Due to continuing issues with VOP_READ/VOP_WRITE ops being called without
[dragonfly.git] / sys / vfs / gnu / ext2fs / ext2_inode.c
CommitLineData
984263bc
MD
1/*
2 * modified for Lites 1.1
3 *
4 * Aug 1995, Godmar Back (gback@cs.utah.edu)
5 * University of Utah, Department of Computer Science
6 */
7/*
8 * Copyright (c) 1982, 1986, 1989, 1993
9 * The Regents of the University of California. All rights reserved.
10 *
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
13 * are met:
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.
19 * 3. All advertising materials mentioning features or use of this software
20 * must display the following acknowledgement:
21 * This product includes software developed by the University of
22 * California, Berkeley and its contributors.
23 * 4. Neither the name of the University nor the names of its contributors
24 * may be used to endorse or promote products derived from this software
25 * without specific prior written permission.
26 *
27 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
28 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
29 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
30 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
31 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
32 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
33 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
34 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
35 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
36 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37 * SUCH DAMAGE.
38 *
39 * @(#)ext2_inode.c 8.5 (Berkeley) 12/30/93
40 * $FreeBSD: src/sys/gnu/ext2fs/ext2_inode.c,v 1.24.2.1 2000/08/03 00:52:57 peter Exp $
1c843a13 41 * $DragonFly: src/sys/vfs/gnu/ext2fs/ext2_inode.c,v 1.15 2006/04/07 06:38:30 dillon Exp $
984263bc
MD
42 */
43
44#include "opt_quota.h"
45
46#include <sys/param.h>
47#include <sys/systm.h>
48#include <sys/mount.h>
49#include <sys/buf.h>
50#include <sys/vnode.h>
51#include <sys/malloc.h>
52
53#include <vm/vm.h>
54#include <vm/vm_extern.h>
55
1f1db49f
MD
56#include "quota.h"
57#include "inode.h"
58#include "ext2mount.h"
984263bc 59
1f2de5d4
MD
60#include "ext2_fs.h"
61#include "ext2_fs_sb.h"
62#include "fs.h"
63#include "ext2_extern.h"
984263bc 64
54078292
MD
65static int ext2_indirtrunc (struct inode *, daddr_t, off_t, daddr_t,
66 int, long *);
984263bc 67
984263bc
MD
68/*
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. If we write now, then clear both IN_MODIFIED and
74 * IN_LAZYMOD to reflect the presumably successful write, and if waitfor is
75 * set, then wait for the write to complete.
76 */
77int
0f7f7a49 78ext2_update(struct vnode *vp, int waitfor)
984263bc 79{
f7aae92f 80 struct ext2_sb_info *fs;
984263bc
MD
81 struct buf *bp;
82 struct inode *ip;
83 int error;
84
1f1db49f 85 ext2_itimes(vp);
984263bc
MD
86 ip = VTOI(vp);
87 if ((ip->i_flag & IN_MODIFIED) == 0)
88 return (0);
89 ip->i_flag &= ~(IN_LAZYMOD | IN_MODIFIED);
90 if (vp->v_mount->mnt_flag & MNT_RDONLY)
91 return (0);
92 fs = ip->i_e2fs;
54078292
MD
93 error = bread(ip->i_devvp,
94 fsbtodoff(fs, ino_to_fsba(fs, ip->i_number)),
95 (int)fs->s_blocksize, &bp);
96 if (error) {
984263bc
MD
97 brelse(bp);
98 return (error);
99 }
100 ext2_di2ei( &ip->i_din, (struct ext2_inode *) ((char *)bp->b_data + EXT2_INODE_SIZE *
101 ino_to_fsbo(fs, ip->i_number)));
102/*
103 if (waitfor && (vp->v_mount->mnt_flag & MNT_ASYNC) == 0)
104 return (bwrite(bp));
105 else {
106*/
107 bdwrite(bp);
108 return (0);
109/*
110 }
111*/
112}
113
114#define SINGLE 0 /* index of single indirect block */
115#define DOUBLE 1 /* index of double indirect block */
116#define TRIPLE 2 /* index of triple indirect block */
117/*
118 * Truncate the inode oip to at most length size, freeing the
119 * disk blocks.
120 */
121int
0f7f7a49
CP
122ext2_truncate(struct vnode *vp, off_t length, int flags, struct ucred *cred,
123 struct thread *td)
984263bc 124{
f7aae92f
RG
125 struct vnode *ovp = vp;
126 daddr_t lastblock;
127 struct inode *oip;
984263bc
MD
128 daddr_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR];
129 daddr_t oldblks[NDADDR + NIADDR], newblks[NDADDR + NIADDR];
f7aae92f 130 struct ext2_sb_info *fs;
984263bc
MD
131 struct buf *bp;
132 int offset, size, level;
133 long count, nblocks, blocksreleased = 0;
f7aae92f 134 int i;
984263bc
MD
135 int aflags, error, allerror;
136 off_t osize;
137/*
138printf("ext2_truncate called %d to %d\n", VTOI(ovp)->i_number, length);
139*/ /*
140 * negative file sizes will totally break the code below and
141 * are not meaningful anyways.
142 */
143 if (length < 0)
144 return EFBIG;
145
146 oip = VTOI(ovp);
147 if (ovp->v_type == VLNK &&
148 oip->i_size < ovp->v_mount->mnt_maxsymlinklen) {
149#if DIAGNOSTIC
150 if (length != 0)
151 panic("ext2_truncate: partial truncate of symlink");
152#endif
153 bzero((char *)&oip->i_shortlink, (u_int)oip->i_size);
154 oip->i_size = 0;
155 oip->i_flag |= IN_CHANGE | IN_UPDATE;
1f1db49f 156 return (EXT2_UPDATE(ovp, 1));
984263bc
MD
157 }
158 if (oip->i_size == length) {
159 oip->i_flag |= IN_CHANGE | IN_UPDATE;
1f1db49f 160 return (EXT2_UPDATE(ovp, 0));
984263bc
MD
161 }
162#if QUOTA
1f1db49f 163 if ((error = ext2_getinoquota(oip)) != 0)
984263bc
MD
164 return (error);
165#endif
166 fs = oip->i_e2fs;
167 osize = oip->i_size;
168 ext2_discard_prealloc(oip);
169 /*
170 * Lengthen the size of the file. We must ensure that the
171 * last byte of the file is allocated. Since the smallest
172 * value of oszie is 0, length will be at least 1.
173 */
174 if (osize < length) {
175 offset = blkoff(fs, length - 1);
176 lbn = lblkno(fs, length - 1);
177 aflags = B_CLRBUF;
178 if (flags & IO_SYNC)
179 aflags |= B_SYNC;
180 vnode_pager_setsize(ovp, length);
181 if ((error = ext2_balloc(oip, lbn, offset + 1, cred, &bp,
182 aflags)) != 0)
183 return (error);
184 oip->i_size = length;
185 if (aflags & IO_SYNC)
186 bwrite(bp);
187 else
188 bawrite(bp);
189 oip->i_flag |= IN_CHANGE | IN_UPDATE;
1f1db49f 190 return (EXT2_UPDATE(ovp, 1));
984263bc
MD
191 }
192 /*
193 * Shorten the size of the file. If the file is not being
194 * truncated to a block boundry, the contents of the
195 * partial block following the end of the file must be
196 * zero'ed in case it ever become accessable again because
197 * of subsequent file growth.
198 */
199 /* I don't understand the comment above */
200 offset = blkoff(fs, length);
201 if (offset == 0) {
202 oip->i_size = length;
203 } else {
204 lbn = lblkno(fs, length);
205 aflags = B_CLRBUF;
206 if (flags & IO_SYNC)
207 aflags |= B_SYNC;
208 if ((error = ext2_balloc(oip, lbn, offset, cred, &bp,
209 aflags)) != 0)
210 return (error);
211 oip->i_size = length;
212 size = blksize(fs, oip, lbn);
213 bzero((char *)bp->b_data + offset, (u_int)(size - offset));
214 allocbuf(bp, size);
215 if (aflags & IO_SYNC)
216 bwrite(bp);
217 else
218 bawrite(bp);
219 }
220 /*
221 * Calculate index into inode's block list of
222 * last direct and indirect blocks (if any)
223 * which we want to keep. Lastblock is -1 when
224 * the file is truncated to 0.
225 */
226 lastblock = lblkno(fs, length + fs->s_blocksize - 1) - 1;
227 lastiblock[SINGLE] = lastblock - NDADDR;
228 lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
229 lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
230 nblocks = btodb(fs->s_blocksize);
231 /*
232 * Update file and block pointers on disk before we start freeing
233 * blocks. If we crash before free'ing blocks below, the blocks
234 * will be returned to the free list. lastiblock values are also
235 * normalized to -1 for calls to ext2_indirtrunc below.
236 */
237 bcopy((caddr_t)&oip->i_db[0], (caddr_t)oldblks, sizeof oldblks);
238 for (level = TRIPLE; level >= SINGLE; level--)
239 if (lastiblock[level] < 0) {
240 oip->i_ib[level] = 0;
241 lastiblock[level] = -1;
242 }
243 for (i = NDADDR - 1; i > lastblock; i--)
244 oip->i_db[i] = 0;
245 oip->i_flag |= IN_CHANGE | IN_UPDATE;
1f1db49f 246 allerror = EXT2_UPDATE(ovp, 1);
984263bc
MD
247
248 /*
249 * Having written the new inode to disk, save its new configuration
250 * and put back the old block pointers long enough to process them.
251 * Note that we save the new block configuration so we can check it
252 * when we are done.
253 */
254 bcopy((caddr_t)&oip->i_db[0], (caddr_t)newblks, sizeof newblks);
255 bcopy((caddr_t)oldblks, (caddr_t)&oip->i_db[0], sizeof oldblks);
256 oip->i_size = osize;
7b95be2a 257 error = vtruncbuf(ovp, td, length, (int)fs->s_blocksize);
984263bc
MD
258 if (error && (allerror == 0))
259 allerror = error;
260
261 /*
262 * Indirect blocks first.
263 */
264 indir_lbn[SINGLE] = -NDADDR;
265 indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
266 indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
267 for (level = TRIPLE; level >= SINGLE; level--) {
268 bn = oip->i_ib[level];
269 if (bn != 0) {
270 error = ext2_indirtrunc(oip, indir_lbn[level],
54078292 271 fsbtodoff(fs, bn), lastiblock[level], level, &count);
984263bc
MD
272 if (error)
273 allerror = error;
274 blocksreleased += count;
275 if (lastiblock[level] < 0) {
276 oip->i_ib[level] = 0;
277 ext2_blkfree(oip, bn, fs->s_frag_size);
278 blocksreleased += nblocks;
279 }
280 }
281 if (lastiblock[level] >= 0)
282 goto done;
283 }
284
285 /*
286 * All whole direct blocks or frags.
287 */
288 for (i = NDADDR - 1; i > lastblock; i--) {
f7aae92f 289 long bsize;
984263bc
MD
290
291 bn = oip->i_db[i];
292 if (bn == 0)
293 continue;
294 oip->i_db[i] = 0;
295 bsize = blksize(fs, oip, i);
296 ext2_blkfree(oip, bn, bsize);
297 blocksreleased += btodb(bsize);
298 }
299 if (lastblock < 0)
300 goto done;
301
302 /*
303 * Finally, look for a change in size of the
304 * last direct block; release any frags.
305 */
306 bn = oip->i_db[lastblock];
307 if (bn != 0) {
308 long oldspace, newspace;
309
310 /*
311 * Calculate amount of space we're giving
312 * back as old block size minus new block size.
313 */
314 oldspace = blksize(fs, oip, lastblock);
315 oip->i_size = length;
316 newspace = blksize(fs, oip, lastblock);
317 if (newspace == 0)
318 panic("itrunc: newspace");
319 if (oldspace - newspace > 0) {
320 /*
321 * Block number of space to be free'd is
322 * the old block # plus the number of frags
323 * required for the storage we're keeping.
324 */
325 bn += numfrags(fs, newspace);
326 ext2_blkfree(oip, bn, oldspace - newspace);
327 blocksreleased += btodb(oldspace - newspace);
328 }
329 }
330done:
331#if DIAGNOSTIC
332 for (level = SINGLE; level <= TRIPLE; level++)
333 if (newblks[NDADDR + level] != oip->i_ib[level])
334 panic("itrunc1");
335 for (i = 0; i < NDADDR; i++)
336 if (newblks[i] != oip->i_db[i])
337 panic("itrunc2");
6bae6177
MD
338 if (length == 0 && (!RB_EMPTY(&ovp->v_rbdirty_tree) ||
339 !RB_EMPTY(&ovp->v_rbclean_tree)))
984263bc
MD
340 panic("itrunc3");
341#endif /* DIAGNOSTIC */
342 /*
343 * Put back the real size.
344 */
345 oip->i_size = length;
346 oip->i_blocks -= blocksreleased;
347 if (oip->i_blocks < 0) /* sanity */
348 oip->i_blocks = 0;
349 oip->i_flag |= IN_CHANGE;
350 vnode_pager_setsize(ovp, length);
351#if QUOTA
1f1db49f 352 ext2_chkdq(oip, -blocksreleased, NOCRED, 0);
984263bc
MD
353#endif
354 return (allerror);
355}
356
357/*
358 * Release blocks associated with the inode ip and stored in the indirect
359 * block bn. Blocks are free'd in LIFO order up to (but not including)
360 * lastbn. If level is greater than SINGLE, the block is an indirect block
361 * and recursive calls to indirtrunc must be used to cleanse other indirect
362 * blocks.
363 *
364 * NB: triple indirect blocks are untested.
365 */
366
367static int
54078292 368ext2_indirtrunc(struct inode *ip, daddr_t lbn, off_t doffset, daddr_t lastbn,
0f7f7a49 369 int level, long *countp)
984263bc 370{
f7aae92f 371 int i;
984263bc 372 struct buf *bp;
f7aae92f
RG
373 struct ext2_sb_info *fs = ip->i_e2fs;
374 daddr_t *bap;
984263bc
MD
375 struct vnode *vp;
376 daddr_t *copy, nb, nlbn, last;
377 long blkcount, factor;
378 int nblocks, blocksreleased = 0;
379 int error = 0, allerror = 0;
380
381 /*
382 * Calculate index in current block of last
383 * block to be kept. -1 indicates the entire
384 * block so we need not calculate the index.
385 */
386 factor = 1;
387 for (i = SINGLE; i < level; i++)
388 factor *= NINDIR(fs);
389 last = lastbn;
390 if (lastbn > 0)
391 last /= factor;
392 nblocks = btodb(fs->s_blocksize);
393 /*
394 * Get buffer of block pointers, zero those entries corresponding
395 * to blocks to be free'd, and update on disk copy first. Since
396 * double(triple) indirect before single(double) indirect, calls
397 * to bmap on these blocks will fail. However, we already have
54078292 398 * the on disk address, so we have to set the bio_offset field
984263bc
MD
399 * explicitly instead of letting bread do everything for us.
400 */
401 vp = ITOV(ip);
54078292 402 bp = getblk(vp, lblktodoff(fs, lbn), (int)fs->s_blocksize, 0, 0);
984263bc 403 if (bp->b_flags & (B_DONE | B_DELWRI)) {
54078292 404 /* nop */
984263bc
MD
405 } else {
406 bp->b_flags |= B_READ;
407 if (bp->b_bcount > bp->b_bufsize)
408 panic("ext2_indirtrunc: bad buffer size");
54078292 409 bp->b_bio2.bio_offset = doffset;
984263bc 410 vfs_busy_pages(bp, 0);
81b5c339 411 vn_strategy(vp, &bp->b_bio1);
984263bc
MD
412 error = biowait(bp);
413 }
414 if (error) {
415 brelse(bp);
416 *countp = 0;
417 return (error);
418 }
419
420 bap = (daddr_t *)bp->b_data;
421 MALLOC(copy, daddr_t *, fs->s_blocksize, M_TEMP, M_WAITOK);
422 bcopy((caddr_t)bap, (caddr_t)copy, (u_int)fs->s_blocksize);
423 bzero((caddr_t)&bap[last + 1],
424 (u_int)(NINDIR(fs) - (last + 1)) * sizeof (daddr_t));
425 if (last == -1)
426 bp->b_flags |= B_INVAL;
427 error = bwrite(bp);
428 if (error)
429 allerror = error;
430 bap = copy;
431
432 /*
433 * Recursively free totally unused blocks.
434 */
435 for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
436 i--, nlbn += factor) {
437 nb = bap[i];
438 if (nb == 0)
439 continue;
440 if (level > SINGLE) {
441 if ((error = ext2_indirtrunc(ip, nlbn,
54078292 442 fsbtodoff(fs, nb), (daddr_t)-1, level - 1, &blkcount)) != 0)
984263bc
MD
443 allerror = error;
444 blocksreleased += blkcount;
445 }
446 ext2_blkfree(ip, nb, fs->s_blocksize);
447 blocksreleased += nblocks;
448 }
449
450 /*
451 * Recursively free last partial block.
452 */
453 if (level > SINGLE && lastbn >= 0) {
454 last = lastbn % factor;
455 nb = bap[i];
456 if (nb != 0) {
54078292
MD
457 error = ext2_indirtrunc(ip, nlbn, fsbtodoff(fs, nb),
458 last, level - 1, &blkcount);
459 if (error)
984263bc
MD
460 allerror = error;
461 blocksreleased += blkcount;
462 }
463 }
464 FREE(copy, M_TEMP);
465 *countp = blocksreleased;
466 return (allerror);
467}
468
469/*
1f1db49f 470 * Last reference to an inode. If necessary, write or delete it.
0f7f7a49 471 *
1f1db49f 472 * ext2_inactive(struct vnode *a_vp, struct thread *a_td)
984263bc
MD
473 */
474int
0f7f7a49 475ext2_inactive(struct vop_inactive_args *ap)
984263bc 476{
1f1db49f
MD
477 struct vnode *vp = ap->a_vp;
478 struct inode *ip = VTOI(vp);
479 struct thread *td = ap->a_td;
480 int mode, error = 0;
481
482 ext2_discard_prealloc(ip);
483 if (prtactive && vp->v_usecount != 1)
484 vprint("ext2_inactive: pushing active", vp);
485
486 /*
487 * Ignore inodes related to stale file handles.
488 */
489 if (ip == NULL || ip->i_mode == 0)
490 goto out;
491 if (ip->i_nlink <= 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
492#ifdef QUOTA
493 if (!ext2_getinoquota(ip))
494 (void)ext2_chkiq(ip, -1, NOCRED, FORCE);
495#endif
496 error = EXT2_TRUNCATE(vp, (off_t)0, 0, NOCRED, td);
497 ip->i_rdev = 0;
498 mode = ip->i_mode;
499 ip->i_mode = 0;
500 ip->i_flag |= IN_CHANGE | IN_UPDATE;
501 EXT2_VFREE(vp, ip->i_number, mode);
502 }
503 if (ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE))
504 EXT2_UPDATE(vp, 0);
505out:
506 /*
507 * If we are done with the inode, reclaim it
508 * so that it can be reused immediately.
509 */
510 if (ip == NULL || ip->i_mode == 0)
511 vrecycle(vp, td);
512 return (error);
513}
514
515/*
516 * Reclaim an inode so that it can be used for other purposes.
517 *
518 * ext2_reclaim(struct vnode *a_vp, struct thread *a_td)
519 */
520int
521ext2_reclaim(struct vop_reclaim_args *ap)
522{
523 struct inode *ip;
524 struct vnode *vp = ap->a_vp;
525#ifdef QUOTA
526 int i;
527#endif
528
529 if (prtactive && vp->v_usecount != 1)
530 vprint("ext2_reclaim: pushing active", vp);
531 ip = VTOI(vp);
532
533 /*
534 * Lazy updates.
535 */
536 if (ip) {
1f1db49f
MD
537 if (ip->i_flag & IN_LAZYMOD) {
538 ip->i_flag |= IN_MODIFIED;
539 EXT2_UPDATE(vp, 0);
540 }
541 }
542#ifdef INVARIANTS
543 if (ip && (ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE))) {
544 printf("WARNING: INODE %ld flags %08x: modified inode being released!\n", (long)ip->i_number, (int)ip->i_flag);
545 ip->i_flag |= IN_MODIFIED;
546 EXT2_UPDATE(vp, 0);
547 }
548#endif
549 /*
550 * Remove the inode from its hash chain and purge namecache
551 * data associated with the vnode.
552 */
553 vp->v_data = NULL;
554 if (ip) {
555 ext2_ihashrem(ip);
556 if (ip->i_devvp) {
557 vrele(ip->i_devvp);
558 ip->i_devvp = 0;
559 }
560#ifdef QUOTA
561 for (i = 0; i < MAXQUOTAS; i++) {
562 if (ip->i_dquot[i] != NODQUOT) {
563 ext2_dqrele(vp, ip->i_dquot[i]);
564 ip->i_dquot[i] = NODQUOT;
565 }
566 }
567#endif
568#ifdef UFS_DIRHASH
569 if (ip->i_dirhash != NULL)
570 ext2dirhash_free(ip);
571#endif
572 free(ip, VFSTOEXT2(vp->v_mount)->um_malloctype);
573 }
574 return (0);
984263bc 575}