xref: /freebsd/sys/ufs/ffs/ffs_inode.c (revision daf1cffce2e07931f27c6c6998652e90df6ba87e)
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  * $FreeBSD$
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/buf.h>
44 #include <sys/vnode.h>
45 #include <sys/kernel.h>
46 #include <sys/malloc.h>
47 #include <sys/resourcevar.h>
48 
49 #include <vm/vm.h>
50 #include <vm/vm_extern.h>
51 
52 #include <ufs/ufs/quota.h>
53 #include <ufs/ufs/ufsmount.h>
54 #include <ufs/ufs/inode.h>
55 #include <ufs/ufs/ufs_extern.h>
56 
57 #include <ufs/ffs/fs.h>
58 #include <ufs/ffs/ffs_extern.h>
59 
60 static int ffs_indirtrunc __P((struct inode *, ufs_daddr_t, ufs_daddr_t,
61 	    ufs_daddr_t, int, long *));
62 
63 /*
64  * Update the access, modified, and inode change times as specified by the
65  * IN_ACCESS, IN_UPDATE, and IN_CHANGE flags respectively.  Write the inode
66  * to disk if the IN_MODIFIED flag is set (it may be set initially, or by
67  * the timestamp update).  The IN_LAZYMOD flag is set to force a write
68  * later if not now.  If we write now, then clear both IN_MODIFIED and
69  * IN_LAZYMOD to reflect the presumably successful write, and if waitfor is
70  * set, then wait for the write to complete.
71  */
72 int
73 ffs_update(vp, waitfor)
74 	struct vnode *vp;
75 	int waitfor;
76 {
77 	register struct fs *fs;
78 	struct buf *bp;
79 	struct inode *ip;
80 	int error;
81 
82 	ufs_itimes(vp);
83 	ip = VTOI(vp);
84 	if ((ip->i_flag & IN_MODIFIED) == 0 && waitfor == 0)
85 		return (0);
86 	ip->i_flag &= ~(IN_LAZYMOD | IN_MODIFIED);
87 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
88 		return (0);
89 	fs = ip->i_fs;
90 	/*
91 	 * Ensure that uid and gid are correct. This is a temporary
92 	 * fix until fsck has been changed to do the update.
93 	 */
94 	if (fs->fs_inodefmt < FS_44INODEFMT) {		/* XXX */
95 		ip->i_din.di_ouid = ip->i_uid;		/* XXX */
96 		ip->i_din.di_ogid = ip->i_gid;		/* XXX */
97 	}						/* XXX */
98 	error = bread(ip->i_devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
99 		(int)fs->fs_bsize, NOCRED, &bp);
100 	if (error) {
101 		brelse(bp);
102 		return (error);
103 	}
104 	if (DOINGSOFTDEP(vp))
105 		softdep_update_inodeblock(ip, bp, waitfor);
106 	else if (ip->i_effnlink != ip->i_nlink)
107 		panic("ffs_update: bad link cnt");
108 	*((struct dinode *)bp->b_data +
109 	    ino_to_fsbo(fs, ip->i_number)) = ip->i_din;
110 	if (waitfor && !DOINGASYNC(vp)) {
111 		return (bwrite(bp));
112 	} else {
113 		if (bp->b_bufsize == fs->fs_bsize)
114 			bp->b_flags |= B_CLUSTEROK;
115 		bdwrite(bp);
116 		return (0);
117 	}
118 }
119 
120 #define	SINGLE	0	/* index of single indirect block */
121 #define	DOUBLE	1	/* index of double indirect block */
122 #define	TRIPLE	2	/* index of triple indirect block */
123 /*
124  * Truncate the inode oip to at most length size, freeing the
125  * disk blocks.
126  */
127 int
128 ffs_truncate(vp, length, flags, cred, p)
129 	struct vnode *vp;
130 	off_t length;
131 	int flags;
132 	struct ucred *cred;
133 	struct proc *p;
134 {
135 	register struct vnode *ovp = vp;
136 	ufs_daddr_t lastblock;
137 	register struct inode *oip;
138 	ufs_daddr_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR];
139 	ufs_daddr_t oldblks[NDADDR + NIADDR], newblks[NDADDR + NIADDR];
140 	register struct fs *fs;
141 	struct buf *bp;
142 	int offset, size, level;
143 	long count, nblocks, blocksreleased = 0;
144 	register int i;
145 	int aflags, error, allerror;
146 	off_t osize;
147 
148 	oip = VTOI(ovp);
149 	if (oip->i_size == length)
150 		return (0);
151 	fs = oip->i_fs;
152 	if (length < 0)
153 		return (EINVAL);
154 	if (length > fs->fs_maxfilesize)
155 		return (EFBIG);
156 	if (ovp->v_type == VLNK &&
157 	    (oip->i_size < ovp->v_mount->mnt_maxsymlinklen || oip->i_din.di_blocks == 0)) {
158 #ifdef DIAGNOSTIC
159 		if (length != 0)
160 			panic("ffs_truncate: partial truncate of symlink");
161 #endif
162 		bzero((char *)&oip->i_shortlink, (u_int)oip->i_size);
163 		oip->i_size = 0;
164 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
165 		return (UFS_UPDATE(ovp, 1));
166 	}
167 	if (oip->i_size == length) {
168 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
169 		return (UFS_UPDATE(ovp, 0));
170 	}
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) {
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 			 */
188 			if ((error = VOP_FSYNC(ovp, cred, MNT_WAIT,
189 			    p)) != 0)
190 				return (error);
191 		} else {
192 #ifdef QUOTA
193 			(void) chkdq(oip, -oip->i_blocks, NOCRED, 0);
194 #endif
195 			softdep_setup_freeblocks(oip, length);
196 			vinvalbuf(ovp, 0, cred, p, 0, 0);
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 		oip->i_size = length;
247 		size = blksize(fs, oip, lbn);
248 		if (ovp->v_type != VDIR)
249 			bzero((char *)bp->b_data + offset,
250 			    (u_int)(size - offset));
251 		/* Kirk's code has reallocbuf(bp, size, 1) here */
252 		allocbuf(bp, size);
253 		if (bp->b_bufsize == fs->fs_bsize)
254 			bp->b_flags |= B_CLUSTEROK;
255 		if (aflags & B_SYNC)
256 			bwrite(bp);
257 		else
258 			bawrite(bp);
259 	}
260 	/*
261 	 * Calculate index into inode's block list of
262 	 * last direct and indirect blocks (if any)
263 	 * which we want to keep.  Lastblock is -1 when
264 	 * the file is truncated to 0.
265 	 */
266 	lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1;
267 	lastiblock[SINGLE] = lastblock - NDADDR;
268 	lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
269 	lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
270 	nblocks = btodb(fs->fs_bsize);
271 	/*
272 	 * Update file and block pointers on disk before we start freeing
273 	 * blocks.  If we crash before free'ing blocks below, the blocks
274 	 * will be returned to the free list.  lastiblock values are also
275 	 * normalized to -1 for calls to ffs_indirtrunc below.
276 	 */
277 	bcopy((caddr_t)&oip->i_db[0], (caddr_t)oldblks, sizeof oldblks);
278 	for (level = TRIPLE; level >= SINGLE; level--)
279 		if (lastiblock[level] < 0) {
280 			oip->i_ib[level] = 0;
281 			lastiblock[level] = -1;
282 		}
283 	for (i = NDADDR - 1; i > lastblock; i--)
284 		oip->i_db[i] = 0;
285 	oip->i_flag |= IN_CHANGE | IN_UPDATE;
286 	allerror = UFS_UPDATE(ovp, ((length > 0) ? 0 : 1));
287 
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 
298 	error = vtruncbuf(ovp, cred, p, length, fs->fs_bsize);
299 	if (error && (allerror == 0))
300 		allerror = error;
301 
302 	/*
303 	 * Indirect blocks first.
304 	 */
305 	indir_lbn[SINGLE] = -NDADDR;
306 	indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
307 	indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
308 	for (level = TRIPLE; level >= SINGLE; level--) {
309 		bn = oip->i_ib[level];
310 		if (bn != 0) {
311 			error = ffs_indirtrunc(oip, indir_lbn[level],
312 			    fsbtodb(fs, bn), lastiblock[level], level, &count);
313 			if (error)
314 				allerror = error;
315 			blocksreleased += count;
316 			if (lastiblock[level] < 0) {
317 				oip->i_ib[level] = 0;
318 				ffs_blkfree(oip, bn, fs->fs_bsize);
319 				blocksreleased += nblocks;
320 			}
321 		}
322 		if (lastiblock[level] >= 0)
323 			goto done;
324 	}
325 
326 	/*
327 	 * All whole direct blocks or frags.
328 	 */
329 	for (i = NDADDR - 1; i > lastblock; i--) {
330 		register long bsize;
331 
332 		bn = oip->i_db[i];
333 		if (bn == 0)
334 			continue;
335 		oip->i_db[i] = 0;
336 		bsize = blksize(fs, oip, i);
337 		ffs_blkfree(oip, bn, bsize);
338 		blocksreleased += btodb(bsize);
339 	}
340 	if (lastblock < 0)
341 		goto done;
342 
343 	/*
344 	 * Finally, look for a change in size of the
345 	 * last direct block; release any frags.
346 	 */
347 	bn = oip->i_db[lastblock];
348 	if (bn != 0) {
349 		long oldspace, newspace;
350 
351 		/*
352 		 * Calculate amount of space we're giving
353 		 * back as old block size minus new block size.
354 		 */
355 		oldspace = blksize(fs, oip, lastblock);
356 		oip->i_size = length;
357 		newspace = blksize(fs, oip, lastblock);
358 		if (newspace == 0)
359 			panic("ffs_truncate: newspace");
360 		if (oldspace - newspace > 0) {
361 			/*
362 			 * Block number of space to be free'd is
363 			 * the old block # plus the number of frags
364 			 * required for the storage we're keeping.
365 			 */
366 			bn += numfrags(fs, newspace);
367 			ffs_blkfree(oip, bn, oldspace - newspace);
368 			blocksreleased += btodb(oldspace - newspace);
369 		}
370 	}
371 done:
372 #ifdef DIAGNOSTIC
373 	for (level = SINGLE; level <= TRIPLE; level++)
374 		if (newblks[NDADDR + level] != oip->i_ib[level])
375 			panic("ffs_truncate1");
376 	for (i = 0; i < NDADDR; i++)
377 		if (newblks[i] != oip->i_db[i])
378 			panic("ffs_truncate2");
379 	if (length == 0 &&
380 	    (!TAILQ_EMPTY(&ovp->v_dirtyblkhd) ||
381 	     !TAILQ_EMPTY(&ovp->v_cleanblkhd)))
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 
390 	if (oip->i_blocks < 0)			/* sanity */
391 		oip->i_blocks = 0;
392 	oip->i_flag |= IN_CHANGE;
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 	ufs_daddr_t lbn, lastbn;
412 	ufs_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 ufs_daddr_t *bap;
420 	struct vnode *vp;
421 	ufs_daddr_t *copy = NULL, 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 		bp->b_flags &= ~(B_ERROR|B_INVAL);
452 		if (bp->b_bcount > bp->b_bufsize)
453 			panic("ffs_indirtrunc: bad buffer size");
454 		bp->b_blkno = dbn;
455 		vfs_busy_pages(bp, 0);
456 		VOP_STRATEGY(bp->b_vp, bp);
457 		error = biowait(bp);
458 	}
459 	if (error) {
460 		brelse(bp);
461 		*countp = 0;
462 		return (error);
463 	}
464 
465 	bap = (ufs_daddr_t *)bp->b_data;
466 	if (lastbn != -1) {
467 		MALLOC(copy, ufs_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 (ufs_daddr_t));
471 		if (DOINGASYNC(vp)) {
472 			bawrite(bp);
473 		} else {
474 			error = bwrite(bp);
475 			if (error)
476 				allerror = error;
477 		}
478 		bap = copy;
479 	}
480 
481 	/*
482 	 * Recursively free totally unused blocks.
483 	 */
484 	for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
485 	    i--, nlbn += factor) {
486 		nb = bap[i];
487 		if (nb == 0)
488 			continue;
489 		if (level > SINGLE) {
490 			if ((error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
491 			    (ufs_daddr_t)-1, level - 1, &blkcount)) != 0)
492 				allerror = error;
493 			blocksreleased += blkcount;
494 		}
495 		ffs_blkfree(ip, nb, fs->fs_bsize);
496 		blocksreleased += nblocks;
497 	}
498 
499 	/*
500 	 * Recursively free last partial block.
501 	 */
502 	if (level > SINGLE && lastbn >= 0) {
503 		last = lastbn % factor;
504 		nb = bap[i];
505 		if (nb != 0) {
506 			error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
507 			    last, level - 1, &blkcount);
508 			if (error)
509 				allerror = error;
510 			blocksreleased += blkcount;
511 		}
512 	}
513 	if (copy != NULL) {
514 		FREE(copy, M_TEMP);
515 	} else {
516 		bp->b_flags |= B_INVAL | B_NOCACHE;
517 		brelse(bp);
518 	}
519 
520 	*countp = blocksreleased;
521 	return (allerror);
522 }
523