proc->thread stage 4: rework the VFS and DEVICE subsystems to take thread
[dragonfly.git] / sys / vfs / ufs / ffs_inode.c
CommitLineData
984263bc
MD
1/*
2 * Copyright (c) 1982, 1986, 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 *
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.
20 *
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.
32 *
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 $
dadab5e9 35 * $DragonFly: src/sys/vfs/ufs/ffs_inode.c,v 1.3 2003/06/25 03:56:11 dillon Exp $
984263bc
MD
36 */
37
38#include "opt_quota.h"
39
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>
50
51#include <vm/vm.h>
52#include <vm/vm_extern.h>
53
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>
58
59#include <ufs/ffs/fs.h>
60#include <ufs/ffs/ffs_extern.h>
61
62static int ffs_indirtrunc __P((struct inode *, ufs_daddr_t, ufs_daddr_t,
63 ufs_daddr_t, int, long *));
64
65/*
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. If we write now, then clear both IN_MODIFIED and
71 * IN_LAZYMOD to reflect the presumably successful write, and if waitfor is
72 * set, then wait for the write to complete.
73 */
74int
75ffs_update(vp, waitfor)
76 struct vnode *vp;
77 int waitfor;
78{
79 register struct fs *fs;
80 struct buf *bp;
81 struct inode *ip;
82 int error;
83
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);
92 /*
93 * Ensure that uid and gid are correct. This is a temporary
94 * fix until fsck has been changed to do the update.
95 */
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 */
99 } /* XXX */
100 error = bread(ip->i_devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
101 (int)fs->fs_bsize, NOCRED, &bp);
102 if (error) {
103 brelse(bp);
104 return (error);
105 }
106 if (DOINGSOFTDEP(vp))
107 softdep_update_inodeblock(ip, bp, waitfor);
108 else if (ip->i_effnlink != ip->i_nlink)
109 panic("ffs_update: bad link cnt");
110 *((struct dinode *)bp->b_data +
111 ino_to_fsbo(fs, ip->i_number)) = ip->i_din;
112 if (waitfor && !DOINGASYNC(vp)) {
113 return (bwrite(bp));
114 } else if (vm_page_count_severe() || buf_dirty_count_severe()) {
115 return (bwrite(bp));
116 } else {
117 if (bp->b_bufsize == fs->fs_bsize)
118 bp->b_flags |= B_CLUSTEROK;
119 bdwrite(bp);
120 return (0);
121 }
122}
123
124#define SINGLE 0 /* index of single indirect block */
125#define DOUBLE 1 /* index of double indirect block */
126#define TRIPLE 2 /* index of triple indirect block */
127/*
128 * Truncate the inode oip to at most length size, freeing the
129 * disk blocks.
130 */
131int
dadab5e9 132ffs_truncate(vp, length, flags, cred, td)
984263bc
MD
133 struct vnode *vp;
134 off_t length;
135 int flags;
136 struct ucred *cred;
dadab5e9 137 struct thread *td;
984263bc
MD
138{
139 register struct vnode *ovp = vp;
140 ufs_daddr_t lastblock;
141 register struct inode *oip;
142 ufs_daddr_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR];
143 ufs_daddr_t oldblks[NDADDR + NIADDR], newblks[NDADDR + NIADDR];
144 register struct fs *fs;
145 struct buf *bp;
146 int offset, size, level;
147 long count, nblocks, blocksreleased = 0;
148 register int i;
149 int aflags, error, allerror;
150 off_t osize;
151
152 oip = VTOI(ovp);
153 fs = oip->i_fs;
154 if (length < 0)
155 return (EINVAL);
156 if (length > fs->fs_maxfilesize)
157 return (EFBIG);
158 if (ovp->v_type == VLNK &&
159 (oip->i_size < ovp->v_mount->mnt_maxsymlinklen || oip->i_din.di_blocks == 0)) {
160#ifdef DIAGNOSTIC
161 if (length != 0)
162 panic("ffs_truncate: partial truncate of symlink");
163#endif
164 bzero((char *)&oip->i_shortlink, (u_int)oip->i_size);
165 oip->i_size = 0;
166 oip->i_flag |= IN_CHANGE | IN_UPDATE;
167 return (UFS_UPDATE(ovp, 1));
168 }
169 if (oip->i_size == length) {
170 oip->i_flag |= IN_CHANGE | IN_UPDATE;
171 return (UFS_UPDATE(ovp, 0));
172 }
173 if (fs->fs_ronly)
174 panic("ffs_truncate: read-only filesystem");
175#ifdef QUOTA
176 error = getinoquota(oip);
177 if (error)
178 return (error);
179#endif
180 ovp->v_lasta = ovp->v_clen = ovp->v_cstart = ovp->v_lastw = 0;
181 if (DOINGSOFTDEP(ovp)) {
182 if (length > 0 || softdep_slowdown(ovp)) {
183 /*
184 * If a file is only partially truncated, then
185 * we have to clean up the data structures
186 * describing the allocation past the truncation
187 * point. Finding and deallocating those structures
188 * is a lot of work. Since partial truncation occurs
189 * rarely, we solve the problem by syncing the file
190 * so that it will have no data structures left.
191 */
192 if ((error = VOP_FSYNC(ovp, cred, MNT_WAIT,
dadab5e9 193 td)) != 0)
984263bc
MD
194 return (error);
195 } else {
196#ifdef QUOTA
197 (void) chkdq(oip, -oip->i_blocks, NOCRED, 0);
198#endif
199 softdep_setup_freeblocks(oip, length);
dadab5e9 200 vinvalbuf(ovp, 0, cred, td, 0, 0);
984263bc
MD
201 oip->i_flag |= IN_CHANGE | IN_UPDATE;
202 return (ffs_update(ovp, 0));
203 }
204 }
205 osize = oip->i_size;
206 /*
207 * Lengthen the size of the file. We must ensure that the
208 * last byte of the file is allocated. Since the smallest
209 * value of osize is 0, length will be at least 1.
210 */
211 if (osize < length) {
212 vnode_pager_setsize(ovp, length);
213 aflags = B_CLRBUF;
214 if (flags & IO_SYNC)
215 aflags |= B_SYNC;
216 error = VOP_BALLOC(ovp, length - 1, 1,
217 cred, aflags, &bp);
218 if (error)
219 return (error);
220 oip->i_size = length;
221 if (bp->b_bufsize == fs->fs_bsize)
222 bp->b_flags |= B_CLUSTEROK;
223 if (aflags & B_SYNC)
224 bwrite(bp);
225 else
226 bawrite(bp);
227 oip->i_flag |= IN_CHANGE | IN_UPDATE;
228 return (UFS_UPDATE(ovp, 1));
229 }
230 /*
231 * Shorten the size of the file. If the file is not being
232 * truncated to a block boundary, the contents of the
233 * partial block following the end of the file must be
234 * zero'ed in case it ever becomes accessible again because
235 * of subsequent file growth. Directories however are not
236 * zero'ed as they should grow back initialized to empty.
237 */
238 offset = blkoff(fs, length);
239 if (offset == 0) {
240 oip->i_size = length;
241 } else {
242 lbn = lblkno(fs, length);
243 aflags = B_CLRBUF;
244 if (flags & IO_SYNC)
245 aflags |= B_SYNC;
246 error = VOP_BALLOC(ovp, length - 1, 1, cred, aflags, &bp);
247 if (error) {
248 return (error);
249 }
250 /*
251 * When we are doing soft updates and the UFS_BALLOC
252 * above fills in a direct block hole with a full sized
253 * block that will be truncated down to a fragment below,
254 * we must flush out the block dependency with an FSYNC
255 * so that we do not get a soft updates inconsistency
256 * when we create the fragment below.
257 */
258 if (DOINGSOFTDEP(ovp) && lbn < NDADDR &&
259 fragroundup(fs, blkoff(fs, length)) < fs->fs_bsize &&
dadab5e9 260 (error = VOP_FSYNC(ovp, cred, MNT_WAIT, td)) != 0) {
984263bc
MD
261 return (error);
262 }
263 oip->i_size = length;
264 size = blksize(fs, oip, lbn);
265 if (ovp->v_type != VDIR)
266 bzero((char *)bp->b_data + offset,
267 (u_int)(size - offset));
268 /* Kirk's code has reallocbuf(bp, size, 1) here */
269 allocbuf(bp, size);
270 if (bp->b_bufsize == fs->fs_bsize)
271 bp->b_flags |= B_CLUSTEROK;
272 if (aflags & B_SYNC)
273 bwrite(bp);
274 else
275 bawrite(bp);
276 }
277 /*
278 * Calculate index into inode's block list of
279 * last direct and indirect blocks (if any)
280 * which we want to keep. Lastblock is -1 when
281 * the file is truncated to 0.
282 */
283 lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1;
284 lastiblock[SINGLE] = lastblock - NDADDR;
285 lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
286 lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
287 nblocks = btodb(fs->fs_bsize);
288 /*
289 * Update file and block pointers on disk before we start freeing
290 * blocks. If we crash before free'ing blocks below, the blocks
291 * will be returned to the free list. lastiblock values are also
292 * normalized to -1 for calls to ffs_indirtrunc below.
293 */
294 bcopy((caddr_t)&oip->i_db[0], (caddr_t)oldblks, sizeof oldblks);
295 for (level = TRIPLE; level >= SINGLE; level--)
296 if (lastiblock[level] < 0) {
297 oip->i_ib[level] = 0;
298 lastiblock[level] = -1;
299 }
300 for (i = NDADDR - 1; i > lastblock; i--)
301 oip->i_db[i] = 0;
302 oip->i_flag |= IN_CHANGE | IN_UPDATE;
303 allerror = UFS_UPDATE(ovp, 1);
304
305 /*
306 * Having written the new inode to disk, save its new configuration
307 * and put back the old block pointers long enough to process them.
308 * Note that we save the new block configuration so we can check it
309 * when we are done.
310 */
311 bcopy((caddr_t)&oip->i_db[0], (caddr_t)newblks, sizeof newblks);
312 bcopy((caddr_t)oldblks, (caddr_t)&oip->i_db[0], sizeof oldblks);
313 oip->i_size = osize;
314
dadab5e9 315 error = vtruncbuf(ovp, cred, td, length, fs->fs_bsize);
984263bc
MD
316 if (error && (allerror == 0))
317 allerror = error;
318
319 /*
320 * Indirect blocks first.
321 */
322 indir_lbn[SINGLE] = -NDADDR;
323 indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
324 indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
325 for (level = TRIPLE; level >= SINGLE; level--) {
326 bn = oip->i_ib[level];
327 if (bn != 0) {
328 error = ffs_indirtrunc(oip, indir_lbn[level],
329 fsbtodb(fs, bn), lastiblock[level], level, &count);
330 if (error)
331 allerror = error;
332 blocksreleased += count;
333 if (lastiblock[level] < 0) {
334 oip->i_ib[level] = 0;
335 ffs_blkfree(oip, bn, fs->fs_bsize);
336 blocksreleased += nblocks;
337 }
338 }
339 if (lastiblock[level] >= 0)
340 goto done;
341 }
342
343 /*
344 * All whole direct blocks or frags.
345 */
346 for (i = NDADDR - 1; i > lastblock; i--) {
347 register long bsize;
348
349 bn = oip->i_db[i];
350 if (bn == 0)
351 continue;
352 oip->i_db[i] = 0;
353 bsize = blksize(fs, oip, i);
354 ffs_blkfree(oip, bn, bsize);
355 blocksreleased += btodb(bsize);
356 }
357 if (lastblock < 0)
358 goto done;
359
360 /*
361 * Finally, look for a change in size of the
362 * last direct block; release any frags.
363 */
364 bn = oip->i_db[lastblock];
365 if (bn != 0) {
366 long oldspace, newspace;
367
368 /*
369 * Calculate amount of space we're giving
370 * back as old block size minus new block size.
371 */
372 oldspace = blksize(fs, oip, lastblock);
373 oip->i_size = length;
374 newspace = blksize(fs, oip, lastblock);
375 if (newspace == 0)
376 panic("ffs_truncate: newspace");
377 if (oldspace - newspace > 0) {
378 /*
379 * Block number of space to be free'd is
380 * the old block # plus the number of frags
381 * required for the storage we're keeping.
382 */
383 bn += numfrags(fs, newspace);
384 ffs_blkfree(oip, bn, oldspace - newspace);
385 blocksreleased += btodb(oldspace - newspace);
386 }
387 }
388done:
389#ifdef DIAGNOSTIC
390 for (level = SINGLE; level <= TRIPLE; level++)
391 if (newblks[NDADDR + level] != oip->i_ib[level])
392 panic("ffs_truncate1");
393 for (i = 0; i < NDADDR; i++)
394 if (newblks[i] != oip->i_db[i])
395 panic("ffs_truncate2");
396 if (length == 0 &&
397 (!TAILQ_EMPTY(&ovp->v_dirtyblkhd) ||
398 !TAILQ_EMPTY(&ovp->v_cleanblkhd)))
399 panic("ffs_truncate3");
400#endif /* DIAGNOSTIC */
401 /*
402 * Put back the real size.
403 */
404 oip->i_size = length;
405 oip->i_blocks -= blocksreleased;
406
407 if (oip->i_blocks < 0) /* sanity */
408 oip->i_blocks = 0;
409 oip->i_flag |= IN_CHANGE;
410#ifdef QUOTA
411 (void) chkdq(oip, -blocksreleased, NOCRED, 0);
412#endif
413 return (allerror);
414}
415
416/*
417 * Release blocks associated with the inode ip and stored in the indirect
418 * block bn. Blocks are free'd in LIFO order up to (but not including)
419 * lastbn. If level is greater than SINGLE, the block is an indirect block
420 * and recursive calls to indirtrunc must be used to cleanse other indirect
421 * blocks.
422 *
423 * NB: triple indirect blocks are untested.
424 */
425static int
426ffs_indirtrunc(ip, lbn, dbn, lastbn, level, countp)
427 register struct inode *ip;
428 ufs_daddr_t lbn, lastbn;
429 ufs_daddr_t dbn;
430 int level;
431 long *countp;
432{
433 register int i;
434 struct buf *bp;
435 register struct fs *fs = ip->i_fs;
436 register ufs_daddr_t *bap;
437 struct vnode *vp;
438 ufs_daddr_t *copy = NULL, nb, nlbn, last;
439 long blkcount, factor;
440 int nblocks, blocksreleased = 0;
441 int error = 0, allerror = 0;
442
443 /*
444 * Calculate index in current block of last
445 * block to be kept. -1 indicates the entire
446 * block so we need not calculate the index.
447 */
448 factor = 1;
449 for (i = SINGLE; i < level; i++)
450 factor *= NINDIR(fs);
451 last = lastbn;
452 if (lastbn > 0)
453 last /= factor;
454 nblocks = btodb(fs->fs_bsize);
455 /*
456 * Get buffer of block pointers, zero those entries corresponding
457 * to blocks to be free'd, and update on disk copy first. Since
458 * double(triple) indirect before single(double) indirect, calls
459 * to bmap on these blocks will fail. However, we already have
460 * the on disk address, so we have to set the b_blkno field
461 * explicitly instead of letting bread do everything for us.
462 */
463 vp = ITOV(ip);
464 bp = getblk(vp, lbn, (int)fs->fs_bsize, 0, 0);
465 if ((bp->b_flags & B_CACHE) == 0) {
466 curproc->p_stats->p_ru.ru_inblock++; /* pay for read */
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_blkno = dbn;
472 vfs_busy_pages(bp, 0);
473 VOP_STRATEGY(bp->b_vp, bp);
474 error = biowait(bp);
475 }
476 if (error) {
477 brelse(bp);
478 *countp = 0;
479 return (error);
480 }
481
482 bap = (ufs_daddr_t *)bp->b_data;
483 if (lastbn != -1) {
484 MALLOC(copy, ufs_daddr_t *, fs->fs_bsize, M_TEMP, M_WAITOK);
485 bcopy((caddr_t)bap, (caddr_t)copy, (u_int)fs->fs_bsize);
486 bzero((caddr_t)&bap[last + 1],
487 (u_int)(NINDIR(fs) - (last + 1)) * sizeof (ufs_daddr_t));
488 if (DOINGASYNC(vp)) {
489 bawrite(bp);
490 } else {
491 error = bwrite(bp);
492 if (error)
493 allerror = error;
494 }
495 bap = copy;
496 }
497
498 /*
499 * Recursively free totally unused blocks.
500 */
501 for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
502 i--, nlbn += factor) {
503 nb = bap[i];
504 if (nb == 0)
505 continue;
506 if (level > SINGLE) {
507 if ((error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
508 (ufs_daddr_t)-1, level - 1, &blkcount)) != 0)
509 allerror = error;
510 blocksreleased += blkcount;
511 }
512 ffs_blkfree(ip, nb, fs->fs_bsize);
513 blocksreleased += nblocks;
514 }
515
516 /*
517 * Recursively free last partial block.
518 */
519 if (level > SINGLE && lastbn >= 0) {
520 last = lastbn % factor;
521 nb = bap[i];
522 if (nb != 0) {
523 error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
524 last, level - 1, &blkcount);
525 if (error)
526 allerror = error;
527 blocksreleased += blkcount;
528 }
529 }
530 if (copy != NULL) {
531 FREE(copy, M_TEMP);
532 } else {
533 bp->b_flags |= B_INVAL | B_NOCACHE;
534 brelse(bp);
535 }
536
537 *countp = blocksreleased;
538 return (allerror);
539}