2 * SPDX-License-Identifier: (BSD-2-Clause-FreeBSD AND BSD-3-Clause)
4 * Copyright (c) 2002 Networks Associates Technology, Inc.
7 * This software was developed for the FreeBSD Project by Marshall
8 * Kirk McKusick and Network Associates Laboratories, the Security
9 * Research Division of Network Associates, Inc. under DARPA/SPAWAR
10 * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
16 * 1. Redistributions of source code must retain the above copyright
17 * notice, this list of conditions and the following disclaimer.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 * notice, this list of conditions and the following disclaimer in the
20 * documentation and/or other materials provided with the distribution.
22 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * Copyright (c) 1982, 1986, 1989, 1993
35 * The Regents of the University of California. All rights reserved.
37 * Redistribution and use in source and binary forms, with or without
38 * modification, are permitted provided that the following conditions
40 * 1. Redistributions of source code must retain the above copyright
41 * notice, this list of conditions and the following disclaimer.
42 * 2. Redistributions in binary form must reproduce the above copyright
43 * notice, this list of conditions and the following disclaimer in the
44 * documentation and/or other materials provided with the distribution.
45 * 3. Neither the name of the University nor the names of its contributors
46 * may be used to endorse or promote products derived from this software
47 * without specific prior written permission.
49 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
50 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
51 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
52 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
53 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
54 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
55 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
56 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
57 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
58 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
61 * @(#)ffs_balloc.c 8.8 (Berkeley) 6/16/95
64 #include <sys/cdefs.h>
65 __FBSDID("$FreeBSD$");
67 #include <sys/param.h>
68 #include <sys/systm.h>
72 #include <sys/mount.h>
73 #include <sys/vnode.h>
74 #include <sys/vmmeter.h>
76 #include <ufs/ufs/quota.h>
77 #include <ufs/ufs/inode.h>
78 #include <ufs/ufs/ufs_extern.h>
79 #include <ufs/ufs/extattr.h>
80 #include <ufs/ufs/ufsmount.h>
82 #include <ufs/ffs/fs.h>
83 #include <ufs/ffs/ffs_extern.h>
86 * Balloc defines the structure of filesystem storage
87 * by allocating the physical blocks on a device given
88 * the inode and the logical block number in a file.
89 * This is the allocation strategy for UFS1. Below is
90 * the allocation strategy for UFS2.
93 ffs_balloc_ufs1(struct vnode *vp, off_t startoffset, int size,
94 struct ucred *cred, int flags, struct buf **bpp)
97 struct ufs1_dinode *dp;
98 ufs_lbn_t lbn, lastlbn;
101 struct buf *bp, *nbp;
103 struct ufsmount *ump;
104 struct indir indirs[UFS_NIADDR + 2];
105 int deallocated, osize, nsize, num, i, error;
107 ufs1_daddr_t *bap, pref;
108 ufs1_daddr_t *allocib, *blkp, *allocblk, allociblk[UFS_NIADDR + 1];
109 ufs2_daddr_t *lbns_remfree, lbns[UFS_NIADDR + 1];
112 int gbflags, reclaimed;
119 lbn = lblkno(fs, startoffset);
120 size = blkoff(fs, startoffset) + size;
122 if (size > fs->fs_bsize)
123 panic("ffs_balloc_ufs1: blk too big");
129 gbflags = (flags & BA_UNMAPPED) != 0 ? GB_UNMAPPED : 0;
131 if (DOINGSOFTDEP(vp))
132 softdep_prealloc(vp, MNT_WAIT);
134 * If the next write will extend the file into a new block,
135 * and the file is currently composed of a fragment
136 * this fragment has to be extended to be a full block.
138 lastlbn = lblkno(fs, ip->i_size);
139 if (lastlbn < UFS_NDADDR && lastlbn < lbn) {
141 osize = blksize(fs, ip, nb);
142 if (osize < fs->fs_bsize && osize > 0) {
144 error = ffs_realloccg(ip, nb, dp->di_db[nb],
145 ffs_blkpref_ufs1(ip, lastlbn, (int)nb,
146 &dp->di_db[0]), osize, (int)fs->fs_bsize, flags,
150 if (DOINGSOFTDEP(vp))
151 softdep_setup_allocdirect(ip, nb,
152 dbtofsb(fs, bp->b_blkno), dp->di_db[nb],
153 fs->fs_bsize, osize, bp);
154 ip->i_size = smalllblktosize(fs, nb + 1);
155 dp->di_size = ip->i_size;
156 dp->di_db[nb] = dbtofsb(fs, bp->b_blkno);
157 ip->i_flag |= IN_CHANGE | IN_UPDATE;
160 else if (DOINGASYNC(vp))
167 * The first UFS_NDADDR blocks are direct blocks
169 if (lbn < UFS_NDADDR) {
170 if (flags & BA_METAONLY)
171 panic("ffs_balloc_ufs1: BA_METAONLY for direct block");
173 if (nb != 0 && ip->i_size >= smalllblktosize(fs, lbn + 1)) {
174 error = bread(vp, lbn, fs->fs_bsize, NOCRED, &bp);
179 bp->b_blkno = fsbtodb(fs, nb);
185 * Consider need to reallocate a fragment.
187 osize = fragroundup(fs, blkoff(fs, ip->i_size));
188 nsize = fragroundup(fs, size);
189 if (nsize <= osize) {
190 error = bread(vp, lbn, osize, NOCRED, &bp);
195 bp->b_blkno = fsbtodb(fs, nb);
198 error = ffs_realloccg(ip, lbn, dp->di_db[lbn],
199 ffs_blkpref_ufs1(ip, lbn, (int)lbn,
200 &dp->di_db[0]), osize, nsize, flags,
204 if (DOINGSOFTDEP(vp))
205 softdep_setup_allocdirect(ip, lbn,
206 dbtofsb(fs, bp->b_blkno), nb,
210 if (ip->i_size < smalllblktosize(fs, lbn + 1))
211 nsize = fragroundup(fs, size);
213 nsize = fs->fs_bsize;
215 error = ffs_alloc(ip, lbn,
216 ffs_blkpref_ufs1(ip, lbn, (int)lbn, &dp->di_db[0]),
217 nsize, flags, cred, &newb);
220 bp = getblk(vp, lbn, nsize, 0, 0, gbflags);
221 bp->b_blkno = fsbtodb(fs, newb);
222 if (flags & BA_CLRBUF)
224 if (DOINGSOFTDEP(vp))
225 softdep_setup_allocdirect(ip, lbn, newb, 0,
228 dp->di_db[lbn] = dbtofsb(fs, bp->b_blkno);
229 ip->i_flag |= IN_CHANGE | IN_UPDATE;
234 * Determine the number of levels of indirection.
237 if ((error = ufs_getlbns(vp, lbn, indirs, &num)) != 0)
241 panic ("ffs_balloc_ufs1: ufs_getlbns returned indirect block");
243 saved_inbdflush = curthread_pflags_set(TDP_INBDFLUSH);
245 * Fetch the first indirect block allocating if necessary.
248 nb = dp->di_ib[indirs[0].in_off];
250 allocblk = allociblk;
254 pref = ffs_blkpref_ufs1(ip, lbn, -indirs[0].in_off - 1,
256 if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
257 flags, cred, &newb)) != 0) {
258 curthread_pflags_restore(saved_inbdflush);
261 pref = newb + fs->fs_frag;
263 MPASS(allocblk < allociblk + nitems(allociblk));
264 MPASS(lbns_remfree < lbns + nitems(lbns));
266 *lbns_remfree++ = indirs[1].in_lbn;
267 bp = getblk(vp, indirs[1].in_lbn, fs->fs_bsize, 0, 0, gbflags);
268 bp->b_blkno = fsbtodb(fs, nb);
270 if (DOINGSOFTDEP(vp)) {
271 softdep_setup_allocdirect(ip,
272 UFS_NDADDR + indirs[0].in_off, newb, 0,
273 fs->fs_bsize, 0, bp);
275 } else if ((flags & IO_SYNC) == 0 && DOINGASYNC(vp)) {
276 if (bp->b_bufsize == fs->fs_bsize)
277 bp->b_flags |= B_CLUSTEROK;
280 if ((error = bwrite(bp)) != 0)
283 allocib = &dp->di_ib[indirs[0].in_off];
285 ip->i_flag |= IN_CHANGE | IN_UPDATE;
288 * Fetch through the indirect blocks, allocating as necessary.
293 indirs[i].in_lbn, (int)fs->fs_bsize, NOCRED, &bp);
298 bap = (ufs1_daddr_t *)bp->b_data;
299 nb = bap[indirs[i].in_off];
300 if ((error = UFS_CHECK_BLKNO(mp, ip->i_number, nb,
301 fs->fs_bsize)) != 0) {
314 * If parent indirect has just been allocated, try to cluster
315 * immediately following it.
318 pref = ffs_blkpref_ufs1(ip, lbn, i - num - 1,
320 if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
321 flags | IO_BUFLOCKED, cred, &newb)) != 0) {
324 if (DOINGSOFTDEP(vp) && ++reclaimed == 1) {
325 softdep_request_cleanup(fs, vp, cred,
330 if (ppsratecheck(&ump->um_last_fullmsg,
331 &ump->um_secs_fullmsg, 1)) {
333 ffs_fserr(fs, ip->i_number, "filesystem full");
334 uprintf("\n%s: write failed, filesystem "
335 "is full\n", fs->fs_fsmnt);
341 pref = newb + fs->fs_frag;
343 MPASS(allocblk < allociblk + nitems(allociblk));
344 MPASS(lbns_remfree < lbns + nitems(lbns));
346 *lbns_remfree++ = indirs[i].in_lbn;
347 nbp = getblk(vp, indirs[i].in_lbn, fs->fs_bsize, 0, 0, 0);
348 nbp->b_blkno = fsbtodb(fs, nb);
350 if (DOINGSOFTDEP(vp)) {
351 softdep_setup_allocindir_meta(nbp, ip, bp,
352 indirs[i - 1].in_off, nb);
354 } else if ((flags & IO_SYNC) == 0 && DOINGASYNC(vp)) {
355 if (nbp->b_bufsize == fs->fs_bsize)
356 nbp->b_flags |= B_CLUSTEROK;
359 if ((error = bwrite(nbp)) != 0) {
364 bap[indirs[i - 1].in_off] = nb;
365 if (allocib == NULL && unwindidx < 0)
368 * If required, write synchronously, otherwise use
371 if (flags & IO_SYNC) {
374 if (bp->b_bufsize == fs->fs_bsize)
375 bp->b_flags |= B_CLUSTEROK;
380 * If asked only for the indirect block, then return it.
382 if (flags & BA_METAONLY) {
383 curthread_pflags_restore(saved_inbdflush);
388 * Get the data block, allocating if necessary.
393 * If allocating metadata at the front of the cylinder
394 * group and parent indirect block has just been allocated,
395 * then cluster next to it if it is the first indirect in
396 * the file. Otherwise it has been allocated in the metadata
397 * area, so we want to find our own place out in the data area.
399 if (pref == 0 || (lbn > UFS_NDADDR && fs->fs_metaspace != 0))
400 pref = ffs_blkpref_ufs1(ip, lbn, indirs[i].in_off,
402 error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
403 flags | IO_BUFLOCKED, cred, &newb);
407 if (DOINGSOFTDEP(vp) && ++reclaimed == 1) {
408 softdep_request_cleanup(fs, vp, cred,
413 if (ppsratecheck(&ump->um_last_fullmsg,
414 &ump->um_secs_fullmsg, 1)) {
416 ffs_fserr(fs, ip->i_number, "filesystem full");
417 uprintf("\n%s: write failed, filesystem "
418 "is full\n", fs->fs_fsmnt);
425 MPASS(allocblk < allociblk + nitems(allociblk));
426 MPASS(lbns_remfree < lbns + nitems(lbns));
428 *lbns_remfree++ = lbn;
429 nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0, gbflags);
430 nbp->b_blkno = fsbtodb(fs, nb);
431 if (flags & BA_CLRBUF)
433 if (DOINGSOFTDEP(vp))
434 softdep_setup_allocindir_page(ip, lbn, bp,
435 indirs[i].in_off, nb, 0, nbp);
436 bap[indirs[i].in_off] = nb;
438 * If required, write synchronously, otherwise use
441 if (flags & IO_SYNC) {
444 if (bp->b_bufsize == fs->fs_bsize)
445 bp->b_flags |= B_CLUSTEROK;
448 curthread_pflags_restore(saved_inbdflush);
453 if (flags & BA_CLRBUF) {
454 int seqcount = (flags & BA_SEQMASK) >> BA_SEQSHIFT;
456 (vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0 &&
457 !(vm_page_count_severe() || buf_dirty_count_severe())) {
458 error = cluster_read(vp, ip->i_size, lbn,
459 (int)fs->fs_bsize, NOCRED,
460 MAXBSIZE, seqcount, gbflags, &nbp);
462 error = bread_gb(vp, lbn, (int)fs->fs_bsize, NOCRED,
470 nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0, gbflags);
471 nbp->b_blkno = fsbtodb(fs, nb);
473 curthread_pflags_restore(saved_inbdflush);
477 curthread_pflags_restore(saved_inbdflush);
479 * If we have failed to allocate any blocks, simply return the error.
480 * This is the usual case and avoids the need to fsync the file.
482 if (allocblk == allociblk && allocib == NULL && unwindidx == -1)
485 * If we have failed part way through block allocation, we
486 * have to deallocate any indirect blocks that we have allocated.
487 * We have to fsync the file before we start to get rid of all
488 * of its dependencies so that we do not leave them dangling.
489 * We have to sync it at the end so that the soft updates code
490 * does not find any untracked changes. Although this is really
491 * slow, running out of disk space is not expected to be a common
492 * occurrence. The error return from fsync is ignored as we already
493 * have an error to return to the user.
495 * XXX Still have to journal the free below
497 (void) ffs_syncvnode(vp, MNT_WAIT, 0);
498 for (deallocated = 0, blkp = allociblk, lbns_remfree = lbns;
499 blkp < allocblk; blkp++, lbns_remfree++) {
501 * We shall not leave the freed blocks on the vnode
502 * buffer object lists.
504 bp = getblk(vp, *lbns_remfree, fs->fs_bsize, 0, 0,
505 GB_NOCREAT | GB_UNMAPPED);
507 KASSERT(bp->b_blkno == fsbtodb(fs, *blkp),
508 ("mismatch1 l %jd %jd b %ju %ju",
509 (intmax_t)bp->b_lblkno, (uintmax_t)*lbns_remfree,
510 (uintmax_t)bp->b_blkno,
511 (uintmax_t)fsbtodb(fs, *blkp)));
512 bp->b_flags |= B_INVAL | B_RELBUF | B_NOCACHE;
513 bp->b_flags &= ~(B_ASYNC | B_CACHE);
516 deallocated += fs->fs_bsize;
518 if (allocib != NULL) {
520 } else if (unwindidx >= 0) {
523 r = bread(vp, indirs[unwindidx].in_lbn,
524 (int)fs->fs_bsize, NOCRED, &bp);
526 panic("Could not unwind indirect block, error %d", r);
529 bap = (ufs1_daddr_t *)bp->b_data;
530 bap[indirs[unwindidx].in_off] = 0;
531 if (flags & IO_SYNC) {
534 if (bp->b_bufsize == fs->fs_bsize)
535 bp->b_flags |= B_CLUSTEROK;
543 * Restore user's disk quota because allocation failed.
545 (void) chkdq(ip, -btodb(deallocated), cred, FORCE);
547 dp->di_blocks -= btodb(deallocated);
548 ip->i_flag |= IN_CHANGE | IN_UPDATE;
550 (void) ffs_syncvnode(vp, MNT_WAIT, 0);
552 * After the buffers are invalidated and on-disk pointers are
553 * cleared, free the blocks.
555 for (blkp = allociblk; blkp < allocblk; blkp++) {
557 if (blkp == allociblk)
559 bp = getblk(vp, *lbns_remfree, fs->fs_bsize, 0, 0,
560 GB_NOCREAT | GB_UNMAPPED);
562 panic("zombie1 %jd %ju %ju",
563 (intmax_t)bp->b_lblkno, (uintmax_t)bp->b_blkno,
564 (uintmax_t)fsbtodb(fs, *blkp));
568 ffs_blkfree(ump, fs, ump->um_devvp, *blkp, fs->fs_bsize,
569 ip->i_number, vp->v_type, NULL, SINGLETON_KEY);
575 * Balloc defines the structure of file system storage
576 * by allocating the physical blocks on a device given
577 * the inode and the logical block number in a file.
578 * This is the allocation strategy for UFS2. Above is
579 * the allocation strategy for UFS1.
582 ffs_balloc_ufs2(struct vnode *vp, off_t startoffset, int size,
583 struct ucred *cred, int flags, struct buf **bpp)
586 struct ufs2_dinode *dp;
587 ufs_lbn_t lbn, lastlbn;
589 struct buf *bp, *nbp;
591 struct ufsmount *ump;
592 struct indir indirs[UFS_NIADDR + 2];
593 ufs2_daddr_t nb, newb, *bap, pref;
594 ufs2_daddr_t *allocib, *blkp, *allocblk, allociblk[UFS_NIADDR + 1];
595 ufs2_daddr_t *lbns_remfree, lbns[UFS_NIADDR + 1];
596 int deallocated, osize, nsize, num, i, error;
599 int gbflags, reclaimed;
606 lbn = lblkno(fs, startoffset);
607 size = blkoff(fs, startoffset) + size;
609 if (size > fs->fs_bsize)
610 panic("ffs_balloc_ufs2: blk too big");
614 gbflags = (flags & BA_UNMAPPED) != 0 ? GB_UNMAPPED : 0;
616 if (DOINGSOFTDEP(vp))
617 softdep_prealloc(vp, MNT_WAIT);
620 * Check for allocating external data.
622 if (flags & IO_EXT) {
623 if (lbn >= UFS_NXADDR)
626 * If the next write will extend the data into a new block,
627 * and the data is currently composed of a fragment
628 * this fragment has to be extended to be a full block.
630 lastlbn = lblkno(fs, dp->di_extsize);
633 osize = sblksize(fs, dp->di_extsize, nb);
634 if (osize < fs->fs_bsize && osize > 0) {
636 error = ffs_realloccg(ip, -1 - nb,
638 ffs_blkpref_ufs2(ip, lastlbn, (int)nb,
639 &dp->di_extb[0]), osize,
640 (int)fs->fs_bsize, flags, cred, &bp);
643 if (DOINGSOFTDEP(vp))
644 softdep_setup_allocext(ip, nb,
645 dbtofsb(fs, bp->b_blkno),
647 fs->fs_bsize, osize, bp);
648 dp->di_extsize = smalllblktosize(fs, nb + 1);
649 dp->di_extb[nb] = dbtofsb(fs, bp->b_blkno);
650 bp->b_xflags |= BX_ALTDATA;
651 ip->i_flag |= IN_CHANGE;
659 * All blocks are direct blocks
661 if (flags & BA_METAONLY)
662 panic("ffs_balloc_ufs2: BA_METAONLY for ext block");
663 nb = dp->di_extb[lbn];
664 if (nb != 0 && dp->di_extsize >= smalllblktosize(fs, lbn + 1)) {
665 error = bread_gb(vp, -1 - lbn, fs->fs_bsize, NOCRED,
671 bp->b_blkno = fsbtodb(fs, nb);
672 bp->b_xflags |= BX_ALTDATA;
678 * Consider need to reallocate a fragment.
680 osize = fragroundup(fs, blkoff(fs, dp->di_extsize));
681 nsize = fragroundup(fs, size);
682 if (nsize <= osize) {
683 error = bread_gb(vp, -1 - lbn, osize, NOCRED,
689 bp->b_blkno = fsbtodb(fs, nb);
690 bp->b_xflags |= BX_ALTDATA;
693 error = ffs_realloccg(ip, -1 - lbn,
695 ffs_blkpref_ufs2(ip, lbn, (int)lbn,
696 &dp->di_extb[0]), osize, nsize, flags,
700 bp->b_xflags |= BX_ALTDATA;
701 if (DOINGSOFTDEP(vp))
702 softdep_setup_allocext(ip, lbn,
703 dbtofsb(fs, bp->b_blkno), nb,
707 if (dp->di_extsize < smalllblktosize(fs, lbn + 1))
708 nsize = fragroundup(fs, size);
710 nsize = fs->fs_bsize;
712 error = ffs_alloc(ip, lbn,
713 ffs_blkpref_ufs2(ip, lbn, (int)lbn, &dp->di_extb[0]),
714 nsize, flags, cred, &newb);
717 bp = getblk(vp, -1 - lbn, nsize, 0, 0, gbflags);
718 bp->b_blkno = fsbtodb(fs, newb);
719 bp->b_xflags |= BX_ALTDATA;
720 if (flags & BA_CLRBUF)
722 if (DOINGSOFTDEP(vp))
723 softdep_setup_allocext(ip, lbn, newb, 0,
726 dp->di_extb[lbn] = dbtofsb(fs, bp->b_blkno);
727 ip->i_flag |= IN_CHANGE;
732 * If the next write will extend the file into a new block,
733 * and the file is currently composed of a fragment
734 * this fragment has to be extended to be a full block.
736 lastlbn = lblkno(fs, ip->i_size);
737 if (lastlbn < UFS_NDADDR && lastlbn < lbn) {
739 osize = blksize(fs, ip, nb);
740 if (osize < fs->fs_bsize && osize > 0) {
742 error = ffs_realloccg(ip, nb, dp->di_db[nb],
743 ffs_blkpref_ufs2(ip, lastlbn, (int)nb,
744 &dp->di_db[0]), osize, (int)fs->fs_bsize,
748 if (DOINGSOFTDEP(vp))
749 softdep_setup_allocdirect(ip, nb,
750 dbtofsb(fs, bp->b_blkno),
752 fs->fs_bsize, osize, bp);
753 ip->i_size = smalllblktosize(fs, nb + 1);
754 dp->di_size = ip->i_size;
755 dp->di_db[nb] = dbtofsb(fs, bp->b_blkno);
756 ip->i_flag |= IN_CHANGE | IN_UPDATE;
764 * The first UFS_NDADDR blocks are direct blocks
766 if (lbn < UFS_NDADDR) {
767 if (flags & BA_METAONLY)
768 panic("ffs_balloc_ufs2: BA_METAONLY for direct block");
770 if (nb != 0 && ip->i_size >= smalllblktosize(fs, lbn + 1)) {
771 error = bread_gb(vp, lbn, fs->fs_bsize, NOCRED,
777 bp->b_blkno = fsbtodb(fs, nb);
783 * Consider need to reallocate a fragment.
785 osize = fragroundup(fs, blkoff(fs, ip->i_size));
786 nsize = fragroundup(fs, size);
787 if (nsize <= osize) {
788 error = bread_gb(vp, lbn, osize, NOCRED,
794 bp->b_blkno = fsbtodb(fs, nb);
797 error = ffs_realloccg(ip, lbn, dp->di_db[lbn],
798 ffs_blkpref_ufs2(ip, lbn, (int)lbn,
799 &dp->di_db[0]), osize, nsize, flags,
803 if (DOINGSOFTDEP(vp))
804 softdep_setup_allocdirect(ip, lbn,
805 dbtofsb(fs, bp->b_blkno), nb,
809 if (ip->i_size < smalllblktosize(fs, lbn + 1))
810 nsize = fragroundup(fs, size);
812 nsize = fs->fs_bsize;
814 error = ffs_alloc(ip, lbn,
815 ffs_blkpref_ufs2(ip, lbn, (int)lbn,
816 &dp->di_db[0]), nsize, flags, cred, &newb);
819 bp = getblk(vp, lbn, nsize, 0, 0, gbflags);
820 bp->b_blkno = fsbtodb(fs, newb);
821 if (flags & BA_CLRBUF)
823 if (DOINGSOFTDEP(vp))
824 softdep_setup_allocdirect(ip, lbn, newb, 0,
827 dp->di_db[lbn] = dbtofsb(fs, bp->b_blkno);
828 ip->i_flag |= IN_CHANGE | IN_UPDATE;
833 * Determine the number of levels of indirection.
836 if ((error = ufs_getlbns(vp, lbn, indirs, &num)) != 0)
840 panic ("ffs_balloc_ufs2: ufs_getlbns returned indirect block");
842 saved_inbdflush = curthread_pflags_set(TDP_INBDFLUSH);
844 * Fetch the first indirect block allocating if necessary.
847 nb = dp->di_ib[indirs[0].in_off];
849 allocblk = allociblk;
853 pref = ffs_blkpref_ufs2(ip, lbn, -indirs[0].in_off - 1,
855 if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
856 flags, cred, &newb)) != 0) {
857 curthread_pflags_restore(saved_inbdflush);
860 pref = newb + fs->fs_frag;
862 MPASS(allocblk < allociblk + nitems(allociblk));
863 MPASS(lbns_remfree < lbns + nitems(lbns));
865 *lbns_remfree++ = indirs[1].in_lbn;
866 bp = getblk(vp, indirs[1].in_lbn, fs->fs_bsize, 0, 0,
868 bp->b_blkno = fsbtodb(fs, nb);
870 if (DOINGSOFTDEP(vp)) {
871 softdep_setup_allocdirect(ip,
872 UFS_NDADDR + indirs[0].in_off, newb, 0,
873 fs->fs_bsize, 0, bp);
875 } else if ((flags & IO_SYNC) == 0 && DOINGASYNC(vp)) {
876 if (bp->b_bufsize == fs->fs_bsize)
877 bp->b_flags |= B_CLUSTEROK;
880 if ((error = bwrite(bp)) != 0)
883 allocib = &dp->di_ib[indirs[0].in_off];
885 ip->i_flag |= IN_CHANGE | IN_UPDATE;
888 * Fetch through the indirect blocks, allocating as necessary.
893 indirs[i].in_lbn, (int)fs->fs_bsize, NOCRED, &bp);
898 bap = (ufs2_daddr_t *)bp->b_data;
899 nb = bap[indirs[i].in_off];
900 if ((error = UFS_CHECK_BLKNO(mp, ip->i_number, nb,
901 fs->fs_bsize)) != 0) {
914 * If parent indirect has just been allocated, try to cluster
915 * immediately following it.
918 pref = ffs_blkpref_ufs2(ip, lbn, i - num - 1,
920 if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
921 flags | IO_BUFLOCKED, cred, &newb)) != 0) {
924 if (DOINGSOFTDEP(vp) && ++reclaimed == 1) {
925 softdep_request_cleanup(fs, vp, cred,
930 if (ppsratecheck(&ump->um_last_fullmsg,
931 &ump->um_secs_fullmsg, 1)) {
933 ffs_fserr(fs, ip->i_number, "filesystem full");
934 uprintf("\n%s: write failed, filesystem "
935 "is full\n", fs->fs_fsmnt);
941 pref = newb + fs->fs_frag;
943 MPASS(allocblk < allociblk + nitems(allociblk));
944 MPASS(lbns_remfree < lbns + nitems(lbns));
946 *lbns_remfree++ = indirs[i].in_lbn;
947 nbp = getblk(vp, indirs[i].in_lbn, fs->fs_bsize, 0, 0,
949 nbp->b_blkno = fsbtodb(fs, nb);
951 if (DOINGSOFTDEP(vp)) {
952 softdep_setup_allocindir_meta(nbp, ip, bp,
953 indirs[i - 1].in_off, nb);
955 } else if ((flags & IO_SYNC) == 0 && DOINGASYNC(vp)) {
956 if (nbp->b_bufsize == fs->fs_bsize)
957 nbp->b_flags |= B_CLUSTEROK;
960 if ((error = bwrite(nbp)) != 0) {
965 bap[indirs[i - 1].in_off] = nb;
966 if (allocib == NULL && unwindidx < 0)
969 * If required, write synchronously, otherwise use
972 if (flags & IO_SYNC) {
975 if (bp->b_bufsize == fs->fs_bsize)
976 bp->b_flags |= B_CLUSTEROK;
981 * If asked only for the indirect block, then return it.
983 if (flags & BA_METAONLY) {
984 curthread_pflags_restore(saved_inbdflush);
989 * Get the data block, allocating if necessary.
994 * If allocating metadata at the front of the cylinder
995 * group and parent indirect block has just been allocated,
996 * then cluster next to it if it is the first indirect in
997 * the file. Otherwise it has been allocated in the metadata
998 * area, so we want to find our own place out in the data area.
1000 if (pref == 0 || (lbn > UFS_NDADDR && fs->fs_metaspace != 0))
1001 pref = ffs_blkpref_ufs2(ip, lbn, indirs[i].in_off,
1003 error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
1004 flags | IO_BUFLOCKED, cred, &newb);
1008 if (DOINGSOFTDEP(vp) && ++reclaimed == 1) {
1009 softdep_request_cleanup(fs, vp, cred,
1014 if (ppsratecheck(&ump->um_last_fullmsg,
1015 &ump->um_secs_fullmsg, 1)) {
1017 ffs_fserr(fs, ip->i_number, "filesystem full");
1018 uprintf("\n%s: write failed, filesystem "
1019 "is full\n", fs->fs_fsmnt);
1026 MPASS(allocblk < allociblk + nitems(allociblk));
1027 MPASS(lbns_remfree < lbns + nitems(lbns));
1029 *lbns_remfree++ = lbn;
1030 nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0, gbflags);
1031 nbp->b_blkno = fsbtodb(fs, nb);
1032 if (flags & BA_CLRBUF)
1033 vfs_bio_clrbuf(nbp);
1034 if (DOINGSOFTDEP(vp))
1035 softdep_setup_allocindir_page(ip, lbn, bp,
1036 indirs[i].in_off, nb, 0, nbp);
1037 bap[indirs[i].in_off] = nb;
1039 * If required, write synchronously, otherwise use
1042 if (flags & IO_SYNC) {
1045 if (bp->b_bufsize == fs->fs_bsize)
1046 bp->b_flags |= B_CLUSTEROK;
1049 curthread_pflags_restore(saved_inbdflush);
1055 * If requested clear invalid portions of the buffer. If we
1056 * have to do a read-before-write (typical if BA_CLRBUF is set),
1057 * try to do some read-ahead in the sequential case to reduce
1058 * the number of I/O transactions.
1060 if (flags & BA_CLRBUF) {
1061 int seqcount = (flags & BA_SEQMASK) >> BA_SEQSHIFT;
1062 if (seqcount != 0 &&
1063 (vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0 &&
1064 !(vm_page_count_severe() || buf_dirty_count_severe())) {
1065 error = cluster_read(vp, ip->i_size, lbn,
1066 (int)fs->fs_bsize, NOCRED,
1067 MAXBSIZE, seqcount, gbflags, &nbp);
1069 error = bread_gb(vp, lbn, (int)fs->fs_bsize,
1070 NOCRED, gbflags, &nbp);
1077 nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0, gbflags);
1078 nbp->b_blkno = fsbtodb(fs, nb);
1080 curthread_pflags_restore(saved_inbdflush);
1084 curthread_pflags_restore(saved_inbdflush);
1086 * If we have failed to allocate any blocks, simply return the error.
1087 * This is the usual case and avoids the need to fsync the file.
1089 if (allocblk == allociblk && allocib == NULL && unwindidx == -1)
1092 * If we have failed part way through block allocation, we
1093 * have to deallocate any indirect blocks that we have allocated.
1094 * We have to fsync the file before we start to get rid of all
1095 * of its dependencies so that we do not leave them dangling.
1096 * We have to sync it at the end so that the soft updates code
1097 * does not find any untracked changes. Although this is really
1098 * slow, running out of disk space is not expected to be a common
1099 * occurrence. The error return from fsync is ignored as we already
1100 * have an error to return to the user.
1102 * XXX Still have to journal the free below
1104 (void) ffs_syncvnode(vp, MNT_WAIT, 0);
1105 for (deallocated = 0, blkp = allociblk, lbns_remfree = lbns;
1106 blkp < allocblk; blkp++, lbns_remfree++) {
1108 * We shall not leave the freed blocks on the vnode
1109 * buffer object lists.
1111 bp = getblk(vp, *lbns_remfree, fs->fs_bsize, 0, 0,
1112 GB_NOCREAT | GB_UNMAPPED);
1114 KASSERT(bp->b_blkno == fsbtodb(fs, *blkp),
1115 ("mismatch2 l %jd %jd b %ju %ju",
1116 (intmax_t)bp->b_lblkno, (uintmax_t)*lbns_remfree,
1117 (uintmax_t)bp->b_blkno,
1118 (uintmax_t)fsbtodb(fs, *blkp)));
1119 bp->b_flags |= B_INVAL | B_RELBUF | B_NOCACHE;
1120 bp->b_flags &= ~(B_ASYNC | B_CACHE);
1123 deallocated += fs->fs_bsize;
1125 if (allocib != NULL) {
1127 } else if (unwindidx >= 0) {
1130 r = bread(vp, indirs[unwindidx].in_lbn,
1131 (int)fs->fs_bsize, NOCRED, &bp);
1133 panic("Could not unwind indirect block, error %d", r);
1136 bap = (ufs2_daddr_t *)bp->b_data;
1137 bap[indirs[unwindidx].in_off] = 0;
1138 if (flags & IO_SYNC) {
1141 if (bp->b_bufsize == fs->fs_bsize)
1142 bp->b_flags |= B_CLUSTEROK;
1150 * Restore user's disk quota because allocation failed.
1152 (void) chkdq(ip, -btodb(deallocated), cred, FORCE);
1154 dp->di_blocks -= btodb(deallocated);
1155 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1157 (void) ffs_syncvnode(vp, MNT_WAIT, 0);
1159 * After the buffers are invalidated and on-disk pointers are
1160 * cleared, free the blocks.
1162 for (blkp = allociblk; blkp < allocblk; blkp++) {
1164 if (blkp == allociblk)
1165 lbns_remfree = lbns;
1166 bp = getblk(vp, *lbns_remfree, fs->fs_bsize, 0, 0,
1167 GB_NOCREAT | GB_UNMAPPED);
1169 panic("zombie2 %jd %ju %ju",
1170 (intmax_t)bp->b_lblkno, (uintmax_t)bp->b_blkno,
1171 (uintmax_t)fsbtodb(fs, *blkp));
1175 ffs_blkfree(ump, fs, ump->um_devvp, *blkp, fs->fs_bsize,
1176 ip->i_number, vp->v_type, NULL, SINGLETON_KEY);