kernel - Fix issue in UFS related to new nvtruncbuf() API use
[dragonfly.git] / sys / vfs / ufs / ffs_inode.c
blob41bb3dd8e4786ee1c023b41bb127eaed9cf143ed
1 /*
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
7 * are met:
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
31 * SUCH DAMAGE.
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 1.24 2007/06/14 02:55:25 dillon Exp $
38 #include "opt_quota.h"
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/mount.h>
43 #include <sys/proc.h>
44 #include <sys/buf.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>
51 #include <vm/vm.h>
52 #include <vm/vm_extern.h>
54 #include "quota.h"
55 #include "ufsmount.h"
56 #include "inode.h"
57 #include "ufs_extern.h"
59 #include "fs.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.
76 int
77 ffs_update(struct vnode *vp, int waitfor)
79 struct fs *fs;
80 struct buf *bp;
81 struct inode *ip;
82 int error;
84 ufs_itimes(vp);
85 ip = VTOI(vp);
86 if ((ip->i_flag & IN_MODIFIED) == 0 && waitfor == 0)
87 return (0);
88 ip->i_flag &= ~(IN_LAZYMOD | IN_MODIFIED);
89 fs = ip->i_fs;
90 if (fs->fs_ronly)
91 return (0);
94 * The vnode type is usually set to VBAD if an unrecoverable I/O
95 * error has occured (such as when reading the inode). Clear the
96 * modified bits but do not write anything out in this case.
98 if (vp->v_type == VBAD)
99 return (0);
101 * Ensure that uid and gid are correct. This is a temporary
102 * fix until fsck has been changed to do the update.
104 if (fs->fs_inodefmt < FS_44INODEFMT) { /* XXX */
105 ip->i_din.di_ouid = ip->i_uid; /* XXX */
106 ip->i_din.di_ogid = ip->i_gid; /* XXX */
107 } /* XXX */
108 error = bread(ip->i_devvp,
109 fsbtodoff(fs, ino_to_fsba(fs, ip->i_number)),
110 (int)fs->fs_bsize, &bp);
111 if (error) {
112 brelse(bp);
113 return (error);
115 if (DOINGSOFTDEP(vp))
116 softdep_update_inodeblock(ip, bp, waitfor);
117 else if (ip->i_effnlink != ip->i_nlink)
118 panic("ffs_update: bad link cnt");
119 *((struct ufs1_dinode *)bp->b_data +
120 ino_to_fsbo(fs, ip->i_number)) = ip->i_din;
121 if (waitfor && !DOINGASYNC(vp)) {
122 return (bwrite(bp));
123 } else if (vm_page_count_severe() || buf_dirty_count_severe()) {
124 return (bwrite(bp));
125 } else {
126 if (bp->b_bufsize == fs->fs_bsize)
127 bp->b_flags |= B_CLUSTEROK;
128 bdwrite(bp);
129 return (0);
133 #define SINGLE 0 /* index of single indirect block */
134 #define DOUBLE 1 /* index of double indirect block */
135 #define TRIPLE 2 /* index of triple indirect block */
137 * Truncate the inode oip to at most length size, freeing the
138 * disk blocks.
141 ffs_truncate(struct vnode *vp, off_t length, int flags, struct ucred *cred)
143 struct vnode *ovp = vp;
144 ufs_daddr_t lastblock;
145 struct inode *oip;
146 ufs_daddr_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR];
147 ufs_daddr_t oldblks[NDADDR + NIADDR], newblks[NDADDR + NIADDR];
148 struct fs *fs;
149 struct buf *bp;
150 int offset, size, level;
151 long count, nblocks, blocksreleased = 0;
152 int i;
153 int aflags, error, allerror;
154 off_t osize;
156 oip = VTOI(ovp);
157 fs = oip->i_fs;
158 if (length < 0)
159 return (EINVAL);
160 if (length > fs->fs_maxfilesize)
161 return (EFBIG);
162 if (ovp->v_type == VLNK &&
163 (oip->i_size < ovp->v_mount->mnt_maxsymlinklen || oip->i_din.di_blocks == 0)) {
164 #ifdef DIAGNOSTIC
165 if (length != 0)
166 panic("ffs_truncate: partial truncate of symlink");
167 #endif /* DIAGNOSTIC */
168 bzero((char *)&oip->i_shortlink, (uint)oip->i_size);
169 oip->i_size = 0;
170 oip->i_flag |= IN_CHANGE | IN_UPDATE;
171 return (ffs_update(ovp, 1));
173 if (oip->i_size == length) {
174 oip->i_flag |= IN_CHANGE | IN_UPDATE;
175 return (ffs_update(ovp, 0));
177 if (fs->fs_ronly)
178 panic("ffs_truncate: read-only filesystem");
179 #ifdef QUOTA
180 error = ufs_getinoquota(oip);
181 if (error)
182 return (error);
183 #endif
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, 0)) != 0)
197 return (error);
198 } else {
199 #ifdef QUOTA
200 (void) ufs_chkdq(oip, -oip->i_blocks, NOCRED, 0);
201 #endif
202 softdep_setup_freeblocks(oip, length);
203 vinvalbuf(ovp, 0, 0, 0);
204 nvnode_pager_setsize(ovp, 0, fs->fs_bsize, 0);
205 oip->i_flag |= IN_CHANGE | IN_UPDATE;
206 return (ffs_update(ovp, 0));
209 osize = oip->i_size;
212 * Lengthen the size of the file. We must ensure that the
213 * last byte of the file is allocated. Since the smallest
214 * value of osize is 0, length will be at least 1.
216 * nvextendbuf() only breads the old buffer. The blocksize
217 * of the new buffer must be specified so it knows how large
218 * to make the VM object.
220 if (osize < length) {
221 nvextendbuf(vp, osize, length,
222 blkoffsize(fs, oip, osize), /* oblksize */
223 blkoffresize(fs, length), /* nblksize */
224 blkoff(fs, osize),
225 blkoff(fs, length),
228 aflags = B_CLRBUF;
229 if (flags & IO_SYNC)
230 aflags |= B_SYNC;
231 /* BALLOC will reallocate the fragment at the old EOF */
232 error = VOP_BALLOC(ovp, length - 1, 1, cred, aflags, &bp);
233 if (error)
234 return (error);
235 oip->i_size = length;
236 if (bp->b_bufsize == fs->fs_bsize)
237 bp->b_flags |= B_CLUSTEROK;
238 if (aflags & B_SYNC)
239 bwrite(bp);
240 else
241 bawrite(bp);
242 oip->i_flag |= IN_CHANGE | IN_UPDATE;
243 return (ffs_update(ovp, 1));
247 * Shorten the size of the file.
249 * NOTE: The block size specified in nvtruncbuf() is the blocksize
250 * of the buffer containing length prior to any reallocation
251 * of the block.
253 allerror = nvtruncbuf(ovp, length, blkoffsize(fs, oip, length),
254 blkoff(fs, length));
255 offset = blkoff(fs, length);
256 if (offset == 0) {
257 oip->i_size = length;
258 } else {
259 lbn = lblkno(fs, length);
260 aflags = B_CLRBUF;
261 if (flags & IO_SYNC)
262 aflags |= B_SYNC;
263 error = VOP_BALLOC(ovp, length - 1, 1, cred, aflags, &bp);
264 if (error)
265 return (error);
268 * When we are doing soft updates and the UFS_BALLOC
269 * above fills in a direct block hole with a full sized
270 * block that will be truncated down to a fragment below,
271 * we must flush out the block dependency with an FSYNC
272 * so that we do not get a soft updates inconsistency
273 * when we create the fragment below.
275 * nvtruncbuf() may have re-dirtied the underlying block
276 * as part of its truncation zeroing code. To avoid a
277 * 'locking against myself' panic in the second fsync we
278 * can simply undirty the bp since the redirtying was
279 * related to areas of the buffer that we are going to
280 * throw away anyway, and we will b*write() the remainder
281 * anyway down below.
283 if (DOINGSOFTDEP(ovp) && lbn < NDADDR &&
284 fragroundup(fs, blkoff(fs, length)) < fs->fs_bsize) {
285 bundirty(bp);
286 error = VOP_FSYNC(ovp, MNT_WAIT, 0);
287 if (error) {
288 bdwrite(bp);
289 return (error);
292 oip->i_size = length;
293 size = blksize(fs, oip, lbn);
294 #if 0
295 /* remove - nvtruncbuf deals with this */
296 if (ovp->v_type != VDIR)
297 bzero((char *)bp->b_data + offset,
298 (uint)(size - offset));
299 #endif
300 /* Kirk's code has reallocbuf(bp, size, 1) here */
301 allocbuf(bp, size);
302 if (bp->b_bufsize == fs->fs_bsize)
303 bp->b_flags |= B_CLUSTEROK;
304 if (aflags & B_SYNC)
305 bwrite(bp);
306 else
307 bawrite(bp);
310 * Calculate index into inode's block list of
311 * last direct and indirect blocks (if any)
312 * which we want to keep. Lastblock is -1 when
313 * the file is truncated to 0.
315 lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1;
316 lastiblock[SINGLE] = lastblock - NDADDR;
317 lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
318 lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
319 nblocks = btodb(fs->fs_bsize);
322 * Update file and block pointers on disk before we start freeing
323 * blocks. If we crash before free'ing blocks below, the blocks
324 * will be returned to the free list. lastiblock values are also
325 * normalized to -1 for calls to ffs_indirtrunc below.
327 bcopy((caddr_t)&oip->i_db[0], (caddr_t)oldblks, sizeof oldblks);
328 for (level = TRIPLE; level >= SINGLE; level--)
329 if (lastiblock[level] < 0) {
330 oip->i_ib[level] = 0;
331 lastiblock[level] = -1;
333 for (i = NDADDR - 1; i > lastblock; i--)
334 oip->i_db[i] = 0;
335 oip->i_flag |= IN_CHANGE | IN_UPDATE;
336 error = ffs_update(ovp, 1);
337 if (error && allerror == 0)
338 allerror = error;
341 * Having written the new inode to disk, save its new configuration
342 * and put back the old block pointers long enough to process them.
343 * Note that we save the new block configuration so we can check it
344 * when we are done.
346 bcopy((caddr_t)&oip->i_db[0], (caddr_t)newblks, sizeof newblks);
347 bcopy((caddr_t)oldblks, (caddr_t)&oip->i_db[0], sizeof oldblks);
348 oip->i_size = osize;
350 if (error && allerror == 0)
351 allerror = error;
354 * Indirect blocks first.
356 indir_lbn[SINGLE] = -NDADDR;
357 indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
358 indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
359 for (level = TRIPLE; level >= SINGLE; level--) {
360 bn = oip->i_ib[level];
361 if (bn != 0) {
362 error = ffs_indirtrunc(oip, indir_lbn[level],
363 fsbtodb(fs, bn), lastiblock[level], level, &count);
364 if (error)
365 allerror = error;
366 blocksreleased += count;
367 if (lastiblock[level] < 0) {
368 oip->i_ib[level] = 0;
369 ffs_blkfree(oip, bn, fs->fs_bsize);
370 blocksreleased += nblocks;
373 if (lastiblock[level] >= 0)
374 goto done;
378 * All whole direct blocks or frags.
380 for (i = NDADDR - 1; i > lastblock; i--) {
381 long bsize;
383 bn = oip->i_db[i];
384 if (bn == 0)
385 continue;
386 oip->i_db[i] = 0;
387 bsize = blksize(fs, oip, i);
388 ffs_blkfree(oip, bn, bsize);
389 blocksreleased += btodb(bsize);
391 if (lastblock < 0)
392 goto done;
395 * Finally, look for a change in size of the
396 * last direct block; release any frags.
398 bn = oip->i_db[lastblock];
399 if (bn != 0) {
400 long oldspace, newspace;
403 * Calculate amount of space we're giving
404 * back as old block size minus new block size.
406 oldspace = blksize(fs, oip, lastblock);
407 oip->i_size = length;
408 newspace = blksize(fs, oip, lastblock);
409 if (newspace == 0)
410 panic("ffs_truncate: newspace");
411 if (oldspace - newspace > 0) {
413 * Block number of space to be free'd is
414 * the old block # plus the number of frags
415 * required for the storage we're keeping.
417 bn += numfrags(fs, newspace);
418 ffs_blkfree(oip, bn, oldspace - newspace);
419 blocksreleased += btodb(oldspace - newspace);
422 done:
423 #ifdef DIAGNOSTIC
424 for (level = SINGLE; level <= TRIPLE; level++)
425 if (newblks[NDADDR + level] != oip->i_ib[level])
426 panic("ffs_truncate1");
427 for (i = 0; i < NDADDR; i++)
428 if (newblks[i] != oip->i_db[i])
429 panic("ffs_truncate2");
430 if (length == 0 && !RB_EMPTY(&ovp->v_rbdirty_tree))
431 panic("ffs_truncate3");
432 #endif /* DIAGNOSTIC */
434 * Put back the real size.
436 oip->i_size = length;
437 oip->i_blocks -= blocksreleased;
439 if (oip->i_blocks < 0) /* sanity */
440 oip->i_blocks = 0;
441 oip->i_flag |= IN_CHANGE;
442 #ifdef QUOTA
443 (void) ufs_chkdq(oip, -blocksreleased, NOCRED, 0);
444 #endif
445 return (allerror);
449 * Release blocks associated with the inode ip and stored in the indirect
450 * block bn. Blocks are free'd in LIFO order up to (but not including)
451 * lastbn. If level is greater than SINGLE, the block is an indirect block
452 * and recursive calls to indirtrunc must be used to cleanse other indirect
453 * blocks.
455 * NB: triple indirect blocks are untested.
457 static int
458 ffs_indirtrunc(struct inode *ip, ufs_daddr_t lbn, ufs_daddr_t dbn,
459 ufs_daddr_t lastbn, int level, long *countp)
461 int i;
462 struct buf *bp;
463 struct fs *fs = ip->i_fs;
464 ufs_daddr_t *bap;
465 struct vnode *vp;
466 ufs_daddr_t *copy = NULL, nb, nlbn, last;
467 long blkcount, factor;
468 int nblocks, blocksreleased = 0;
469 int error = 0, allerror = 0;
472 * Calculate index in current block of last
473 * block to be kept. -1 indicates the entire
474 * block so we need not calculate the index.
476 factor = 1;
477 for (i = SINGLE; i < level; i++)
478 factor *= NINDIR(fs);
479 last = lastbn;
480 if (lastbn > 0)
481 last /= factor;
482 nblocks = btodb(fs->fs_bsize);
484 * Get buffer of block pointers, zero those entries corresponding
485 * to blocks to be free'd, and update on disk copy first. Since
486 * double(triple) indirect before single(double) indirect, calls
487 * to bmap on these blocks will fail. However, we already have
488 * the on disk address, so we have to set the bio_offset field
489 * explicitly instead of letting bread do everything for us.
491 vp = ITOV(ip);
492 bp = getblk(vp, lblktodoff(fs, lbn), (int)fs->fs_bsize, 0, 0);
493 if ((bp->b_flags & B_CACHE) == 0) {
494 bp->b_flags &= ~(B_ERROR|B_INVAL);
495 bp->b_cmd = BUF_CMD_READ;
496 if (bp->b_bcount > bp->b_bufsize)
497 panic("ffs_indirtrunc: bad buffer size");
499 * BIO is bio2 which chains back to bio1. We wait
500 * on bio1.
502 bp->b_bio2.bio_offset = dbtodoff(fs, dbn);
503 bp->b_bio1.bio_done = biodone_sync;
504 bp->b_bio1.bio_flags |= BIO_SYNC;
505 vfs_busy_pages(vp, bp);
507 * Access the block device layer using the device vnode
508 * and the translated block number (bio2) instead of the
509 * file vnode (vp) and logical block number (bio1).
511 * Even though we are bypassing the vnode layer, we still
512 * want the vnode state to indicate that an I/O on its behalf
513 * is in progress.
515 bio_start_transaction(&bp->b_bio1, &vp->v_track_read);
516 vn_strategy(ip->i_devvp, &bp->b_bio2);
517 error = biowait(&bp->b_bio1, "biord");
519 if (error) {
520 brelse(bp);
521 *countp = 0;
522 return (error);
525 bap = (ufs_daddr_t *)bp->b_data;
526 if (lastbn != -1) {
527 MALLOC(copy, ufs_daddr_t *, fs->fs_bsize, M_TEMP, M_WAITOK);
528 bcopy((caddr_t)bap, (caddr_t)copy, (uint)fs->fs_bsize);
529 bzero((caddr_t)&bap[last + 1],
530 (uint)(NINDIR(fs) - (last + 1)) * sizeof (ufs_daddr_t));
531 if (DOINGASYNC(vp)) {
532 bawrite(bp);
533 } else {
534 error = bwrite(bp);
535 if (error)
536 allerror = error;
538 bap = copy;
542 * Recursively free totally unused blocks.
544 for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
545 i--, nlbn += factor) {
546 nb = bap[i];
547 if (nb == 0)
548 continue;
549 if (level > SINGLE) {
550 if ((error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
551 (ufs_daddr_t)-1, level - 1, &blkcount)) != 0)
552 allerror = error;
553 blocksreleased += blkcount;
555 ffs_blkfree(ip, nb, fs->fs_bsize);
556 blocksreleased += nblocks;
560 * Recursively free last partial block.
562 if (level > SINGLE && lastbn >= 0) {
563 last = lastbn % factor;
564 nb = bap[i];
565 if (nb != 0) {
566 error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
567 last, level - 1, &blkcount);
568 if (error)
569 allerror = error;
570 blocksreleased += blkcount;
573 if (copy != NULL) {
574 FREE(copy, M_TEMP);
575 } else {
576 bp->b_flags |= B_INVAL | B_NOCACHE;
577 brelse(bp);
580 *countp = blocksreleased;
581 return (allerror);