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 * 4. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * @(#)ffs_inode.c 8.13 (Berkeley) 4/21/95
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
35 #include "opt_quota.h"
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/mount.h>
43 #include <sys/vnode.h>
44 #include <sys/malloc.h>
45 #include <sys/resourcevar.h>
46 #include <sys/vmmeter.h>
50 #include <vm/vm_extern.h>
51 #include <vm/vm_object.h>
53 #include <ufs/ufs/extattr.h>
54 #include <ufs/ufs/quota.h>
55 #include <ufs/ufs/ufsmount.h>
56 #include <ufs/ufs/inode.h>
57 #include <ufs/ufs/ufs_extern.h>
59 #include <ufs/ffs/fs.h>
60 #include <ufs/ffs/ffs_extern.h>
62 static int ffs_indirtrunc(struct inode *, ufs2_daddr_t, ufs2_daddr_t,
63 ufs2_daddr_t, int, ufs2_daddr_t *);
66 * Update the access, modified, and inode change times as specified by the
67 * IN_ACCESS, IN_UPDATE, and IN_CHANGE flags respectively. Write the inode
68 * to disk if the IN_MODIFIED flag is set (it may be set initially, or by
69 * the timestamp update). The IN_LAZYMOD flag is set to force a write
70 * later if not now. The IN_LAZYACCESS is set instead of IN_MODIFIED if the fs
71 * is currently being suspended (or is suspended) and vnode has been accessed.
72 * If we write now, then clear IN_MODIFIED, IN_LAZYACCESS and IN_LAZYMOD to
73 * reflect the presumably successful write, and if waitfor is set, then wait
74 * for the write to complete.
77 ffs_update(vp, waitfor)
86 ASSERT_VOP_ELOCKED(vp, "ffs_update");
89 if ((ip->i_flag & IN_MODIFIED) == 0 && waitfor == 0)
91 ip->i_flag &= ~(IN_LAZYACCESS | IN_LAZYMOD | IN_MODIFIED);
93 if (fs->fs_ronly && ip->i_ump->um_fsckpid == 0)
96 * If we are updating a snapshot and another process is currently
97 * writing the buffer containing the inode for this snapshot then
98 * a deadlock can occur when it tries to check the snapshot to see
99 * if that block needs to be copied. Thus when updating a snapshot
100 * we check to see if the buffer is already locked, and if it is
101 * we drop the snapshot lock until the buffer has been written
102 * and is available to us. We have to grab a reference to the
103 * snapshot vnode to prevent it from being removed while we are
104 * waiting for the buffer.
108 flags = GB_LOCK_NOWAIT;
110 error = breadn_flags(ip->i_devvp,
111 fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
112 (int) fs->fs_bsize, 0, 0, 0, NOCRED, flags, &bp);
114 if (error != EBUSY) {
118 KASSERT((IS_SNAPSHOT(ip)), ("EBUSY from non-snapshot"));
120 * Wait for our inode block to become available.
122 * Hold a reference to the vnode to protect against
123 * ffs_snapgone(). Since we hold a reference, it can only
124 * get reclaimed (VI_DOOMED flag) in a forcible downgrade
125 * or unmount. For an unmount, the entire filesystem will be
126 * gone, so we cannot attempt to touch anything associated
127 * with it while the vnode is unlocked; all we can do is
128 * pause briefly and try again. If when we relock the vnode
129 * we discover that it has been reclaimed, updating it is no
130 * longer necessary and we can just return an error.
135 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
137 if ((vp->v_iflag & VI_DOOMED) != 0)
141 if (DOINGSOFTDEP(vp))
142 softdep_update_inodeblock(ip, bp, waitfor);
143 else if (ip->i_effnlink != ip->i_nlink)
144 panic("ffs_update: bad link cnt");
145 if (ip->i_ump->um_fstype == UFS1)
146 *((struct ufs1_dinode *)bp->b_data +
147 ino_to_fsbo(fs, ip->i_number)) = *ip->i_din1;
149 *((struct ufs2_dinode *)bp->b_data +
150 ino_to_fsbo(fs, ip->i_number)) = *ip->i_din2;
151 if ((waitfor && !DOINGASYNC(vp)) ||
152 (vm_page_count_severe() || buf_dirty_count_severe())) {
155 if (bp->b_bufsize == fs->fs_bsize)
156 bp->b_flags |= B_CLUSTEROK;
163 #define SINGLE 0 /* index of single indirect block */
164 #define DOUBLE 1 /* index of double indirect block */
165 #define TRIPLE 2 /* index of triple indirect block */
167 * Truncate the inode ip to at most length size, freeing the
171 ffs_truncate(vp, length, flags, cred, td)
179 ufs2_daddr_t bn, lbn, lastblock, lastiblock[NIADDR], indir_lbn[NIADDR];
180 ufs2_daddr_t oldblks[NDADDR + NIADDR], newblks[NDADDR + NIADDR];
181 ufs2_daddr_t count, blocksreleased = 0, datablocks;
185 struct ufsmount *ump;
186 int softdeptrunc, journaltrunc;
187 int needextclean, extblocks;
188 int offset, size, level, nblocks;
189 int i, error, allerror;
197 ASSERT_VOP_LOCKED(vp, "ffs_truncate");
201 if (length > fs->fs_maxfilesize)
204 error = getinoquota(ip);
209 * Historically clients did not have to specify which data
210 * they were truncating. So, if not specified, we assume
211 * traditional behavior, e.g., just the normal data.
213 if ((flags & (IO_EXT | IO_NORMAL)) == 0)
215 if (!DOINGSOFTDEP(vp) && !DOINGASYNC(vp))
218 * If we are truncating the extended-attributes, and cannot
219 * do it with soft updates, then do it slowly here. If we are
220 * truncating both the extended attributes and the file contents
221 * (e.g., the file is being unlinked), then pick it off with
222 * soft updates below.
227 journaltrunc = DOINGSUJ(vp);
228 if (journaltrunc == 0 && DOINGSOFTDEP(vp) && length == 0)
229 softdeptrunc = !softdep_slowdown(vp);
231 datablocks = DIP(ip, i_blocks);
232 if (fs->fs_magic == FS_UFS2_MAGIC && ip->i_din2->di_extsize > 0) {
233 extblocks = btodb(fragroundup(fs, ip->i_din2->di_extsize));
234 datablocks -= extblocks;
236 if ((flags & IO_EXT) && extblocks > 0) {
238 panic("ffs_truncate: partial trunc of extdata");
239 if (softdeptrunc || journaltrunc) {
240 if ((flags & IO_NORMAL) == 0)
244 if ((error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0)
247 (void) chkdq(ip, -extblocks, NOCRED, 0);
249 vinvalbuf(vp, V_ALT, 0, 0);
251 OFF_TO_IDX(lblktosize(fs, -extblocks)), 0);
252 osize = ip->i_din2->di_extsize;
253 ip->i_din2->di_blocks -= extblocks;
254 ip->i_din2->di_extsize = 0;
255 for (i = 0; i < NXADDR; i++) {
256 oldblks[i] = ip->i_din2->di_extb[i];
257 ip->i_din2->di_extb[i] = 0;
259 ip->i_flag |= IN_CHANGE;
260 if ((error = ffs_update(vp, !DOINGASYNC(vp))))
262 for (i = 0; i < NXADDR; i++) {
265 ffs_blkfree(ump, fs, ip->i_devvp, oldblks[i],
266 sblksize(fs, osize, i), ip->i_number,
271 if ((flags & IO_NORMAL) == 0)
273 if (vp->v_type == VLNK &&
274 (ip->i_size < vp->v_mount->mnt_maxsymlinklen ||
278 panic("ffs_truncate: partial truncate of symlink");
280 bzero(SHORTLINK(ip), (u_int)ip->i_size);
282 DIP_SET(ip, i_size, 0);
283 ip->i_flag |= IN_CHANGE | IN_UPDATE;
286 return (ffs_update(vp, !DOINGASYNC(vp)));
288 if (ip->i_size == length) {
289 ip->i_flag |= IN_CHANGE | IN_UPDATE;
292 return (ffs_update(vp, 0));
295 panic("ffs_truncate: read-only filesystem");
298 vp->v_lasta = vp->v_clen = vp->v_cstart = vp->v_lastw = 0;
301 * Lengthen the size of the file. We must ensure that the
302 * last byte of the file is allocated. Since the smallest
303 * value of osize is 0, length will be at least 1.
305 if (osize < length) {
306 vnode_pager_setsize(vp, length);
308 error = UFS_BALLOC(vp, length - 1, 1, cred, flags, &bp);
310 vnode_pager_setsize(vp, osize);
314 DIP_SET(ip, i_size, length);
315 if (bp->b_bufsize == fs->fs_bsize)
316 bp->b_flags |= B_CLUSTEROK;
319 else if (DOINGASYNC(vp))
323 ip->i_flag |= IN_CHANGE | IN_UPDATE;
324 return (ffs_update(vp, !DOINGASYNC(vp)));
326 if (DOINGSOFTDEP(vp)) {
327 if (softdeptrunc == 0 && journaltrunc == 0) {
329 * If a file is only partially truncated, then
330 * we have to clean up the data structures
331 * describing the allocation past the truncation
332 * point. Finding and deallocating those structures
333 * is a lot of work. Since partial truncation occurs
334 * rarely, we solve the problem by syncing the file
335 * so that it will have no data structures left.
337 if ((error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0)
340 flags = IO_NORMAL | (needextclean ? IO_EXT: 0);
342 softdep_journal_freeblocks(ip, cred, length,
345 softdep_setup_freeblocks(ip, length, flags);
346 ASSERT_VOP_LOCKED(vp, "ffs_truncate1");
347 if (journaltrunc == 0) {
348 ip->i_flag |= IN_CHANGE | IN_UPDATE;
349 error = ffs_update(vp, 0);
355 * Shorten the size of the file. If the file is not being
356 * truncated to a block boundary, the contents of the
357 * partial block following the end of the file must be
358 * zero'ed in case it ever becomes accessible again because
359 * of subsequent file growth. Directories however are not
360 * zero'ed as they should grow back initialized to empty.
362 offset = blkoff(fs, length);
365 DIP_SET(ip, i_size, length);
367 lbn = lblkno(fs, length);
369 error = UFS_BALLOC(vp, length - 1, 1, cred, flags, &bp);
373 * When we are doing soft updates and the UFS_BALLOC
374 * above fills in a direct block hole with a full sized
375 * block that will be truncated down to a fragment below,
376 * we must flush out the block dependency with an FSYNC
377 * so that we do not get a soft updates inconsistency
378 * when we create the fragment below.
380 if (DOINGSOFTDEP(vp) && lbn < NDADDR &&
381 fragroundup(fs, blkoff(fs, length)) < fs->fs_bsize &&
382 (error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0)
385 DIP_SET(ip, i_size, length);
386 size = blksize(fs, ip, lbn);
387 if (vp->v_type != VDIR)
388 bzero((char *)bp->b_data + offset,
389 (u_int)(size - offset));
390 /* Kirk's code has reallocbuf(bp, size, 1) here */
392 if (bp->b_bufsize == fs->fs_bsize)
393 bp->b_flags |= B_CLUSTEROK;
396 else if (DOINGASYNC(vp))
402 * Calculate index into inode's block list of
403 * last direct and indirect blocks (if any)
404 * which we want to keep. Lastblock is -1 when
405 * the file is truncated to 0.
407 lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1;
408 lastiblock[SINGLE] = lastblock - NDADDR;
409 lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
410 lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
411 nblocks = btodb(fs->fs_bsize);
413 * Update file and block pointers on disk before we start freeing
414 * blocks. If we crash before free'ing blocks below, the blocks
415 * will be returned to the free list. lastiblock values are also
416 * normalized to -1 for calls to ffs_indirtrunc below.
418 for (level = TRIPLE; level >= SINGLE; level--) {
419 oldblks[NDADDR + level] = DIP(ip, i_ib[level]);
420 if (lastiblock[level] < 0) {
421 DIP_SET(ip, i_ib[level], 0);
422 lastiblock[level] = -1;
425 for (i = 0; i < NDADDR; i++) {
426 oldblks[i] = DIP(ip, i_db[i]);
428 DIP_SET(ip, i_db[i], 0);
430 ip->i_flag |= IN_CHANGE | IN_UPDATE;
431 allerror = ffs_update(vp, !DOINGASYNC(vp));
434 * Having written the new inode to disk, save its new configuration
435 * and put back the old block pointers long enough to process them.
436 * Note that we save the new block configuration so we can check it
439 for (i = 0; i < NDADDR; i++) {
440 newblks[i] = DIP(ip, i_db[i]);
441 DIP_SET(ip, i_db[i], oldblks[i]);
443 for (i = 0; i < NIADDR; i++) {
444 newblks[NDADDR + i] = DIP(ip, i_ib[i]);
445 DIP_SET(ip, i_ib[i], oldblks[NDADDR + i]);
448 DIP_SET(ip, i_size, osize);
450 error = vtruncbuf(vp, cred, td, length, fs->fs_bsize);
451 if (error && (allerror == 0))
455 * Indirect blocks first.
457 indir_lbn[SINGLE] = -NDADDR;
458 indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
459 indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
460 for (level = TRIPLE; level >= SINGLE; level--) {
461 bn = DIP(ip, i_ib[level]);
463 error = ffs_indirtrunc(ip, indir_lbn[level],
464 fsbtodb(fs, bn), lastiblock[level], level, &count);
467 blocksreleased += count;
468 if (lastiblock[level] < 0) {
469 DIP_SET(ip, i_ib[level], 0);
470 ffs_blkfree(ump, fs, ip->i_devvp, bn,
471 fs->fs_bsize, ip->i_number,
473 blocksreleased += nblocks;
476 if (lastiblock[level] >= 0)
481 * All whole direct blocks or frags.
483 for (i = NDADDR - 1; i > lastblock; i--) {
486 bn = DIP(ip, i_db[i]);
489 DIP_SET(ip, i_db[i], 0);
490 bsize = blksize(fs, ip, i);
491 ffs_blkfree(ump, fs, ip->i_devvp, bn, bsize, ip->i_number,
493 blocksreleased += btodb(bsize);
499 * Finally, look for a change in size of the
500 * last direct block; release any frags.
502 bn = DIP(ip, i_db[lastblock]);
504 long oldspace, newspace;
507 * Calculate amount of space we're giving
508 * back as old block size minus new block size.
510 oldspace = blksize(fs, ip, lastblock);
512 DIP_SET(ip, i_size, length);
513 newspace = blksize(fs, ip, lastblock);
515 panic("ffs_truncate: newspace");
516 if (oldspace - newspace > 0) {
518 * Block number of space to be free'd is
519 * the old block # plus the number of frags
520 * required for the storage we're keeping.
522 bn += numfrags(fs, newspace);
523 ffs_blkfree(ump, fs, ip->i_devvp, bn,
524 oldspace - newspace, ip->i_number, vp->v_type, NULL);
525 blocksreleased += btodb(oldspace - newspace);
530 for (level = SINGLE; level <= TRIPLE; level++)
531 if (newblks[NDADDR + level] != DIP(ip, i_ib[level]))
532 panic("ffs_truncate1");
533 for (i = 0; i < NDADDR; i++)
534 if (newblks[i] != DIP(ip, i_db[i]))
535 panic("ffs_truncate2");
538 (fs->fs_magic != FS_UFS2_MAGIC || ip->i_din2->di_extsize == 0) &&
539 (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
540 panic("ffs_truncate3");
542 #endif /* INVARIANTS */
544 * Put back the real size.
547 DIP_SET(ip, i_size, length);
548 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) - blocksreleased);
550 if (DIP(ip, i_blocks) < 0) /* sanity */
551 DIP_SET(ip, i_blocks, 0);
552 ip->i_flag |= IN_CHANGE;
554 (void) chkdq(ip, -blocksreleased, NOCRED, 0);
560 softdep_journal_freeblocks(ip, cred, length, IO_EXT);
562 softdep_setup_freeblocks(ip, length, IO_EXT);
563 return (ffs_update(vp, !DOINGASYNC(vp)));
567 * Release blocks associated with the inode ip and stored in the indirect
568 * block bn. Blocks are free'd in LIFO order up to (but not including)
569 * lastbn. If level is greater than SINGLE, the block is an indirect block
570 * and recursive calls to indirtrunc must be used to cleanse other indirect
574 ffs_indirtrunc(ip, lbn, dbn, lastbn, level, countp)
576 ufs2_daddr_t lbn, lastbn;
579 ufs2_daddr_t *countp;
582 struct fs *fs = ip->i_fs;
585 int i, nblocks, error = 0, allerror = 0;
586 ufs2_daddr_t nb, nlbn, last;
587 ufs2_daddr_t blkcount, factor, blocksreleased = 0;
588 ufs1_daddr_t *bap1 = NULL;
589 ufs2_daddr_t *bap2 = NULL;
590 # define BAP(ip, i) (((ip)->i_ump->um_fstype == UFS1) ? bap1[i] : bap2[i])
593 * Calculate index in current block of last
594 * block to be kept. -1 indicates the entire
595 * block so we need not calculate the index.
597 factor = lbn_offset(fs, level);
601 nblocks = btodb(fs->fs_bsize);
603 * Get buffer of block pointers, zero those entries corresponding
604 * to blocks to be free'd, and update on disk copy first. Since
605 * double(triple) indirect before single(double) indirect, calls
606 * to bmap on these blocks will fail. However, we already have
607 * the on disk address, so we have to set the b_blkno field
608 * explicitly instead of letting bread do everything for us.
611 bp = getblk(vp, lbn, (int)fs->fs_bsize, 0, 0, 0);
612 if ((bp->b_flags & B_CACHE) == 0) {
613 curthread->td_ru.ru_inblock++; /* pay for read */
614 bp->b_iocmd = BIO_READ;
615 bp->b_flags &= ~B_INVAL;
616 bp->b_ioflags &= ~BIO_ERROR;
617 if (bp->b_bcount > bp->b_bufsize)
618 panic("ffs_indirtrunc: bad buffer size");
620 vfs_busy_pages(bp, 0);
621 bp->b_iooffset = dbtob(bp->b_blkno);
631 if (ip->i_ump->um_fstype == UFS1)
632 bap1 = (ufs1_daddr_t *)bp->b_data;
634 bap2 = (ufs2_daddr_t *)bp->b_data;
636 copy = malloc(fs->fs_bsize, M_TEMP, M_WAITOK);
637 bcopy((caddr_t)bp->b_data, copy, (u_int)fs->fs_bsize);
638 for (i = last + 1; i < NINDIR(fs); i++)
639 if (ip->i_ump->um_fstype == UFS1)
643 if (DOINGASYNC(vp)) {
650 if (ip->i_ump->um_fstype == UFS1)
651 bap1 = (ufs1_daddr_t *)copy;
653 bap2 = (ufs2_daddr_t *)copy;
657 * Recursively free totally unused blocks.
659 for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
660 i--, nlbn += factor) {
664 if (level > SINGLE) {
665 if ((error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
666 (ufs2_daddr_t)-1, level - 1, &blkcount)) != 0)
668 blocksreleased += blkcount;
670 ffs_blkfree(ip->i_ump, fs, ip->i_devvp, nb, fs->fs_bsize,
671 ip->i_number, vp->v_type, NULL);
672 blocksreleased += nblocks;
676 * Recursively free last partial block.
678 if (level > SINGLE && lastbn >= 0) {
679 last = lastbn % factor;
682 error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
683 last, level - 1, &blkcount);
686 blocksreleased += blkcount;
692 bp->b_flags |= B_INVAL | B_NOCACHE;
696 *countp = blocksreleased;
701 ffs_rdonly(struct inode *ip)
704 return (ip->i_ump->um_fs->fs_ronly != 0);