xref: /freebsd/sys/ufs/ffs/ffs_inode.c (revision 0ea3482342b4d7d6e71f3007ce4dafe445c639fd)
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	8.5 (Berkeley) 12/30/93
34  * $Id: ffs_inode.c,v 1.15 1995/08/16 13:16:58 davidg Exp $
35  */
36 
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/mount.h>
40 #include <sys/proc.h>
41 #include <sys/file.h>
42 #include <sys/buf.h>
43 #include <sys/vnode.h>
44 #include <sys/kernel.h>
45 #include <sys/malloc.h>
46 #include <sys/resourcevar.h>
47 
48 #include <vm/vm.h>
49 
50 #include <ufs/ufs/quota.h>
51 #include <ufs/ufs/inode.h>
52 #include <ufs/ufs/ufsmount.h>
53 #include <ufs/ufs/ufs_extern.h>
54 
55 #include <ufs/ffs/fs.h>
56 #include <ufs/ffs/ffs_extern.h>
57 
58 static int ffs_indirtrunc __P((struct inode *, daddr_t, daddr_t, daddr_t, int,
59 	    long *));
60 
61 int
62 ffs_init()
63 {
64 	return (ufs_init());
65 }
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. The IN_MODIFIED
70  * flag is used to specify that the inode needs to be updated even if none
71  * of the times needs to be updated. The access and modified times are taken
72  * from the second and third parameters; the inode change time is always
73  * taken from the current time. If waitfor is set, then wait for the disk
74  * write of the inode to complete.
75  */
76 int
77 ffs_update(ap)
78 	struct vop_update_args /* {
79 		struct vnode *a_vp;
80 		struct timeval *a_access;
81 		struct timeval *a_modify;
82 		int a_waitfor;
83 	} */ *ap;
84 {
85 	register struct fs *fs;
86 	struct buf *bp;
87 	struct inode *ip;
88 	int error;
89 	time_t tv_sec;
90 
91 	ip = VTOI(ap->a_vp);
92 	if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY) {
93 		ip->i_flag &=
94 		    ~(IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE);
95 		return (0);
96 	}
97 	if ((ip->i_flag &
98 	    (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0)
99 		return (0);
100 	/*
101 	 * Use a copy of the current time to get consistent timestamps
102 	 * (a_access and a_modify are sometimes aliases for &time).
103 	 *
104 	 * XXX in 2.0, a_access and a_modify are often pointers to the
105 	 * same copy of `time'.  This is not as good.  Some callers forget
106 	 * to make a copy; others make a copy too early (before the i/o
107 	 * has completed)...
108 	 *
109 	 * XXX there should be a function or macro for reading the time
110 	 * (e.g., some machines may require splclock()).
111 	 */
112 	tv_sec = time.tv_sec;
113 	if (ip->i_flag & IN_ACCESS)
114 		ip->i_atime.ts_sec =
115 		    (ap->a_access == &time ? tv_sec : ap->a_access->tv_sec);
116 	if (ip->i_flag & IN_UPDATE) {
117 		ip->i_mtime.ts_sec =
118 		    (ap->a_modify == &time ? tv_sec : ap->a_modify->tv_sec);
119 		ip->i_modrev++;
120 	}
121 	if (ip->i_flag & IN_CHANGE)
122 		ip->i_ctime.ts_sec = tv_sec;
123 	ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE);
124 	fs = ip->i_fs;
125 	/*
126 	 * Ensure that uid and gid are correct. This is a temporary
127 	 * fix until fsck has been changed to do the update.
128 	 */
129 	if (fs->fs_inodefmt < FS_44INODEFMT) {		/* XXX */
130 		ip->i_din.di_ouid = ip->i_uid;		/* XXX */
131 		ip->i_din.di_ogid = ip->i_gid;		/* XXX */
132 	}						/* XXX */
133 	error = bread(ip->i_devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
134 		(int)fs->fs_bsize, NOCRED, &bp);
135 	if (error) {
136 		brelse(bp);
137 		return (error);
138 	}
139 	*((struct dinode *)bp->b_data +
140 	    ino_to_fsbo(fs, ip->i_number)) = ip->i_din;
141 	if (ap->a_waitfor && (ap->a_vp->v_mount->mnt_flag & MNT_ASYNC) == 0)
142 		return (bwrite(bp));
143 	else {
144 		bdwrite(bp);
145 		return (0);
146 	}
147 }
148 
149 #define	SINGLE	0	/* index of single indirect block */
150 #define	DOUBLE	1	/* index of double indirect block */
151 #define	TRIPLE	2	/* index of triple indirect block */
152 /*
153  * Truncate the inode oip to at most length size, freeing the
154  * disk blocks.
155  */
156 int
157 ffs_truncate(ap)
158 	struct vop_truncate_args /* {
159 		struct vnode *a_vp;
160 		off_t a_length;
161 		int a_flags;
162 		struct ucred *a_cred;
163 		struct proc *a_p;
164 	} */ *ap;
165 {
166 	register struct vnode *ovp = ap->a_vp;
167 	register daddr_t lastblock;
168 	register struct inode *oip;
169 	daddr_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR];
170 	daddr_t oldblks[NDADDR + NIADDR], newblks[NDADDR + NIADDR];
171 	off_t length = ap->a_length;
172 	register struct fs *fs;
173 	struct buf *bp;
174 	int offset, size, level;
175 	long count, nblocks, vflags, blocksreleased = 0;
176 	struct timeval tv;
177 	register int i;
178 	int aflags, error, allerror;
179 	off_t osize;
180 
181 	oip = VTOI(ovp);
182 	fs = oip->i_fs;
183 	if (length < 0 || length > fs->fs_maxfilesize)
184 		return (EINVAL);
185 	tv = time;
186 	if (ovp->v_type == VLNK &&
187 	    (oip->i_size < ovp->v_mount->mnt_maxsymlinklen || oip->i_din.di_blocks == 0)) {
188 #ifdef DIAGNOSTIC
189 		if (length != 0)
190 			panic("ffs_truncate: partial truncate of symlink");
191 #endif
192 		bzero((char *)&oip->i_shortlink, (u_int)oip->i_size);
193 		oip->i_size = 0;
194 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
195 		return (VOP_UPDATE(ovp, &tv, &tv, 1));
196 	}
197 	if (oip->i_size == length) {
198 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
199 		return (VOP_UPDATE(ovp, &tv, &tv, 0));
200 	}
201 #ifdef QUOTA
202 	error = getinoquota(oip);
203 	if (error)
204 		return (error);
205 #endif
206 	osize = oip->i_size;
207 	/*
208 	 * Lengthen the size of the file. We must ensure that the
209 	 * last byte of the file is allocated. Since the smallest
210 	 * value of osize is 0, length will be at least 1.
211 	 */
212 	if (osize < length) {
213 		offset = blkoff(fs, length - 1);
214 		lbn = lblkno(fs, length - 1);
215 		aflags = B_CLRBUF;
216 		if (ap->a_flags & IO_SYNC)
217 			aflags |= B_SYNC;
218 		error = ffs_balloc(oip, lbn, offset + 1, ap->a_cred,
219 		    &bp, aflags);
220 		if (error)
221 			return (error);
222 		oip->i_size = length;
223 		if (aflags & B_SYNC)
224 			bwrite(bp);
225 		else if (ovp->v_mount->mnt_flag & MNT_ASYNC)
226 			bdwrite(bp);
227 		else
228 			bawrite(bp);
229 		vnode_pager_setsize(ovp, (u_long)length);
230 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
231 		return (VOP_UPDATE(ovp, &tv, &tv, 1));
232 	}
233 	/*
234 	 * Shorten the size of the file. If the file is not being
235 	 * truncated to a block boundry, the contents of the
236 	 * partial block following the end of the file must be
237 	 * zero'ed in case it ever become accessable again because
238 	 * of subsequent file growth.
239 	 */
240 	offset = blkoff(fs, length);
241 	if (offset == 0) {
242 		oip->i_size = length;
243 	} else {
244 		lbn = lblkno(fs, length);
245 		aflags = B_CLRBUF;
246 		if (ap->a_flags & IO_SYNC)
247 			aflags |= B_SYNC;
248 		error = ffs_balloc(oip, lbn, offset, ap->a_cred, &bp, aflags);
249 		if (error)
250 			return (error);
251 		oip->i_size = length;
252 		size = blksize(fs, oip, lbn);
253 		bzero((char *)bp->b_data + offset, (u_int)(size - offset));
254 		allocbuf(bp, size);
255 		if (aflags & B_SYNC)
256 			bwrite(bp);
257 		else if (ovp->v_mount->mnt_flag & MNT_ASYNC)
258 			bdwrite(bp);
259 		else
260 			bawrite(bp);
261 	}
262 	/*
263 	 * Calculate index into inode's block list of
264 	 * last direct and indirect blocks (if any)
265 	 * which we want to keep.  Lastblock is -1 when
266 	 * the file is truncated to 0.
267 	 */
268 	lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1;
269 	lastiblock[SINGLE] = lastblock - NDADDR;
270 	lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
271 	lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
272 	nblocks = btodb(fs->fs_bsize);
273 	/*
274 	 * Update file and block pointers on disk before we start freeing
275 	 * blocks.  If we crash before free'ing blocks below, the blocks
276 	 * will be returned to the free list.  lastiblock values are also
277 	 * normalized to -1 for calls to ffs_indirtrunc below.
278 	 */
279 	bcopy((caddr_t)&oip->i_db[0], (caddr_t)oldblks, sizeof oldblks);
280 	for (level = TRIPLE; level >= SINGLE; level--)
281 		if (lastiblock[level] < 0) {
282 			oip->i_ib[level] = 0;
283 			lastiblock[level] = -1;
284 		}
285 	for (i = NDADDR - 1; i > lastblock; i--)
286 		oip->i_db[i] = 0;
287 	oip->i_flag |= IN_CHANGE | IN_UPDATE;
288 	error = VOP_UPDATE(ovp, &tv, &tv, 1);
289 	if (error)
290 		allerror = error;
291 	/*
292 	 * Having written the new inode to disk, save its new configuration
293 	 * and put back the old block pointers long enough to process them.
294 	 * Note that we save the new block configuration so we can check it
295 	 * when we are done.
296 	 */
297 	bcopy((caddr_t)&oip->i_db[0], (caddr_t)newblks, sizeof newblks);
298 	bcopy((caddr_t)oldblks, (caddr_t)&oip->i_db[0], sizeof oldblks);
299 	oip->i_size = osize;
300 	vflags = ((length > 0) ? V_SAVE : 0) | V_SAVEMETA;
301 	allerror = vinvalbuf(ovp, vflags, ap->a_cred, ap->a_p, 0, 0);
302 
303 	/*
304 	 * Indirect blocks first.
305 	 */
306 	indir_lbn[SINGLE] = -NDADDR;
307 	indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
308 	indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
309 	for (level = TRIPLE; level >= SINGLE; level--) {
310 		bn = oip->i_ib[level];
311 		if (bn != 0) {
312 			error = ffs_indirtrunc(oip, indir_lbn[level],
313 			    fsbtodb(fs, bn), lastiblock[level], level, &count);
314 			if (error)
315 				allerror = error;
316 			blocksreleased += count;
317 			if (lastiblock[level] < 0) {
318 				oip->i_ib[level] = 0;
319 				ffs_blkfree(oip, bn, fs->fs_bsize);
320 				blocksreleased += nblocks;
321 			}
322 		}
323 		if (lastiblock[level] >= 0)
324 			goto done;
325 	}
326 
327 	/*
328 	 * All whole direct blocks or frags.
329 	 */
330 	for (i = NDADDR - 1; i > lastblock; i--) {
331 		register long bsize;
332 
333 		bn = oip->i_db[i];
334 		if (bn == 0)
335 			continue;
336 		oip->i_db[i] = 0;
337 		bsize = blksize(fs, oip, i);
338 		ffs_blkfree(oip, bn, bsize);
339 		blocksreleased += btodb(bsize);
340 	}
341 	if (lastblock < 0)
342 		goto done;
343 
344 	/*
345 	 * Finally, look for a change in size of the
346 	 * last direct block; release any frags.
347 	 */
348 	bn = oip->i_db[lastblock];
349 	if (bn != 0) {
350 		long oldspace, newspace;
351 
352 		/*
353 		 * Calculate amount of space we're giving
354 		 * back as old block size minus new block size.
355 		 */
356 		oldspace = blksize(fs, oip, lastblock);
357 		oip->i_size = length;
358 		newspace = blksize(fs, oip, lastblock);
359 		if (newspace == 0)
360 			panic("ffs_truncate: newspace");
361 		if (oldspace - newspace > 0) {
362 			/*
363 			 * Block number of space to be free'd is
364 			 * the old block # plus the number of frags
365 			 * required for the storage we're keeping.
366 			 */
367 			bn += numfrags(fs, newspace);
368 			ffs_blkfree(oip, bn, oldspace - newspace);
369 			blocksreleased += btodb(oldspace - newspace);
370 		}
371 	}
372 done:
373 #ifdef DIAGNOSTIC
374 	for (level = SINGLE; level <= TRIPLE; level++)
375 		if (newblks[NDADDR + level] != oip->i_ib[level])
376 			panic("ffs_truncate1");
377 	for (i = 0; i < NDADDR; i++)
378 		if (newblks[i] != oip->i_db[i])
379 			panic("ffs_truncate2");
380 	if (length == 0 &&
381 	    (ovp->v_dirtyblkhd.lh_first || ovp->v_cleanblkhd.lh_first))
382 		panic("ffs_truncate3");
383 #endif /* DIAGNOSTIC */
384 	/*
385 	 * Put back the real size.
386 	 */
387 	oip->i_size = length;
388 	oip->i_blocks -= blocksreleased;
389 	if (oip->i_blocks < 0)			/* sanity */
390 		oip->i_blocks = 0;
391 	oip->i_flag |= IN_CHANGE;
392 	vnode_pager_setsize(ovp, (u_long)length);
393 #ifdef QUOTA
394 	(void) chkdq(oip, -blocksreleased, NOCRED, 0);
395 #endif
396 	return (allerror);
397 }
398 
399 /*
400  * Release blocks associated with the inode ip and stored in the indirect
401  * block bn.  Blocks are free'd in LIFO order up to (but not including)
402  * lastbn.  If level is greater than SINGLE, the block is an indirect block
403  * and recursive calls to indirtrunc must be used to cleanse other indirect
404  * blocks.
405  *
406  * NB: triple indirect blocks are untested.
407  */
408 static int
409 ffs_indirtrunc(ip, lbn, dbn, lastbn, level, countp)
410 	register struct inode *ip;
411 	daddr_t lbn, lastbn;
412 	daddr_t dbn;
413 	int level;
414 	long *countp;
415 {
416 	register int i;
417 	struct buf *bp;
418 	register struct fs *fs = ip->i_fs;
419 	register daddr_t *bap;
420 	struct vnode *vp;
421 	daddr_t *copy, nb, nlbn, last;
422 	long blkcount, factor;
423 	int nblocks, blocksreleased = 0;
424 	int error = 0, allerror = 0;
425 
426 	/*
427 	 * Calculate index in current block of last
428 	 * block to be kept.  -1 indicates the entire
429 	 * block so we need not calculate the index.
430 	 */
431 	factor = 1;
432 	for (i = SINGLE; i < level; i++)
433 		factor *= NINDIR(fs);
434 	last = lastbn;
435 	if (lastbn > 0)
436 		last /= factor;
437 	nblocks = btodb(fs->fs_bsize);
438 	/*
439 	 * Get buffer of block pointers, zero those entries corresponding
440 	 * to blocks to be free'd, and update on disk copy first.  Since
441 	 * double(triple) indirect before single(double) indirect, calls
442 	 * to bmap on these blocks will fail.  However, we already have
443 	 * the on disk address, so we have to set the b_blkno field
444 	 * explicitly instead of letting bread do everything for us.
445 	 */
446 	vp = ITOV(ip);
447 	bp = getblk(vp, lbn, (int)fs->fs_bsize, 0, 0);
448 	if ((bp->b_flags & B_CACHE) == 0) {
449 		curproc->p_stats->p_ru.ru_inblock++;	/* pay for read */
450 		bp->b_flags |= B_READ;
451 		if (bp->b_bcount > bp->b_bufsize)
452 			panic("ffs_indirtrunc: bad buffer size");
453 		bp->b_blkno = dbn;
454 		vfs_busy_pages(bp, 0);
455 		VOP_STRATEGY(bp);
456 		error = biowait(bp);
457 	}
458 	if (error) {
459 		brelse(bp);
460 		*countp = 0;
461 		return (error);
462 	}
463 
464 	bap = (daddr_t *)bp->b_data;
465 	MALLOC(copy, daddr_t *, fs->fs_bsize, M_TEMP, M_WAITOK);
466 	bcopy((caddr_t)bap, (caddr_t)copy, (u_int)fs->fs_bsize);
467 	bzero((caddr_t)&bap[last + 1],
468 	  (u_int)(NINDIR(fs) - (last + 1)) * sizeof (daddr_t));
469 	if (last == -1)
470 		bp->b_flags |= B_INVAL;
471 	error = bwrite(bp);
472 	if (error)
473 		allerror = error;
474 	bap = copy;
475 
476 	/*
477 	 * Recursively free totally unused blocks.
478 	 */
479 	for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
480 	    i--, nlbn += factor) {
481 		nb = bap[i];
482 		if (nb == 0)
483 			continue;
484 		if (level > SINGLE) {
485 			error = ffs_indirtrunc(ip, nlbn,
486 			    fsbtodb(fs, nb), (daddr_t)-1, level - 1, &blkcount);
487 			if (error)
488 				allerror = error;
489 			blocksreleased += blkcount;
490 		}
491 		ffs_blkfree(ip, nb, fs->fs_bsize);
492 		blocksreleased += nblocks;
493 	}
494 
495 	/*
496 	 * Recursively free last partial block.
497 	 */
498 	if (level > SINGLE && lastbn >= 0) {
499 		last = lastbn % factor;
500 		nb = bap[i];
501 		if (nb != 0) {
502 			error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
503 			    last, level - 1, &blkcount);
504 			if (error)
505 				allerror = error;
506 			blocksreleased += blkcount;
507 		}
508 	}
509 	FREE(copy, M_TEMP);
510 	*countp = blocksreleased;
511 	return (allerror);
512 }
513