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_balloc.c 8.8 (Berkeley) 6/16/95
62 #include <sys/cdefs.h>
63 __FBSDID("$FreeBSD$");
65 #include <sys/param.h>
66 #include <sys/systm.h>
70 #include <sys/mount.h>
71 #include <sys/vnode.h>
73 #include <ufs/ufs/quota.h>
74 #include <ufs/ufs/inode.h>
75 #include <ufs/ufs/ufs_extern.h>
76 #include <ufs/ufs/extattr.h>
77 #include <ufs/ufs/ufsmount.h>
79 #include <ufs/ffs/fs.h>
80 #include <ufs/ffs/ffs_extern.h>
83 * Balloc defines the structure of filesystem storage
84 * by allocating the physical blocks on a device given
85 * the inode and the logical block number in a file.
86 * This is the allocation strategy for UFS1. Below is
87 * the allocation strategy for UFS2.
90 ffs_balloc_ufs1(struct vnode *vp, off_t startoffset, int size,
91 struct ucred *cred, int flags, struct buf **bpp)
94 struct ufs1_dinode *dp;
95 ufs_lbn_t lbn, lastlbn;
100 struct indir indirs[NIADDR + 2];
101 int deallocated, osize, nsize, num, i, error;
103 ufs1_daddr_t *bap, pref;
104 ufs1_daddr_t *allocib, *blkp, *allocblk, allociblk[NIADDR + 1];
105 ufs2_daddr_t *lbns_remfree, lbns[NIADDR + 1];
108 static struct timeval lastfail;
110 int gbflags, reclaimed;
116 lbn = lblkno(fs, startoffset);
117 size = blkoff(fs, startoffset) + size;
119 if (size > fs->fs_bsize)
120 panic("ffs_balloc_ufs1: blk too big");
126 gbflags = (flags & BA_UNMAPPED) != 0 ? GB_UNMAPPED : 0;
128 if (DOINGSOFTDEP(vp))
129 softdep_prealloc(vp, MNT_WAIT);
131 * If the next write will extend the file into a new block,
132 * and the file is currently composed of a fragment
133 * this fragment has to be extended to be a full block.
135 lastlbn = lblkno(fs, ip->i_size);
136 if (lastlbn < NDADDR && lastlbn < lbn) {
138 osize = blksize(fs, ip, nb);
139 if (osize < fs->fs_bsize && osize > 0) {
141 error = ffs_realloccg(ip, nb, dp->di_db[nb],
142 ffs_blkpref_ufs1(ip, lastlbn, (int)nb,
143 &dp->di_db[0]), osize, (int)fs->fs_bsize, flags,
147 if (DOINGSOFTDEP(vp))
148 softdep_setup_allocdirect(ip, nb,
149 dbtofsb(fs, bp->b_blkno), dp->di_db[nb],
150 fs->fs_bsize, osize, bp);
151 ip->i_size = smalllblktosize(fs, nb + 1);
152 dp->di_size = ip->i_size;
153 dp->di_db[nb] = dbtofsb(fs, bp->b_blkno);
154 ip->i_flag |= IN_CHANGE | IN_UPDATE;
162 * The first NDADDR blocks are direct blocks
165 if (flags & BA_METAONLY)
166 panic("ffs_balloc_ufs1: BA_METAONLY for direct block");
168 if (nb != 0 && ip->i_size >= smalllblktosize(fs, lbn + 1)) {
169 error = bread(vp, lbn, fs->fs_bsize, NOCRED, &bp);
174 bp->b_blkno = fsbtodb(fs, nb);
180 * Consider need to reallocate a fragment.
182 osize = fragroundup(fs, blkoff(fs, ip->i_size));
183 nsize = fragroundup(fs, size);
184 if (nsize <= osize) {
185 error = bread(vp, lbn, osize, NOCRED, &bp);
190 bp->b_blkno = fsbtodb(fs, nb);
193 error = ffs_realloccg(ip, lbn, dp->di_db[lbn],
194 ffs_blkpref_ufs1(ip, lbn, (int)lbn,
195 &dp->di_db[0]), osize, nsize, flags,
199 if (DOINGSOFTDEP(vp))
200 softdep_setup_allocdirect(ip, lbn,
201 dbtofsb(fs, bp->b_blkno), nb,
205 if (ip->i_size < smalllblktosize(fs, lbn + 1))
206 nsize = fragroundup(fs, size);
208 nsize = fs->fs_bsize;
210 error = ffs_alloc(ip, lbn,
211 ffs_blkpref_ufs1(ip, lbn, (int)lbn, &dp->di_db[0]),
212 nsize, flags, cred, &newb);
215 bp = getblk(vp, lbn, nsize, 0, 0, gbflags);
216 bp->b_blkno = fsbtodb(fs, newb);
217 if (flags & BA_CLRBUF)
219 if (DOINGSOFTDEP(vp))
220 softdep_setup_allocdirect(ip, lbn, newb, 0,
223 dp->di_db[lbn] = dbtofsb(fs, bp->b_blkno);
224 ip->i_flag |= IN_CHANGE | IN_UPDATE;
229 * Determine the number of levels of indirection.
232 if ((error = ufs_getlbns(vp, lbn, indirs, &num)) != 0)
236 panic ("ffs_balloc_ufs1: ufs_getlbns returned indirect block");
238 saved_inbdflush = curthread_pflags_set(TDP_INBDFLUSH);
240 * Fetch the first indirect block allocating if necessary.
243 nb = dp->di_ib[indirs[0].in_off];
245 allocblk = allociblk;
249 pref = ffs_blkpref_ufs1(ip, lbn, -indirs[0].in_off - 1,
251 if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
252 flags, cred, &newb)) != 0) {
253 curthread_pflags_restore(saved_inbdflush);
256 pref = newb + fs->fs_frag;
258 MPASS(allocblk < allociblk + nitems(allociblk));
259 MPASS(lbns_remfree < lbns + nitems(lbns));
261 *lbns_remfree++ = indirs[1].in_lbn;
262 bp = getblk(vp, indirs[1].in_lbn, fs->fs_bsize, 0, 0, gbflags);
263 bp->b_blkno = fsbtodb(fs, nb);
265 if (DOINGSOFTDEP(vp)) {
266 softdep_setup_allocdirect(ip, NDADDR + indirs[0].in_off,
267 newb, 0, fs->fs_bsize, 0, bp);
271 * Write synchronously so that indirect blocks
272 * never point at garbage.
276 else if ((error = bwrite(bp)) != 0)
279 allocib = &dp->di_ib[indirs[0].in_off];
281 ip->i_flag |= IN_CHANGE | IN_UPDATE;
284 * Fetch through the indirect blocks, allocating as necessary.
289 indirs[i].in_lbn, (int)fs->fs_bsize, NOCRED, &bp);
294 bap = (ufs1_daddr_t *)bp->b_data;
295 nb = bap[indirs[i].in_off];
305 * If parent indirect has just been allocated, try to cluster
306 * immediately following it.
309 pref = ffs_blkpref_ufs1(ip, lbn, i - num - 1,
311 if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
312 flags | IO_BUFLOCKED, cred, &newb)) != 0) {
314 if (DOINGSOFTDEP(vp) && ++reclaimed == 1) {
316 softdep_request_cleanup(fs, vp, cred,
321 if (ppsratecheck(&lastfail, &curfail, 1)) {
322 ffs_fserr(fs, ip->i_number, "filesystem full");
323 uprintf("\n%s: write failed, filesystem "
324 "is full\n", fs->fs_fsmnt);
328 pref = newb + fs->fs_frag;
330 MPASS(allocblk < allociblk + nitems(allociblk));
331 MPASS(lbns_remfree < lbns + nitems(lbns));
333 *lbns_remfree++ = indirs[i].in_lbn;
334 nbp = getblk(vp, indirs[i].in_lbn, fs->fs_bsize, 0, 0, 0);
335 nbp->b_blkno = fsbtodb(fs, nb);
337 if (DOINGSOFTDEP(vp)) {
338 softdep_setup_allocindir_meta(nbp, ip, bp,
339 indirs[i - 1].in_off, nb);
343 * Write synchronously so that indirect blocks
344 * never point at garbage.
346 if ((error = bwrite(nbp)) != 0) {
351 bap[indirs[i - 1].in_off] = nb;
352 if (allocib == NULL && unwindidx < 0)
355 * If required, write synchronously, otherwise use
358 if (flags & IO_SYNC) {
361 if (bp->b_bufsize == fs->fs_bsize)
362 bp->b_flags |= B_CLUSTEROK;
367 * If asked only for the indirect block, then return it.
369 if (flags & BA_METAONLY) {
370 curthread_pflags_restore(saved_inbdflush);
375 * Get the data block, allocating if necessary.
380 * If allocating metadata at the front of the cylinder
381 * group and parent indirect block has just been allocated,
382 * then cluster next to it if it is the first indirect in
383 * the file. Otherwise it has been allocated in the metadata
384 * area, so we want to find our own place out in the data area.
386 if (pref == 0 || (lbn > NDADDR && fs->fs_metaspace != 0))
387 pref = ffs_blkpref_ufs1(ip, lbn, indirs[i].in_off,
389 error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
390 flags | IO_BUFLOCKED, cred, &newb);
393 if (DOINGSOFTDEP(vp) && ++reclaimed == 1) {
395 softdep_request_cleanup(fs, vp, cred,
400 if (ppsratecheck(&lastfail, &curfail, 1)) {
401 ffs_fserr(fs, ip->i_number, "filesystem full");
402 uprintf("\n%s: write failed, filesystem "
403 "is full\n", fs->fs_fsmnt);
408 MPASS(allocblk < allociblk + nitems(allociblk));
409 MPASS(lbns_remfree < lbns + nitems(lbns));
411 *lbns_remfree++ = lbn;
412 nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0, gbflags);
413 nbp->b_blkno = fsbtodb(fs, nb);
414 if (flags & BA_CLRBUF)
416 if (DOINGSOFTDEP(vp))
417 softdep_setup_allocindir_page(ip, lbn, bp,
418 indirs[i].in_off, nb, 0, nbp);
419 bap[indirs[i].in_off] = nb;
421 * If required, write synchronously, otherwise use
424 if (flags & IO_SYNC) {
427 if (bp->b_bufsize == fs->fs_bsize)
428 bp->b_flags |= B_CLUSTEROK;
431 curthread_pflags_restore(saved_inbdflush);
436 if (flags & BA_CLRBUF) {
437 int seqcount = (flags & BA_SEQMASK) >> BA_SEQSHIFT;
439 (vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0 &&
440 !(vm_page_count_severe() || buf_dirty_count_severe())) {
441 error = cluster_read(vp, ip->i_size, lbn,
442 (int)fs->fs_bsize, NOCRED,
443 MAXBSIZE, seqcount, gbflags, &nbp);
445 error = bread_gb(vp, lbn, (int)fs->fs_bsize, NOCRED,
453 nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0, gbflags);
454 nbp->b_blkno = fsbtodb(fs, nb);
456 curthread_pflags_restore(saved_inbdflush);
460 curthread_pflags_restore(saved_inbdflush);
462 * If we have failed to allocate any blocks, simply return the error.
463 * This is the usual case and avoids the need to fsync the file.
465 if (allocblk == allociblk && allocib == NULL && unwindidx == -1)
468 * If we have failed part way through block allocation, we
469 * have to deallocate any indirect blocks that we have allocated.
470 * We have to fsync the file before we start to get rid of all
471 * of its dependencies so that we do not leave them dangling.
472 * We have to sync it at the end so that the soft updates code
473 * does not find any untracked changes. Although this is really
474 * slow, running out of disk space is not expected to be a common
475 * occurrence. The error return from fsync is ignored as we already
476 * have an error to return to the user.
478 * XXX Still have to journal the free below
480 (void) ffs_syncvnode(vp, MNT_WAIT, 0);
481 for (deallocated = 0, blkp = allociblk, lbns_remfree = lbns;
482 blkp < allocblk; blkp++, lbns_remfree++) {
484 * We shall not leave the freed blocks on the vnode
485 * buffer object lists.
487 bp = getblk(vp, *lbns_remfree, fs->fs_bsize, 0, 0, GB_NOCREAT);
489 bp->b_flags |= (B_INVAL | B_RELBUF);
490 bp->b_flags &= ~B_ASYNC;
493 deallocated += fs->fs_bsize;
495 if (allocib != NULL) {
497 } else if (unwindidx >= 0) {
500 r = bread(vp, indirs[unwindidx].in_lbn,
501 (int)fs->fs_bsize, NOCRED, &bp);
503 panic("Could not unwind indirect block, error %d", r);
506 bap = (ufs1_daddr_t *)bp->b_data;
507 bap[indirs[unwindidx].in_off] = 0;
508 if (flags & IO_SYNC) {
511 if (bp->b_bufsize == fs->fs_bsize)
512 bp->b_flags |= B_CLUSTEROK;
520 * Restore user's disk quota because allocation failed.
522 (void) chkdq(ip, -btodb(deallocated), cred, FORCE);
524 dp->di_blocks -= btodb(deallocated);
525 ip->i_flag |= IN_CHANGE | IN_UPDATE;
527 (void) ffs_syncvnode(vp, MNT_WAIT, 0);
529 * After the buffers are invalidated and on-disk pointers are
530 * cleared, free the blocks.
532 for (blkp = allociblk; blkp < allocblk; blkp++) {
533 ffs_blkfree(ump, fs, ip->i_devvp, *blkp, fs->fs_bsize,
534 ip->i_number, vp->v_type, NULL);
540 * Balloc defines the structure of file system storage
541 * by allocating the physical blocks on a device given
542 * the inode and the logical block number in a file.
543 * This is the allocation strategy for UFS2. Above is
544 * the allocation strategy for UFS1.
547 ffs_balloc_ufs2(struct vnode *vp, off_t startoffset, int size,
548 struct ucred *cred, int flags, struct buf **bpp)
551 struct ufs2_dinode *dp;
552 ufs_lbn_t lbn, lastlbn;
554 struct buf *bp, *nbp;
555 struct ufsmount *ump;
556 struct indir indirs[NIADDR + 2];
557 ufs2_daddr_t nb, newb, *bap, pref;
558 ufs2_daddr_t *allocib, *blkp, *allocblk, allociblk[NIADDR + 1];
559 ufs2_daddr_t *lbns_remfree, lbns[NIADDR + 1];
560 int deallocated, osize, nsize, num, i, error;
563 static struct timeval lastfail;
565 int gbflags, reclaimed;
571 lbn = lblkno(fs, startoffset);
572 size = blkoff(fs, startoffset) + size;
574 if (size > fs->fs_bsize)
575 panic("ffs_balloc_ufs2: blk too big");
579 gbflags = (flags & BA_UNMAPPED) != 0 ? GB_UNMAPPED : 0;
581 if (DOINGSOFTDEP(vp))
582 softdep_prealloc(vp, MNT_WAIT);
585 * Check for allocating external data.
587 if (flags & IO_EXT) {
591 * If the next write will extend the data into a new block,
592 * and the data is currently composed of a fragment
593 * this fragment has to be extended to be a full block.
595 lastlbn = lblkno(fs, dp->di_extsize);
598 osize = sblksize(fs, dp->di_extsize, nb);
599 if (osize < fs->fs_bsize && osize > 0) {
601 error = ffs_realloccg(ip, -1 - nb,
603 ffs_blkpref_ufs2(ip, lastlbn, (int)nb,
604 &dp->di_extb[0]), osize,
605 (int)fs->fs_bsize, flags, cred, &bp);
608 if (DOINGSOFTDEP(vp))
609 softdep_setup_allocext(ip, nb,
610 dbtofsb(fs, bp->b_blkno),
612 fs->fs_bsize, osize, bp);
613 dp->di_extsize = smalllblktosize(fs, nb + 1);
614 dp->di_extb[nb] = dbtofsb(fs, bp->b_blkno);
615 bp->b_xflags |= BX_ALTDATA;
616 ip->i_flag |= IN_CHANGE;
624 * All blocks are direct blocks
626 if (flags & BA_METAONLY)
627 panic("ffs_balloc_ufs2: BA_METAONLY for ext block");
628 nb = dp->di_extb[lbn];
629 if (nb != 0 && dp->di_extsize >= smalllblktosize(fs, lbn + 1)) {
630 error = bread_gb(vp, -1 - lbn, fs->fs_bsize, NOCRED,
636 bp->b_blkno = fsbtodb(fs, nb);
637 bp->b_xflags |= BX_ALTDATA;
643 * Consider need to reallocate a fragment.
645 osize = fragroundup(fs, blkoff(fs, dp->di_extsize));
646 nsize = fragroundup(fs, size);
647 if (nsize <= osize) {
648 error = bread_gb(vp, -1 - lbn, osize, NOCRED,
654 bp->b_blkno = fsbtodb(fs, nb);
655 bp->b_xflags |= BX_ALTDATA;
658 error = ffs_realloccg(ip, -1 - lbn,
660 ffs_blkpref_ufs2(ip, lbn, (int)lbn,
661 &dp->di_extb[0]), osize, nsize, flags,
665 bp->b_xflags |= BX_ALTDATA;
666 if (DOINGSOFTDEP(vp))
667 softdep_setup_allocext(ip, lbn,
668 dbtofsb(fs, bp->b_blkno), nb,
672 if (dp->di_extsize < smalllblktosize(fs, lbn + 1))
673 nsize = fragroundup(fs, size);
675 nsize = fs->fs_bsize;
677 error = ffs_alloc(ip, lbn,
678 ffs_blkpref_ufs2(ip, lbn, (int)lbn, &dp->di_extb[0]),
679 nsize, flags, cred, &newb);
682 bp = getblk(vp, -1 - lbn, nsize, 0, 0, gbflags);
683 bp->b_blkno = fsbtodb(fs, newb);
684 bp->b_xflags |= BX_ALTDATA;
685 if (flags & BA_CLRBUF)
687 if (DOINGSOFTDEP(vp))
688 softdep_setup_allocext(ip, lbn, newb, 0,
691 dp->di_extb[lbn] = dbtofsb(fs, bp->b_blkno);
692 ip->i_flag |= IN_CHANGE;
697 * If the next write will extend the file into a new block,
698 * and the file is currently composed of a fragment
699 * this fragment has to be extended to be a full block.
701 lastlbn = lblkno(fs, ip->i_size);
702 if (lastlbn < NDADDR && lastlbn < lbn) {
704 osize = blksize(fs, ip, nb);
705 if (osize < fs->fs_bsize && osize > 0) {
707 error = ffs_realloccg(ip, nb, dp->di_db[nb],
708 ffs_blkpref_ufs2(ip, lastlbn, (int)nb,
709 &dp->di_db[0]), osize, (int)fs->fs_bsize,
713 if (DOINGSOFTDEP(vp))
714 softdep_setup_allocdirect(ip, nb,
715 dbtofsb(fs, bp->b_blkno),
717 fs->fs_bsize, osize, bp);
718 ip->i_size = smalllblktosize(fs, nb + 1);
719 dp->di_size = ip->i_size;
720 dp->di_db[nb] = dbtofsb(fs, bp->b_blkno);
721 ip->i_flag |= IN_CHANGE | IN_UPDATE;
729 * The first NDADDR blocks are direct blocks
732 if (flags & BA_METAONLY)
733 panic("ffs_balloc_ufs2: BA_METAONLY for direct block");
735 if (nb != 0 && ip->i_size >= smalllblktosize(fs, lbn + 1)) {
736 error = bread_gb(vp, lbn, fs->fs_bsize, NOCRED,
742 bp->b_blkno = fsbtodb(fs, nb);
748 * Consider need to reallocate a fragment.
750 osize = fragroundup(fs, blkoff(fs, ip->i_size));
751 nsize = fragroundup(fs, size);
752 if (nsize <= osize) {
753 error = bread_gb(vp, lbn, osize, NOCRED,
759 bp->b_blkno = fsbtodb(fs, nb);
762 error = ffs_realloccg(ip, lbn, dp->di_db[lbn],
763 ffs_blkpref_ufs2(ip, lbn, (int)lbn,
764 &dp->di_db[0]), osize, nsize, flags,
768 if (DOINGSOFTDEP(vp))
769 softdep_setup_allocdirect(ip, lbn,
770 dbtofsb(fs, bp->b_blkno), nb,
774 if (ip->i_size < smalllblktosize(fs, lbn + 1))
775 nsize = fragroundup(fs, size);
777 nsize = fs->fs_bsize;
779 error = ffs_alloc(ip, lbn,
780 ffs_blkpref_ufs2(ip, lbn, (int)lbn,
781 &dp->di_db[0]), nsize, flags, cred, &newb);
784 bp = getblk(vp, lbn, nsize, 0, 0, gbflags);
785 bp->b_blkno = fsbtodb(fs, newb);
786 if (flags & BA_CLRBUF)
788 if (DOINGSOFTDEP(vp))
789 softdep_setup_allocdirect(ip, lbn, newb, 0,
792 dp->di_db[lbn] = dbtofsb(fs, bp->b_blkno);
793 ip->i_flag |= IN_CHANGE | IN_UPDATE;
798 * Determine the number of levels of indirection.
801 if ((error = ufs_getlbns(vp, lbn, indirs, &num)) != 0)
805 panic ("ffs_balloc_ufs2: ufs_getlbns returned indirect block");
807 saved_inbdflush = curthread_pflags_set(TDP_INBDFLUSH);
809 * Fetch the first indirect block allocating if necessary.
812 nb = dp->di_ib[indirs[0].in_off];
814 allocblk = allociblk;
818 pref = ffs_blkpref_ufs2(ip, lbn, -indirs[0].in_off - 1,
820 if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
821 flags, cred, &newb)) != 0) {
822 curthread_pflags_restore(saved_inbdflush);
825 pref = newb + fs->fs_frag;
827 MPASS(allocblk < allociblk + nitems(allociblk));
828 MPASS(lbns_remfree < lbns + nitems(lbns));
830 *lbns_remfree++ = indirs[1].in_lbn;
831 bp = getblk(vp, indirs[1].in_lbn, fs->fs_bsize, 0, 0,
833 bp->b_blkno = fsbtodb(fs, nb);
835 if (DOINGSOFTDEP(vp)) {
836 softdep_setup_allocdirect(ip, NDADDR + indirs[0].in_off,
837 newb, 0, fs->fs_bsize, 0, bp);
841 * Write synchronously so that indirect blocks
842 * never point at garbage.
846 else if ((error = bwrite(bp)) != 0)
849 allocib = &dp->di_ib[indirs[0].in_off];
851 ip->i_flag |= IN_CHANGE | IN_UPDATE;
854 * Fetch through the indirect blocks, allocating as necessary.
859 indirs[i].in_lbn, (int)fs->fs_bsize, NOCRED, &bp);
864 bap = (ufs2_daddr_t *)bp->b_data;
865 nb = bap[indirs[i].in_off];
875 * If parent indirect has just been allocated, try to cluster
876 * immediately following it.
879 pref = ffs_blkpref_ufs2(ip, lbn, i - num - 1,
881 if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
882 flags | IO_BUFLOCKED, cred, &newb)) != 0) {
884 if (DOINGSOFTDEP(vp) && ++reclaimed == 1) {
886 softdep_request_cleanup(fs, vp, cred,
891 if (ppsratecheck(&lastfail, &curfail, 1)) {
892 ffs_fserr(fs, ip->i_number, "filesystem full");
893 uprintf("\n%s: write failed, filesystem "
894 "is full\n", fs->fs_fsmnt);
898 pref = newb + fs->fs_frag;
900 MPASS(allocblk < allociblk + nitems(allociblk));
901 MPASS(lbns_remfree < lbns + nitems(lbns));
903 *lbns_remfree++ = indirs[i].in_lbn;
904 nbp = getblk(vp, indirs[i].in_lbn, fs->fs_bsize, 0, 0,
906 nbp->b_blkno = fsbtodb(fs, nb);
908 if (DOINGSOFTDEP(vp)) {
909 softdep_setup_allocindir_meta(nbp, ip, bp,
910 indirs[i - 1].in_off, nb);
914 * Write synchronously so that indirect blocks
915 * never point at garbage.
917 if ((error = bwrite(nbp)) != 0) {
922 bap[indirs[i - 1].in_off] = nb;
923 if (allocib == NULL && unwindidx < 0)
926 * If required, write synchronously, otherwise use
929 if (flags & IO_SYNC) {
932 if (bp->b_bufsize == fs->fs_bsize)
933 bp->b_flags |= B_CLUSTEROK;
938 * If asked only for the indirect block, then return it.
940 if (flags & BA_METAONLY) {
941 curthread_pflags_restore(saved_inbdflush);
946 * Get the data block, allocating if necessary.
951 * If allocating metadata at the front of the cylinder
952 * group and parent indirect block has just been allocated,
953 * then cluster next to it if it is the first indirect in
954 * the file. Otherwise it has been allocated in the metadata
955 * area, so we want to find our own place out in the data area.
957 if (pref == 0 || (lbn > NDADDR && fs->fs_metaspace != 0))
958 pref = ffs_blkpref_ufs2(ip, lbn, indirs[i].in_off,
960 error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
961 flags | IO_BUFLOCKED, cred, &newb);
964 if (DOINGSOFTDEP(vp) && ++reclaimed == 1) {
966 softdep_request_cleanup(fs, vp, cred,
971 if (ppsratecheck(&lastfail, &curfail, 1)) {
972 ffs_fserr(fs, ip->i_number, "filesystem full");
973 uprintf("\n%s: write failed, filesystem "
974 "is full\n", fs->fs_fsmnt);
979 MPASS(allocblk < allociblk + nitems(allociblk));
980 MPASS(lbns_remfree < lbns + nitems(lbns));
982 *lbns_remfree++ = lbn;
983 nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0, gbflags);
984 nbp->b_blkno = fsbtodb(fs, nb);
985 if (flags & BA_CLRBUF)
987 if (DOINGSOFTDEP(vp))
988 softdep_setup_allocindir_page(ip, lbn, bp,
989 indirs[i].in_off, nb, 0, nbp);
990 bap[indirs[i].in_off] = nb;
992 * If required, write synchronously, otherwise use
995 if (flags & IO_SYNC) {
998 if (bp->b_bufsize == fs->fs_bsize)
999 bp->b_flags |= B_CLUSTEROK;
1002 curthread_pflags_restore(saved_inbdflush);
1008 * If requested clear invalid portions of the buffer. If we
1009 * have to do a read-before-write (typical if BA_CLRBUF is set),
1010 * try to do some read-ahead in the sequential case to reduce
1011 * the number of I/O transactions.
1013 if (flags & BA_CLRBUF) {
1014 int seqcount = (flags & BA_SEQMASK) >> BA_SEQSHIFT;
1015 if (seqcount != 0 &&
1016 (vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0 &&
1017 !(vm_page_count_severe() || buf_dirty_count_severe())) {
1018 error = cluster_read(vp, ip->i_size, lbn,
1019 (int)fs->fs_bsize, NOCRED,
1020 MAXBSIZE, seqcount, gbflags, &nbp);
1022 error = bread_gb(vp, lbn, (int)fs->fs_bsize,
1023 NOCRED, gbflags, &nbp);
1030 nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0, gbflags);
1031 nbp->b_blkno = fsbtodb(fs, nb);
1033 curthread_pflags_restore(saved_inbdflush);
1037 curthread_pflags_restore(saved_inbdflush);
1039 * If we have failed to allocate any blocks, simply return the error.
1040 * This is the usual case and avoids the need to fsync the file.
1042 if (allocblk == allociblk && allocib == NULL && unwindidx == -1)
1045 * If we have failed part way through block allocation, we
1046 * have to deallocate any indirect blocks that we have allocated.
1047 * We have to fsync the file before we start to get rid of all
1048 * of its dependencies so that we do not leave them dangling.
1049 * We have to sync it at the end so that the soft updates code
1050 * does not find any untracked changes. Although this is really
1051 * slow, running out of disk space is not expected to be a common
1052 * occurrence. The error return from fsync is ignored as we already
1053 * have an error to return to the user.
1055 * XXX Still have to journal the free below
1057 (void) ffs_syncvnode(vp, MNT_WAIT, 0);
1058 for (deallocated = 0, blkp = allociblk, lbns_remfree = lbns;
1059 blkp < allocblk; blkp++, lbns_remfree++) {
1061 * We shall not leave the freed blocks on the vnode
1062 * buffer object lists.
1064 bp = getblk(vp, *lbns_remfree, fs->fs_bsize, 0, 0, GB_NOCREAT);
1066 bp->b_flags |= (B_INVAL | B_RELBUF);
1067 bp->b_flags &= ~B_ASYNC;
1070 deallocated += fs->fs_bsize;
1072 if (allocib != NULL) {
1074 } else if (unwindidx >= 0) {
1077 r = bread(vp, indirs[unwindidx].in_lbn,
1078 (int)fs->fs_bsize, NOCRED, &bp);
1080 panic("Could not unwind indirect block, error %d", r);
1083 bap = (ufs2_daddr_t *)bp->b_data;
1084 bap[indirs[unwindidx].in_off] = 0;
1085 if (flags & IO_SYNC) {
1088 if (bp->b_bufsize == fs->fs_bsize)
1089 bp->b_flags |= B_CLUSTEROK;
1097 * Restore user's disk quota because allocation failed.
1099 (void) chkdq(ip, -btodb(deallocated), cred, FORCE);
1101 dp->di_blocks -= btodb(deallocated);
1102 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1104 (void) ffs_syncvnode(vp, MNT_WAIT, 0);
1106 * After the buffers are invalidated and on-disk pointers are
1107 * cleared, free the blocks.
1109 for (blkp = allociblk; blkp < allocblk; blkp++) {
1110 ffs_blkfree(ump, fs, ip->i_devvp, *blkp, fs->fs_bsize,
1111 ip->i_number, vp->v_type, NULL);