xref: /freebsd/sys/ufs/ffs/ffs_inode.c (revision 5521ff5a4d1929056e7ffc982fac3341ca54df7c)
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/bio.h>
44 #include <sys/buf.h>
45 #include <sys/vnode.h>
46 #include <sys/malloc.h>
47 #include <sys/resourcevar.h>
48 #include <sys/vmmeter.h>
49 #include <sys/stat.h>
50 
51 #include <vm/vm.h>
52 #include <vm/vm_extern.h>
53 
54 #include <ufs/ufs/extattr.h>
55 #include <ufs/ufs/quota.h>
56 #include <ufs/ufs/ufsmount.h>
57 #include <ufs/ufs/inode.h>
58 #include <ufs/ufs/ufs_extern.h>
59 
60 #include <ufs/ffs/fs.h>
61 #include <ufs/ffs/ffs_extern.h>
62 
63 static int ffs_indirtrunc __P((struct inode *, ufs_daddr_t, ufs_daddr_t,
64 	    ufs_daddr_t, int, long *));
65 
66 /*
67  * Update the access, modified, and inode change times as specified by the
68  * IN_ACCESS, IN_UPDATE, and IN_CHANGE flags respectively.  Write the inode
69  * to disk if the IN_MODIFIED flag is set (it may be set initially, or by
70  * the timestamp update).  The IN_LAZYMOD flag is set to force a write
71  * later if not now.  If we write now, then clear both IN_MODIFIED and
72  * IN_LAZYMOD to reflect the presumably successful write, and if waitfor is
73  * set, then wait for the write to complete.
74  */
75 int
76 ffs_update(vp, waitfor)
77 	struct vnode *vp;
78 	int waitfor;
79 {
80 	register struct fs *fs;
81 	struct buf *bp;
82 	struct inode *ip;
83 	int error;
84 
85 	ufs_itimes(vp);
86 	ip = VTOI(vp);
87 	if ((ip->i_flag & IN_MODIFIED) == 0 && waitfor == 0)
88 		return (0);
89 	ip->i_flag &= ~(IN_LAZYMOD | IN_MODIFIED);
90 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
91 		return (0);
92 	fs = ip->i_fs;
93 	/*
94 	 * Ensure that uid and gid are correct. This is a temporary
95 	 * fix until fsck has been changed to do the update.
96 	 */
97 	if (fs->fs_inodefmt < FS_44INODEFMT) {		/* XXX */
98 		ip->i_din.di_ouid = ip->i_uid;		/* XXX */
99 		ip->i_din.di_ogid = ip->i_gid;		/* XXX */
100 	}						/* XXX */
101 	error = bread(ip->i_devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
102 		(int)fs->fs_bsize, NOCRED, &bp);
103 	if (error) {
104 		brelse(bp);
105 		return (error);
106 	}
107 	if (DOINGSOFTDEP(vp))
108 		softdep_update_inodeblock(ip, bp, waitfor);
109 	else if (ip->i_effnlink != ip->i_nlink)
110 		panic("ffs_update: bad link cnt");
111 	*((struct dinode *)bp->b_data +
112 	    ino_to_fsbo(fs, ip->i_number)) = ip->i_din;
113 	if (waitfor && !DOINGASYNC(vp)) {
114 		return (bwrite(bp));
115 	} else if (vm_page_count_severe() || buf_dirty_count_severe()) {
116 		return (bwrite(bp));
117 	} else {
118 		if (bp->b_bufsize == fs->fs_bsize)
119 			bp->b_flags |= B_CLUSTEROK;
120 		bdwrite(bp);
121 		return (0);
122 	}
123 }
124 
125 #define	SINGLE	0	/* index of single indirect block */
126 #define	DOUBLE	1	/* index of double indirect block */
127 #define	TRIPLE	2	/* index of triple indirect block */
128 /*
129  * Truncate the inode oip to at most length size, freeing the
130  * disk blocks.
131  */
132 int
133 ffs_truncate(vp, length, flags, cred, p)
134 	struct vnode *vp;
135 	off_t length;
136 	int flags;
137 	struct ucred *cred;
138 	struct proc *p;
139 {
140 	register struct vnode *ovp = vp;
141 	ufs_daddr_t lastblock;
142 	register struct inode *oip;
143 	ufs_daddr_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR];
144 	ufs_daddr_t oldblks[NDADDR + NIADDR], newblks[NDADDR + NIADDR];
145 	register struct fs *fs;
146 	struct buf *bp;
147 	int offset, size, level;
148 	long count, nblocks, blocksreleased = 0;
149 	register int i;
150 	int aflags, error, allerror;
151 	off_t osize;
152 
153 	oip = VTOI(ovp);
154 	fs = oip->i_fs;
155 	if (length < 0)
156 		return (EINVAL);
157 	if (length > fs->fs_maxfilesize)
158 		return (EFBIG);
159 	if (ovp->v_type == VLNK &&
160 	    (oip->i_size < ovp->v_mount->mnt_maxsymlinklen || oip->i_din.di_blocks == 0)) {
161 #ifdef DIAGNOSTIC
162 		if (length != 0)
163 			panic("ffs_truncate: partial truncate of symlink");
164 #endif
165 		bzero((char *)&oip->i_shortlink, (u_int)oip->i_size);
166 		oip->i_size = 0;
167 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
168 		return (UFS_UPDATE(ovp, 1));
169 	}
170 	if (oip->i_size == length) {
171 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
172 		return (UFS_UPDATE(ovp, 0));
173 	}
174 #ifdef QUOTA
175 	error = getinoquota(oip);
176 	if (error)
177 		return (error);
178 #endif
179 	if ((oip->i_flags & SF_SNAPSHOT) != 0)
180 		ffs_snapremove(ovp);
181 	ovp->v_lasta = ovp->v_clen = ovp->v_cstart = ovp->v_lastw = 0;
182 	if (DOINGSOFTDEP(ovp)) {
183 		if (length > 0 || softdep_slowdown(ovp)) {
184 			/*
185 			 * If a file is only partially truncated, then
186 			 * we have to clean up the data structures
187 			 * describing the allocation past the truncation
188 			 * point. Finding and deallocating those structures
189 			 * is a lot of work. Since partial truncation occurs
190 			 * rarely, we solve the problem by syncing the file
191 			 * so that it will have no data structures left.
192 			 */
193 			if ((error = VOP_FSYNC(ovp, cred, MNT_WAIT,
194 			    p)) != 0)
195 				return (error);
196 			if (oip->i_flag & IN_SPACECOUNTED)
197 				fs->fs_pendingblocks -= oip->i_blocks;
198 		} else {
199 #ifdef QUOTA
200 			(void) chkdq(oip, -oip->i_blocks, NOCRED, 0);
201 #endif
202 			softdep_setup_freeblocks(oip, length);
203 			vinvalbuf(ovp, 0, cred, p, 0, 0);
204 			oip->i_flag |= IN_CHANGE | IN_UPDATE;
205 			return (ffs_update(ovp, 0));
206 		}
207 	}
208 	osize = oip->i_size;
209 	/*
210 	 * Lengthen the size of the file. We must ensure that the
211 	 * last byte of the file is allocated. Since the smallest
212 	 * value of osize is 0, length will be at least 1.
213 	 */
214 	if (osize < length) {
215 		vnode_pager_setsize(ovp, length);
216 		aflags = B_CLRBUF;
217 		if (flags & IO_SYNC)
218 			aflags |= B_SYNC;
219 		error = UFS_BALLOC(ovp, length - 1, 1,
220 		    cred, aflags, &bp);
221 		if (error)
222 			return (error);
223 		oip->i_size = length;
224 		if (bp->b_bufsize == fs->fs_bsize)
225 			bp->b_flags |= B_CLUSTEROK;
226 		if (aflags & B_SYNC)
227 			bwrite(bp);
228 		else
229 			bawrite(bp);
230 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
231 		return (UFS_UPDATE(ovp, 1));
232 	}
233 	/*
234 	 * Shorten the size of the file. If the file is not being
235 	 * truncated to a block boundary, the contents of the
236 	 * partial block following the end of the file must be
237 	 * zero'ed in case it ever becomes accessible again because
238 	 * of subsequent file growth. Directories however are not
239 	 * zero'ed as they should grow back initialized to empty.
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 (flags & IO_SYNC)
248 			aflags |= B_SYNC;
249 		error = UFS_BALLOC(ovp, length - 1, 1, cred, aflags, &bp);
250 		if (error) {
251 			return (error);
252 		}
253 		oip->i_size = length;
254 		size = blksize(fs, oip, lbn);
255 		if (ovp->v_type != VDIR)
256 			bzero((char *)bp->b_data + offset,
257 			    (u_int)(size - offset));
258 		/* Kirk's code has reallocbuf(bp, size, 1) here */
259 		allocbuf(bp, size);
260 		if (bp->b_bufsize == fs->fs_bsize)
261 			bp->b_flags |= B_CLUSTEROK;
262 		if (aflags & B_SYNC)
263 			bwrite(bp);
264 		else
265 			bawrite(bp);
266 	}
267 	/*
268 	 * Calculate index into inode's block list of
269 	 * last direct and indirect blocks (if any)
270 	 * which we want to keep.  Lastblock is -1 when
271 	 * the file is truncated to 0.
272 	 */
273 	lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1;
274 	lastiblock[SINGLE] = lastblock - NDADDR;
275 	lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
276 	lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
277 	nblocks = btodb(fs->fs_bsize);
278 	/*
279 	 * Update file and block pointers on disk before we start freeing
280 	 * blocks.  If we crash before free'ing blocks below, the blocks
281 	 * will be returned to the free list.  lastiblock values are also
282 	 * normalized to -1 for calls to ffs_indirtrunc below.
283 	 */
284 	bcopy((caddr_t)&oip->i_db[0], (caddr_t)oldblks, sizeof oldblks);
285 	for (level = TRIPLE; level >= SINGLE; level--)
286 		if (lastiblock[level] < 0) {
287 			oip->i_ib[level] = 0;
288 			lastiblock[level] = -1;
289 		}
290 	for (i = NDADDR - 1; i > lastblock; i--)
291 		oip->i_db[i] = 0;
292 	oip->i_flag |= IN_CHANGE | IN_UPDATE;
293 	allerror = UFS_UPDATE(ovp, 1);
294 
295 	/*
296 	 * Having written the new inode to disk, save its new configuration
297 	 * and put back the old block pointers long enough to process them.
298 	 * Note that we save the new block configuration so we can check it
299 	 * when we are done.
300 	 */
301 	bcopy((caddr_t)&oip->i_db[0], (caddr_t)newblks, sizeof newblks);
302 	bcopy((caddr_t)oldblks, (caddr_t)&oip->i_db[0], sizeof oldblks);
303 	oip->i_size = osize;
304 
305 	error = vtruncbuf(ovp, cred, p, length, fs->fs_bsize);
306 	if (error && (allerror == 0))
307 		allerror = error;
308 
309 	/*
310 	 * Indirect blocks first.
311 	 */
312 	indir_lbn[SINGLE] = -NDADDR;
313 	indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
314 	indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
315 	for (level = TRIPLE; level >= SINGLE; level--) {
316 		bn = oip->i_ib[level];
317 		if (bn != 0) {
318 			error = ffs_indirtrunc(oip, indir_lbn[level],
319 			    fsbtodb(fs, bn), lastiblock[level], level, &count);
320 			if (error)
321 				allerror = error;
322 			blocksreleased += count;
323 			if (lastiblock[level] < 0) {
324 				oip->i_ib[level] = 0;
325 				ffs_blkfree(oip, bn, fs->fs_bsize);
326 				blocksreleased += nblocks;
327 			}
328 		}
329 		if (lastiblock[level] >= 0)
330 			goto done;
331 	}
332 
333 	/*
334 	 * All whole direct blocks or frags.
335 	 */
336 	for (i = NDADDR - 1; i > lastblock; i--) {
337 		register long bsize;
338 
339 		bn = oip->i_db[i];
340 		if (bn == 0)
341 			continue;
342 		oip->i_db[i] = 0;
343 		bsize = blksize(fs, oip, i);
344 		ffs_blkfree(oip, bn, bsize);
345 		blocksreleased += btodb(bsize);
346 	}
347 	if (lastblock < 0)
348 		goto done;
349 
350 	/*
351 	 * Finally, look for a change in size of the
352 	 * last direct block; release any frags.
353 	 */
354 	bn = oip->i_db[lastblock];
355 	if (bn != 0) {
356 		long oldspace, newspace;
357 
358 		/*
359 		 * Calculate amount of space we're giving
360 		 * back as old block size minus new block size.
361 		 */
362 		oldspace = blksize(fs, oip, lastblock);
363 		oip->i_size = length;
364 		newspace = blksize(fs, oip, lastblock);
365 		if (newspace == 0)
366 			panic("ffs_truncate: newspace");
367 		if (oldspace - newspace > 0) {
368 			/*
369 			 * Block number of space to be free'd is
370 			 * the old block # plus the number of frags
371 			 * required for the storage we're keeping.
372 			 */
373 			bn += numfrags(fs, newspace);
374 			ffs_blkfree(oip, bn, oldspace - newspace);
375 			blocksreleased += btodb(oldspace - newspace);
376 		}
377 	}
378 done:
379 #ifdef DIAGNOSTIC
380 	for (level = SINGLE; level <= TRIPLE; level++)
381 		if (newblks[NDADDR + level] != oip->i_ib[level])
382 			panic("ffs_truncate1");
383 	for (i = 0; i < NDADDR; i++)
384 		if (newblks[i] != oip->i_db[i])
385 			panic("ffs_truncate2");
386 	if (length == 0 &&
387 	    (!TAILQ_EMPTY(&ovp->v_dirtyblkhd) ||
388 	     !TAILQ_EMPTY(&ovp->v_cleanblkhd)))
389 		panic("ffs_truncate3");
390 #endif /* DIAGNOSTIC */
391 	/*
392 	 * Put back the real size.
393 	 */
394 	oip->i_size = length;
395 	oip->i_blocks -= blocksreleased;
396 
397 	if (oip->i_blocks < 0)			/* sanity */
398 		oip->i_blocks = 0;
399 	oip->i_flag |= IN_CHANGE;
400 #ifdef QUOTA
401 	(void) chkdq(oip, -blocksreleased, NOCRED, 0);
402 #endif
403 	return (allerror);
404 }
405 
406 /*
407  * Release blocks associated with the inode ip and stored in the indirect
408  * block bn.  Blocks are free'd in LIFO order up to (but not including)
409  * lastbn.  If level is greater than SINGLE, the block is an indirect block
410  * and recursive calls to indirtrunc must be used to cleanse other indirect
411  * blocks.
412  *
413  * NB: triple indirect blocks are untested.
414  */
415 static int
416 ffs_indirtrunc(ip, lbn, dbn, lastbn, level, countp)
417 	register struct inode *ip;
418 	ufs_daddr_t lbn, lastbn;
419 	ufs_daddr_t dbn;
420 	int level;
421 	long *countp;
422 {
423 	register int i;
424 	struct buf *bp;
425 	register struct fs *fs = ip->i_fs;
426 	register ufs_daddr_t *bap;
427 	struct vnode *vp;
428 	ufs_daddr_t *copy = NULL, nb, nlbn, last;
429 	long blkcount, factor;
430 	int nblocks, blocksreleased = 0;
431 	int error = 0, allerror = 0;
432 
433 	/*
434 	 * Calculate index in current block of last
435 	 * block to be kept.  -1 indicates the entire
436 	 * block so we need not calculate the index.
437 	 */
438 	factor = 1;
439 	for (i = SINGLE; i < level; i++)
440 		factor *= NINDIR(fs);
441 	last = lastbn;
442 	if (lastbn > 0)
443 		last /= factor;
444 	nblocks = btodb(fs->fs_bsize);
445 	/*
446 	 * Get buffer of block pointers, zero those entries corresponding
447 	 * to blocks to be free'd, and update on disk copy first.  Since
448 	 * double(triple) indirect before single(double) indirect, calls
449 	 * to bmap on these blocks will fail.  However, we already have
450 	 * the on disk address, so we have to set the b_blkno field
451 	 * explicitly instead of letting bread do everything for us.
452 	 */
453 	vp = ITOV(ip);
454 	bp = getblk(vp, lbn, (int)fs->fs_bsize, 0, 0);
455 	if ((bp->b_flags & B_CACHE) == 0) {
456 		curproc->p_stats->p_ru.ru_inblock++;	/* pay for read */
457 		bp->b_iocmd = BIO_READ;
458 		bp->b_flags &= ~B_INVAL;
459 		bp->b_ioflags &= ~BIO_ERROR;
460 		if (bp->b_bcount > bp->b_bufsize)
461 			panic("ffs_indirtrunc: bad buffer size");
462 		bp->b_blkno = dbn;
463 		vfs_busy_pages(bp, 0);
464 		BUF_STRATEGY(bp);
465 		error = bufwait(bp);
466 	}
467 	if (error) {
468 		brelse(bp);
469 		*countp = 0;
470 		return (error);
471 	}
472 
473 	bap = (ufs_daddr_t *)bp->b_data;
474 	if (lastbn != -1) {
475 		MALLOC(copy, ufs_daddr_t *, fs->fs_bsize, M_TEMP, M_WAITOK);
476 		bcopy((caddr_t)bap, (caddr_t)copy, (u_int)fs->fs_bsize);
477 		bzero((caddr_t)&bap[last + 1],
478 		    (u_int)(NINDIR(fs) - (last + 1)) * sizeof (ufs_daddr_t));
479 		if (DOINGASYNC(vp)) {
480 			bawrite(bp);
481 		} else {
482 			error = bwrite(bp);
483 			if (error)
484 				allerror = error;
485 		}
486 		bap = copy;
487 	}
488 
489 	/*
490 	 * Recursively free totally unused blocks.
491 	 */
492 	for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
493 	    i--, nlbn += factor) {
494 		nb = bap[i];
495 		if (nb == 0)
496 			continue;
497 		if (level > SINGLE) {
498 			if ((error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
499 			    (ufs_daddr_t)-1, level - 1, &blkcount)) != 0)
500 				allerror = error;
501 			blocksreleased += blkcount;
502 		}
503 		ffs_blkfree(ip, nb, fs->fs_bsize);
504 		blocksreleased += nblocks;
505 	}
506 
507 	/*
508 	 * Recursively free last partial block.
509 	 */
510 	if (level > SINGLE && lastbn >= 0) {
511 		last = lastbn % factor;
512 		nb = bap[i];
513 		if (nb != 0) {
514 			error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
515 			    last, level - 1, &blkcount);
516 			if (error)
517 				allerror = error;
518 			blocksreleased += blkcount;
519 		}
520 	}
521 	if (copy != NULL) {
522 		FREE(copy, M_TEMP);
523 	} else {
524 		bp->b_flags |= B_INVAL | B_NOCACHE;
525 		brelse(bp);
526 	}
527 
528 	*countp = blocksreleased;
529 	return (allerror);
530 }
531