to wait for another pass of the elevator
1895 * algorithm before it will be read). We prefer to fail and
1896 * be recalled to try an allocation in the next cylinder group.
1898 if (dtog(fs, bpref) != cg)
1899 bpref = cgdata(fs, cg);
1901 bpref = blknum(fs, bpref);
1902 bpref = fragstoblks(fs, dtogd(fs, bpref));
1903 mapp = &cg_clustersfree(cgp)[bpref / NBBY];
1905 bit = 1 << (bpref % NBBY);
1906 for (run = 0, got = bpref; got < cgp->cg_nclusterblks; got++) {
1907 if ((map & bit) == 0) {
1914 if ((got & (NBBY - 1)) != (NBBY - 1)) {
1921 if (got >= cgp->cg_nclusterblks)
1924 * Allocate the cluster that we have found.
1926 blksfree = cg_blksfree(cgp);
1927 for (i = 1; i <= len; i++)
1928 if (!ffs_isblock(fs, blksfree, got - run + i))
1929 panic("ffs_clusteralloc: map mismatch");
1930 bno = cgbase(fs, cg) + blkstofrags(fs, got - run + 1);
1931 if (dtog(fs, bno) != cg)
1932 panic("ffs_clusteralloc: allocated out of group");
1933 len = blkstofrags(fs, len);
1935 for (i = 0; i < len; i += fs->fs_frag)
1936 if (ffs_alloccgblk(ip, bp, bno + i, fs->fs_bsize) != bno + i)
1937 panic("ffs_clusteralloc: lost block");
1938 ACTIVECLEAR(fs, cg);
1950 static inline struct buf *
1951 getinobuf(struct inode *ip, u_int cg, u_int32_t cginoblk, int gbflags)
1956 return (getblk(ip->i_devvp, fsbtodb(fs, ino_to_fsba(fs,
1957 cg * fs->fs_ipg + cginoblk)), (int)fs->fs_bsize, 0, 0,
1962 * Determine whether an inode can be allocated.
1964 * Check to see if an inode is available, and if it is,
1965 * allocate it using the following policy:
1966 * 1) allocate the requested inode.
1967 * 2) allocate the next available inode after the requested
1968 * inode in the specified cylinder group.
1971 ffs_nodealloccg(ip, cg, ipref, mode, unused)
1980 struct buf *bp, *ibp;
1981 struct ufsmount *ump;
1982 u_int8_t *inosused, *loc;
1983 struct ufs2_dinode *dp2;
1984 int error, start, len, i;
1985 u_int32_t old_initediblk;
1990 if (fs->fs_cs(fs, cg).cs_nifree == 0)
1993 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
1994 (int)fs->fs_cgsize, NOCRED, &bp);
2000 cgp = (struct cg *)bp->b_data;
2002 if (!cg_chkmagic(cgp) || cgp->cg_cs.cs_nifree == 0) {
2007 bp->b_xflags |= BX_BKGRDWRITE;
2008 inosused = cg_inosused(cgp);
2010 ipref %= fs->fs_ipg;
2011 if (isclr(inosused, ipref))
2014 start = cgp->cg_irotor / NBBY;
2015 len = howmany(fs->fs_ipg - cgp->cg_irotor, NBBY);
2016 loc = memcchr(&inosused[start], 0xff, len);
2020 loc = memcchr(&inosused[start], 0xff, len);
2022 printf("cg = %d, irotor = %ld, fs = %s\n",
2023 cg, (long)cgp->cg_irotor, fs->fs_fsmnt);
2024 panic("ffs_nodealloccg: map corrupted");
2028 ipref = (loc - inosused) * NBBY + ffs(~*loc) - 1;
2031 * Check to see if we need to initialize more inodes.
2033 if (fs->fs_magic == FS_UFS2_MAGIC &&
2034 ipref + INOPB(fs) > cgp->cg_initediblk &&
2035 cgp->cg_initediblk < cgp->cg_niblk) {
2036 old_initediblk = cgp->cg_initediblk;
2039 * Free the cylinder group lock before writing the
2040 * initialized inode block. Entering the
2041 * babarrierwrite() with the cylinder group lock
2042 * causes lock order violation between the lock and
2045 * Another thread can decide to initialize the same
2046 * inode block, but whichever thread first gets the
2047 * cylinder group lock after writing the newly
2048 * allocated inode block will update it and the other
2049 * will realize that it has lost and leave the
2050 * cylinder group unchanged.
2052 ibp = getinobuf(ip, cg, old_initediblk, GB_LOCK_NOWAIT);
2056 * The inode block buffer is already owned by
2057 * another thread, which must initialize it.
2058 * Wait on the buffer to allow another thread
2059 * to finish the updates, with dropped cg
2060 * buffer lock, then retry.
2062 ibp = getinobuf(ip, cg, old_initediblk, 0);
2067 bzero(ibp->b_data, (int)fs->fs_bsize);
2068 dp2 = (struct ufs2_dinode *)(ibp->b_data);
2069 for (i = 0; i < INOPB(fs); i++) {
2070 dp2->di_gen = arc4random() / 2 + 1;
2074 * Rather than adding a soft updates dependency to ensure
2075 * that the new inode block is written before it is claimed
2076 * by the cylinder group map, we just do a barrier write
2077 * here. The barrier write will ensure that the inode block
2078 * gets written before the updated cylinder group map can be
2079 * written. The barrier write should only slow down bulk
2080 * loading of newly created filesystems.
2082 babarrierwrite(ibp);
2085 * After the inode block is written, try to update the
2086 * cg initediblk pointer. If another thread beat us
2087 * to it, then leave it unchanged as the other thread
2088 * has already set it correctly.
2090 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
2091 (int)fs->fs_cgsize, NOCRED, &bp);
2093 ACTIVECLEAR(fs, cg);
2099 cgp = (struct cg *)bp->b_data;
2100 if (cgp->cg_initediblk == old_initediblk)
2101 cgp->cg_initediblk += INOPB(fs);
2104 cgp->cg_old_time = cgp->cg_time = time_second;
2105 cgp->cg_irotor = ipref;
2107 ACTIVECLEAR(fs, cg);
2108 setbit(inosused, ipref);
2109 cgp->cg_cs.cs_nifree--;
2110 fs->fs_cstotal.cs_nifree--;
2111 fs->fs_cs(fs, cg).cs_nifree--;
2113 if ((mode & IFMT) == IFDIR) {
2114 cgp->cg_cs.cs_ndir++;
2115 fs->fs_cstotal.cs_ndir++;
2116 fs->fs_cs(fs, cg).cs_ndir++;
2119 if (DOINGSOFTDEP(ITOV(ip)))
2120 softdep_setup_inomapdep(bp, ip, cg * fs->fs_ipg + ipref, mode);
2122 return ((ino_t)(cg * fs->fs_ipg + ipref));
2126 * Free a block or fragment.
2128 * The specified block or fragment is placed back in the
2129 * free map. If a fragment is deallocated, a possible
2130 * block reassembly is checked.
2133 ffs_blkfree_cg(ump, fs, devvp, bno, size, inum, dephd)
2134 struct ufsmount *ump;
2136 struct vnode *devvp;
2140 struct workhead *dephd;
2145 ufs1_daddr_t fragno, cgbno;
2146 ufs2_daddr_t cgblkno;
2147 int i, blk, frags, bbase;
2153 if (devvp->v_type == VREG) {
2154 /* devvp is a snapshot */
2155 dev = VTOI(devvp)->i_devvp->v_rdev;
2156 cgblkno = fragstoblks(fs, cgtod(fs, cg));
2158 /* devvp is a normal disk device */
2159 dev = devvp->v_rdev;
2160 cgblkno = fsbtodb(fs, cgtod(fs, cg));
2161 ASSERT_VOP_LOCKED(devvp, "ffs_blkfree_cg");
2164 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0 ||
2165 fragnum(fs, bno) + numfrags(fs, size) > fs->fs_frag) {
2166 printf("dev=%s, bno = %jd, bsize = %ld, size = %ld, fs = %s\n",
2167 devtoname(dev), (intmax_t)bno, (long)fs->fs_bsize,
2168 size, fs->fs_fsmnt);
2169 panic("ffs_blkfree_cg: bad size");
2172 if ((u_int)bno >= fs->fs_size) {
2173 printf("bad block %jd, ino %lu\n", (intmax_t)bno,
2175 ffs_fserr(fs, inum, "bad block");
2178 if (bread(devvp, cgblkno, (int)fs->fs_cgsize, NOCRED, &bp)) {
2182 cgp = (struct cg *)bp->b_data;
2183 if (!cg_chkmagic(cgp)) {
2187 bp->b_xflags |= BX_BKGRDWRITE;
2188 cgp->cg_old_time = cgp->cg_time = time_second;
2189 cgbno = dtogd(fs, bno);
2190 blksfree = cg_blksfree(cgp);
2192 if (size == fs->fs_bsize) {
2193 fragno = fragstoblks(fs, cgbno);
2194 if (!ffs_isfreeblock(fs, blksfree, fragno)) {
2195 if (devvp->v_type == VREG) {
2197 /* devvp is a snapshot */
2201 printf("dev = %s, block = %jd, fs = %s\n",
2202 devtoname(dev), (intmax_t)bno, fs->fs_fsmnt);
2203 panic("ffs_blkfree_cg: freeing free block");
2205 ffs_setblock(fs, blksfree, fragno);
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++;
2211 bbase = cgbno - fragnum(fs, cgbno);
2213 * decrement the counts associated with the old frags
2215 blk = blkmap(fs, blksfree, bbase);
2216 ffs_fragacct(fs, blk, cgp->cg_frsum, -1);
2218 * deallocate the fragment
2220 frags = numfrags(fs, size);
2221 for (i = 0; i < frags; i++) {
2222 if (isset(blksfree, cgbno + i)) {
2223 printf("dev = %s, block = %jd, fs = %s\n",
2224 devtoname(dev), (intmax_t)(bno + i),
2226 panic("ffs_blkfree_cg: freeing free frag");
2228 setbit(blksfree, cgbno + i);
2230 cgp->cg_cs.cs_nffree += i;
2231 fs->fs_cstotal.cs_nffree += i;
2232 fs->fs_cs(fs, cg).cs_nffree += i;
2234 * add back in counts associated with the new frags
2236 blk = blkmap(fs, blksfree, bbase);
2237 ffs_fragacct(fs, blk, cgp->cg_frsum, 1);
2239 * if a complete block has been reassembled, account for it
2241 fragno = fragstoblks(fs, bbase);
2242 if (ffs_isblock(fs, blksfree, fragno)) {
2243 cgp->cg_cs.cs_nffree -= fs->fs_frag;
2244 fs->fs_cstotal.cs_nffree -= fs->fs_frag;
2245 fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag;
2246 ffs_clusteracct(fs, cgp, fragno, 1);
2247 cgp->cg_cs.cs_nbfree++;
2248 fs->fs_cstotal.cs_nbfree++;
2249 fs->fs_cs(fs, cg).cs_nbfree++;
2253 ACTIVECLEAR(fs, cg);
2256 if (MOUNTEDSOFTDEP(mp) && devvp->v_type != VREG)
2257 softdep_setup_blkfree(UFSTOVFS(ump), bp, bno,
2258 numfrags(fs, size), dephd);
2262 TASKQUEUE_DEFINE_THREAD(ffs_trim);
2264 struct ffs_blkfree_trim_params {
2266 struct ufsmount *ump;
2267 struct vnode *devvp;
2271 struct workhead *pdephd;
2272 struct workhead dephd;
2276 ffs_blkfree_trim_task(ctx, pending)
2280 struct ffs_blkfree_trim_params *tp;
2283 ffs_blkfree_cg(tp->ump, tp->ump->um_fs, tp->devvp, tp->bno, tp->size,
2284 tp->inum, tp->pdephd);
2285 vn_finished_secondary_write(UFSTOVFS(tp->ump));
2290 ffs_blkfree_trim_completed(bip)
2293 struct ffs_blkfree_trim_params *tp;
2295 tp = bip->bio_caller2;
2297 TASK_INIT(&tp->task, 0, ffs_blkfree_trim_task, tp);
2298 taskqueue_enqueue(taskqueue_ffs_trim, &tp->task);
2302 ffs_blkfree(ump, fs, devvp, bno, size, inum, vtype, dephd)
2303 struct ufsmount *ump;
2305 struct vnode *devvp;
2310 struct workhead *dephd;
2314 struct ffs_blkfree_trim_params *tp;
2317 * Check to see if a snapshot wants to claim the block.
2318 * Check that devvp is a normal disk device, not a snapshot,
2319 * it has a snapshot(s) associated with it, and one of the
2320 * snapshots wants to claim the block.
2322 if (devvp->v_type != VREG &&
2323 (devvp->v_vflag & VV_COPYONWRITE) &&
2324 ffs_snapblkfree(fs, devvp, bno, size, inum, vtype, dephd)) {
2328 * Nothing to delay if TRIM is disabled, or the operation is
2329 * performed on the snapshot.
2331 if (!ump->um_candelete || devvp->v_type == VREG) {
2332 ffs_blkfree_cg(ump, fs, devvp, bno, size, inum, dephd);
2337 * Postpone the set of the free bit in the cg bitmap until the
2338 * BIO_DELETE is completed. Otherwise, due to disk queue
2339 * reordering, TRIM might be issued after we reuse the block
2340 * and write some new data into it.
2342 tp = malloc(sizeof(struct ffs_blkfree_trim_params), M_TEMP, M_WAITOK);
2348 if (dephd != NULL) {
2349 LIST_INIT(&tp->dephd);
2350 LIST_SWAP(dephd, &tp->dephd, worklist, wk_list);
2351 tp->pdephd = &tp->dephd;
2355 bip = g_alloc_bio();
2356 bip->bio_cmd = BIO_DELETE;
2357 bip->bio_offset = dbtob(fsbtodb(fs, bno));
2358 bip->bio_done = ffs_blkfree_trim_completed;
2359 bip->bio_length = size;
2360 bip->bio_caller2 = tp;
2363 vn_start_secondary_write(NULL, &mp, 0);
2364 g_io_request(bip, (struct g_consumer *)devvp->v_bufobj.bo_private);
2369 * Verify allocation of a block or fragment. Returns true if block or
2370 * fragment is allocated, false if it is free.
2373 ffs_checkblk(ip, bno, size)
2382 int i, error, frags, free;
2386 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) {
2387 printf("bsize = %ld, size = %ld, fs = %s\n",
2388 (long)fs->fs_bsize, size, fs->fs_fsmnt);
2389 panic("ffs_checkblk: bad size");
2391 if ((u_int)bno >= fs->fs_size)
2392 panic("ffs_checkblk: bad block %jd", (intmax_t)bno);
2393 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, dtog(fs, bno))),
2394 (int)fs->fs_cgsize, NOCRED, &bp);
2396 panic("ffs_checkblk: cg bread failed");
2397 cgp = (struct cg *)bp->b_data;
2398 if (!cg_chkmagic(cgp))
2399 panic("ffs_checkblk: cg magic mismatch");
2400 bp->b_xflags |= BX_BKGRDWRITE;
2401 blksfree = cg_blksfree(cgp);
2402 cgbno = dtogd(fs, bno);
2403 if (size == fs->fs_bsize) {
2404 free = ffs_isblock(fs, blksfree, fragstoblks(fs, cgbno));
2406 frags = numfrags(fs, size);
2407 for (free = 0, i = 0; i < frags; i++)
2408 if (isset(blksfree, cgbno + i))
2410 if (free != 0 && free != frags)
2411 panic("ffs_checkblk: partially free fragment");
2416 #endif /* INVARIANTS */
2422 ffs_vfree(pvp, ino, mode)
2429 if (DOINGSOFTDEP(pvp)) {
2430 softdep_freefile(pvp, ino, mode);
2434 return (ffs_freefile(ip->i_ump, ip->i_fs, ip->i_devvp, ino, mode,
2439 * Do the actual free operation.
2440 * The specified inode is placed back in the free map.
2443 ffs_freefile(ump, fs, devvp, ino, mode, wkhd)
2444 struct ufsmount *ump;
2446 struct vnode *devvp;
2449 struct workhead *wkhd;
2459 cg = ino_to_cg(fs, ino);
2460 if (devvp->v_type == VREG) {
2461 /* devvp is a snapshot */
2462 dev = VTOI(devvp)->i_devvp->v_rdev;
2463 cgbno = fragstoblks(fs, cgtod(fs, cg));
2465 /* devvp is a normal disk device */
2466 dev = devvp->v_rdev;
2467 cgbno = fsbtodb(fs, cgtod(fs, cg));
2469 if (ino >= fs->fs_ipg * fs->fs_ncg)
2470 panic("ffs_freefile: range: dev = %s, ino = %ju, fs = %s",
2471 devtoname(dev), (uintmax_t)ino, fs->fs_fsmnt);
2472 if ((error = bread(devvp, cgbno, (int)fs->fs_cgsize, NOCRED, &bp))) {
2476 cgp = (struct cg *)bp->b_data;
2477 if (!cg_chkmagic(cgp)) {
2481 bp->b_xflags |= BX_BKGRDWRITE;
2482 cgp->cg_old_time = cgp->cg_time = time_second;
2483 inosused = cg_inosused(cgp);
2485 if (isclr(inosused, ino)) {
2486 printf("dev = %s, ino = %ju, fs = %s\n", devtoname(dev),
2487 (uintmax_t)(ino + cg * fs->fs_ipg), fs->fs_fsmnt);
2488 if (fs->fs_ronly == 0)
2489 panic("ffs_freefile: freeing free inode");
2491 clrbit(inosused, ino);
2492 if (ino < cgp->cg_irotor)
2493 cgp->cg_irotor = ino;
2494 cgp->cg_cs.cs_nifree++;
2496 fs->fs_cstotal.cs_nifree++;
2497 fs->fs_cs(fs, cg).cs_nifree++;
2498 if ((mode & IFMT) == IFDIR) {
2499 cgp->cg_cs.cs_ndir--;
2500 fs->fs_cstotal.cs_ndir--;
2501 fs->fs_cs(fs, cg).cs_ndir--;
2504 ACTIVECLEAR(fs, cg);
2506 if (MOUNTEDSOFTDEP(UFSTOVFS(ump)) && devvp->v_type != VREG)
2507 softdep_setup_inofree(UFSTOVFS(ump), bp,
2508 ino + cg * fs->fs_ipg, wkhd);
2514 * Check to see if a file is free.
2517 ffs_checkfreefile(fs, devvp, ino)
2519 struct vnode *devvp;
2529 cg = ino_to_cg(fs, ino);
2530 if (devvp->v_type == VREG) {
2531 /* devvp is a snapshot */
2532 cgbno = fragstoblks(fs, cgtod(fs, cg));
2534 /* devvp is a normal disk device */
2535 cgbno = fsbtodb(fs, cgtod(fs, cg));
2537 if (ino >= fs->fs_ipg * fs->fs_ncg)
2539 if (bread(devvp, cgbno, (int)fs->fs_cgsize, NOCRED, &bp)) {
2543 cgp = (struct cg *)bp->b_data;
2544 if (!cg_chkmagic(cgp)) {
2548 inosused = cg_inosused(cgp);
2550 ret = isclr(inosused, ino);
2556 * Find a block of the specified size in the specified cylinder group.
2558 * It is a panic if a request is made to find a block if none are
2562 ffs_mapsearch(fs, cgp, bpref, allocsiz)
2569 int start, len, loc, i;
2570 int blk, field, subfield, pos;
2574 * find the fragment by searching through the free block
2575 * map for an appropriate bit pattern
2578 start = dtogd(fs, bpref) / NBBY;
2580 start = cgp->cg_frotor / NBBY;
2581 blksfree = cg_blksfree(cgp);
2582 len = howmany(fs->fs_fpg, NBBY) - start;
2583 loc = scanc((u_int)len, (u_char *)&blksfree[start],
2584 fragtbl[fs->fs_frag],
2585 (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
2589 loc = scanc((u_int)len, (u_char *)&blksfree[0],
2590 fragtbl[fs->fs_frag],
2591 (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
2593 printf("start = %d, len = %d, fs = %s\n",
2594 start, len, fs->fs_fsmnt);
2595 panic("ffs_alloccg: map corrupted");
2599 bno = (start + len - loc) * NBBY;
2600 cgp->cg_frotor = bno;
2602 * found the byte in the map
2603 * sift through the bits to find the selected frag
2605 for (i = bno + NBBY; bno < i; bno += fs->fs_frag) {
2606 blk = blkmap(fs, blksfree, bno);
2608 field = around[allocsiz];
2609 subfield = inside[allocsiz];
2610 for (pos = 0; pos <= fs->fs_frag - allocsiz; pos++) {
2611 if ((blk & field) == subfield)
2617 printf("bno = %lu, fs = %s\n", (u_long)bno, fs->fs_fsmnt);
2618 panic("ffs_alloccg: block not in map");
2623 * Fserr prints the name of a filesystem with an error diagnostic.
2625 * The form of the error message is:
2629 ffs_fserr(fs, inum, cp)
2634 struct thread *td = curthread; /* XXX */
2635 struct proc *p = td->td_proc;
2637 log(LOG_ERR, "pid %d (%s), uid %d inumber %ju on %s: %s\n",
2638 p->p_pid, p->p_comm, td->td_ucred->cr_uid, (uintmax_t)inum,
2643 * This function provides the capability for the fsck program to
2644 * update an active filesystem. Fourteen operations are provided:
2646 * adjrefcnt(inode, amt) - adjusts the reference count on the
2647 * specified inode by the specified amount. Under normal
2648 * operation the count should always go down. Decrementing
2649 * the count to zero will cause the inode to be freed.
2650 * adjblkcnt(inode, amt) - adjust the number of blocks used by the
2651 * inode by the specified amount.
2652 * adjndir, adjbfree, adjifree, adjffree, adjnumclusters(amt) -
2653 * adjust the superblock summary.
2654 * freedirs(inode, count) - directory inodes [inode..inode + count - 1]
2655 * are marked as free. Inodes should never have to be marked
2657 * freefiles(inode, count) - file inodes [inode..inode + count - 1]
2658 * are marked as free. Inodes should never have to be marked
2660 * freeblks(blockno, size) - blocks [blockno..blockno + size - 1]
2661 * are marked as free. Blocks should never have to be marked
2663 * setflags(flags, set/clear) - the fs_flags field has the specified
2664 * flags set (second parameter +1) or cleared (second parameter -1).
2665 * setcwd(dirinode) - set the current directory to dirinode in the
2666 * filesystem associated with the snapshot.
2667 * setdotdot(oldvalue, newvalue) - Verify that the inode number for ".."
2668 * in the current directory is oldvalue then change it to newvalue.
2669 * unlink(nameptr, oldvalue) - Verify that the inode number associated
2670 * with nameptr in the current directory is oldvalue then unlink it.
2672 * The following functions may only be used on a quiescent filesystem
2673 * by the soft updates journal. They are not safe to be run on an active
2676 * setinode(inode, dip) - the specified disk inode is replaced with the
2677 * contents pointed to by dip.
2678 * setbufoutput(fd, flags) - output associated with the specified file
2679 * descriptor (which must reference the character device supporting
2680 * the filesystem) switches from using physio to running through the
2681 * buffer cache when flags is set to 1. The descriptor reverts to
2682 * physio for output when flags is set to zero.
2685 static int sysctl_ffs_fsck(SYSCTL_HANDLER_ARGS);
2687 SYSCTL_PROC(_vfs_ffs, FFS_ADJ_REFCNT, adjrefcnt, CTLFLAG_WR|CTLTYPE_STRUCT,
2688 0, 0, sysctl_ffs_fsck, "S,fsck", "Adjust Inode Reference Count");
2690 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_BLKCNT, adjblkcnt, CTLFLAG_WR,
2691 sysctl_ffs_fsck, "Adjust Inode Used Blocks Count");
2693 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NDIR, adjndir, CTLFLAG_WR,
2694 sysctl_ffs_fsck, "Adjust number of directories");
2696 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NBFREE, adjnbfree, CTLFLAG_WR,
2697 sysctl_ffs_fsck, "Adjust number of free blocks");
2699 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NIFREE, adjnifree, CTLFLAG_WR,
2700 sysctl_ffs_fsck, "Adjust number of free inodes");
2702 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NFFREE, adjnffree, CTLFLAG_WR,
2703 sysctl_ffs_fsck, "Adjust number of free frags");
2705 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NUMCLUSTERS, adjnumclusters, CTLFLAG_WR,
2706 sysctl_ffs_fsck, "Adjust number of free clusters");
2708 static SYSCTL_NODE(_vfs_ffs, FFS_DIR_FREE, freedirs, CTLFLAG_WR,
2709 sysctl_ffs_fsck, "Free Range of Directory Inodes");
2711 static SYSCTL_NODE(_vfs_ffs, FFS_FILE_FREE, freefiles, CTLFLAG_WR,
2712 sysctl_ffs_fsck, "Free Range of File Inodes");
2714 static SYSCTL_NODE(_vfs_ffs, FFS_BLK_FREE, freeblks, CTLFLAG_WR,
2715 sysctl_ffs_fsck, "Free Range of Blocks");
2717 static SYSCTL_NODE(_vfs_ffs, FFS_SET_FLAGS, setflags, CTLFLAG_WR,
2718 sysctl_ffs_fsck, "Change Filesystem Flags");
2720 static SYSCTL_NODE(_vfs_ffs, FFS_SET_CWD, setcwd, CTLFLAG_WR,
2721 sysctl_ffs_fsck, "Set Current Working Directory");
2723 static SYSCTL_NODE(_vfs_ffs, FFS_SET_DOTDOT, setdotdot, CTLFLAG_WR,
2724 sysctl_ffs_fsck, "Change Value of .. Entry");
2726 static SYSCTL_NODE(_vfs_ffs, FFS_UNLINK, unlink, CTLFLAG_WR,
2727 sysctl_ffs_fsck, "Unlink a Duplicate Name");
2729 static SYSCTL_NODE(_vfs_ffs, FFS_SET_INODE, setinode, CTLFLAG_WR,
2730 sysctl_ffs_fsck, "Update an On-Disk Inode");
2732 static SYSCTL_NODE(_vfs_ffs, FFS_SET_BUFOUTPUT, setbufoutput, CTLFLAG_WR,
2733 sysctl_ffs_fsck, "Set Buffered Writing for Descriptor");
2737 static int fsckcmds = 0;
2738 SYSCTL_INT(_debug, OID_AUTO, fsckcmds, CTLFLAG_RW, &fsckcmds, 0, "");
2741 static int buffered_write(struct file *, struct uio *, struct ucred *,
2742 int, struct thread *);
2745 sysctl_ffs_fsck(SYSCTL_HANDLER_ARGS)
2747 struct thread *td = curthread;
2748 struct fsck_cmd cmd;
2749 struct ufsmount *ump;
2750 struct vnode *vp, *vpold, *dvp, *fdvp;
2751 struct inode *ip, *dp;
2755 long blkcnt, blksize;
2756 struct filedesc *fdp;
2757 struct file *fp, *vfp;
2758 cap_rights_t rights;
2759 int filetype, error;
2760 static struct fileops *origops, bufferedops;
2762 if (req->newlen > sizeof cmd)
2764 if ((error = SYSCTL_IN(req, &cmd, sizeof cmd)) != 0)
2766 if (cmd.version != FFS_CMD_VERSION)
2767 return (ERPCMISMATCH);
2768 if ((error = getvnode(td->td_proc->p_fd, cmd.handle,
2769 cap_rights_init(&rights, CAP_FSCK), &fp)) != 0)
2772 if (vp->v_type != VREG && vp->v_type != VDIR) {
2776 vn_start_write(vp, &mp, V_WAIT);
2777 if (mp == 0 || strncmp(mp->mnt_stat.f_fstypename, "ufs", MFSNAMELEN)) {
2778 vn_finished_write(mp);
2783 if ((mp->mnt_flag & MNT_RDONLY) &&
2784 ump->um_fsckpid != td->td_proc->p_pid) {
2785 vn_finished_write(mp);
2792 switch (oidp->oid_number) {
2797 printf("%s: %s flags\n", mp->mnt_stat.f_mntonname,
2798 cmd.size > 0 ? "set" : "clear");
2801 fs->fs_flags |= (long)cmd.value;
2803 fs->fs_flags &= ~(long)cmd.value;
2806 case FFS_ADJ_REFCNT:
2809 printf("%s: adjust inode %jd link count by %jd\n",
2810 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value,
2811 (intmax_t)cmd.size);
2814 if ((error = ffs_vget(mp, (ino_t)cmd.value, LK_EXCLUSIVE, &vp)))
2817 ip->i_nlink += cmd.size;
2818 DIP_SET(ip, i_nlink, ip->i_nlink);
2819 ip->i_effnlink += cmd.size;
2820 ip->i_flag |= IN_CHANGE | IN_MODIFIED;
2821 error = ffs_update(vp, 1);
2822 if (DOINGSOFTDEP(vp))
2823 softdep_change_linkcnt(ip);
2827 case FFS_ADJ_BLKCNT:
2830 printf("%s: adjust inode %jd block count by %jd\n",
2831 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value,
2832 (intmax_t)cmd.size);
2835 if ((error = ffs_vget(mp, (ino_t)cmd.value, LK_EXCLUSIVE, &vp)))
2838 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + cmd.size);
2839 ip->i_flag |= IN_CHANGE | IN_MODIFIED;
2840 error = ffs_update(vp, 1);
2852 printf("%s: free %s inode %ju\n",
2853 mp->mnt_stat.f_mntonname,
2854 filetype == IFDIR ? "directory" : "file",
2855 (uintmax_t)cmd.value);
2857 printf("%s: free %s inodes %ju-%ju\n",
2858 mp->mnt_stat.f_mntonname,
2859 filetype == IFDIR ? "directory" : "file",
2860 (uintmax_t)cmd.value,
2861 (uintmax_t)(cmd.value + cmd.size - 1));
2864 while (cmd.size > 0) {
2865 if ((error = ffs_freefile(ump, fs, ump->um_devvp,
2866 cmd.value, filetype, NULL)))
2877 printf("%s: free block %jd\n",
2878 mp->mnt_stat.f_mntonname,
2879 (intmax_t)cmd.value);
2881 printf("%s: free blocks %jd-%jd\n",
2882 mp->mnt_stat.f_mntonname,
2883 (intmax_t)cmd.value,
2884 (intmax_t)cmd.value + cmd.size - 1);
2889 blksize = fs->fs_frag - (blkno % fs->fs_frag);
2890 while (blkcnt > 0) {
2891 if (blksize > blkcnt)
2893 ffs_blkfree(ump, fs, ump->um_devvp, blkno,
2894 blksize * fs->fs_fsize, ROOTINO, VDIR, NULL);
2897 blksize = fs->fs_frag;
2902 * Adjust superblock summaries. fsck(8) is expected to
2903 * submit deltas when necessary.
2908 printf("%s: adjust number of directories by %jd\n",
2909 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value);
2912 fs->fs_cstotal.cs_ndir += cmd.value;
2915 case FFS_ADJ_NBFREE:
2918 printf("%s: adjust number of free blocks by %+jd\n",
2919 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value);
2922 fs->fs_cstotal.cs_nbfree += cmd.value;
2925 case FFS_ADJ_NIFREE:
2928 printf("%s: adjust number of free inodes by %+jd\n",
2929 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value);
2932 fs->fs_cstotal.cs_nifree += cmd.value;
2935 case FFS_ADJ_NFFREE:
2938 printf("%s: adjust number of free frags by %+jd\n",
2939 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value);
2942 fs->fs_cstotal.cs_nffree += cmd.value;
2945 case FFS_ADJ_NUMCLUSTERS:
2948 printf("%s: adjust number of free clusters by %+jd\n",
2949 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value);
2952 fs->fs_cstotal.cs_numclusters += cmd.value;
2958 printf("%s: set current directory to inode %jd\n",
2959 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value);
2962 if ((error = ffs_vget(mp, (ino_t)cmd.value, LK_SHARED, &vp)))
2964 AUDIT_ARG_VNODE1(vp);
2965 if ((error = change_dir(vp, td)) != 0) {
2970 fdp = td->td_proc->p_fd;
2971 FILEDESC_XLOCK(fdp);
2972 vpold = fdp->fd_cdir;
2974 FILEDESC_XUNLOCK(fdp);
2978 case FFS_SET_DOTDOT:
2981 printf("%s: change .. in cwd from %jd to %jd\n",
2982 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value,
2983 (intmax_t)cmd.size);
2987 * First we have to get and lock the parent directory
2988 * to which ".." points.
2990 error = ffs_vget(mp, (ino_t)cmd.value, LK_EXCLUSIVE, &fdvp);
2994 * Now we get and lock the child directory containing "..".
2996 FILEDESC_SLOCK(td->td_proc->p_fd);
2997 dvp = td->td_proc->p_fd->fd_cdir;
2998 FILEDESC_SUNLOCK(td->td_proc->p_fd);
2999 if ((error = vget(dvp, LK_EXCLUSIVE, td)) != 0) {
3004 dp->i_offset = 12; /* XXX mastertemplate.dot_reclen */
3005 error = ufs_dirrewrite(dp, VTOI(fdvp), (ino_t)cmd.size,
3018 if (copyinstr((char *)(intptr_t)cmd.value, buf,32,NULL))
3019 strncpy(buf, "Name_too_long", 32);
3020 printf("%s: unlink %s (inode %jd)\n",
3021 mp->mnt_stat.f_mntonname, buf, (intmax_t)cmd.size);
3025 * kern_unlinkat will do its own start/finish writes and
3026 * they do not nest, so drop ours here. Setting mp == NULL
3027 * indicates that vn_finished_write is not needed down below.
3029 vn_finished_write(mp);
3031 error = kern_unlinkat(td, AT_FDCWD, (char *)(intptr_t)cmd.value,
3032 UIO_USERSPACE, (ino_t)cmd.size);
3036 if (ump->um_fsckpid != td->td_proc->p_pid) {
3042 printf("%s: update inode %jd\n",
3043 mp->mnt_stat.f_mntonname, (intmax_t)cmd.value);
3046 if ((error = ffs_vget(mp, (ino_t)cmd.value, LK_EXCLUSIVE, &vp)))
3048 AUDIT_ARG_VNODE1(vp);
3050 if (ip->i_ump->um_fstype == UFS1)
3051 error = copyin((void *)(intptr_t)cmd.size, ip->i_din1,
3052 sizeof(struct ufs1_dinode));
3054 error = copyin((void *)(intptr_t)cmd.size, ip->i_din2,
3055 sizeof(struct ufs2_dinode));
3060 ip->i_flag |= IN_CHANGE | IN_MODIFIED;
3061 error = ffs_update(vp, 1);
3065 case FFS_SET_BUFOUTPUT:
3066 if (ump->um_fsckpid != td->td_proc->p_pid) {
3070 if (VTOI(vp)->i_ump != ump) {
3076 printf("%s: %s buffered output for descriptor %jd\n",
3077 mp->mnt_stat.f_mntonname,
3078 cmd.size == 1 ? "enable" : "disable",
3079 (intmax_t)cmd.value);
3082 if ((error = getvnode(td->td_proc->p_fd, cmd.value,
3083 cap_rights_init(&rights, CAP_FSCK), &vfp)) != 0)
3085 if (vfp->f_vnode->v_type != VCHR) {
3090 if (origops == NULL) {
3091 origops = vfp->f_ops;
3092 bcopy((void *)origops, (void *)&bufferedops,
3093 sizeof(bufferedops));
3094 bufferedops.fo_write = buffered_write;
3097 atomic_store_rel_ptr((volatile uintptr_t *)&vfp->f_ops,
3098 (uintptr_t)&bufferedops);
3100 atomic_store_rel_ptr((volatile uintptr_t *)&vfp->f_ops,
3101 (uintptr_t)origops);
3108 printf("Invalid request %d from fsck\n",
3117 vn_finished_write(mp);
3122 * Function to switch a descriptor to use the buffer cache to stage
3123 * its I/O. This is needed so that writes to the filesystem device
3124 * will give snapshots a chance to copy modified blocks for which it
3125 * needs to retain copies.
3128 buffered_write(fp, uio, active_cred, flags, td)
3131 struct ucred *active_cred;
3135 struct vnode *devvp, *vp;
3139 struct filedesc *fdp;
3144 * The devvp is associated with the /dev filesystem. To discover
3145 * the filesystem with which the device is associated, we depend
3146 * on the application setting the current directory to a location
3147 * within the filesystem being written. Yes, this is an ugly hack.
3149 devvp = fp->f_vnode;
3150 if (!vn_isdisk(devvp, NULL))
3152 fdp = td->td_proc->p_fd;
3153 FILEDESC_SLOCK(fdp);
3156 FILEDESC_SUNLOCK(fdp);
3157 vn_lock(vp, LK_SHARED | LK_RETRY);
3159 * Check that the current directory vnode indeed belongs to
3160 * UFS before trying to dereference UFS-specific v_data fields.
3162 if (vp->v_op != &ffs_vnodeops1 && vp->v_op != &ffs_vnodeops2) {
3167 if (ip->i_devvp != devvp) {
3173 foffset_lock_uio(fp, uio, flags);
3174 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
3177 printf("%s: buffered write for block %jd\n",
3178 fs->fs_fsmnt, (intmax_t)btodb(uio->uio_offset));
3182 * All I/O must be contained within a filesystem block, start on
3183 * a fragment boundary, and be a multiple of fragments in length.
3185 if (uio->uio_resid > fs->fs_bsize - (uio->uio_offset % fs->fs_bsize) ||
3186 fragoff(fs, uio->uio_offset) != 0 ||
3187 fragoff(fs, uio->uio_resid) != 0) {
3191 lbn = numfrags(fs, uio->uio_offset);
3192 bp = getblk(devvp, lbn, uio->uio_resid, 0, 0, 0);
3193 bp->b_flags |= B_RELBUF;
3194 if ((error = uiomove((char *)bp->b_data, uio->uio_resid, uio)) != 0) {
3200 VOP_UNLOCK(devvp, 0);
3201 foffset_unlock_uio(fp, uio, flags | FOF_NEXTOFF);