xref: /openbsd/sys/ufs/ffs/ffs_inode.c (revision 91f110e0)
1 /*	$OpenBSD: ffs_inode.c,v 1.68 2014/03/19 04:17:33 guenther Exp $	*/
2 /*	$NetBSD: ffs_inode.c,v 1.10 1996/05/11 18:27:19 mycroft Exp $	*/
3 
4 /*
5  * Copyright (c) 1982, 1986, 1989, 1993
6  *	The Regents of the University of California.  All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  *	@(#)ffs_inode.c	8.8 (Berkeley) 10/19/94
33  */
34 
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/mount.h>
38 #include <sys/proc.h>
39 #include <sys/file.h>
40 #include <sys/buf.h>
41 #include <sys/vnode.h>
42 #include <sys/kernel.h>
43 #include <sys/malloc.h>
44 #include <sys/resourcevar.h>
45 
46 #include <uvm/uvm_extern.h>
47 
48 #include <ufs/ufs/quota.h>
49 #include <ufs/ufs/inode.h>
50 #include <ufs/ufs/ufsmount.h>
51 #include <ufs/ufs/ufs_extern.h>
52 
53 #include <ufs/ffs/fs.h>
54 #include <ufs/ffs/ffs_extern.h>
55 
56 int ffs_indirtrunc(struct inode *, daddr_t, daddr_t, daddr_t, int, long *);
57 
58 /*
59  * Update the access, modified, and inode change times as specified by the
60  * IN_ACCESS, IN_UPDATE, and IN_CHANGE flags respectively. The IN_MODIFIED
61  * flag is used to specify that the inode needs to be updated but that the
62  * times have already been set.  The IN_LAZYMOD flag is used to specify
63  * that the inode needs to be updated at some point, by reclaim if not
64  * in the course of other changes; this is used to defer writes just to
65  * update device timestamps.  If waitfor is set, then wait for the disk
66  * write of the inode to complete.
67  */
68 int
69 ffs_update(struct inode *ip, int waitfor)
70 {
71 	struct vnode *vp;
72 	struct fs *fs;
73 	struct buf *bp;
74 	int error;
75 
76 	vp = ITOV(ip);
77 	ufs_itimes(vp);
78 
79 	if ((ip->i_flag & IN_MODIFIED) == 0 && waitfor != MNT_WAIT)
80 		return (0);
81 
82 	ip->i_flag &= ~(IN_MODIFIED | IN_LAZYMOD);
83 	fs = ip->i_fs;
84 
85 	/*
86 	 * Ensure that uid and gid are correct. This is a temporary
87 	 * fix until fsck has been changed to do the update.
88 	 */
89 	if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_inodefmt < FS_44INODEFMT) {
90 		ip->i_din1->di_ouid = ip->i_ffs1_uid;
91 		ip->i_din1->di_ogid = ip->i_ffs1_gid;
92 	}
93 
94 	error = bread(ip->i_devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
95 	    (int)fs->fs_bsize, &bp);
96 	if (error) {
97 		brelse(bp);
98 		return (error);
99 	}
100 
101 	if (DOINGSOFTDEP(vp))
102 		softdep_update_inodeblock(ip, bp, waitfor);
103 	else if (ip->i_effnlink != DIP(ip, nlink))
104 		panic("ffs_update: bad link cnt");
105 
106 #ifdef FFS2
107 	if (ip->i_ump->um_fstype == UM_UFS2)
108 		*((struct ufs2_dinode *)bp->b_data +
109 		    ino_to_fsbo(fs, ip->i_number)) = *ip->i_din2;
110 	else
111 #endif
112 		*((struct ufs1_dinode *)bp->b_data +
113 		    ino_to_fsbo(fs, ip->i_number)) = *ip->i_din1;
114 
115 	if (waitfor && !DOINGASYNC(vp)) {
116 		return (bwrite(bp));
117 	} else {
118 		bdwrite(bp);
119 		return (0);
120 	}
121 }
122 
123 #define	SINGLE	0	/* index of single indirect block */
124 #define	DOUBLE	1	/* index of double indirect block */
125 #define	TRIPLE	2	/* index of triple indirect block */
126 
127 /*
128  * Truncate the inode oip to at most length size, freeing the
129  * disk blocks.
130  */
131 int
132 ffs_truncate(struct inode *oip, off_t length, int flags, struct ucred *cred)
133 {
134 	struct vnode *ovp;
135 	daddr_t lastblock;
136 	daddr_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR];
137 	daddr_t oldblks[NDADDR + NIADDR], newblks[NDADDR + NIADDR];
138 	struct fs *fs;
139 	struct buf *bp;
140 	int offset, size, level;
141 	long count, nblocks, vflags, blocksreleased = 0;
142 	int i, aflags, error, allerror;
143 	off_t osize;
144 
145 	if (length < 0)
146 		return (EINVAL);
147 	ovp = ITOV(oip);
148 
149 	if (ovp->v_type != VREG &&
150 	    ovp->v_type != VDIR &&
151 	    ovp->v_type != VLNK)
152 		return (0);
153 
154 	if (DIP(oip, size) == length)
155 		return (0);
156 
157 	if (ovp->v_type == VLNK &&
158 	    (DIP(oip, size) < ovp->v_mount->mnt_maxsymlinklen ||
159 	     (ovp->v_mount->mnt_maxsymlinklen == 0 &&
160 	      oip->i_din1->di_blocks == 0))) {
161 #ifdef DIAGNOSTIC
162 		if (length != 0)
163 			panic("ffs_truncate: partial truncate of symlink");
164 #endif
165 		memset(SHORTLINK(oip), 0, (size_t) DIP(oip, size));
166 		DIP_ASSIGN(oip, size, 0);
167 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
168 		return (UFS_UPDATE(oip, MNT_WAIT));
169 	}
170 
171 	if ((error = getinoquota(oip)) != 0)
172 		return (error);
173 
174 	uvm_vnp_setsize(ovp, length);
175 	oip->i_ci.ci_lasta = oip->i_ci.ci_clen
176 	    = oip->i_ci.ci_cstart = oip->i_ci.ci_lastw = 0;
177 
178 	if (DOINGSOFTDEP(ovp)) {
179 		if (length > 0 || softdep_slowdown(ovp)) {
180 			/*
181 			 * If a file is only partially truncated, then
182 			 * we have to clean up the data structures
183 			 * describing the allocation past the truncation
184 			 * point. Finding and deallocating those structures
185 			 * is a lot of work. Since partial truncation occurs
186 			 * rarely, we solve the problem by syncing the file
187 			 * so that it will have no data structures left.
188 			 */
189 			if ((error = VOP_FSYNC(ovp, cred, MNT_WAIT,
190 					       curproc)) != 0)
191 				return (error);
192 		} else {
193 			(void)ufs_quota_free_blocks(oip, DIP(oip, blocks),
194 			    NOCRED);
195 			softdep_setup_freeblocks(oip, length);
196 			(void) vinvalbuf(ovp, 0, cred, curproc, 0, 0);
197 			oip->i_flag |= IN_CHANGE | IN_UPDATE;
198 			return (UFS_UPDATE(oip, 0));
199 		}
200 	}
201 
202 	fs = oip->i_fs;
203 	osize = DIP(oip, size);
204 	/*
205 	 * Lengthen the size of the file. We must ensure that the
206 	 * last byte of the file is allocated. Since the smallest
207 	 * value of osize is 0, length will be at least 1.
208 	 */
209 	if (osize < length) {
210 		if (length > fs->fs_maxfilesize)
211 			return (EFBIG);
212 		aflags = B_CLRBUF;
213 		if (flags & IO_SYNC)
214 			aflags |= B_SYNC;
215 		error = UFS_BUF_ALLOC(oip, length - 1, 1,
216 				   cred, aflags, &bp);
217 		if (error)
218 			return (error);
219 		DIP_ASSIGN(oip, size, length);
220 		uvm_vnp_setsize(ovp, length);
221 		(void) uvm_vnp_uncache(ovp);
222 		if (aflags & B_SYNC)
223 			bwrite(bp);
224 		else
225 			bawrite(bp);
226 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
227 		return (UFS_UPDATE(oip, MNT_WAIT));
228 	}
229 	uvm_vnp_setsize(ovp, length);
230 
231 	/*
232 	 * Shorten the size of the file. If the file is not being
233 	 * truncated to a block boundary, the contents of the
234 	 * partial block following the end of the file must be
235 	 * zero'ed in case it ever becomes accessible again because
236 	 * of subsequent file growth. Directories however are not
237 	 * zero'ed as they should grow back initialized to empty.
238 	 */
239 	offset = blkoff(fs, length);
240 	if (offset == 0) {
241 		DIP_ASSIGN(oip, size, length);
242 	} else {
243 		lbn = lblkno(fs, length);
244 		aflags = B_CLRBUF;
245 		if (flags & IO_SYNC)
246 			aflags |= B_SYNC;
247 		error = UFS_BUF_ALLOC(oip, length - 1, 1,
248 				   cred, aflags, &bp);
249 		if (error)
250 			return (error);
251 		/*
252 		 * When we are doing soft updates and the UFS_BALLOC
253 		 * above fills in a direct block hole with a full sized
254 		 * block that will be truncated down to a fragment below,
255 		 * we must flush out the block dependency with an FSYNC
256 		 * so that we do not get a soft updates inconsistency
257 		 * when we create the fragment below.
258 		 */
259 		if (DOINGSOFTDEP(ovp) && lbn < NDADDR &&
260 		    fragroundup(fs, blkoff(fs, length)) < fs->fs_bsize &&
261 		    (error = VOP_FSYNC(ovp, cred, MNT_WAIT, curproc)) != 0)
262 			return (error);
263 		DIP_ASSIGN(oip, size, length);
264 		size = blksize(fs, oip, lbn);
265 		(void) uvm_vnp_uncache(ovp);
266 		if (ovp->v_type != VDIR)
267 			memset(bp->b_data + offset, 0, size - offset);
268 		bp->b_bcount = size;
269 		if (aflags & B_SYNC)
270 			bwrite(bp);
271 		else
272 			bawrite(bp);
273 	}
274 	/*
275 	 * Calculate index into inode's block list of
276 	 * last direct and indirect blocks (if any)
277 	 * which we want to keep.  Lastblock is -1 when
278 	 * the file is truncated to 0.
279 	 */
280 	lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1;
281 	lastiblock[SINGLE] = lastblock - NDADDR;
282 	lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
283 	lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
284 	nblocks = btodb(fs->fs_bsize);
285 
286 	/*
287 	 * Update file and block pointers on disk before we start freeing
288 	 * blocks.  If we crash before free'ing blocks below, the blocks
289 	 * will be returned to the free list.  lastiblock values are also
290 	 * normalized to -1 for calls to ffs_indirtrunc below.
291 	 */
292 	for (level = TRIPLE; level >= SINGLE; level--) {
293 		oldblks[NDADDR + level] = DIP(oip, ib[level]);
294 		if (lastiblock[level] < 0) {
295 			DIP_ASSIGN(oip, ib[level], 0);
296 			lastiblock[level] = -1;
297 		}
298 	}
299 
300 	for (i = 0; i < NDADDR; i++) {
301 		oldblks[i] = DIP(oip, db[i]);
302 		if (i > lastblock)
303 			DIP_ASSIGN(oip, db[i], 0);
304 	}
305 
306 	oip->i_flag |= IN_CHANGE | IN_UPDATE;
307 	if ((error = UFS_UPDATE(oip, MNT_WAIT)) != 0)
308 		allerror = error;
309 
310 	/*
311 	 * Having written the new inode to disk, save its new configuration
312 	 * and put back the old block pointers long enough to process them.
313 	 * Note that we save the new block configuration so we can check it
314 	 * when we are done.
315 	 */
316 	for (i = 0; i < NDADDR; i++) {
317 		newblks[i] = DIP(oip, db[i]);
318 		DIP_ASSIGN(oip, db[i], oldblks[i]);
319 	}
320 
321 	for (i = 0; i < NIADDR; i++) {
322 		newblks[NDADDR + i] = DIP(oip, ib[i]);
323 		DIP_ASSIGN(oip, ib[i], oldblks[NDADDR + i]);
324 	}
325 
326 	DIP_ASSIGN(oip, size, osize);
327 	vflags = ((length > 0) ? V_SAVE : 0) | V_SAVEMETA;
328 	allerror = vinvalbuf(ovp, vflags, cred, curproc, 0, 0);
329 
330 	/*
331 	 * Indirect blocks first.
332 	 */
333 	indir_lbn[SINGLE] = -NDADDR;
334 	indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
335 	indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
336 	for (level = TRIPLE; level >= SINGLE; level--) {
337 		bn = DIP(oip, ib[level]);
338 		if (bn != 0) {
339 			error = ffs_indirtrunc(oip, indir_lbn[level],
340 			    fsbtodb(fs, bn), lastiblock[level], level, &count);
341 			if (error)
342 				allerror = error;
343 			blocksreleased += count;
344 			if (lastiblock[level] < 0) {
345 				DIP_ASSIGN(oip, ib[level], 0);
346 				ffs_blkfree(oip, bn, fs->fs_bsize);
347 				blocksreleased += nblocks;
348 			}
349 		}
350 		if (lastiblock[level] >= 0)
351 			goto done;
352 	}
353 
354 	/*
355 	 * All whole direct blocks or frags.
356 	 */
357 	for (i = NDADDR - 1; i > lastblock; i--) {
358 		long bsize;
359 
360 		bn = DIP(oip, db[i]);
361 		if (bn == 0)
362 			continue;
363 
364 		DIP_ASSIGN(oip, db[i], 0);
365 		bsize = blksize(fs, oip, i);
366 		ffs_blkfree(oip, bn, bsize);
367 		blocksreleased += btodb(bsize);
368 	}
369 	if (lastblock < 0)
370 		goto done;
371 
372 	/*
373 	 * Finally, look for a change in size of the
374 	 * last direct block; release any frags.
375 	 */
376 	bn = DIP(oip, db[lastblock]);
377 	if (bn != 0) {
378 		long oldspace, newspace;
379 
380 		/*
381 		 * Calculate amount of space we're giving
382 		 * back as old block size minus new block size.
383 		 */
384 		oldspace = blksize(fs, oip, lastblock);
385 		DIP_ASSIGN(oip, size, length);
386 		newspace = blksize(fs, oip, lastblock);
387 		if (newspace == 0)
388 			panic("ffs_truncate: newspace");
389 		if (oldspace - newspace > 0) {
390 			/*
391 			 * Block number of space to be free'd is
392 			 * the old block # plus the number of frags
393 			 * required for the storage we're keeping.
394 			 */
395 			bn += numfrags(fs, newspace);
396 			ffs_blkfree(oip, bn, oldspace - newspace);
397 			blocksreleased += btodb(oldspace - newspace);
398 		}
399 	}
400 done:
401 #ifdef DIAGNOSTIC
402 	for (level = SINGLE; level <= TRIPLE; level++)
403 		if (newblks[NDADDR + level] != DIP(oip, ib[level]))
404 			panic("ffs_truncate1");
405 	for (i = 0; i < NDADDR; i++)
406 		if (newblks[i] != DIP(oip, db[i]))
407 			panic("ffs_truncate2");
408 #endif /* DIAGNOSTIC */
409 	/*
410 	 * Put back the real size.
411 	 */
412 	DIP_ASSIGN(oip, size, length);
413 	if (DIP(oip, blocks) >= blocksreleased)
414 		DIP_ADD(oip, blocks, -blocksreleased);
415 	else	/* sanity */
416 		DIP_ASSIGN(oip, blocks, 0);
417 	oip->i_flag |= IN_CHANGE;
418 	(void)ufs_quota_free_blocks(oip, blocksreleased, NOCRED);
419 	return (allerror);
420 }
421 
422 #ifdef FFS2
423 #define BAP(ip, i) (((ip)->i_ump->um_fstype == UM_UFS2) ? bap2[i] : bap1[i])
424 #define BAP_ASSIGN(ip, i, value)					\
425 	do {								\
426 		if ((ip)->i_ump->um_fstype == UM_UFS2)			\
427 			bap2[i] = (value);				\
428 		else							\
429 			bap1[i] = (value);				\
430 	} while (0)
431 #else
432 #define BAP(ip, i) bap1[i]
433 #define BAP_ASSIGN(ip, i, value) do { bap1[i] = (value); } while (0)
434 #endif /* FFS2 */
435 
436 /*
437  * Release blocks associated with the inode ip and stored in the indirect
438  * block bn.  Blocks are free'd in LIFO order up to (but not including)
439  * lastbn.  If level is greater than SINGLE, the block is an indirect block
440  * and recursive calls to indirtrunc must be used to cleanse other indirect
441  * blocks.
442  *
443  * NB: triple indirect blocks are untested.
444  */
445 int
446 ffs_indirtrunc(struct inode *ip, daddr_t lbn, daddr_t dbn,
447     daddr_t lastbn, int level, long *countp)
448 {
449 	int i;
450 	struct buf *bp;
451 	struct fs *fs = ip->i_fs;
452 	struct vnode *vp;
453 	void *copy = NULL;
454 	daddr_t nb, nlbn, last;
455 	long blkcount, factor;
456 	int nblocks, blocksreleased = 0;
457 	int error = 0, allerror = 0;
458 	int32_t *bap1 = NULL;
459 #ifdef FFS2
460 	int64_t *bap2 = NULL;
461 #endif
462 
463 	/*
464 	 * Calculate index in current block of last
465 	 * block to be kept.  -1 indicates the entire
466 	 * block so we need not calculate the index.
467 	 */
468 	factor = 1;
469 	for (i = SINGLE; i < level; i++)
470 		factor *= NINDIR(fs);
471 	last = lastbn;
472 	if (lastbn > 0)
473 		last /= factor;
474 	nblocks = btodb(fs->fs_bsize);
475 	/*
476 	 * Get buffer of block pointers, zero those entries corresponding
477 	 * to blocks to be free'd, and update on disk copy first.  Since
478 	 * double(triple) indirect before single(double) indirect, calls
479 	 * to bmap on these blocks will fail.  However, we already have
480 	 * the on disk address, so we have to set the b_blkno field
481 	 * explicitly instead of letting bread do everything for us.
482 	 */
483 	vp = ITOV(ip);
484 	bp = getblk(vp, lbn, (int)fs->fs_bsize, 0, 0);
485 	if (!(bp->b_flags & (B_DONE | B_DELWRI))) {
486 		curproc->p_ru.ru_inblock++;		/* pay for read */
487 		bcstats.pendingreads++;
488 		bcstats.numreads++;
489 		bp->b_flags |= B_READ;
490 		if (bp->b_bcount > bp->b_bufsize)
491 			panic("ffs_indirtrunc: bad buffer size");
492 		bp->b_blkno = dbn;
493 		VOP_STRATEGY(bp);
494 		error = biowait(bp);
495 	}
496 	if (error) {
497 		brelse(bp);
498 		*countp = 0;
499 		return (error);
500 	}
501 
502 #ifdef FFS2
503 	if (ip->i_ump->um_fstype == UM_UFS2)
504 		bap2 = (int64_t *)bp->b_data;
505 	else
506 #endif
507 		bap1 = (int32_t *)bp->b_data;
508 
509 	if (lastbn != -1) {
510 		copy = malloc(fs->fs_bsize, M_TEMP, M_WAITOK);
511 		memcpy(copy, bp->b_data, fs->fs_bsize);
512 
513 		for (i = last + 1; i < NINDIR(fs); i++)
514 			BAP_ASSIGN(ip, i, 0);
515 
516 		if (!DOINGASYNC(vp)) {
517 			error = bwrite(bp);
518 			if (error)
519 				allerror = error;
520 		} else {
521 			bawrite(bp);
522 		}
523 
524 #ifdef FFS2
525 		if (ip->i_ump->um_fstype == UM_UFS2)
526 			bap2 = (int64_t *)copy;
527 		else
528 #endif
529 			bap1 = (int32_t *)copy;
530 	}
531 
532 	/*
533 	 * Recursively free totally unused blocks.
534 	 */
535 	for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
536 	    i--, nlbn += factor) {
537 		nb = BAP(ip, i);
538 		if (nb == 0)
539 			continue;
540 		if (level > SINGLE) {
541 			error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
542 					       -1, level - 1, &blkcount);
543 			if (error)
544 				allerror = error;
545 			blocksreleased += blkcount;
546 		}
547 		ffs_blkfree(ip, nb, fs->fs_bsize);
548 		blocksreleased += nblocks;
549 	}
550 
551 	/*
552 	 * Recursively free last partial block.
553 	 */
554 	if (level > SINGLE && lastbn >= 0) {
555 		last = lastbn % factor;
556 		nb = BAP(ip, i);
557 		if (nb != 0) {
558 			error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
559 					       last, level - 1, &blkcount);
560 			if (error)
561 				allerror = error;
562 			blocksreleased += blkcount;
563 		}
564 	}
565 	if (copy != NULL) {
566 		free(copy, M_TEMP);
567 	} else {
568 		bp->b_flags |= B_INVAL;
569 		brelse(bp);
570 	}
571 
572 	*countp = blocksreleased;
573 	return (allerror);
574 }
575