xref: /freebsd/sys/ufs/ffs/ffs_inode.c (revision 70fe064ad7cab6c0444b91622f60ec6a462f308a)
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 	fs = ip->i_fs;
91 	if (fs->fs_ronly)
92 		return (0);
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, td)
134 	struct vnode *vp;
135 	off_t length;
136 	int flags;
137 	struct ucred *cred;
138 	struct thread *td;
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 	if (fs->fs_ronly)
175 		panic("ffs_truncate: read-only filesystem");
176 #ifdef QUOTA
177 	error = getinoquota(oip);
178 	if (error)
179 		return (error);
180 #endif
181 	if ((oip->i_flags & SF_SNAPSHOT) != 0)
182 		ffs_snapremove(ovp);
183 	ovp->v_lasta = ovp->v_clen = ovp->v_cstart = ovp->v_lastw = 0;
184 	if (DOINGSOFTDEP(ovp)) {
185 		if (length > 0 || softdep_slowdown(ovp)) {
186 			/*
187 			 * If a file is only partially truncated, then
188 			 * we have to clean up the data structures
189 			 * describing the allocation past the truncation
190 			 * point. Finding and deallocating those structures
191 			 * is a lot of work. Since partial truncation occurs
192 			 * rarely, we solve the problem by syncing the file
193 			 * so that it will have no data structures left.
194 			 */
195 			if ((error = VOP_FSYNC(ovp, cred, MNT_WAIT,
196 			    td)) != 0)
197 				return (error);
198 			if (oip->i_flag & IN_SPACECOUNTED)
199 				fs->fs_pendingblocks -= oip->i_blocks;
200 		} else {
201 #ifdef QUOTA
202 			(void) chkdq(oip, -oip->i_blocks, NOCRED, 0);
203 #endif
204 			softdep_setup_freeblocks(oip, length);
205 			vinvalbuf(ovp, 0, cred, td, 0, 0);
206 			oip->i_flag |= IN_CHANGE | IN_UPDATE;
207 			return (ffs_update(ovp, 0));
208 		}
209 	}
210 	osize = oip->i_size;
211 	/*
212 	 * Lengthen the size of the file. We must ensure that the
213 	 * last byte of the file is allocated. Since the smallest
214 	 * value of osize is 0, length will be at least 1.
215 	 */
216 	if (osize < length) {
217 		vnode_pager_setsize(ovp, length);
218 		aflags = B_CLRBUF;
219 		if (flags & IO_SYNC)
220 			aflags |= B_SYNC;
221 		error = UFS_BALLOC(ovp, length - 1, 1,
222 		    cred, aflags, &bp);
223 		if (error)
224 			return (error);
225 		oip->i_size = length;
226 		if (bp->b_bufsize == fs->fs_bsize)
227 			bp->b_flags |= B_CLUSTEROK;
228 		if (aflags & B_SYNC)
229 			bwrite(bp);
230 		else
231 			bawrite(bp);
232 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
233 		return (UFS_UPDATE(ovp, 1));
234 	}
235 	/*
236 	 * Shorten the size of the file. If the file is not being
237 	 * truncated to a block boundary, the contents of the
238 	 * partial block following the end of the file must be
239 	 * zero'ed in case it ever becomes accessible again because
240 	 * of subsequent file growth. Directories however are not
241 	 * zero'ed as they should grow back initialized to empty.
242 	 */
243 	offset = blkoff(fs, length);
244 	if (offset == 0) {
245 		oip->i_size = length;
246 	} else {
247 		lbn = lblkno(fs, length);
248 		aflags = B_CLRBUF;
249 		if (flags & IO_SYNC)
250 			aflags |= B_SYNC;
251 		error = UFS_BALLOC(ovp, length - 1, 1, cred, aflags, &bp);
252 		if (error) {
253 			return (error);
254 		}
255 		/*
256 		 * When we are doing soft updates and the UFS_BALLOC
257 		 * above fills in a direct block hole with a full sized
258 		 * block that will be truncated down to a fragment below,
259 		 * we must flush out the block dependency with an FSYNC
260 		 * so that we do not get a soft updates inconsistency
261 		 * when we create the fragment below.
262 		 */
263 		if (DOINGSOFTDEP(ovp) && lbn < NDADDR &&
264 		    fragroundup(fs, blkoff(fs, length)) < fs->fs_bsize &&
265 		    (error = VOP_FSYNC(ovp, cred, MNT_WAIT, td)) != 0)
266 			return (error);
267 		oip->i_size = length;
268 		size = blksize(fs, oip, lbn);
269 		if (ovp->v_type != VDIR)
270 			bzero((char *)bp->b_data + offset,
271 			    (u_int)(size - offset));
272 		/* Kirk's code has reallocbuf(bp, size, 1) here */
273 		allocbuf(bp, size);
274 		if (bp->b_bufsize == fs->fs_bsize)
275 			bp->b_flags |= B_CLUSTEROK;
276 		if (aflags & B_SYNC)
277 			bwrite(bp);
278 		else
279 			bawrite(bp);
280 	}
281 	/*
282 	 * Calculate index into inode's block list of
283 	 * last direct and indirect blocks (if any)
284 	 * which we want to keep.  Lastblock is -1 when
285 	 * the file is truncated to 0.
286 	 */
287 	lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1;
288 	lastiblock[SINGLE] = lastblock - NDADDR;
289 	lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
290 	lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
291 	nblocks = btodb(fs->fs_bsize);
292 	/*
293 	 * Update file and block pointers on disk before we start freeing
294 	 * blocks.  If we crash before free'ing blocks below, the blocks
295 	 * will be returned to the free list.  lastiblock values are also
296 	 * normalized to -1 for calls to ffs_indirtrunc below.
297 	 */
298 	bcopy((caddr_t)&oip->i_db[0], (caddr_t)oldblks, sizeof oldblks);
299 	for (level = TRIPLE; level >= SINGLE; level--)
300 		if (lastiblock[level] < 0) {
301 			oip->i_ib[level] = 0;
302 			lastiblock[level] = -1;
303 		}
304 	for (i = NDADDR - 1; i > lastblock; i--)
305 		oip->i_db[i] = 0;
306 	oip->i_flag |= IN_CHANGE | IN_UPDATE;
307 	allerror = UFS_UPDATE(ovp, 1);
308 
309 	/*
310 	 * Having written the new inode to disk, save its new configuration
311 	 * and put back the old block pointers long enough to process them.
312 	 * Note that we save the new block configuration so we can check it
313 	 * when we are done.
314 	 */
315 	bcopy((caddr_t)&oip->i_db[0], (caddr_t)newblks, sizeof newblks);
316 	bcopy((caddr_t)oldblks, (caddr_t)&oip->i_db[0], sizeof oldblks);
317 	oip->i_size = osize;
318 
319 	error = vtruncbuf(ovp, cred, td, length, fs->fs_bsize);
320 	if (error && (allerror == 0))
321 		allerror = error;
322 
323 	/*
324 	 * Indirect blocks first.
325 	 */
326 	indir_lbn[SINGLE] = -NDADDR;
327 	indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
328 	indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
329 	for (level = TRIPLE; level >= SINGLE; level--) {
330 		bn = oip->i_ib[level];
331 		if (bn != 0) {
332 			error = ffs_indirtrunc(oip, indir_lbn[level],
333 			    fsbtodb(fs, bn), lastiblock[level], level, &count);
334 			if (error)
335 				allerror = error;
336 			blocksreleased += count;
337 			if (lastiblock[level] < 0) {
338 				oip->i_ib[level] = 0;
339 				ffs_blkfree(oip, bn, fs->fs_bsize);
340 				blocksreleased += nblocks;
341 			}
342 		}
343 		if (lastiblock[level] >= 0)
344 			goto done;
345 	}
346 
347 	/*
348 	 * All whole direct blocks or frags.
349 	 */
350 	for (i = NDADDR - 1; i > lastblock; i--) {
351 		register long bsize;
352 
353 		bn = oip->i_db[i];
354 		if (bn == 0)
355 			continue;
356 		oip->i_db[i] = 0;
357 		bsize = blksize(fs, oip, i);
358 		ffs_blkfree(oip, bn, bsize);
359 		blocksreleased += btodb(bsize);
360 	}
361 	if (lastblock < 0)
362 		goto done;
363 
364 	/*
365 	 * Finally, look for a change in size of the
366 	 * last direct block; release any frags.
367 	 */
368 	bn = oip->i_db[lastblock];
369 	if (bn != 0) {
370 		long oldspace, newspace;
371 
372 		/*
373 		 * Calculate amount of space we're giving
374 		 * back as old block size minus new block size.
375 		 */
376 		oldspace = blksize(fs, oip, lastblock);
377 		oip->i_size = length;
378 		newspace = blksize(fs, oip, lastblock);
379 		if (newspace == 0)
380 			panic("ffs_truncate: newspace");
381 		if (oldspace - newspace > 0) {
382 			/*
383 			 * Block number of space to be free'd is
384 			 * the old block # plus the number of frags
385 			 * required for the storage we're keeping.
386 			 */
387 			bn += numfrags(fs, newspace);
388 			ffs_blkfree(oip, bn, oldspace - newspace);
389 			blocksreleased += btodb(oldspace - newspace);
390 		}
391 	}
392 done:
393 #ifdef DIAGNOSTIC
394 	for (level = SINGLE; level <= TRIPLE; level++)
395 		if (newblks[NDADDR + level] != oip->i_ib[level])
396 			panic("ffs_truncate1");
397 	for (i = 0; i < NDADDR; i++)
398 		if (newblks[i] != oip->i_db[i])
399 			panic("ffs_truncate2");
400 	if (length == 0 &&
401 	    (!TAILQ_EMPTY(&ovp->v_dirtyblkhd) ||
402 	     !TAILQ_EMPTY(&ovp->v_cleanblkhd)))
403 		panic("ffs_truncate3");
404 #endif /* DIAGNOSTIC */
405 	/*
406 	 * Put back the real size.
407 	 */
408 	oip->i_size = length;
409 	oip->i_blocks -= blocksreleased;
410 
411 	if (oip->i_blocks < 0)			/* sanity */
412 		oip->i_blocks = 0;
413 	oip->i_flag |= IN_CHANGE;
414 #ifdef QUOTA
415 	(void) chkdq(oip, -blocksreleased, NOCRED, 0);
416 #endif
417 	return (allerror);
418 }
419 
420 /*
421  * Release blocks associated with the inode ip and stored in the indirect
422  * block bn.  Blocks are free'd in LIFO order up to (but not including)
423  * lastbn.  If level is greater than SINGLE, the block is an indirect block
424  * and recursive calls to indirtrunc must be used to cleanse other indirect
425  * blocks.
426  *
427  * NB: triple indirect blocks are untested.
428  */
429 static int
430 ffs_indirtrunc(ip, lbn, dbn, lastbn, level, countp)
431 	register struct inode *ip;
432 	ufs_daddr_t lbn, lastbn;
433 	ufs_daddr_t dbn;
434 	int level;
435 	long *countp;
436 {
437 	register int i;
438 	struct buf *bp;
439 	register struct fs *fs = ip->i_fs;
440 	register ufs_daddr_t *bap;
441 	struct vnode *vp;
442 	ufs_daddr_t *copy = NULL, nb, nlbn, last;
443 	long blkcount, factor;
444 	int nblocks, blocksreleased = 0;
445 	int error = 0, allerror = 0;
446 
447 	/*
448 	 * Calculate index in current block of last
449 	 * block to be kept.  -1 indicates the entire
450 	 * block so we need not calculate the index.
451 	 */
452 	factor = 1;
453 	for (i = SINGLE; i < level; i++)
454 		factor *= NINDIR(fs);
455 	last = lastbn;
456 	if (lastbn > 0)
457 		last /= factor;
458 	nblocks = btodb(fs->fs_bsize);
459 	/*
460 	 * Get buffer of block pointers, zero those entries corresponding
461 	 * to blocks to be free'd, and update on disk copy first.  Since
462 	 * double(triple) indirect before single(double) indirect, calls
463 	 * to bmap on these blocks will fail.  However, we already have
464 	 * the on disk address, so we have to set the b_blkno field
465 	 * explicitly instead of letting bread do everything for us.
466 	 */
467 	vp = ITOV(ip);
468 	bp = getblk(vp, lbn, (int)fs->fs_bsize, 0, 0);
469 	if ((bp->b_flags & B_CACHE) == 0) {
470 		curproc->p_stats->p_ru.ru_inblock++;	/* pay for read */
471 		bp->b_iocmd = BIO_READ;
472 		bp->b_flags &= ~B_INVAL;
473 		bp->b_ioflags &= ~BIO_ERROR;
474 		if (bp->b_bcount > bp->b_bufsize)
475 			panic("ffs_indirtrunc: bad buffer size");
476 		bp->b_blkno = dbn;
477 		vfs_busy_pages(bp, 0);
478 		BUF_STRATEGY(bp);
479 		error = bufwait(bp);
480 	}
481 	if (error) {
482 		brelse(bp);
483 		*countp = 0;
484 		return (error);
485 	}
486 
487 	bap = (ufs_daddr_t *)bp->b_data;
488 	if (lastbn != -1) {
489 		MALLOC(copy, ufs_daddr_t *, fs->fs_bsize, M_TEMP, M_WAITOK);
490 		bcopy((caddr_t)bap, (caddr_t)copy, (u_int)fs->fs_bsize);
491 		bzero((caddr_t)&bap[last + 1],
492 		    (u_int)(NINDIR(fs) - (last + 1)) * sizeof (ufs_daddr_t));
493 		if (DOINGASYNC(vp)) {
494 			bawrite(bp);
495 		} else {
496 			error = bwrite(bp);
497 			if (error)
498 				allerror = error;
499 		}
500 		bap = copy;
501 	}
502 
503 	/*
504 	 * Recursively free totally unused blocks.
505 	 */
506 	for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
507 	    i--, nlbn += factor) {
508 		nb = bap[i];
509 		if (nb == 0)
510 			continue;
511 		if (level > SINGLE) {
512 			if ((error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
513 			    (ufs_daddr_t)-1, level - 1, &blkcount)) != 0)
514 				allerror = error;
515 			blocksreleased += blkcount;
516 		}
517 		ffs_blkfree(ip, nb, fs->fs_bsize);
518 		blocksreleased += nblocks;
519 	}
520 
521 	/*
522 	 * Recursively free last partial block.
523 	 */
524 	if (level > SINGLE && lastbn >= 0) {
525 		last = lastbn % factor;
526 		nb = bap[i];
527 		if (nb != 0) {
528 			error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
529 			    last, level - 1, &blkcount);
530 			if (error)
531 				allerror = error;
532 			blocksreleased += blkcount;
533 		}
534 	}
535 	if (copy != NULL) {
536 		FREE(copy, M_TEMP);
537 	} else {
538 		bp->b_flags |= B_INVAL | B_NOCACHE;
539 		brelse(bp);
540 	}
541 
542 	*countp = blocksreleased;
543 	return (allerror);
544 }
545