xref: /freebsd/sys/ufs/ffs/ffs_inode.c (revision a8445737e740901f5f2c8d24c12ef7fc8b00134e)
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 1.46 1998/07/04 20:45:38 julian 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, access, modify, waitfor)
74 	struct vnode *vp;
75 	struct timeval *access;
76 	struct timeval *modify;
77 	int waitfor;
78 {
79 	register struct fs *fs;
80 	struct buf *bp;
81 	struct inode *ip;
82 	int error;
83 
84 	ip = VTOI(vp);
85 	if (((ip->i_flag &
86 	      (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0) &&
87 	    (waitfor != MNT_WAIT))
88 		return (0);
89 	ufs_itimes(vp);
90 	ip->i_flag &= ~(IN_LAZYMOD | IN_MODIFIED);
91 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
92 		return (0);
93 	fs = ip->i_fs;
94 	/*
95 	 * Ensure that uid and gid are correct. This is a temporary
96 	 * fix until fsck has been changed to do the update.
97 	 */
98 	if (fs->fs_inodefmt < FS_44INODEFMT) {		/* XXX */
99 		ip->i_din.di_ouid = ip->i_uid;		/* XXX */
100 		ip->i_din.di_ogid = ip->i_gid;		/* XXX */
101 	}						/* XXX */
102 	error = bread(ip->i_devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
103 		(int)fs->fs_bsize, NOCRED, &bp);
104 	if (error) {
105 		brelse(bp);
106 		return (error);
107 	}
108 	if (DOINGSOFTDEP(vp))
109 		softdep_update_inodeblock(ip, bp, waitfor);
110 	else if (ip->i_effnlink != ip->i_nlink)
111 		panic("ffs_update: bad link cnt");
112 	*((struct dinode *)bp->b_data +
113 	    ino_to_fsbo(fs, ip->i_number)) = ip->i_din;
114 	if (waitfor && (vp->v_mount->mnt_flag & MNT_ASYNC) == 0) {
115 		return (bwrite(bp));
116 	} else {
117 		if (bp->b_bufsize == fs->fs_bsize)
118 			bp->b_flags |= B_CLUSTEROK;
119 		bdwrite(bp);
120 		return (0);
121 	}
122 }
123 
124 #define	SINGLE	0	/* index of single indirect block */
125 #define	DOUBLE	1	/* index of double indirect block */
126 #define	TRIPLE	2	/* index of triple indirect block */
127 /*
128  * Truncate the inode oip to at most length size, freeing the
129  * disk blocks.
130  */
131 int
132 ffs_truncate(vp, length, flags, cred, p)
133 	struct vnode *vp;
134 	off_t length;
135 	int flags;
136 	struct ucred *cred;
137 	struct proc *p;
138 {
139 	register struct vnode *ovp = vp;
140 	ufs_daddr_t lastblock;
141 	register struct inode *oip;
142 	ufs_daddr_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR];
143 	ufs_daddr_t oldblks[NDADDR + NIADDR], newblks[NDADDR + NIADDR];
144 	register struct fs *fs;
145 	struct buf *bp;
146 	int offset, size, level;
147 	long count, nblocks, vflags, blocksreleased = 0;
148 	struct timeval tv;
149 	register int i;
150 	int aflags, error, allerror;
151 	off_t osize;
152 
153 	oip = VTOI(ovp);
154 	if (oip->i_size == length)
155 		return (0);
156 	fs = oip->i_fs;
157 	if (length < 0)
158 		return (EINVAL);
159 	if (length > fs->fs_maxfilesize)
160 		return (EFBIG);
161 	getmicrotime(&tv);
162 	if (ovp->v_type == VLNK &&
163 	    (oip->i_size < ovp->v_mount->mnt_maxsymlinklen || oip->i_din.di_blocks == 0)) {
164 #ifdef DIAGNOSTIC
165 		if (length != 0)
166 			panic("ffs_truncate: partial truncate of symlink");
167 #endif
168 		bzero((char *)&oip->i_shortlink, (u_int)oip->i_size);
169 		oip->i_size = 0;
170 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
171 		return (UFS_UPDATE(ovp, &tv, &tv, 1));
172 	}
173 	if (oip->i_size == length) {
174 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
175 		return (UFS_UPDATE(ovp, &tv, &tv, 0));
176 	}
177 #ifdef QUOTA
178 	error = getinoquota(oip);
179 	if (error)
180 		return (error);
181 #endif
182 	ovp->v_lasta = ovp->v_clen = ovp->v_cstart = ovp->v_lastw = 0;
183 	if (DOINGSOFTDEP(ovp)) {
184 		if (length > 0) {
185 			/*
186 			 * If a file is only partially truncated, then
187 			 * we have to clean up the data structures
188 			 * describing the allocation past the truncation
189 			 * point. Finding and deallocating those structures
190 			 * is a lot of work. Since partial truncation occurs
191 			 * rarely, we solve the problem by syncing the file
192 			 * so that it will have no data structures left.
193 			 */
194 			if ((error = VOP_FSYNC(ovp, cred, MNT_WAIT,
195 			    p)) != 0)
196 				return (error);
197 		} else {
198 #ifdef QUOTA
199 			(void) chkdq(oip, -oip->i_blocks, NOCRED, 0);
200 #endif
201 			softdep_setup_freeblocks(oip, length);
202 			vinvalbuf(ovp, 0, cred, p, 0, 0);
203 			oip->i_flag |= IN_CHANGE | IN_UPDATE;
204 			return (ffs_update(ovp, &tv, &tv, 0));
205 		}
206 	}
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 		vnode_pager_setsize(ovp, length);
215 		aflags = B_CLRBUF;
216 		if (flags & IO_SYNC)
217 			aflags |= B_SYNC;
218 		error = VOP_BALLOC(ovp, length - 1, 1,
219 		    cred, aflags, &bp);
220 		if (error)
221 			return (error);
222 		oip->i_size = length;
223 		if (bp->b_bufsize == fs->fs_bsize)
224 			bp->b_flags |= B_CLUSTEROK;
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 (UFS_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 boundary, the contents of the
237 	 * partial block following the end of the file must be
238 	 * zero'ed in case it ever becomes accessible again because
239 	 * of subsequent file growth. Directories however are not
240 	 * zero'ed as they should grow back initialized to empty.
241 	 */
242 	offset = blkoff(fs, length);
243 	if (offset == 0) {
244 		oip->i_size = length;
245 	} else {
246 		lbn = lblkno(fs, length);
247 		aflags = B_CLRBUF;
248 		if (flags & IO_SYNC)
249 			aflags |= B_SYNC;
250 		error = VOP_BALLOC(ovp, length - 1, 1, cred, aflags, &bp);
251 		if (error) {
252 			return (error);
253 		}
254 		oip->i_size = length;
255 		size = blksize(fs, oip, lbn);
256 		if (ovp->v_type != VDIR)
257 			bzero((char *)bp->b_data + offset,
258 			    (u_int)(size - offset));
259 		/* Kirk's code has reallocbuf(bp, size, 1) here */
260 		allocbuf(bp, size);
261 		if (bp->b_bufsize == fs->fs_bsize)
262 			bp->b_flags |= B_CLUSTEROK;
263 		if (aflags & B_SYNC)
264 			bwrite(bp);
265 		else if (ovp->v_mount->mnt_flag & MNT_ASYNC)
266 			bdwrite(bp);
267 		else
268 			bawrite(bp);
269 	}
270 	/*
271 	 * Calculate index into inode's block list of
272 	 * last direct and indirect blocks (if any)
273 	 * which we want to keep.  Lastblock is -1 when
274 	 * the file is truncated to 0.
275 	 */
276 	lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1;
277 	lastiblock[SINGLE] = lastblock - NDADDR;
278 	lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
279 	lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
280 	nblocks = btodb(fs->fs_bsize);
281 	/*
282 	 * Update file and block pointers on disk before we start freeing
283 	 * blocks.  If we crash before free'ing blocks below, the blocks
284 	 * will be returned to the free list.  lastiblock values are also
285 	 * normalized to -1 for calls to ffs_indirtrunc below.
286 	 */
287 	bcopy((caddr_t)&oip->i_db[0], (caddr_t)oldblks, sizeof oldblks);
288 	for (level = TRIPLE; level >= SINGLE; level--)
289 		if (lastiblock[level] < 0) {
290 			oip->i_ib[level] = 0;
291 			lastiblock[level] = -1;
292 		}
293 	for (i = NDADDR - 1; i > lastblock; i--)
294 		oip->i_db[i] = 0;
295 	oip->i_flag |= IN_CHANGE | IN_UPDATE;
296 	allerror = UFS_UPDATE(ovp, &tv, &tv, ((length > 0) ? 0 : 1));
297 
298 	/*
299 	 * Having written the new inode to disk, save its new configuration
300 	 * and put back the old block pointers long enough to process them.
301 	 * Note that we save the new block configuration so we can check it
302 	 * when we are done.
303 	 */
304 	bcopy((caddr_t)&oip->i_db[0], (caddr_t)newblks, sizeof newblks);
305 	bcopy((caddr_t)oldblks, (caddr_t)&oip->i_db[0], sizeof oldblks);
306 	oip->i_size = osize;
307 
308 	error = vtruncbuf(ovp, cred, p, length, fs->fs_bsize);
309 	if (error && (allerror == 0))
310 		allerror = error;
311 
312 	/*
313 	 * Indirect blocks first.
314 	 */
315 	indir_lbn[SINGLE] = -NDADDR;
316 	indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
317 	indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
318 	for (level = TRIPLE; level >= SINGLE; level--) {
319 		bn = oip->i_ib[level];
320 		if (bn != 0) {
321 			error = ffs_indirtrunc(oip, indir_lbn[level],
322 			    fsbtodb(fs, bn), lastiblock[level], level, &count);
323 			if (error)
324 				allerror = error;
325 			blocksreleased += count;
326 			if (lastiblock[level] < 0) {
327 				oip->i_ib[level] = 0;
328 				ffs_blkfree(oip, bn, fs->fs_bsize);
329 				blocksreleased += nblocks;
330 			}
331 		}
332 		if (lastiblock[level] >= 0)
333 			goto done;
334 	}
335 
336 	/*
337 	 * All whole direct blocks or frags.
338 	 */
339 	for (i = NDADDR - 1; i > lastblock; i--) {
340 		register long bsize;
341 
342 		bn = oip->i_db[i];
343 		if (bn == 0)
344 			continue;
345 		oip->i_db[i] = 0;
346 		bsize = blksize(fs, oip, i);
347 		ffs_blkfree(oip, bn, bsize);
348 		blocksreleased += btodb(bsize);
349 	}
350 	if (lastblock < 0)
351 		goto done;
352 
353 	/*
354 	 * Finally, look for a change in size of the
355 	 * last direct block; release any frags.
356 	 */
357 	bn = oip->i_db[lastblock];
358 	if (bn != 0) {
359 		long oldspace, newspace;
360 
361 		/*
362 		 * Calculate amount of space we're giving
363 		 * back as old block size minus new block size.
364 		 */
365 		oldspace = blksize(fs, oip, lastblock);
366 		oip->i_size = length;
367 		newspace = blksize(fs, oip, lastblock);
368 		if (newspace == 0)
369 			panic("ffs_truncate: newspace");
370 		if (oldspace - newspace > 0) {
371 			/*
372 			 * Block number of space to be free'd is
373 			 * the old block # plus the number of frags
374 			 * required for the storage we're keeping.
375 			 */
376 			bn += numfrags(fs, newspace);
377 			ffs_blkfree(oip, bn, oldspace - newspace);
378 			blocksreleased += btodb(oldspace - newspace);
379 		}
380 	}
381 done:
382 #ifdef DIAGNOSTIC
383 	for (level = SINGLE; level <= TRIPLE; level++)
384 		if (newblks[NDADDR + level] != oip->i_ib[level])
385 			panic("ffs_truncate1");
386 	for (i = 0; i < NDADDR; i++)
387 		if (newblks[i] != oip->i_db[i])
388 			panic("ffs_truncate2");
389 	if (length == 0 &&
390 	    (ovp->v_dirtyblkhd.lh_first || ovp->v_cleanblkhd.lh_first))
391 		panic("ffs_truncate3");
392 #endif /* DIAGNOSTIC */
393 	/*
394 	 * Put back the real size.
395 	 */
396 	oip->i_size = length;
397 	oip->i_blocks -= blocksreleased;
398 
399 	if (oip->i_blocks < 0)			/* sanity */
400 		oip->i_blocks = 0;
401 	oip->i_flag |= IN_CHANGE;
402 #ifdef QUOTA
403 	(void) chkdq(oip, -blocksreleased, NOCRED, 0);
404 #endif
405 	return (allerror);
406 }
407 
408 /*
409  * Release blocks associated with the inode ip and stored in the indirect
410  * block bn.  Blocks are free'd in LIFO order up to (but not including)
411  * lastbn.  If level is greater than SINGLE, the block is an indirect block
412  * and recursive calls to indirtrunc must be used to cleanse other indirect
413  * blocks.
414  *
415  * NB: triple indirect blocks are untested.
416  */
417 static int
418 ffs_indirtrunc(ip, lbn, dbn, lastbn, level, countp)
419 	register struct inode *ip;
420 	ufs_daddr_t lbn, lastbn;
421 	ufs_daddr_t dbn;
422 	int level;
423 	long *countp;
424 {
425 	register int i;
426 	struct buf *bp;
427 	register struct fs *fs = ip->i_fs;
428 	register ufs_daddr_t *bap;
429 	struct vnode *vp;
430 	ufs_daddr_t *copy = NULL, nb, nlbn, last;
431 	long blkcount, factor;
432 	int nblocks, blocksreleased = 0;
433 	int error = 0, allerror = 0;
434 
435 	/*
436 	 * Calculate index in current block of last
437 	 * block to be kept.  -1 indicates the entire
438 	 * block so we need not calculate the index.
439 	 */
440 	factor = 1;
441 	for (i = SINGLE; i < level; i++)
442 		factor *= NINDIR(fs);
443 	last = lastbn;
444 	if (lastbn > 0)
445 		last /= factor;
446 	nblocks = btodb(fs->fs_bsize);
447 	/*
448 	 * Get buffer of block pointers, zero those entries corresponding
449 	 * to blocks to be free'd, and update on disk copy first.  Since
450 	 * double(triple) indirect before single(double) indirect, calls
451 	 * to bmap on these blocks will fail.  However, we already have
452 	 * the on disk address, so we have to set the b_blkno field
453 	 * explicitly instead of letting bread do everything for us.
454 	 */
455 	vp = ITOV(ip);
456 	bp = getblk(vp, lbn, (int)fs->fs_bsize, 0, 0);
457 	if ((bp->b_flags & B_CACHE) == 0) {
458 		curproc->p_stats->p_ru.ru_inblock++;	/* pay for read */
459 		bp->b_flags |= B_READ;
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 		VOP_STRATEGY(bp->b_vp, bp);
465 		error = biowait(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 ((vp->v_mount->mnt_flag & MNT_ASYNC) == 0) {
480 			error = bwrite(bp);
481 			if (error)
482 				allerror = error;
483 		} else {
484 			bawrite(bp);
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))
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