xref: /freebsd/sys/ufs/ffs/ffs_inode.c (revision 17ee9d00bc1ae1e598c38f25826f861e4bc6c3ce)
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 1.10 1994/12/27 14:44:42 bde 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/trace.h>
47 #include <sys/resourcevar.h>
48 
49 #include <vm/vm.h>
50 
51 #include <ufs/ufs/quota.h>
52 #include <ufs/ufs/inode.h>
53 #include <ufs/ufs/ufsmount.h>
54 #include <ufs/ufs/ufs_extern.h>
55 
56 #include <ufs/ffs/fs.h>
57 #include <ufs/ffs/ffs_extern.h>
58 
59 static int ffs_indirtrunc __P((struct inode *, daddr_t, daddr_t, daddr_t, int,
60 	    long *));
61 
62 int
63 ffs_init()
64 {
65 	return (ufs_init());
66 }
67 
68 /*
69  * Update the access, modified, and inode change times as specified by the
70  * IN_ACCESS, IN_UPDATE, and IN_CHANGE flags respectively. The IN_MODIFIED
71  * flag is used to specify that the inode needs to be updated even if none
72  * of the times needs to be updated. The access and modified times are taken
73  * from the second and third parameters; the inode change time is always
74  * taken from the current time. If waitfor is set, then wait for the disk
75  * write of the inode to complete.
76  */
77 int
78 ffs_update(ap)
79 	struct vop_update_args /* {
80 		struct vnode *a_vp;
81 		struct timeval *a_access;
82 		struct timeval *a_modify;
83 		int a_waitfor;
84 	} */ *ap;
85 {
86 	register struct fs *fs;
87 	struct buf *bp;
88 	struct inode *ip;
89 	int error;
90 	time_t tv_sec;
91 
92 	ip = VTOI(ap->a_vp);
93 	if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY) {
94 		ip->i_flag &=
95 		    ~(IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE);
96 		return (0);
97 	}
98 	if ((ip->i_flag &
99 	    (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0)
100 		return (0);
101 	/*
102 	 * Use a copy of the current time to get consistent timestamps
103 	 * (a_access and a_modify are sometimes aliases for &time).
104 	 *
105 	 * XXX in 2.0, a_access and a_modify are often pointers to the
106 	 * same copy of `time'.  This is not as good.  Some callers forget
107 	 * to make a copy; others make a copy too early (before the i/o
108 	 * has completed)...
109 	 *
110 	 * XXX there should be a function or macro for reading the time
111 	 * (e.g., some machines may require splclock()).
112 	 */
113 	tv_sec = time.tv_sec;
114 	if (ip->i_flag & IN_ACCESS)
115 		ip->i_atime.ts_sec =
116 		    (ap->a_access == &time ? tv_sec : ap->a_access->tv_sec);
117 	if (ip->i_flag & IN_UPDATE) {
118 		ip->i_mtime.ts_sec =
119 		    (ap->a_modify == &time ? tv_sec : ap->a_modify->tv_sec);
120 		ip->i_modrev++;
121 	}
122 	if (ip->i_flag & IN_CHANGE)
123 		ip->i_ctime.ts_sec = tv_sec;
124 	ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE);
125 	fs = ip->i_fs;
126 	/*
127 	 * Ensure that uid and gid are correct. This is a temporary
128 	 * fix until fsck has been changed to do the update.
129 	 */
130 	if (fs->fs_inodefmt < FS_44INODEFMT) {		/* XXX */
131 		ip->i_din.di_ouid = ip->i_uid;		/* XXX */
132 		ip->i_din.di_ogid = ip->i_gid;		/* XXX */
133 	}						/* XXX */
134 	error = bread(ip->i_devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
135 		(int)fs->fs_bsize, NOCRED, &bp);
136 	if (error) {
137 		brelse(bp);
138 		return (error);
139 	}
140 	*((struct dinode *)bp->b_data +
141 	    ino_to_fsbo(fs, ip->i_number)) = ip->i_din;
142 	if (ap->a_waitfor)
143 		return (bwrite(bp));
144 	else {
145 		bdwrite(bp);
146 		return (0);
147 	}
148 }
149 
150 #define	SINGLE	0	/* index of single indirect block */
151 #define	DOUBLE	1	/* index of double indirect block */
152 #define	TRIPLE	2	/* index of triple indirect block */
153 /*
154  * Truncate the inode oip to at most length size, freeing the
155  * disk blocks.
156  */
157 int
158 ffs_truncate(ap)
159 	struct vop_truncate_args /* {
160 		struct vnode *a_vp;
161 		off_t a_length;
162 		int a_flags;
163 		struct ucred *a_cred;
164 		struct proc *a_p;
165 	} */ *ap;
166 {
167 	register struct vnode *ovp = ap->a_vp;
168 	register daddr_t lastblock;
169 	register struct inode *oip;
170 	daddr_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR];
171 	daddr_t oldblks[NDADDR + NIADDR], newblks[NDADDR + NIADDR];
172 	off_t length = ap->a_length;
173 	register struct fs *fs;
174 	struct buf *bp;
175 	int offset, size, level;
176 	long count, nblocks, vflags, blocksreleased = 0;
177 	struct timeval tv;
178 	register int i;
179 	int aflags, error, allerror;
180 	off_t osize;
181 
182 	oip = VTOI(ovp);
183 	fs = oip->i_fs;
184 	if (length < 0 || length > fs->fs_maxfilesize)
185 		return (EINVAL);
186 	tv = time;
187 	if (ovp->v_type == VLNK &&
188 	    (oip->i_size < ovp->v_mount->mnt_maxsymlinklen || oip->i_din.di_blocks == 0)) {
189 #ifdef DIAGNOSTIC
190 		if (length != 0)
191 			panic("ffs_truncate: partial truncate of symlink");
192 #endif
193 		bzero((char *)&oip->i_shortlink, (u_int)oip->i_size);
194 		oip->i_size = 0;
195 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
196 		return (VOP_UPDATE(ovp, &tv, &tv, 1));
197 	}
198 	if (oip->i_size == length) {
199 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
200 		return (VOP_UPDATE(ovp, &tv, &tv, 0));
201 	}
202 #ifdef QUOTA
203 	error = getinoquota(oip);
204 	if (error)
205 		return (error);
206 #endif
207 	osize = oip->i_size;
208 	/*
209 	 * Lengthen the size of the file. We must ensure that the
210 	 * last byte of the file is allocated. Since the smallest
211 	 * value of osize is 0, length will be at least 1.
212 	 */
213 	if (osize < length) {
214 		offset = blkoff(fs, length - 1);
215 		lbn = lblkno(fs, length - 1);
216 		aflags = B_CLRBUF;
217 		if (ap->a_flags & IO_SYNC)
218 			aflags |= B_SYNC;
219 		error = ffs_balloc(oip, lbn, offset + 1, ap->a_cred,
220 		    &bp, aflags);
221 		if (error)
222 			return (error);
223 		oip->i_size = length;
224 		if (aflags & IO_SYNC)
225 			bwrite(bp);
226 		else
227 			bawrite(bp);
228 		vnode_pager_setsize(ovp, (u_long)length);
229 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
230 		return (VOP_UPDATE(ovp, &tv, &tv, 1));
231 	}
232 	/*
233 	 * Shorten the size of the file. If the file is not being
234 	 * truncated to a block boundry, the contents of the
235 	 * partial block following the end of the file must be
236 	 * zero'ed in case it ever become accessable again because
237 	 * of subsequent file growth.
238 	 */
239 	offset = blkoff(fs, length);
240 	if (offset == 0) {
241 		oip->i_size = length;
242 	} else {
243 		lbn = lblkno(fs, length);
244 		aflags = B_CLRBUF;
245 		if (ap->a_flags & IO_SYNC)
246 			aflags |= B_SYNC;
247 		error = ffs_balloc(oip, lbn, offset, ap->a_cred, &bp, aflags);
248 		if (error)
249 			return (error);
250 		oip->i_size = length;
251 		size = blksize(fs, oip, lbn);
252 		bzero((char *)bp->b_data + offset, (u_int)(size - offset));
253 		allocbuf(bp, size, 0);
254 		if (aflags & IO_SYNC)
255 			bwrite(bp);
256 		else
257 			bawrite(bp);
258 	}
259 	/*
260 	 * Calculate index into inode's block list of
261 	 * last direct and indirect blocks (if any)
262 	 * which we want to keep.  Lastblock is -1 when
263 	 * the file is truncated to 0.
264 	 */
265 	lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1;
266 	lastiblock[SINGLE] = lastblock - NDADDR;
267 	lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
268 	lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
269 	nblocks = btodb(fs->fs_bsize);
270 	/*
271 	 * Update file and block pointers on disk before we start freeing
272 	 * blocks.  If we crash before free'ing blocks below, the blocks
273 	 * will be returned to the free list.  lastiblock values are also
274 	 * normalized to -1 for calls to ffs_indirtrunc below.
275 	 */
276 	bcopy((caddr_t)&oip->i_db[0], (caddr_t)oldblks, sizeof oldblks);
277 	for (level = TRIPLE; level >= SINGLE; level--)
278 		if (lastiblock[level] < 0) {
279 			oip->i_ib[level] = 0;
280 			lastiblock[level] = -1;
281 		}
282 	for (i = NDADDR - 1; i > lastblock; i--)
283 		oip->i_db[i] = 0;
284 	oip->i_flag |= IN_CHANGE | IN_UPDATE;
285 	error = VOP_UPDATE(ovp, &tv, &tv, 1);
286 	if (error)
287 		allerror = error;
288 	/*
289 	 * Having written the new inode to disk, save its new configuration
290 	 * and put back the old block pointers long enough to process them.
291 	 * Note that we save the new block configuration so we can check it
292 	 * when we are done.
293 	 */
294 	bcopy((caddr_t)&oip->i_db[0], (caddr_t)newblks, sizeof newblks);
295 	bcopy((caddr_t)oldblks, (caddr_t)&oip->i_db[0], sizeof oldblks);
296 	oip->i_size = osize;
297 	vflags = ((length > 0) ? V_SAVE : 0) | V_SAVEMETA;
298 	allerror = vinvalbuf(ovp, vflags, ap->a_cred, ap->a_p, 0, 0);
299 
300 	/*
301 	 * Indirect blocks first.
302 	 */
303 	indir_lbn[SINGLE] = -NDADDR;
304 	indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
305 	indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
306 	for (level = TRIPLE; level >= SINGLE; level--) {
307 		bn = oip->i_ib[level];
308 		if (bn != 0) {
309 			error = ffs_indirtrunc(oip, indir_lbn[level],
310 			    fsbtodb(fs, bn), lastiblock[level], level, &count);
311 			if (error)
312 				allerror = error;
313 			blocksreleased += count;
314 			if (lastiblock[level] < 0) {
315 				oip->i_ib[level] = 0;
316 				ffs_blkfree(oip, bn, fs->fs_bsize);
317 				blocksreleased += nblocks;
318 			}
319 		}
320 		if (lastiblock[level] >= 0)
321 			goto done;
322 	}
323 
324 	/*
325 	 * All whole direct blocks or frags.
326 	 */
327 	for (i = NDADDR - 1; i > lastblock; i--) {
328 		register long bsize;
329 
330 		bn = oip->i_db[i];
331 		if (bn == 0)
332 			continue;
333 		oip->i_db[i] = 0;
334 		bsize = blksize(fs, oip, i);
335 		ffs_blkfree(oip, bn, bsize);
336 		blocksreleased += btodb(bsize);
337 	}
338 	if (lastblock < 0)
339 		goto done;
340 
341 	/*
342 	 * Finally, look for a change in size of the
343 	 * last direct block; release any frags.
344 	 */
345 	bn = oip->i_db[lastblock];
346 	if (bn != 0) {
347 		long oldspace, newspace;
348 
349 		/*
350 		 * Calculate amount of space we're giving
351 		 * back as old block size minus new block size.
352 		 */
353 		oldspace = blksize(fs, oip, lastblock);
354 		oip->i_size = length;
355 		newspace = blksize(fs, oip, lastblock);
356 		if (newspace == 0)
357 			panic("ffs_truncate: newspace");
358 		if (oldspace - newspace > 0) {
359 			/*
360 			 * Block number of space to be free'd is
361 			 * the old block # plus the number of frags
362 			 * required for the storage we're keeping.
363 			 */
364 			bn += numfrags(fs, newspace);
365 			ffs_blkfree(oip, bn, oldspace - newspace);
366 			blocksreleased += btodb(oldspace - newspace);
367 		}
368 	}
369 done:
370 #ifdef DIAGNOSTIC
371 	for (level = SINGLE; level <= TRIPLE; level++)
372 		if (newblks[NDADDR + level] != oip->i_ib[level])
373 			panic("ffs_truncate1");
374 	for (i = 0; i < NDADDR; i++)
375 		if (newblks[i] != oip->i_db[i])
376 			panic("ffs_truncate2");
377 	if (length == 0 &&
378 	    (ovp->v_dirtyblkhd.lh_first || ovp->v_cleanblkhd.lh_first))
379 		panic("ffs_truncate3");
380 #endif /* DIAGNOSTIC */
381 	/*
382 	 * Put back the real size.
383 	 */
384 	oip->i_size = length;
385 	oip->i_blocks -= blocksreleased;
386 	if (oip->i_blocks < 0)			/* sanity */
387 		oip->i_blocks = 0;
388 	oip->i_flag |= IN_CHANGE;
389 	vnode_pager_setsize(ovp, (u_long)length);
390 #ifdef QUOTA
391 	(void) chkdq(oip, -blocksreleased, NOCRED, 0);
392 #endif
393 	return (allerror);
394 }
395 
396 /*
397  * Release blocks associated with the inode ip and stored in the indirect
398  * block bn.  Blocks are free'd in LIFO order up to (but not including)
399  * lastbn.  If level is greater than SINGLE, the block is an indirect block
400  * and recursive calls to indirtrunc must be used to cleanse other indirect
401  * blocks.
402  *
403  * NB: triple indirect blocks are untested.
404  */
405 static int
406 ffs_indirtrunc(ip, lbn, dbn, lastbn, level, countp)
407 	register struct inode *ip;
408 	daddr_t lbn, lastbn;
409 	daddr_t dbn;
410 	int level;
411 	long *countp;
412 {
413 	register int i;
414 	struct buf *bp;
415 	register struct fs *fs = ip->i_fs;
416 	register daddr_t *bap;
417 	struct vnode *vp;
418 	daddr_t *copy, nb, nlbn, last;
419 	long blkcount, factor;
420 	int nblocks, blocksreleased = 0;
421 	int error = 0, allerror = 0;
422 
423 	/*
424 	 * Calculate index in current block of last
425 	 * block to be kept.  -1 indicates the entire
426 	 * block so we need not calculate the index.
427 	 */
428 	factor = 1;
429 	for (i = SINGLE; i < level; i++)
430 		factor *= NINDIR(fs);
431 	last = lastbn;
432 	if (lastbn > 0)
433 		last /= factor;
434 	nblocks = btodb(fs->fs_bsize);
435 	/*
436 	 * Get buffer of block pointers, zero those entries corresponding
437 	 * to blocks to be free'd, and update on disk copy first.  Since
438 	 * double(triple) indirect before single(double) indirect, calls
439 	 * to bmap on these blocks will fail.  However, we already have
440 	 * the on disk address, so we have to set the b_blkno field
441 	 * explicitly instead of letting bread do everything for us.
442 	 */
443 	vp = ITOV(ip);
444 	bp = getblk(vp, lbn, (int)fs->fs_bsize, 0, 0);
445 	/* if (bp->b_flags & (B_DONE | B_DELWRI)) { */
446 	if (bp->b_flags & B_CACHE) {
447 		/* Braces must be here in case trace evaluates to nothing. */
448 		trace(TR_BREADHIT, pack(vp, fs->fs_bsize), lbn);
449 	} else {
450 		trace(TR_BREADMISS, pack(vp, fs->fs_bsize), lbn);
451 		curproc->p_stats->p_ru.ru_inblock++;	/* pay for read */
452 		bp->b_flags |= B_READ;
453 		if (bp->b_bcount > bp->b_bufsize)
454 			panic("ffs_indirtrunc: bad buffer size");
455 		bp->b_blkno = dbn;
456 		vfs_busy_pages(bp, 0);
457 		VOP_STRATEGY(bp);
458 		error = biowait(bp);
459 	}
460 	if (error) {
461 		brelse(bp);
462 		*countp = 0;
463 		return (error);
464 	}
465 
466 	bap = (daddr_t *)bp->b_data;
467 	MALLOC(copy, daddr_t *, fs->fs_bsize, M_TEMP, M_WAITOK);
468 	bcopy((caddr_t)bap, (caddr_t)copy, (u_int)fs->fs_bsize);
469 	bzero((caddr_t)&bap[last + 1],
470 	  (u_int)(NINDIR(fs) - (last + 1)) * sizeof (daddr_t));
471 	if (last == -1)
472 		bp->b_flags |= B_INVAL;
473 	error = bwrite(bp);
474 	if (error)
475 		allerror = error;
476 	bap = copy;
477 
478 	/*
479 	 * Recursively free totally unused blocks.
480 	 */
481 	for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
482 	    i--, nlbn += factor) {
483 		nb = bap[i];
484 		if (nb == 0)
485 			continue;
486 		if (level > SINGLE) {
487 			error = ffs_indirtrunc(ip, nlbn,
488 			    fsbtodb(fs, nb), (daddr_t)-1, level - 1, &blkcount);
489 			if (error)
490 				allerror = error;
491 			blocksreleased += blkcount;
492 		}
493 		ffs_blkfree(ip, nb, fs->fs_bsize);
494 		blocksreleased += nblocks;
495 	}
496 
497 	/*
498 	 * Recursively free last partial block.
499 	 */
500 	if (level > SINGLE && lastbn >= 0) {
501 		last = lastbn % factor;
502 		nb = bap[i];
503 		if (nb != 0) {
504 			error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
505 			    last, level - 1, &blkcount);
506 			if (error)
507 				allerror = error;
508 			blocksreleased += blkcount;
509 		}
510 	}
511 	FREE(copy, M_TEMP);
512 	*countp = blocksreleased;
513 	return (allerror);
514 }
515