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>
41 #include <sys/malloc.h>
42 #include <sys/mount.h>
44 #include <sys/racct.h>
45 #include <sys/random.h>
46 #include <sys/resourcevar.h>
47 #include <sys/rwlock.h>
49 #include <sys/vmmeter.h>
50 #include <sys/vnode.h>
53 #include <vm/vm_extern.h>
54 #include <vm/vm_object.h>
56 #include <ufs/ufs/extattr.h>
57 #include <ufs/ufs/quota.h>
58 #include <ufs/ufs/ufsmount.h>
59 #include <ufs/ufs/inode.h>
60 #include <ufs/ufs/ufs_extern.h>
62 #include <ufs/ffs/fs.h>
63 #include <ufs/ffs/ffs_extern.h>
65 static int ffs_indirtrunc(struct inode *, ufs2_daddr_t, ufs2_daddr_t,
66 ufs2_daddr_t, int, ufs2_daddr_t *);
69 * Update the access, modified, and inode change times as specified by the
70 * IN_ACCESS, IN_UPDATE, and IN_CHANGE flags respectively. Write the inode
71 * to disk if the IN_MODIFIED flag is set (it may be set initially, or by
72 * the timestamp update). The IN_LAZYMOD flag is set to force a write
73 * later if not now. The IN_LAZYACCESS is set instead of IN_MODIFIED if the fs
74 * is currently being suspended (or is suspended) and vnode has been accessed.
75 * If we write now, then clear IN_MODIFIED, IN_LAZYACCESS and IN_LAZYMOD to
76 * reflect the presumably successful write, and if waitfor is set, then wait
77 * for the write to complete.
80 ffs_update(vp, waitfor)
89 ASSERT_VOP_ELOCKED(vp, "ffs_update");
92 if ((ip->i_flag & IN_MODIFIED) == 0 && waitfor == 0)
94 ip->i_flag &= ~(IN_LAZYACCESS | IN_LAZYMOD | IN_MODIFIED);
96 if (fs->fs_ronly && ITOUMP(ip)->um_fsckpid == 0)
99 * If we are updating a snapshot and another process is currently
100 * writing the buffer containing the inode for this snapshot then
101 * a deadlock can occur when it tries to check the snapshot to see
102 * if that block needs to be copied. Thus when updating a snapshot
103 * we check to see if the buffer is already locked, and if it is
104 * we drop the snapshot lock until the buffer has been written
105 * and is available to us. We have to grab a reference to the
106 * snapshot vnode to prevent it from being removed while we are
107 * waiting for the buffer.
111 flags = GB_LOCK_NOWAIT;
113 error = breadn_flags(ITODEVVP(ip),
114 fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
115 (int) fs->fs_bsize, 0, 0, 0, NOCRED, flags, &bp);
119 KASSERT((IS_SNAPSHOT(ip)), ("EBUSY from non-snapshot"));
121 * Wait for our inode block to become available.
123 * Hold a reference to the vnode to protect against
124 * ffs_snapgone(). Since we hold a reference, it can only
125 * get reclaimed (VI_DOOMED flag) in a forcible downgrade
126 * or unmount. For an unmount, the entire filesystem will be
127 * gone, so we cannot attempt to touch anything associated
128 * with it while the vnode is unlocked; all we can do is
129 * pause briefly and try again. If when we relock the vnode
130 * we discover that it has been reclaimed, updating it is no
131 * longer necessary and we can just return an error.
136 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
138 if ((vp->v_iflag & VI_DOOMED) != 0)
142 if (DOINGSOFTDEP(vp))
143 softdep_update_inodeblock(ip, bp, waitfor);
144 else if (ip->i_effnlink != ip->i_nlink)
145 panic("ffs_update: bad link cnt");
147 *((struct ufs1_dinode *)bp->b_data +
148 ino_to_fsbo(fs, ip->i_number)) = *ip->i_din1;
149 /* XXX: FIX? The entropy here is desirable, but the harvesting may be expensive */
150 random_harvest_queue(&(ip->i_din1), sizeof(ip->i_din1), 1, RANDOM_FS_ATIME);
152 *((struct ufs2_dinode *)bp->b_data +
153 ino_to_fsbo(fs, ip->i_number)) = *ip->i_din2;
154 /* XXX: FIX? The entropy here is desirable, but the harvesting may be expensive */
155 random_harvest_queue(&(ip->i_din2), sizeof(ip->i_din2), 1, RANDOM_FS_ATIME);
159 else if (vm_page_count_severe() || buf_dirty_count_severe()) {
163 if (bp->b_bufsize == fs->fs_bsize)
164 bp->b_flags |= B_CLUSTEROK;
171 #define SINGLE 0 /* index of single indirect block */
172 #define DOUBLE 1 /* index of double indirect block */
173 #define TRIPLE 2 /* index of triple indirect block */
175 * Truncate the inode ip to at most length size, freeing the
179 ffs_truncate(vp, length, flags, cred)
186 ufs2_daddr_t bn, lbn, lastblock, lastiblock[NIADDR], indir_lbn[NIADDR];
187 ufs2_daddr_t oldblks[NDADDR + NIADDR], newblks[NDADDR + NIADDR];
188 ufs2_daddr_t count, blocksreleased = 0, datablocks, blkno;
192 struct ufsmount *ump;
193 int softdeptrunc, journaltrunc;
194 int needextclean, extblocks;
195 int offset, size, level, nblocks;
196 int i, error, allerror, indiroff, waitforupdate;
200 ump = VFSTOUFS(vp->v_mount);
204 ASSERT_VOP_LOCKED(vp, "ffs_truncate");
208 if (length > fs->fs_maxfilesize)
211 error = getinoquota(ip);
216 * Historically clients did not have to specify which data
217 * they were truncating. So, if not specified, we assume
218 * traditional behavior, e.g., just the normal data.
220 if ((flags & (IO_EXT | IO_NORMAL)) == 0)
222 if (!DOINGSOFTDEP(vp) && !DOINGASYNC(vp))
224 waitforupdate = (flags & IO_SYNC) != 0 || !DOINGASYNC(vp);
226 * If we are truncating the extended-attributes, and cannot
227 * do it with soft updates, then do it slowly here. If we are
228 * truncating both the extended attributes and the file contents
229 * (e.g., the file is being unlinked), then pick it off with
230 * soft updates below.
235 journaltrunc = DOINGSUJ(vp);
236 if (journaltrunc == 0 && DOINGSOFTDEP(vp) && length == 0)
237 softdeptrunc = !softdep_slowdown(vp);
239 datablocks = DIP(ip, i_blocks);
240 if (fs->fs_magic == FS_UFS2_MAGIC && ip->i_din2->di_extsize > 0) {
241 extblocks = btodb(fragroundup(fs, ip->i_din2->di_extsize));
242 datablocks -= extblocks;
244 if ((flags & IO_EXT) && extblocks > 0) {
246 panic("ffs_truncate: partial trunc of extdata");
247 if (softdeptrunc || journaltrunc) {
248 if ((flags & IO_NORMAL) == 0)
252 if ((error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0)
255 (void) chkdq(ip, -extblocks, NOCRED, 0);
257 vinvalbuf(vp, V_ALT, 0, 0);
259 OFF_TO_IDX(lblktosize(fs, -extblocks)), 0);
260 osize = ip->i_din2->di_extsize;
261 ip->i_din2->di_blocks -= extblocks;
262 ip->i_din2->di_extsize = 0;
263 for (i = 0; i < NXADDR; i++) {
264 oldblks[i] = ip->i_din2->di_extb[i];
265 ip->i_din2->di_extb[i] = 0;
267 ip->i_flag |= IN_CHANGE;
268 if ((error = ffs_update(vp, waitforupdate)))
270 for (i = 0; i < NXADDR; i++) {
273 ffs_blkfree(ump, fs, ITODEVVP(ip), oldblks[i],
274 sblksize(fs, osize, i), ip->i_number,
279 if ((flags & IO_NORMAL) == 0)
281 if (vp->v_type == VLNK &&
282 (ip->i_size < vp->v_mount->mnt_maxsymlinklen ||
286 panic("ffs_truncate: partial truncate of symlink");
288 bzero(SHORTLINK(ip), (u_int)ip->i_size);
290 DIP_SET(ip, i_size, 0);
291 ip->i_flag |= IN_CHANGE | IN_UPDATE;
294 return (ffs_update(vp, waitforupdate));
296 if (ip->i_size == length) {
297 ip->i_flag |= IN_CHANGE | IN_UPDATE;
300 return (ffs_update(vp, 0));
303 panic("ffs_truncate: read-only filesystem");
306 vp->v_lasta = vp->v_clen = vp->v_cstart = vp->v_lastw = 0;
309 * Lengthen the size of the file. We must ensure that the
310 * last byte of the file is allocated. Since the smallest
311 * value of osize is 0, length will be at least 1.
313 if (osize < length) {
314 vnode_pager_setsize(vp, length);
316 error = UFS_BALLOC(vp, length - 1, 1, cred, flags, &bp);
318 vnode_pager_setsize(vp, osize);
322 DIP_SET(ip, i_size, length);
323 if (bp->b_bufsize == fs->fs_bsize)
324 bp->b_flags |= B_CLUSTEROK;
327 else if (DOINGASYNC(vp))
331 ip->i_flag |= IN_CHANGE | IN_UPDATE;
332 return (ffs_update(vp, waitforupdate));
335 * Lookup block number for a given offset. Zero length files
336 * have no blocks, so return a blkno of -1.
338 lbn = lblkno(fs, length - 1);
341 } else if (lbn < NDADDR) {
342 blkno = DIP(ip, i_db[lbn]);
344 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn), fs->fs_bsize,
345 cred, BA_METAONLY, &bp);
348 indiroff = (lbn - NDADDR) % NINDIR(fs);
350 blkno = ((ufs1_daddr_t *)(bp->b_data))[indiroff];
352 blkno = ((ufs2_daddr_t *)(bp->b_data))[indiroff];
354 * If the block number is non-zero, then the indirect block
355 * must have been previously allocated and need not be written.
356 * If the block number is zero, then we may have allocated
357 * the indirect block and hence need to write it out.
361 else if (flags & IO_SYNC)
367 * If the block number at the new end of the file is zero,
368 * then we must allocate it to ensure that the last block of
369 * the file is allocated. Soft updates does not handle this
370 * case, so here we have to clean up the soft updates data
371 * structures describing the allocation past the truncation
372 * point. Finding and deallocating those structures is a lot of
373 * work. Since partial truncation with a hole at the end occurs
374 * rarely, we solve the problem by syncing the file so that it
375 * will have no soft updates data structures left.
377 if (blkno == 0 && (error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0)
379 if (blkno != 0 && DOINGSOFTDEP(vp)) {
380 if (softdeptrunc == 0 && journaltrunc == 0) {
382 * If soft updates cannot handle this truncation,
383 * clean up soft dependency data structures and
384 * fall through to the synchronous truncation.
386 if ((error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0)
389 flags = IO_NORMAL | (needextclean ? IO_EXT: 0);
391 softdep_journal_freeblocks(ip, cred, length,
394 softdep_setup_freeblocks(ip, length, flags);
395 ASSERT_VOP_LOCKED(vp, "ffs_truncate1");
396 if (journaltrunc == 0) {
397 ip->i_flag |= IN_CHANGE | IN_UPDATE;
398 error = ffs_update(vp, 0);
404 * Shorten the size of the file. If the last block of the
405 * shortened file is unallocated, we must allocate it.
406 * Additionally, if the file is not being truncated to a
407 * block boundary, the contents of the partial block
408 * following the end of the file must be zero'ed in
409 * case it ever becomes accessible again because of
410 * subsequent file growth. Directories however are not
411 * zero'ed as they should grow back initialized to empty.
413 offset = blkoff(fs, length);
414 if (blkno != 0 && offset == 0) {
416 DIP_SET(ip, i_size, length);
418 lbn = lblkno(fs, length);
420 error = UFS_BALLOC(vp, length - 1, 1, cred, flags, &bp);
424 * When we are doing soft updates and the UFS_BALLOC
425 * above fills in a direct block hole with a full sized
426 * block that will be truncated down to a fragment below,
427 * we must flush out the block dependency with an FSYNC
428 * so that we do not get a soft updates inconsistency
429 * when we create the fragment below.
431 if (DOINGSOFTDEP(vp) && lbn < NDADDR &&
432 fragroundup(fs, blkoff(fs, length)) < fs->fs_bsize &&
433 (error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0)
436 DIP_SET(ip, i_size, length);
437 size = blksize(fs, ip, lbn);
438 if (vp->v_type != VDIR && offset != 0)
439 bzero((char *)bp->b_data + offset,
440 (u_int)(size - offset));
441 /* Kirk's code has reallocbuf(bp, size, 1) here */
443 if (bp->b_bufsize == fs->fs_bsize)
444 bp->b_flags |= B_CLUSTEROK;
447 else if (DOINGASYNC(vp))
453 * Calculate index into inode's block list of
454 * last direct and indirect blocks (if any)
455 * which we want to keep. Lastblock is -1 when
456 * the file is truncated to 0.
458 lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1;
459 lastiblock[SINGLE] = lastblock - NDADDR;
460 lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
461 lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
462 nblocks = btodb(fs->fs_bsize);
464 * Update file and block pointers on disk before we start freeing
465 * blocks. If we crash before free'ing blocks below, the blocks
466 * will be returned to the free list. lastiblock values are also
467 * normalized to -1 for calls to ffs_indirtrunc below.
469 for (level = TRIPLE; level >= SINGLE; level--) {
470 oldblks[NDADDR + level] = DIP(ip, i_ib[level]);
471 if (lastiblock[level] < 0) {
472 DIP_SET(ip, i_ib[level], 0);
473 lastiblock[level] = -1;
476 for (i = 0; i < NDADDR; i++) {
477 oldblks[i] = DIP(ip, i_db[i]);
479 DIP_SET(ip, i_db[i], 0);
481 ip->i_flag |= IN_CHANGE | IN_UPDATE;
482 allerror = ffs_update(vp, waitforupdate);
485 * Having written the new inode to disk, save its new configuration
486 * and put back the old block pointers long enough to process them.
487 * Note that we save the new block configuration so we can check it
490 for (i = 0; i < NDADDR; i++) {
491 newblks[i] = DIP(ip, i_db[i]);
492 DIP_SET(ip, i_db[i], oldblks[i]);
494 for (i = 0; i < NIADDR; i++) {
495 newblks[NDADDR + i] = DIP(ip, i_ib[i]);
496 DIP_SET(ip, i_ib[i], oldblks[NDADDR + i]);
499 DIP_SET(ip, i_size, osize);
501 error = vtruncbuf(vp, cred, length, fs->fs_bsize);
502 if (error && (allerror == 0))
506 * Indirect blocks first.
508 indir_lbn[SINGLE] = -NDADDR;
509 indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
510 indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
511 for (level = TRIPLE; level >= SINGLE; level--) {
512 bn = DIP(ip, i_ib[level]);
514 error = ffs_indirtrunc(ip, indir_lbn[level],
515 fsbtodb(fs, bn), lastiblock[level], level, &count);
518 blocksreleased += count;
519 if (lastiblock[level] < 0) {
520 DIP_SET(ip, i_ib[level], 0);
521 ffs_blkfree(ump, fs, ump->um_devvp, bn,
522 fs->fs_bsize, ip->i_number,
524 blocksreleased += nblocks;
527 if (lastiblock[level] >= 0)
532 * All whole direct blocks or frags.
534 for (i = NDADDR - 1; i > lastblock; i--) {
537 bn = DIP(ip, i_db[i]);
540 DIP_SET(ip, i_db[i], 0);
541 bsize = blksize(fs, ip, i);
542 ffs_blkfree(ump, fs, ump->um_devvp, bn, bsize, ip->i_number,
544 blocksreleased += btodb(bsize);
550 * Finally, look for a change in size of the
551 * last direct block; release any frags.
553 bn = DIP(ip, i_db[lastblock]);
555 long oldspace, newspace;
558 * Calculate amount of space we're giving
559 * back as old block size minus new block size.
561 oldspace = blksize(fs, ip, lastblock);
563 DIP_SET(ip, i_size, length);
564 newspace = blksize(fs, ip, lastblock);
566 panic("ffs_truncate: newspace");
567 if (oldspace - newspace > 0) {
569 * Block number of space to be free'd is
570 * the old block # plus the number of frags
571 * required for the storage we're keeping.
573 bn += numfrags(fs, newspace);
574 ffs_blkfree(ump, fs, ump->um_devvp, bn,
575 oldspace - newspace, ip->i_number, vp->v_type, NULL);
576 blocksreleased += btodb(oldspace - newspace);
581 for (level = SINGLE; level <= TRIPLE; level++)
582 if (newblks[NDADDR + level] != DIP(ip, i_ib[level]))
583 panic("ffs_truncate1");
584 for (i = 0; i < NDADDR; i++)
585 if (newblks[i] != DIP(ip, i_db[i]))
586 panic("ffs_truncate2");
589 (fs->fs_magic != FS_UFS2_MAGIC || ip->i_din2->di_extsize == 0) &&
590 (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
591 panic("ffs_truncate3");
593 #endif /* INVARIANTS */
595 * Put back the real size.
598 DIP_SET(ip, i_size, length);
599 if (DIP(ip, i_blocks) >= blocksreleased)
600 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) - blocksreleased);
602 DIP_SET(ip, i_blocks, 0);
603 ip->i_flag |= IN_CHANGE;
605 (void) chkdq(ip, -blocksreleased, NOCRED, 0);
611 softdep_journal_freeblocks(ip, cred, length, IO_EXT);
613 softdep_setup_freeblocks(ip, length, IO_EXT);
614 return (ffs_update(vp, waitforupdate));
618 * Release blocks associated with the inode ip and stored in the indirect
619 * block bn. Blocks are free'd in LIFO order up to (but not including)
620 * lastbn. If level is greater than SINGLE, the block is an indirect block
621 * and recursive calls to indirtrunc must be used to cleanse other indirect
625 ffs_indirtrunc(ip, lbn, dbn, lastbn, level, countp)
627 ufs2_daddr_t lbn, lastbn;
630 ufs2_daddr_t *countp;
636 int i, nblocks, error = 0, allerror = 0;
637 ufs2_daddr_t nb, nlbn, last;
638 ufs2_daddr_t blkcount, factor, blocksreleased = 0;
639 ufs1_daddr_t *bap1 = NULL;
640 ufs2_daddr_t *bap2 = NULL;
641 #define BAP(ip, i) (I_IS_UFS1(ip) ? bap1[i] : bap2[i])
646 * Calculate index in current block of last
647 * block to be kept. -1 indicates the entire
648 * block so we need not calculate the index.
650 factor = lbn_offset(fs, level);
654 nblocks = btodb(fs->fs_bsize);
656 * Get buffer of block pointers, zero those entries corresponding
657 * to blocks to be free'd, and update on disk copy first. Since
658 * double(triple) indirect before single(double) indirect, calls
659 * to bmap on these blocks will fail. However, we already have
660 * the on disk address, so we have to set the b_blkno field
661 * explicitly instead of letting bread do everything for us.
664 bp = getblk(vp, lbn, (int)fs->fs_bsize, 0, 0, 0);
665 if ((bp->b_flags & B_CACHE) == 0) {
669 racct_add_buf(curproc, bp, 0);
670 PROC_UNLOCK(curproc);
673 curthread->td_ru.ru_inblock++; /* pay for read */
674 bp->b_iocmd = BIO_READ;
675 bp->b_flags &= ~B_INVAL;
676 bp->b_ioflags &= ~BIO_ERROR;
677 if (bp->b_bcount > bp->b_bufsize)
678 panic("ffs_indirtrunc: bad buffer size");
680 vfs_busy_pages(bp, 0);
681 bp->b_iooffset = dbtob(bp->b_blkno);
692 bap1 = (ufs1_daddr_t *)bp->b_data;
694 bap2 = (ufs2_daddr_t *)bp->b_data;
696 copy = malloc(fs->fs_bsize, M_TEMP, M_WAITOK);
697 bcopy((caddr_t)bp->b_data, copy, (u_int)fs->fs_bsize);
698 for (i = last + 1; i < NINDIR(fs); i++)
703 if (DOINGASYNC(vp)) {
711 bap1 = (ufs1_daddr_t *)copy;
713 bap2 = (ufs2_daddr_t *)copy;
717 * Recursively free totally unused blocks.
719 for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
720 i--, nlbn += factor) {
724 if (level > SINGLE) {
725 if ((error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
726 (ufs2_daddr_t)-1, level - 1, &blkcount)) != 0)
728 blocksreleased += blkcount;
730 ffs_blkfree(ITOUMP(ip), fs, ITODEVVP(ip), nb, fs->fs_bsize,
731 ip->i_number, vp->v_type, NULL);
732 blocksreleased += nblocks;
736 * Recursively free last partial block.
738 if (level > SINGLE && lastbn >= 0) {
739 last = lastbn % factor;
742 error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
743 last, level - 1, &blkcount);
746 blocksreleased += blkcount;
752 bp->b_flags |= B_INVAL | B_NOCACHE;
756 *countp = blocksreleased;
761 ffs_rdonly(struct inode *ip)
764 return (ITOFS(ip)->fs_ronly != 0);