Properly assert the state of the MP lock in the async syscall message
[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 $
6bae6177 35 * $DragonFly: src/sys/vfs/ufs/ffs_inode.c,v 1.14 2005/04/15 19:08:32 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
1f2de5d4
MD
54#include "quota.h"
55#include "ufsmount.h"
56#include "inode.h"
57#include "ufs_extern.h"
984263bc 58
1f2de5d4
MD
59#include "fs.h"
60#include "ffs_extern.h"
984263bc 61
12e4aaff
MD
62#include <vm/vm_page2.h>
63
a6ee311a
RG
64static int ffs_indirtrunc (struct inode *, ufs_daddr_t, ufs_daddr_t,
65 ufs_daddr_t, int, long *);
984263bc
MD
66
67/*
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.
75 */
76int
0973c589 77ffs_update(struct vnode *vp, int waitfor)
984263bc 78{
3ff2135f 79 struct fs *fs;
984263bc
MD
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)),
3b568787 101 (int)fs->fs_bsize, &bp);
984263bc
MD
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
0973c589
CP
132ffs_truncate(struct vnode *vp, off_t length, int flags, struct ucred *cred,
133 struct thread *td)
984263bc 134{
3ff2135f 135 struct vnode *ovp = vp;
984263bc 136 ufs_daddr_t lastblock;
3ff2135f 137 struct inode *oip;
984263bc
MD
138 ufs_daddr_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR];
139 ufs_daddr_t oldblks[NDADDR + NIADDR], newblks[NDADDR + NIADDR];
3ff2135f 140 struct fs *fs;
984263bc
MD
141 struct buf *bp;
142 int offset, size, level;
143 long count, nblocks, blocksreleased = 0;
3ff2135f 144 int i;
984263bc
MD
145 int aflags, error, allerror;
146 off_t osize;
147
148 oip = VTOI(ovp);
149 fs = oip->i_fs;
150 if (length < 0)
151 return (EINVAL);
152 if (length > fs->fs_maxfilesize)
153 return (EFBIG);
154 if (ovp->v_type == VLNK &&
155 (oip->i_size < ovp->v_mount->mnt_maxsymlinklen || oip->i_din.di_blocks == 0)) {
156#ifdef DIAGNOSTIC
157 if (length != 0)
158 panic("ffs_truncate: partial truncate of symlink");
f719c866
DR
159#endif /* DIAGNOSTIC */
160 bzero((char *)&oip->i_shortlink, (uint)oip->i_size);
984263bc
MD
161 oip->i_size = 0;
162 oip->i_flag |= IN_CHANGE | IN_UPDATE;
163 return (UFS_UPDATE(ovp, 1));
164 }
165 if (oip->i_size == length) {
166 oip->i_flag |= IN_CHANGE | IN_UPDATE;
167 return (UFS_UPDATE(ovp, 0));
168 }
169 if (fs->fs_ronly)
170 panic("ffs_truncate: read-only filesystem");
171#ifdef QUOTA
172 error = getinoquota(oip);
173 if (error)
174 return (error);
175#endif
176 ovp->v_lasta = ovp->v_clen = ovp->v_cstart = ovp->v_lastw = 0;
177 if (DOINGSOFTDEP(ovp)) {
178 if (length > 0 || softdep_slowdown(ovp)) {
179 /*
180 * If a file is only partially truncated, then
181 * we have to clean up the data structures
182 * describing the allocation past the truncation
183 * point. Finding and deallocating those structures
184 * is a lot of work. Since partial truncation occurs
185 * rarely, we solve the problem by syncing the file
186 * so that it will have no data structures left.
187 */
3b568787 188 if ((error = VOP_FSYNC(ovp, MNT_WAIT, td)) != 0)
984263bc
MD
189 return (error);
190 } else {
191#ifdef QUOTA
192 (void) chkdq(oip, -oip->i_blocks, NOCRED, 0);
193#endif
194 softdep_setup_freeblocks(oip, length);
3b568787 195 vinvalbuf(ovp, 0, td, 0, 0);
0e8fa33c 196 vnode_pager_setsize(ovp, 0);
984263bc
MD
197 oip->i_flag |= IN_CHANGE | IN_UPDATE;
198 return (ffs_update(ovp, 0));
199 }
200 }
201 osize = oip->i_size;
202 /*
203 * Lengthen the size of the file. We must ensure that the
204 * last byte of the file is allocated. Since the smallest
205 * value of osize is 0, length will be at least 1.
206 */
207 if (osize < length) {
208 vnode_pager_setsize(ovp, length);
209 aflags = B_CLRBUF;
210 if (flags & IO_SYNC)
211 aflags |= B_SYNC;
212 error = VOP_BALLOC(ovp, length - 1, 1,
213 cred, aflags, &bp);
214 if (error)
215 return (error);
216 oip->i_size = length;
217 if (bp->b_bufsize == fs->fs_bsize)
218 bp->b_flags |= B_CLUSTEROK;
219 if (aflags & B_SYNC)
220 bwrite(bp);
221 else
222 bawrite(bp);
223 oip->i_flag |= IN_CHANGE | IN_UPDATE;
224 return (UFS_UPDATE(ovp, 1));
225 }
226 /*
227 * Shorten the size of the file. If the file is not being
228 * truncated to a block boundary, the contents of the
229 * partial block following the end of the file must be
230 * zero'ed in case it ever becomes accessible again because
231 * of subsequent file growth. Directories however are not
232 * zero'ed as they should grow back initialized to empty.
233 */
234 offset = blkoff(fs, length);
235 if (offset == 0) {
236 oip->i_size = length;
237 } else {
238 lbn = lblkno(fs, length);
239 aflags = B_CLRBUF;
240 if (flags & IO_SYNC)
241 aflags |= B_SYNC;
242 error = VOP_BALLOC(ovp, length - 1, 1, cred, aflags, &bp);
243 if (error) {
244 return (error);
245 }
246 /*
247 * When we are doing soft updates and the UFS_BALLOC
248 * above fills in a direct block hole with a full sized
249 * block that will be truncated down to a fragment below,
250 * we must flush out the block dependency with an FSYNC
251 * so that we do not get a soft updates inconsistency
252 * when we create the fragment below.
253 */
254 if (DOINGSOFTDEP(ovp) && lbn < NDADDR &&
255 fragroundup(fs, blkoff(fs, length)) < fs->fs_bsize &&
3b568787 256 (error = VOP_FSYNC(ovp, MNT_WAIT, td)) != 0) {
984263bc
MD
257 return (error);
258 }
259 oip->i_size = length;
260 size = blksize(fs, oip, lbn);
261 if (ovp->v_type != VDIR)
262 bzero((char *)bp->b_data + offset,
f719c866 263 (uint)(size - offset));
984263bc
MD
264 /* Kirk's code has reallocbuf(bp, size, 1) here */
265 allocbuf(bp, size);
266 if (bp->b_bufsize == fs->fs_bsize)
267 bp->b_flags |= B_CLUSTEROK;
268 if (aflags & B_SYNC)
269 bwrite(bp);
270 else
271 bawrite(bp);
272 }
273 /*
274 * Calculate index into inode's block list of
275 * last direct and indirect blocks (if any)
276 * which we want to keep. Lastblock is -1 when
277 * the file is truncated to 0.
278 */
279 lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1;
280 lastiblock[SINGLE] = lastblock - NDADDR;
281 lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
282 lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
283 nblocks = btodb(fs->fs_bsize);
0f1737b3 284
984263bc
MD
285 /*
286 * Update file and block pointers on disk before we start freeing
287 * blocks. If we crash before free'ing blocks below, the blocks
288 * will be returned to the free list. lastiblock values are also
289 * normalized to -1 for calls to ffs_indirtrunc below.
290 */
291 bcopy((caddr_t)&oip->i_db[0], (caddr_t)oldblks, sizeof oldblks);
292 for (level = TRIPLE; level >= SINGLE; level--)
293 if (lastiblock[level] < 0) {
294 oip->i_ib[level] = 0;
295 lastiblock[level] = -1;
296 }
297 for (i = NDADDR - 1; i > lastblock; i--)
298 oip->i_db[i] = 0;
299 oip->i_flag |= IN_CHANGE | IN_UPDATE;
300 allerror = UFS_UPDATE(ovp, 1);
301
302 /*
303 * Having written the new inode to disk, save its new configuration
304 * and put back the old block pointers long enough to process them.
305 * Note that we save the new block configuration so we can check it
306 * when we are done.
307 */
308 bcopy((caddr_t)&oip->i_db[0], (caddr_t)newblks, sizeof newblks);
309 bcopy((caddr_t)oldblks, (caddr_t)&oip->i_db[0], sizeof oldblks);
310 oip->i_size = osize;
311
3b568787 312 error = vtruncbuf(ovp, td, length, fs->fs_bsize);
984263bc
MD
313 if (error && (allerror == 0))
314 allerror = error;
315
316 /*
317 * Indirect blocks first.
318 */
319 indir_lbn[SINGLE] = -NDADDR;
320 indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
321 indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
322 for (level = TRIPLE; level >= SINGLE; level--) {
323 bn = oip->i_ib[level];
324 if (bn != 0) {
325 error = ffs_indirtrunc(oip, indir_lbn[level],
326 fsbtodb(fs, bn), lastiblock[level], level, &count);
327 if (error)
328 allerror = error;
329 blocksreleased += count;
330 if (lastiblock[level] < 0) {
331 oip->i_ib[level] = 0;
332 ffs_blkfree(oip, bn, fs->fs_bsize);
333 blocksreleased += nblocks;
334 }
335 }
336 if (lastiblock[level] >= 0)
337 goto done;
338 }
339
340 /*
341 * All whole direct blocks or frags.
342 */
343 for (i = NDADDR - 1; i > lastblock; i--) {
3ff2135f 344 long bsize;
984263bc
MD
345
346 bn = oip->i_db[i];
347 if (bn == 0)
348 continue;
349 oip->i_db[i] = 0;
350 bsize = blksize(fs, oip, i);
351 ffs_blkfree(oip, bn, bsize);
352 blocksreleased += btodb(bsize);
353 }
354 if (lastblock < 0)
355 goto done;
356
357 /*
358 * Finally, look for a change in size of the
359 * last direct block; release any frags.
360 */
361 bn = oip->i_db[lastblock];
362 if (bn != 0) {
363 long oldspace, newspace;
364
365 /*
366 * Calculate amount of space we're giving
367 * back as old block size minus new block size.
368 */
369 oldspace = blksize(fs, oip, lastblock);
370 oip->i_size = length;
371 newspace = blksize(fs, oip, lastblock);
372 if (newspace == 0)
373 panic("ffs_truncate: newspace");
374 if (oldspace - newspace > 0) {
375 /*
376 * Block number of space to be free'd is
377 * the old block # plus the number of frags
378 * required for the storage we're keeping.
379 */
380 bn += numfrags(fs, newspace);
381 ffs_blkfree(oip, bn, oldspace - newspace);
382 blocksreleased += btodb(oldspace - newspace);
383 }
384 }
385done:
386#ifdef DIAGNOSTIC
387 for (level = SINGLE; level <= TRIPLE; level++)
388 if (newblks[NDADDR + level] != oip->i_ib[level])
389 panic("ffs_truncate1");
390 for (i = 0; i < NDADDR; i++)
391 if (newblks[i] != oip->i_db[i])
392 panic("ffs_truncate2");
393 if (length == 0 &&
6bae6177
MD
394 (!RB_EMPTY(&ovp->v_rbdirty_tree) ||
395 !RB_EMPTY(&ovp->v_rbclean_tree)))
984263bc
MD
396 panic("ffs_truncate3");
397#endif /* DIAGNOSTIC */
398 /*
399 * Put back the real size.
400 */
401 oip->i_size = length;
402 oip->i_blocks -= blocksreleased;
403
404 if (oip->i_blocks < 0) /* sanity */
405 oip->i_blocks = 0;
406 oip->i_flag |= IN_CHANGE;
407#ifdef QUOTA
408 (void) chkdq(oip, -blocksreleased, NOCRED, 0);
409#endif
410 return (allerror);
411}
412
413/*
414 * Release blocks associated with the inode ip and stored in the indirect
415 * block bn. Blocks are free'd in LIFO order up to (but not including)
416 * lastbn. If level is greater than SINGLE, the block is an indirect block
417 * and recursive calls to indirtrunc must be used to cleanse other indirect
418 * blocks.
419 *
420 * NB: triple indirect blocks are untested.
421 */
422static int
0973c589
CP
423ffs_indirtrunc(struct inode *ip, ufs_daddr_t lbn, ufs_daddr_t dbn,
424 ufs_daddr_t lastbn, int level, long *countp)
984263bc 425{
3ff2135f 426 int i;
984263bc 427 struct buf *bp;
3ff2135f
RG
428 struct fs *fs = ip->i_fs;
429 ufs_daddr_t *bap;
984263bc
MD
430 struct vnode *vp;
431 ufs_daddr_t *copy = NULL, nb, nlbn, last;
432 long blkcount, factor;
433 int nblocks, blocksreleased = 0;
434 int error = 0, allerror = 0;
435
436 /*
437 * Calculate index in current block of last
438 * block to be kept. -1 indicates the entire
439 * block so we need not calculate the index.
440 */
441 factor = 1;
442 for (i = SINGLE; i < level; i++)
443 factor *= NINDIR(fs);
444 last = lastbn;
445 if (lastbn > 0)
446 last /= factor;
447 nblocks = btodb(fs->fs_bsize);
448 /*
449 * Get buffer of block pointers, zero those entries corresponding
450 * to blocks to be free'd, and update on disk copy first. Since
451 * double(triple) indirect before single(double) indirect, calls
452 * to bmap on these blocks will fail. However, we already have
453 * the on disk address, so we have to set the b_blkno field
454 * explicitly instead of letting bread do everything for us.
455 */
456 vp = ITOV(ip);
457 bp = getblk(vp, lbn, (int)fs->fs_bsize, 0, 0);
458 if ((bp->b_flags & B_CACHE) == 0) {
984263bc
MD
459 bp->b_flags |= B_READ;
460 bp->b_flags &= ~(B_ERROR|B_INVAL);
461 if (bp->b_bcount > bp->b_bufsize)
462 panic("ffs_indirtrunc: bad buffer size");
463 bp->b_blkno = dbn;
464 vfs_busy_pages(bp, 0);
465 VOP_STRATEGY(bp->b_vp, bp);
466 error = biowait(bp);
467 }
468 if (error) {
469 brelse(bp);
470 *countp = 0;
471 return (error);
472 }
473
474 bap = (ufs_daddr_t *)bp->b_data;
475 if (lastbn != -1) {
476 MALLOC(copy, ufs_daddr_t *, fs->fs_bsize, M_TEMP, M_WAITOK);
f719c866 477 bcopy((caddr_t)bap, (caddr_t)copy, (uint)fs->fs_bsize);
984263bc 478 bzero((caddr_t)&bap[last + 1],
f719c866 479 (uint)(NINDIR(fs) - (last + 1)) * sizeof (ufs_daddr_t));
984263bc
MD
480 if (DOINGASYNC(vp)) {
481 bawrite(bp);
482 } else {
483 error = bwrite(bp);
484 if (error)
485 allerror = error;
486 }
487 bap = copy;
488 }
489
490 /*
491 * Recursively free totally unused blocks.
492 */
493 for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
494 i--, nlbn += factor) {
495 nb = bap[i];
496 if (nb == 0)
497 continue;
498 if (level > SINGLE) {
499 if ((error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
500 (ufs_daddr_t)-1, level - 1, &blkcount)) != 0)
501 allerror = error;
502 blocksreleased += blkcount;
503 }
504 ffs_blkfree(ip, nb, fs->fs_bsize);
505 blocksreleased += nblocks;
506 }
507
508 /*
509 * Recursively free last partial block.
510 */
511 if (level > SINGLE && lastbn >= 0) {
512 last = lastbn % factor;
513 nb = bap[i];
514 if (nb != 0) {
515 error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
516 last, level - 1, &blkcount);
517 if (error)
518 allerror = error;
519 blocksreleased += blkcount;
520 }
521 }
522 if (copy != NULL) {
523 FREE(copy, M_TEMP);
524 } else {
525 bp->b_flags |= B_INVAL | B_NOCACHE;
526 brelse(bp);
527 }
528
529 *countp = blocksreleased;
530 return (allerror);
531}