xref: /freebsd/sys/ufs/ffs/ffs_inode.c (revision 952d112864d8008aa87278a30a539d888a8493cd)
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  * $Id: ffs_inode.c,v 1.25 1997/02/22 09:47:03 peter Exp $
35  */
36 
37 #include "opt_quota.h"
38 
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/mount.h>
42 #include <sys/proc.h>
43 #include <sys/file.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 
50 #include <vm/vm.h>
51 #include <vm/vm_param.h>
52 #include <vm/vm_extern.h>
53 
54 #include <ufs/ufs/quota.h>
55 #include <ufs/ufs/inode.h>
56 #include <ufs/ufs/ufsmount.h>
57 #include <ufs/ufs/ufs_extern.h>
58 
59 #include <ufs/ffs/fs.h>
60 #include <ufs/ffs/ffs_extern.h>
61 
62 static 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. The IN_MODIFIED
68  * flag is used to specify that the inode needs to be updated even if none
69  * of the times needs to be updated. The access and modified times are taken
70  * from the second and third parameters; the inode change time is always
71  * taken from the current time. If waitfor is set, then wait for the disk
72  * write of the inode to complete.
73  */
74 int
75 ffs_update(ap)
76 	struct vop_update_args /* {
77 		struct vnode *a_vp;
78 		struct timeval *a_access;
79 		struct timeval *a_modify;
80 		int a_waitfor;
81 	} */ *ap;
82 {
83 	register struct fs *fs;
84 	struct buf *bp;
85 	struct inode *ip;
86 	int error;
87 	time_t tv_sec;
88 
89 	ip = VTOI(ap->a_vp);
90 	if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY) {
91 		ip->i_flag &=
92 		    ~(IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE);
93 		return (0);
94 	}
95 	if ((ip->i_flag &
96 	    (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0)
97 		return (0);
98 	/*
99 	 * Use a copy of the current time to get consistent timestamps
100 	 * (a_access and a_modify are sometimes aliases for &time).
101 	 *
102 	 * XXX in 2.0, a_access and a_modify are often pointers to the
103 	 * same copy of `time'.  This is not as good.  Some callers forget
104 	 * to make a copy; others make a copy too early (before the i/o
105 	 * has completed)...
106 	 *
107 	 * XXX there should be a function or macro for reading the time
108 	 * (e.g., some machines may require splclock()).
109 	 */
110 	tv_sec = time.tv_sec;
111 	if (ip->i_flag & IN_ACCESS)
112 		ip->i_atime =
113 		    (ap->a_access == &time ? tv_sec : ap->a_access->tv_sec);
114 	if (ip->i_flag & IN_UPDATE) {
115 		ip->i_mtime =
116 		    (ap->a_modify == &time ? tv_sec : ap->a_modify->tv_sec);
117 		ip->i_modrev++;
118 	}
119 	if (ip->i_flag & IN_CHANGE)
120 		ip->i_ctime = tv_sec;
121 	ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE);
122 	fs = ip->i_fs;
123 	/*
124 	 * Ensure that uid and gid are correct. This is a temporary
125 	 * fix until fsck has been changed to do the update.
126 	 */
127 	if (fs->fs_inodefmt < FS_44INODEFMT) {		/* XXX */
128 		ip->i_din.di_ouid = ip->i_uid;		/* XXX */
129 		ip->i_din.di_ogid = ip->i_gid;		/* XXX */
130 	}						/* XXX */
131 	error = bread(ip->i_devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
132 		(int)fs->fs_bsize, NOCRED, &bp);
133 	if (error) {
134 		brelse(bp);
135 		return (error);
136 	}
137 	*((struct dinode *)bp->b_data +
138 	    ino_to_fsbo(fs, ip->i_number)) = ip->i_din;
139 	if (ap->a_waitfor && (ap->a_vp->v_mount->mnt_flag & MNT_ASYNC) == 0)
140 		return (bwrite(bp));
141 	else {
142 		bp->b_flags |= B_CLUSTEROK;
143 		bdwrite(bp);
144 		return (0);
145 	}
146 }
147 
148 #define	SINGLE	0	/* index of single indirect block */
149 #define	DOUBLE	1	/* index of double indirect block */
150 #define	TRIPLE	2	/* index of triple indirect block */
151 /*
152  * Truncate the inode oip to at most length size, freeing the
153  * disk blocks.
154  */
155 int
156 ffs_truncate(ap)
157 	struct vop_truncate_args /* {
158 		struct vnode *a_vp;
159 		off_t a_length;
160 		int a_flags;
161 		struct ucred *a_cred;
162 		struct proc *a_p;
163 	} */ *ap;
164 {
165 	register struct vnode *ovp = ap->a_vp;
166 	ufs_daddr_t lastblock;
167 	register struct inode *oip;
168 	ufs_daddr_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR];
169 	ufs_daddr_t oldblks[NDADDR + NIADDR], newblks[NDADDR + NIADDR];
170 	off_t length = ap->a_length;
171 	register struct fs *fs;
172 	struct buf *bp;
173 	int offset, size, level;
174 	long count, nblocks, vflags, blocksreleased = 0;
175 	struct timeval tv;
176 	register int i;
177 	int aflags, error, allerror;
178 	off_t osize;
179 
180 	oip = VTOI(ovp);
181 	fs = oip->i_fs;
182 	if (length < 0)
183 		return (EINVAL);
184 	if (length > fs->fs_maxfilesize)
185 		return (EFBIG);
186 	gettime(&tv);
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 		vnode_pager_setsize(ovp, length);
225 		if (aflags & B_SYNC)
226 			bwrite(bp);
227 		else if (ovp->v_mount->mnt_flag & MNT_ASYNC)
228 			bdwrite(bp);
229 		else
230 			bawrite(bp);
231 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
232 		return (VOP_UPDATE(ovp, &tv, &tv, 1));
233 	}
234 	/*
235 	 * Shorten the size of the file. If the file is not being
236 	 * truncated to a block boundry, the contents of the
237 	 * partial block following the end of the file must be
238 	 * zero'ed in case it ever become accessable again because
239 	 * of subsequent file growth.
240 	 */
241 	offset = blkoff(fs, length);
242 	if (offset == 0) {
243 		oip->i_size = length;
244 	} else {
245 		lbn = lblkno(fs, length);
246 		aflags = B_CLRBUF;
247 		if (ap->a_flags & IO_SYNC)
248 			aflags |= B_SYNC;
249 		error = ffs_balloc(oip, lbn, offset, ap->a_cred, &bp, aflags);
250 		if (error)
251 			return (error);
252 		oip->i_size = length;
253 		size = blksize(fs, oip, lbn);
254 		bzero((char *)bp->b_data + offset, (u_int)(size - offset));
255 		allocbuf(bp, size);
256 		if (aflags & B_SYNC)
257 			bwrite(bp);
258 		else if (ovp->v_mount->mnt_flag & MNT_ASYNC)
259 			bdwrite(bp);
260 		else
261 			bawrite(bp);
262 	}
263 	vnode_pager_setsize(ovp, length);
264 	/*
265 	 * Calculate index into inode's block list of
266 	 * last direct and indirect blocks (if any)
267 	 * which we want to keep.  Lastblock is -1 when
268 	 * the file is truncated to 0.
269 	 */
270 	lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1;
271 	lastiblock[SINGLE] = lastblock - NDADDR;
272 	lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
273 	lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
274 	nblocks = btodb(fs->fs_bsize);
275 	/*
276 	 * Update file and block pointers on disk before we start freeing
277 	 * blocks.  If we crash before free'ing blocks below, the blocks
278 	 * will be returned to the free list.  lastiblock values are also
279 	 * normalized to -1 for calls to ffs_indirtrunc below.
280 	 */
281 	bcopy((caddr_t)&oip->i_db[0], (caddr_t)oldblks, sizeof oldblks);
282 	for (level = TRIPLE; level >= SINGLE; level--)
283 		if (lastiblock[level] < 0) {
284 			oip->i_ib[level] = 0;
285 			lastiblock[level] = -1;
286 		}
287 	for (i = NDADDR - 1; i > lastblock; i--)
288 		oip->i_db[i] = 0;
289 	oip->i_flag |= IN_CHANGE | IN_UPDATE;
290 	error = VOP_UPDATE(ovp, &tv, &tv, ((length > 0) ? 0 : 1));
291 	if (error)
292 		allerror = error;
293 	/*
294 	 * Having written the new inode to disk, save its new configuration
295 	 * and put back the old block pointers long enough to process them.
296 	 * Note that we save the new block configuration so we can check it
297 	 * when we are done.
298 	 */
299 	bcopy((caddr_t)&oip->i_db[0], (caddr_t)newblks, sizeof newblks);
300 	bcopy((caddr_t)oldblks, (caddr_t)&oip->i_db[0], sizeof oldblks);
301 	oip->i_size = osize;
302 	vflags = ((length > 0) ? V_SAVE : 0) | V_SAVEMETA;
303 	allerror = vinvalbuf(ovp, vflags, ap->a_cred, ap->a_p, 0, 0);
304 
305 	/*
306 	 * Indirect blocks first.
307 	 */
308 	indir_lbn[SINGLE] = -NDADDR;
309 	indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
310 	indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
311 	for (level = TRIPLE; level >= SINGLE; level--) {
312 		bn = oip->i_ib[level];
313 		if (bn != 0) {
314 			error = ffs_indirtrunc(oip, indir_lbn[level],
315 			    fsbtodb(fs, bn), lastiblock[level], level, &count);
316 			if (error)
317 				allerror = error;
318 			blocksreleased += count;
319 			if (lastiblock[level] < 0) {
320 				oip->i_ib[level] = 0;
321 				ffs_blkfree(oip, bn, fs->fs_bsize);
322 				blocksreleased += nblocks;
323 			}
324 		}
325 		if (lastiblock[level] >= 0)
326 			goto done;
327 	}
328 
329 	/*
330 	 * All whole direct blocks or frags.
331 	 */
332 	for (i = NDADDR - 1; i > lastblock; i--) {
333 		register long bsize;
334 
335 		bn = oip->i_db[i];
336 		if (bn == 0)
337 			continue;
338 		oip->i_db[i] = 0;
339 		bsize = blksize(fs, oip, i);
340 		ffs_blkfree(oip, bn, bsize);
341 		blocksreleased += btodb(bsize);
342 	}
343 	if (lastblock < 0)
344 		goto done;
345 
346 	/*
347 	 * Finally, look for a change in size of the
348 	 * last direct block; release any frags.
349 	 */
350 	bn = oip->i_db[lastblock];
351 	if (bn != 0) {
352 		long oldspace, newspace;
353 
354 		/*
355 		 * Calculate amount of space we're giving
356 		 * back as old block size minus new block size.
357 		 */
358 		oldspace = blksize(fs, oip, lastblock);
359 		oip->i_size = length;
360 		newspace = blksize(fs, oip, lastblock);
361 		if (newspace == 0)
362 			panic("ffs_truncate: newspace");
363 		if (oldspace - newspace > 0) {
364 			/*
365 			 * Block number of space to be free'd is
366 			 * the old block # plus the number of frags
367 			 * required for the storage we're keeping.
368 			 */
369 			bn += numfrags(fs, newspace);
370 			ffs_blkfree(oip, bn, oldspace - newspace);
371 			blocksreleased += btodb(oldspace - newspace);
372 		}
373 	}
374 done:
375 #ifdef DIAGNOSTIC
376 	for (level = SINGLE; level <= TRIPLE; level++)
377 		if (newblks[NDADDR + level] != oip->i_ib[level])
378 			panic("ffs_truncate1");
379 	for (i = 0; i < NDADDR; i++)
380 		if (newblks[i] != oip->i_db[i])
381 			panic("ffs_truncate2");
382 	if (length == 0 &&
383 	    (ovp->v_dirtyblkhd.lh_first || ovp->v_cleanblkhd.lh_first))
384 		panic("ffs_truncate3");
385 #endif /* DIAGNOSTIC */
386 	/*
387 	 * Put back the real size.
388 	 */
389 	oip->i_size = length;
390 	oip->i_blocks -= blocksreleased;
391 	if (oip->i_blocks < 0)			/* sanity */
392 		oip->i_blocks = 0;
393 	oip->i_flag |= IN_CHANGE;
394 	vnode_pager_setsize(ovp, length);
395 #ifdef QUOTA
396 	(void) chkdq(oip, -blocksreleased, NOCRED, 0);
397 #endif
398 	return (allerror);
399 }
400 
401 /*
402  * Release blocks associated with the inode ip and stored in the indirect
403  * block bn.  Blocks are free'd in LIFO order up to (but not including)
404  * lastbn.  If level is greater than SINGLE, the block is an indirect block
405  * and recursive calls to indirtrunc must be used to cleanse other indirect
406  * blocks.
407  *
408  * NB: triple indirect blocks are untested.
409  */
410 static int
411 ffs_indirtrunc(ip, lbn, dbn, lastbn, level, countp)
412 	register struct inode *ip;
413 	ufs_daddr_t lbn, lastbn;
414 	ufs_daddr_t dbn;
415 	int level;
416 	long *countp;
417 {
418 	register int i;
419 	struct buf *bp;
420 	register struct fs *fs = ip->i_fs;
421 	register ufs_daddr_t *bap;
422 	struct vnode *vp;
423 	ufs_daddr_t *copy = NULL, nb, nlbn, last;
424 	long blkcount, factor;
425 	int nblocks, blocksreleased = 0;
426 	int error = 0, allerror = 0;
427 
428 	/*
429 	 * Calculate index in current block of last
430 	 * block to be kept.  -1 indicates the entire
431 	 * block so we need not calculate the index.
432 	 */
433 	factor = 1;
434 	for (i = SINGLE; i < level; i++)
435 		factor *= NINDIR(fs);
436 	last = lastbn;
437 	if (lastbn > 0)
438 		last /= factor;
439 	nblocks = btodb(fs->fs_bsize);
440 	/*
441 	 * Get buffer of block pointers, zero those entries corresponding
442 	 * to blocks to be free'd, and update on disk copy first.  Since
443 	 * double(triple) indirect before single(double) indirect, calls
444 	 * to bmap on these blocks will fail.  However, we already have
445 	 * the on disk address, so we have to set the b_blkno field
446 	 * explicitly instead of letting bread do everything for us.
447 	 */
448 	vp = ITOV(ip);
449 	bp = getblk(vp, lbn, (int)fs->fs_bsize, 0, 0);
450 	if ((bp->b_flags & B_CACHE) == 0) {
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 = (ufs_daddr_t *)bp->b_data;
467 	if (lastbn != -1) {
468 		MALLOC(copy, ufs_daddr_t *, fs->fs_bsize, M_TEMP, M_WAITOK);
469 		bcopy((caddr_t)bap, (caddr_t)copy, (u_int)fs->fs_bsize);
470 		bzero((caddr_t)&bap[last + 1],
471 		    (u_int)(NINDIR(fs) - (last + 1)) * sizeof (ufs_daddr_t));
472 		if ((vp->v_mount->mnt_flag & MNT_ASYNC) == 0) {
473 			error = bwrite(bp);
474 			if (error)
475 				allerror = error;
476 		} else {
477 			bawrite(bp);
478 		}
479 		bap = copy;
480 	}
481 
482 	/*
483 	 * Recursively free totally unused blocks.
484 	 */
485 	for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
486 	    i--, nlbn += factor) {
487 		nb = bap[i];
488 		if (nb == 0)
489 			continue;
490 		if (level > SINGLE) {
491 			if (error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
492 			    (ufs_daddr_t)-1, level - 1, &blkcount))
493 				allerror = error;
494 			blocksreleased += blkcount;
495 		}
496 		ffs_blkfree(ip, nb, fs->fs_bsize);
497 		blocksreleased += nblocks;
498 	}
499 
500 	/*
501 	 * Recursively free last partial block.
502 	 */
503 	if (level > SINGLE && lastbn >= 0) {
504 		last = lastbn % factor;
505 		nb = bap[i];
506 		if (nb != 0) {
507 			error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
508 			    last, level - 1, &blkcount);
509 			if (error)
510 				allerror = error;
511 			blocksreleased += blkcount;
512 		}
513 	}
514 	if (copy != NULL) {
515 		FREE(copy, M_TEMP);
516 	} else {
517 		bp->b_flags |= B_INVAL | B_NOCACHE;
518 		brelse(bp);
519 	}
520 
521 	*countp = blocksreleased;
522 	return (allerror);
523 }
524