xref: /freebsd/sys/ufs/ffs/ffs_inode.c (revision 6e8394b8baa7d5d9153ab90de6824bcd19b3b4e1)
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 1.53 1999/01/28 00:57:54 dillon 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/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 && (vp->v_mount->mnt_flag & MNT_ASYNC) == 0) {
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 if (ovp->v_mount->mnt_flag & MNT_ASYNC)
222 			bdwrite(bp);
223 		else
224 			bawrite(bp);
225 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
226 		return (UFS_UPDATE(ovp, 1));
227 	}
228 	/*
229 	 * Shorten the size of the file. If the file is not being
230 	 * truncated to a block boundary, the contents of the
231 	 * partial block following the end of the file must be
232 	 * zero'ed in case it ever becomes accessible again because
233 	 * of subsequent file growth. Directories however are not
234 	 * zero'ed as they should grow back initialized to empty.
235 	 */
236 	offset = blkoff(fs, length);
237 	if (offset == 0) {
238 		oip->i_size = length;
239 	} else {
240 		lbn = lblkno(fs, length);
241 		aflags = B_CLRBUF;
242 		if (flags & IO_SYNC)
243 			aflags |= B_SYNC;
244 		error = VOP_BALLOC(ovp, length - 1, 1, cred, aflags, &bp);
245 		if (error) {
246 			return (error);
247 		}
248 		oip->i_size = length;
249 		size = blksize(fs, oip, lbn);
250 		if (ovp->v_type != VDIR)
251 			bzero((char *)bp->b_data + offset,
252 			    (u_int)(size - offset));
253 		/* Kirk's code has reallocbuf(bp, size, 1) here */
254 		allocbuf(bp, size);
255 		if (bp->b_bufsize == fs->fs_bsize)
256 			bp->b_flags |= B_CLUSTEROK;
257 		if (aflags & B_SYNC)
258 			bwrite(bp);
259 		else if (ovp->v_mount->mnt_flag & MNT_ASYNC)
260 			bdwrite(bp);
261 		else
262 			bawrite(bp);
263 	}
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 	allerror = UFS_UPDATE(ovp, ((length > 0) ? 0 : 1));
291 
292 	/*
293 	 * Having written the new inode to disk, save its new configuration
294 	 * and put back the old block pointers long enough to process them.
295 	 * Note that we save the new block configuration so we can check it
296 	 * when we are done.
297 	 */
298 	bcopy((caddr_t)&oip->i_db[0], (caddr_t)newblks, sizeof newblks);
299 	bcopy((caddr_t)oldblks, (caddr_t)&oip->i_db[0], sizeof oldblks);
300 	oip->i_size = osize;
301 
302 	error = vtruncbuf(ovp, cred, p, length, fs->fs_bsize);
303 	if (error && (allerror == 0))
304 		allerror = error;
305 
306 	/*
307 	 * Indirect blocks first.
308 	 */
309 	indir_lbn[SINGLE] = -NDADDR;
310 	indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
311 	indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
312 	for (level = TRIPLE; level >= SINGLE; level--) {
313 		bn = oip->i_ib[level];
314 		if (bn != 0) {
315 			error = ffs_indirtrunc(oip, indir_lbn[level],
316 			    fsbtodb(fs, bn), lastiblock[level], level, &count);
317 			if (error)
318 				allerror = error;
319 			blocksreleased += count;
320 			if (lastiblock[level] < 0) {
321 				oip->i_ib[level] = 0;
322 				ffs_blkfree(oip, bn, fs->fs_bsize);
323 				blocksreleased += nblocks;
324 			}
325 		}
326 		if (lastiblock[level] >= 0)
327 			goto done;
328 	}
329 
330 	/*
331 	 * All whole direct blocks or frags.
332 	 */
333 	for (i = NDADDR - 1; i > lastblock; i--) {
334 		register long bsize;
335 
336 		bn = oip->i_db[i];
337 		if (bn == 0)
338 			continue;
339 		oip->i_db[i] = 0;
340 		bsize = blksize(fs, oip, i);
341 		ffs_blkfree(oip, bn, bsize);
342 		blocksreleased += btodb(bsize);
343 	}
344 	if (lastblock < 0)
345 		goto done;
346 
347 	/*
348 	 * Finally, look for a change in size of the
349 	 * last direct block; release any frags.
350 	 */
351 	bn = oip->i_db[lastblock];
352 	if (bn != 0) {
353 		long oldspace, newspace;
354 
355 		/*
356 		 * Calculate amount of space we're giving
357 		 * back as old block size minus new block size.
358 		 */
359 		oldspace = blksize(fs, oip, lastblock);
360 		oip->i_size = length;
361 		newspace = blksize(fs, oip, lastblock);
362 		if (newspace == 0)
363 			panic("ffs_truncate: newspace");
364 		if (oldspace - newspace > 0) {
365 			/*
366 			 * Block number of space to be free'd is
367 			 * the old block # plus the number of frags
368 			 * required for the storage we're keeping.
369 			 */
370 			bn += numfrags(fs, newspace);
371 			ffs_blkfree(oip, bn, oldspace - newspace);
372 			blocksreleased += btodb(oldspace - newspace);
373 		}
374 	}
375 done:
376 #ifdef DIAGNOSTIC
377 	for (level = SINGLE; level <= TRIPLE; level++)
378 		if (newblks[NDADDR + level] != oip->i_ib[level])
379 			panic("ffs_truncate1");
380 	for (i = 0; i < NDADDR; i++)
381 		if (newblks[i] != oip->i_db[i])
382 			panic("ffs_truncate2");
383 	if (length == 0 &&
384 	    (!TAILQ_EMPTY(&ovp->v_dirtyblkhd) ||
385 	     !TAILQ_EMPTY(&ovp->v_cleanblkhd)))
386 		panic("ffs_truncate3");
387 #endif /* DIAGNOSTIC */
388 	/*
389 	 * Put back the real size.
390 	 */
391 	oip->i_size = length;
392 	oip->i_blocks -= blocksreleased;
393 
394 	if (oip->i_blocks < 0)			/* sanity */
395 		oip->i_blocks = 0;
396 	oip->i_flag |= IN_CHANGE;
397 #ifdef QUOTA
398 	(void) chkdq(oip, -blocksreleased, NOCRED, 0);
399 #endif
400 	return (allerror);
401 }
402 
403 /*
404  * Release blocks associated with the inode ip and stored in the indirect
405  * block bn.  Blocks are free'd in LIFO order up to (but not including)
406  * lastbn.  If level is greater than SINGLE, the block is an indirect block
407  * and recursive calls to indirtrunc must be used to cleanse other indirect
408  * blocks.
409  *
410  * NB: triple indirect blocks are untested.
411  */
412 static int
413 ffs_indirtrunc(ip, lbn, dbn, lastbn, level, countp)
414 	register struct inode *ip;
415 	ufs_daddr_t lbn, lastbn;
416 	ufs_daddr_t dbn;
417 	int level;
418 	long *countp;
419 {
420 	register int i;
421 	struct buf *bp;
422 	register struct fs *fs = ip->i_fs;
423 	register ufs_daddr_t *bap;
424 	struct vnode *vp;
425 	ufs_daddr_t *copy = NULL, nb, nlbn, last;
426 	long blkcount, factor;
427 	int nblocks, blocksreleased = 0;
428 	int error = 0, allerror = 0;
429 
430 	/*
431 	 * Calculate index in current block of last
432 	 * block to be kept.  -1 indicates the entire
433 	 * block so we need not calculate the index.
434 	 */
435 	factor = 1;
436 	for (i = SINGLE; i < level; i++)
437 		factor *= NINDIR(fs);
438 	last = lastbn;
439 	if (lastbn > 0)
440 		last /= factor;
441 	nblocks = btodb(fs->fs_bsize);
442 	/*
443 	 * Get buffer of block pointers, zero those entries corresponding
444 	 * to blocks to be free'd, and update on disk copy first.  Since
445 	 * double(triple) indirect before single(double) indirect, calls
446 	 * to bmap on these blocks will fail.  However, we already have
447 	 * the on disk address, so we have to set the b_blkno field
448 	 * explicitly instead of letting bread do everything for us.
449 	 */
450 	vp = ITOV(ip);
451 	bp = getblk(vp, lbn, (int)fs->fs_bsize, 0, 0);
452 	if ((bp->b_flags & B_CACHE) == 0) {
453 		curproc->p_stats->p_ru.ru_inblock++;	/* pay for read */
454 		bp->b_flags |= B_READ;
455 		bp->b_flags &= ~(B_ERROR|B_INVAL);
456 		if (bp->b_bcount > bp->b_bufsize)
457 			panic("ffs_indirtrunc: bad buffer size");
458 		bp->b_blkno = dbn;
459 		vfs_busy_pages(bp, 0);
460 		VOP_STRATEGY(bp->b_vp, bp);
461 		error = biowait(bp);
462 	}
463 	if (error) {
464 		brelse(bp);
465 		*countp = 0;
466 		return (error);
467 	}
468 
469 	bap = (ufs_daddr_t *)bp->b_data;
470 	if (lastbn != -1) {
471 		MALLOC(copy, ufs_daddr_t *, fs->fs_bsize, M_TEMP, M_WAITOK);
472 		bcopy((caddr_t)bap, (caddr_t)copy, (u_int)fs->fs_bsize);
473 		bzero((caddr_t)&bap[last + 1],
474 		    (u_int)(NINDIR(fs) - (last + 1)) * sizeof (ufs_daddr_t));
475 		if ((vp->v_mount->mnt_flag & MNT_ASYNC) == 0) {
476 			error = bwrite(bp);
477 			if (error)
478 				allerror = error;
479 		} else {
480 			bawrite(bp);
481 		}
482 		bap = copy;
483 	}
484 
485 	/*
486 	 * Recursively free totally unused blocks.
487 	 */
488 	for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
489 	    i--, nlbn += factor) {
490 		nb = bap[i];
491 		if (nb == 0)
492 			continue;
493 		if (level > SINGLE) {
494 			if ((error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
495 			    (ufs_daddr_t)-1, level - 1, &blkcount)) != 0)
496 				allerror = error;
497 			blocksreleased += blkcount;
498 		}
499 		ffs_blkfree(ip, nb, fs->fs_bsize);
500 		blocksreleased += nblocks;
501 	}
502 
503 	/*
504 	 * Recursively free last partial block.
505 	 */
506 	if (level > SINGLE && lastbn >= 0) {
507 		last = lastbn % factor;
508 		nb = bap[i];
509 		if (nb != 0) {
510 			error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
511 			    last, level - 1, &blkcount);
512 			if (error)
513 				allerror = error;
514 			blocksreleased += blkcount;
515 		}
516 	}
517 	if (copy != NULL) {
518 		FREE(copy, M_TEMP);
519 	} else {
520 		bp->b_flags |= B_INVAL | B_NOCACHE;
521 		brelse(bp);
522 	}
523 
524 	*countp = blocksreleased;
525 	return (allerror);
526 }
527