2 * Copyright (c) 1982, 1986, 1989, 1993
3 * The Regents of the University of California. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 3. All advertising materials mentioning features or use of this software
14 * must display the following acknowledgement:
15 * This product includes software developed by the University of
16 * California, Berkeley and its contributors.
17 * 4. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 * @(#)ffs_inode.c 8.13 (Berkeley) 4/21/95
34 * $FreeBSD: src/sys/ufs/ffs/ffs_inode.c,v 1.56.2.5 2002/02/05 18:35:03 dillon Exp $
35 * $DragonFly: src/sys/vfs/ufs/ffs_inode.c,v 1.18 2006/04/03 02:02:37 dillon Exp $
38 #include "opt_quota.h"
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/mount.h>
45 #include <sys/vnode.h>
46 #include <sys/kernel.h>
47 #include <sys/malloc.h>
48 #include <sys/resourcevar.h>
49 #include <sys/vmmeter.h>
52 #include <vm/vm_extern.h>
57 #include "ufs_extern.h"
60 #include "ffs_extern.h"
62 #include <vm/vm_page2.h>
64 static int ffs_indirtrunc (struct inode *, ufs_daddr_t, ufs_daddr_t,
65 ufs_daddr_t, int, long *);
68 * Update the access, modified, and inode change times as specified by the
69 * IN_ACCESS, IN_UPDATE, and IN_CHANGE flags respectively. Write the inode
70 * to disk if the IN_MODIFIED flag is set (it may be set initially, or by
71 * the timestamp update). The IN_LAZYMOD flag is set to force a write
72 * later if not now. If we write now, then clear both IN_MODIFIED and
73 * IN_LAZYMOD to reflect the presumably successful write, and if waitfor is
74 * set, then wait for the write to complete.
77 ffs_update(struct vnode *vp, int waitfor)
86 if ((ip->i_flag & IN_MODIFIED) == 0 && waitfor == 0)
88 ip->i_flag &= ~(IN_LAZYMOD | IN_MODIFIED);
93 * Ensure that uid and gid are correct. This is a temporary
94 * fix until fsck has been changed to do the update.
96 if (fs->fs_inodefmt < FS_44INODEFMT) { /* XXX */
97 ip->i_din.di_ouid = ip->i_uid; /* XXX */
98 ip->i_din.di_ogid = ip->i_gid; /* XXX */
100 error = bread(ip->i_devvp,
101 fsbtodoff(fs, ino_to_fsba(fs, ip->i_number)),
102 (int)fs->fs_bsize, &bp);
107 if (DOINGSOFTDEP(vp))
108 softdep_update_inodeblock(ip, bp, waitfor);
109 else if (ip->i_effnlink != ip->i_nlink)
110 panic("ffs_update: bad link cnt");
111 *((struct ufs1_dinode *)bp->b_data +
112 ino_to_fsbo(fs, ip->i_number)) = ip->i_din;
113 if (waitfor && !DOINGASYNC(vp)) {
115 } else if (vm_page_count_severe() || buf_dirty_count_severe()) {
118 if (bp->b_bufsize == fs->fs_bsize)
119 bp->b_flags |= B_CLUSTEROK;
125 #define SINGLE 0 /* index of single indirect block */
126 #define DOUBLE 1 /* index of double indirect block */
127 #define TRIPLE 2 /* index of triple indirect block */
129 * Truncate the inode oip to at most length size, freeing the
133 ffs_truncate(struct vnode *vp, off_t length, int flags, struct ucred *cred,
136 struct vnode *ovp = vp;
137 ufs_daddr_t lastblock;
139 ufs_daddr_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR];
140 ufs_daddr_t oldblks[NDADDR + NIADDR], newblks[NDADDR + NIADDR];
143 int offset, size, level;
144 long count, nblocks, blocksreleased = 0;
146 int aflags, error, allerror;
153 if (length > fs->fs_maxfilesize)
155 if (ovp->v_type == VLNK &&
156 (oip->i_size < ovp->v_mount->mnt_maxsymlinklen || oip->i_din.di_blocks == 0)) {
159 panic("ffs_truncate: partial truncate of symlink");
160 #endif /* DIAGNOSTIC */
161 bzero((char *)&oip->i_shortlink, (uint)oip->i_size);
163 oip->i_flag |= IN_CHANGE | IN_UPDATE;
164 return (UFS_UPDATE(ovp, 1));
166 if (oip->i_size == length) {
167 oip->i_flag |= IN_CHANGE | IN_UPDATE;
168 return (UFS_UPDATE(ovp, 0));
171 panic("ffs_truncate: read-only filesystem");
173 error = ufs_getinoquota(oip);
178 * truncation can occur for a variety of reasons where an OPEN has
179 * not been performed. truncate(), rmdir(), and remove() being
180 * examples. Vnode-based buffer cache ops require a VM object.
182 if (vp->v_object == NULL)
184 ovp->v_lasta = ovp->v_clen = ovp->v_cstart = ovp->v_lastw = 0;
185 if (DOINGSOFTDEP(ovp)) {
186 if (length > 0 || softdep_slowdown(ovp)) {
188 * If a file is only partially truncated, then
189 * we have to clean up the data structures
190 * describing the allocation past the truncation
191 * point. Finding and deallocating those structures
192 * is a lot of work. Since partial truncation occurs
193 * rarely, we solve the problem by syncing the file
194 * so that it will have no data structures left.
196 if ((error = VOP_FSYNC(ovp, MNT_WAIT, td)) != 0)
200 (void) ufs_chkdq(oip, -oip->i_blocks, NOCRED, 0);
202 softdep_setup_freeblocks(oip, length);
203 vinvalbuf(ovp, 0, td, 0, 0);
204 vnode_pager_setsize(ovp, 0);
205 oip->i_flag |= IN_CHANGE | IN_UPDATE;
206 return (ffs_update(ovp, 0));
211 * Lengthen the size of the file. We must ensure that the
212 * last byte of the file is allocated. Since the smallest
213 * value of osize is 0, length will be at least 1.
215 if (osize < length) {
216 vnode_pager_setsize(ovp, length);
220 error = VOP_BALLOC(ovp, length - 1, 1,
224 oip->i_size = length;
225 if (bp->b_bufsize == fs->fs_bsize)
226 bp->b_flags |= B_CLUSTEROK;
231 oip->i_flag |= IN_CHANGE | IN_UPDATE;
232 return (UFS_UPDATE(ovp, 1));
235 * Shorten the size of the file. If the file is not being
236 * truncated to a block boundary, the contents of the
237 * partial block following the end of the file must be
238 * zero'ed in case it ever becomes accessible again because
239 * of subsequent file growth. Directories however are not
240 * zero'ed as they should grow back initialized to empty.
242 offset = blkoff(fs, length);
244 oip->i_size = length;
246 lbn = lblkno(fs, length);
250 error = VOP_BALLOC(ovp, length - 1, 1, cred, aflags, &bp);
255 * When we are doing soft updates and the UFS_BALLOC
256 * above fills in a direct block hole with a full sized
257 * block that will be truncated down to a fragment below,
258 * we must flush out the block dependency with an FSYNC
259 * so that we do not get a soft updates inconsistency
260 * when we create the fragment below.
262 if (DOINGSOFTDEP(ovp) && lbn < NDADDR &&
263 fragroundup(fs, blkoff(fs, length)) < fs->fs_bsize &&
264 (error = VOP_FSYNC(ovp, MNT_WAIT, td)) != 0) {
267 oip->i_size = length;
268 size = blksize(fs, oip, lbn);
269 if (ovp->v_type != VDIR)
270 bzero((char *)bp->b_data + offset,
271 (uint)(size - offset));
272 /* Kirk's code has reallocbuf(bp, size, 1) here */
274 if (bp->b_bufsize == fs->fs_bsize)
275 bp->b_flags |= B_CLUSTEROK;
282 * Calculate index into inode's block list of
283 * last direct and indirect blocks (if any)
284 * which we want to keep. Lastblock is -1 when
285 * the file is truncated to 0.
287 lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1;
288 lastiblock[SINGLE] = lastblock - NDADDR;
289 lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
290 lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
291 nblocks = btodb(fs->fs_bsize);
294 * Update file and block pointers on disk before we start freeing
295 * blocks. If we crash before free'ing blocks below, the blocks
296 * will be returned to the free list. lastiblock values are also
297 * normalized to -1 for calls to ffs_indirtrunc below.
299 bcopy((caddr_t)&oip->i_db[0], (caddr_t)oldblks, sizeof oldblks);
300 for (level = TRIPLE; level >= SINGLE; level--)
301 if (lastiblock[level] < 0) {
302 oip->i_ib[level] = 0;
303 lastiblock[level] = -1;
305 for (i = NDADDR - 1; i > lastblock; i--)
307 oip->i_flag |= IN_CHANGE | IN_UPDATE;
308 allerror = UFS_UPDATE(ovp, 1);
311 * Having written the new inode to disk, save its new configuration
312 * and put back the old block pointers long enough to process them.
313 * Note that we save the new block configuration so we can check it
316 bcopy((caddr_t)&oip->i_db[0], (caddr_t)newblks, sizeof newblks);
317 bcopy((caddr_t)oldblks, (caddr_t)&oip->i_db[0], sizeof oldblks);
320 error = vtruncbuf(ovp, td, length, fs->fs_bsize);
321 if (error && (allerror == 0))
325 * Indirect blocks first.
327 indir_lbn[SINGLE] = -NDADDR;
328 indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
329 indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
330 for (level = TRIPLE; level >= SINGLE; level--) {
331 bn = oip->i_ib[level];
333 error = ffs_indirtrunc(oip, indir_lbn[level],
334 fsbtodb(fs, bn), lastiblock[level], level, &count);
337 blocksreleased += count;
338 if (lastiblock[level] < 0) {
339 oip->i_ib[level] = 0;
340 ffs_blkfree(oip, bn, fs->fs_bsize);
341 blocksreleased += nblocks;
344 if (lastiblock[level] >= 0)
349 * All whole direct blocks or frags.
351 for (i = NDADDR - 1; i > lastblock; i--) {
358 bsize = blksize(fs, oip, i);
359 ffs_blkfree(oip, bn, bsize);
360 blocksreleased += btodb(bsize);
366 * Finally, look for a change in size of the
367 * last direct block; release any frags.
369 bn = oip->i_db[lastblock];
371 long oldspace, newspace;
374 * Calculate amount of space we're giving
375 * back as old block size minus new block size.
377 oldspace = blksize(fs, oip, lastblock);
378 oip->i_size = length;
379 newspace = blksize(fs, oip, lastblock);
381 panic("ffs_truncate: newspace");
382 if (oldspace - newspace > 0) {
384 * Block number of space to be free'd is
385 * the old block # plus the number of frags
386 * required for the storage we're keeping.
388 bn += numfrags(fs, newspace);
389 ffs_blkfree(oip, bn, oldspace - newspace);
390 blocksreleased += btodb(oldspace - newspace);
395 for (level = SINGLE; level <= TRIPLE; level++)
396 if (newblks[NDADDR + level] != oip->i_ib[level])
397 panic("ffs_truncate1");
398 for (i = 0; i < NDADDR; i++)
399 if (newblks[i] != oip->i_db[i])
400 panic("ffs_truncate2");
402 (!RB_EMPTY(&ovp->v_rbdirty_tree) ||
403 !RB_EMPTY(&ovp->v_rbclean_tree)))
404 panic("ffs_truncate3");
405 #endif /* DIAGNOSTIC */
407 * Put back the real size.
409 oip->i_size = length;
410 oip->i_blocks -= blocksreleased;
412 if (oip->i_blocks < 0) /* sanity */
414 oip->i_flag |= IN_CHANGE;
416 (void) ufs_chkdq(oip, -blocksreleased, NOCRED, 0);
422 * Release blocks associated with the inode ip and stored in the indirect
423 * block bn. Blocks are free'd in LIFO order up to (but not including)
424 * lastbn. If level is greater than SINGLE, the block is an indirect block
425 * and recursive calls to indirtrunc must be used to cleanse other indirect
428 * NB: triple indirect blocks are untested.
431 ffs_indirtrunc(struct inode *ip, ufs_daddr_t lbn, ufs_daddr_t dbn,
432 ufs_daddr_t lastbn, int level, long *countp)
436 struct fs *fs = ip->i_fs;
439 ufs_daddr_t *copy = NULL, nb, nlbn, last;
440 long blkcount, factor;
441 int nblocks, blocksreleased = 0;
442 int error = 0, allerror = 0;
445 * Calculate index in current block of last
446 * block to be kept. -1 indicates the entire
447 * block so we need not calculate the index.
450 for (i = SINGLE; i < level; i++)
451 factor *= NINDIR(fs);
455 nblocks = btodb(fs->fs_bsize);
457 * Get buffer of block pointers, zero those entries corresponding
458 * to blocks to be free'd, and update on disk copy first. Since
459 * double(triple) indirect before single(double) indirect, calls
460 * to bmap on these blocks will fail. However, we already have
461 * the on disk address, so we have to set the bio_offset field
462 * explicitly instead of letting bread do everything for us.
465 bp = getblk(vp, lblktodoff(fs, lbn), (int)fs->fs_bsize, 0, 0);
466 if ((bp->b_flags & B_CACHE) == 0) {
467 bp->b_flags |= B_READ;
468 bp->b_flags &= ~(B_ERROR|B_INVAL);
469 if (bp->b_bcount > bp->b_bufsize)
470 panic("ffs_indirtrunc: bad buffer size");
471 bp->b_bio2.bio_offset = dbtodoff(fs, dbn);
472 vfs_busy_pages(bp, 0);
474 * Access the block device layer using the device vnode
475 * and the translated block number (bio2) instead of the
476 * file vnode (vp) and logical block number (bio1).
478 * Even though we are bypassing the vnode layer, we still
479 * want the vnode state to indicate that an I/O on its behalf
482 bio_start_transaction(&bp->b_bio1, &vp->v_track_read);
483 vn_strategy(ip->i_devvp, &bp->b_bio2);
492 bap = (ufs_daddr_t *)bp->b_data;
494 MALLOC(copy, ufs_daddr_t *, fs->fs_bsize, M_TEMP, M_WAITOK);
495 bcopy((caddr_t)bap, (caddr_t)copy, (uint)fs->fs_bsize);
496 bzero((caddr_t)&bap[last + 1],
497 (uint)(NINDIR(fs) - (last + 1)) * sizeof (ufs_daddr_t));
498 if (DOINGASYNC(vp)) {
509 * Recursively free totally unused blocks.
511 for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
512 i--, nlbn += factor) {
516 if (level > SINGLE) {
517 if ((error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
518 (ufs_daddr_t)-1, level - 1, &blkcount)) != 0)
520 blocksreleased += blkcount;
522 ffs_blkfree(ip, nb, fs->fs_bsize);
523 blocksreleased += nblocks;
527 * Recursively free last partial block.
529 if (level > SINGLE && lastbn >= 0) {
530 last = lastbn % factor;
533 error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
534 last, level - 1, &blkcount);
537 blocksreleased += blkcount;
543 bp->b_flags |= B_INVAL | B_NOCACHE;
547 *countp = blocksreleased;