xref: /openbsd-src/sys/ufs/ffs/ffs_inode.c (revision f2da64fbbbf1b03f09f390ab01267c93dfd77c4c)
1 /*	$OpenBSD: ffs_inode.c,v 1.76 2016/02/27 18:50:38 natano 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/buf.h>
40 #include <sys/vnode.h>
41 #include <sys/kernel.h>
42 #include <sys/malloc.h>
43 #include <sys/resourcevar.h>
44 
45 #include <ufs/ufs/quota.h>
46 #include <ufs/ufs/inode.h>
47 #include <ufs/ufs/ufsmount.h>
48 #include <ufs/ufs/ufs_extern.h>
49 
50 #include <ufs/ffs/fs.h>
51 #include <ufs/ffs/ffs_extern.h>
52 
53 int ffs_indirtrunc(struct inode *, daddr_t, daddr_t, daddr_t, int, long *);
54 
55 /*
56  * Update the access, modified, and inode change times as specified by the
57  * IN_ACCESS, IN_UPDATE, and IN_CHANGE flags respectively. The IN_MODIFIED
58  * flag is used to specify that the inode needs to be updated but that the
59  * times have already been set.  The IN_LAZYMOD flag is used to specify
60  * that the inode needs to be updated at some point, by reclaim if not
61  * in the course of other changes; this is used to defer writes just to
62  * update device timestamps.  If waitfor is set, then wait for the disk
63  * write of the inode to complete.
64  */
65 int
66 ffs_update(struct inode *ip, int waitfor)
67 {
68 	struct vnode *vp;
69 	struct fs *fs;
70 	struct buf *bp;
71 	int error;
72 
73 	vp = ITOV(ip);
74 	ufs_itimes(vp);
75 
76 	if ((ip->i_flag & IN_MODIFIED) == 0 && waitfor == 0)
77 		return (0);
78 
79 	ip->i_flag &= ~(IN_MODIFIED | IN_LAZYMOD);
80 	fs = ip->i_fs;
81 
82 	/*
83 	 * Ensure that uid and gid are correct. This is a temporary
84 	 * fix until fsck has been changed to do the update.
85 	 */
86 	if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_inodefmt < FS_44INODEFMT) {
87 		ip->i_din1->di_ouid = ip->i_ffs1_uid;
88 		ip->i_din1->di_ogid = ip->i_ffs1_gid;
89 	}
90 
91 	error = bread(ip->i_devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
92 	    (int)fs->fs_bsize, &bp);
93 	if (error) {
94 		brelse(bp);
95 		return (error);
96 	}
97 
98 	if (DOINGSOFTDEP(vp))
99 		softdep_update_inodeblock(ip, bp, waitfor);
100 	else if (ip->i_effnlink != DIP(ip, nlink))
101 		panic("ffs_update: bad link cnt");
102 
103 #ifdef FFS2
104 	if (ip->i_ump->um_fstype == UM_UFS2)
105 		*((struct ufs2_dinode *)bp->b_data +
106 		    ino_to_fsbo(fs, ip->i_number)) = *ip->i_din2;
107 	else
108 #endif
109 		*((struct ufs1_dinode *)bp->b_data +
110 		    ino_to_fsbo(fs, ip->i_number)) = *ip->i_din1;
111 
112 	if (waitfor && !DOINGASYNC(vp)) {
113 		return (bwrite(bp));
114 	} else {
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 /*
125  * Truncate the inode oip to at most length size, freeing the
126  * disk blocks.
127  */
128 int
129 ffs_truncate(struct inode *oip, off_t length, int flags, struct ucred *cred)
130 {
131 	struct vnode *ovp;
132 	daddr_t lastblock;
133 	daddr_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR];
134 	daddr_t oldblks[NDADDR + NIADDR], newblks[NDADDR + NIADDR];
135 	struct fs *fs;
136 	struct buf *bp;
137 	int offset, size, level;
138 	long count, nblocks, vflags, blocksreleased = 0;
139 	int i, aflags, error, allerror;
140 	off_t osize;
141 
142 	if (length < 0)
143 		return (EINVAL);
144 	ovp = ITOV(oip);
145 
146 	if (ovp->v_type != VREG &&
147 	    ovp->v_type != VDIR &&
148 	    ovp->v_type != VLNK)
149 		return (0);
150 
151 	if (DIP(oip, size) == length)
152 		return (0);
153 
154 	if (ovp->v_type == VLNK &&
155 	    (DIP(oip, size) < oip->i_ump->um_maxsymlinklen ||
156 	     (oip->i_ump->um_maxsymlinklen == 0 &&
157 	      oip->i_din1->di_blocks == 0))) {
158 #ifdef DIAGNOSTIC
159 		if (length != 0)
160 			panic("ffs_truncate: partial truncate of symlink");
161 #endif
162 		memset(SHORTLINK(oip), 0, (size_t) DIP(oip, size));
163 		DIP_ASSIGN(oip, size, 0);
164 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
165 		return (UFS_UPDATE(oip, 1));
166 	}
167 
168 	if ((error = getinoquota(oip)) != 0)
169 		return (error);
170 
171 	uvm_vnp_setsize(ovp, length);
172 	oip->i_ci.ci_lasta = oip->i_ci.ci_clen
173 	    = oip->i_ci.ci_cstart = oip->i_ci.ci_lastw = 0;
174 
175 	if (DOINGSOFTDEP(ovp)) {
176 		if (length > 0 || softdep_slowdown(ovp)) {
177 			/*
178 			 * If a file is only partially truncated, then
179 			 * we have to clean up the data structures
180 			 * describing the allocation past the truncation
181 			 * point. Finding and deallocating those structures
182 			 * is a lot of work. Since partial truncation occurs
183 			 * rarely, we solve the problem by syncing the file
184 			 * so that it will have no data structures left.
185 			 */
186 			if ((error = VOP_FSYNC(ovp, cred, MNT_WAIT,
187 					       curproc)) != 0)
188 				return (error);
189 		} else {
190 			(void)ufs_quota_free_blocks(oip, DIP(oip, blocks),
191 			    NOCRED);
192 			softdep_setup_freeblocks(oip, length);
193 			(void) vinvalbuf(ovp, 0, cred, curproc, 0, 0);
194 			oip->i_flag |= IN_CHANGE | IN_UPDATE;
195 			return (UFS_UPDATE(oip, 0));
196 		}
197 	}
198 
199 	fs = oip->i_fs;
200 	osize = DIP(oip, size);
201 	/*
202 	 * Lengthen the size of the file. We must ensure that the
203 	 * last byte of the file is allocated. Since the smallest
204 	 * value of osize is 0, length will be at least 1.
205 	 */
206 	if (osize < length) {
207 		if (length > fs->fs_maxfilesize)
208 			return (EFBIG);
209 		aflags = B_CLRBUF;
210 		if (flags & IO_SYNC)
211 			aflags |= B_SYNC;
212 		error = UFS_BUF_ALLOC(oip, length - 1, 1,
213 				   cred, aflags, &bp);
214 		if (error)
215 			return (error);
216 		DIP_ASSIGN(oip, size, length);
217 		uvm_vnp_setsize(ovp, length);
218 		(void) uvm_vnp_uncache(ovp);
219 		if (aflags & B_SYNC)
220 			bwrite(bp);
221 		else
222 			bawrite(bp);
223 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
224 		return (UFS_UPDATE(oip, 1));
225 	}
226 	uvm_vnp_setsize(ovp, length);
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 		DIP_ASSIGN(oip, size, length);
239 	} else {
240 		lbn = lblkno(fs, length);
241 		aflags = B_CLRBUF;
242 		if (flags & IO_SYNC)
243 			aflags |= B_SYNC;
244 		error = UFS_BUF_ALLOC(oip, length - 1, 1,
245 				   cred, aflags, &bp);
246 		if (error)
247 			return (error);
248 		/*
249 		 * When we are doing soft updates and the UFS_BALLOC
250 		 * above fills in a direct block hole with a full sized
251 		 * block that will be truncated down to a fragment below,
252 		 * we must flush out the block dependency with an FSYNC
253 		 * so that we do not get a soft updates inconsistency
254 		 * when we create the fragment below.
255 		 */
256 		if (DOINGSOFTDEP(ovp) && lbn < NDADDR &&
257 		    fragroundup(fs, blkoff(fs, length)) < fs->fs_bsize &&
258 		    (error = VOP_FSYNC(ovp, cred, MNT_WAIT, curproc)) != 0)
259 			return (error);
260 		DIP_ASSIGN(oip, size, length);
261 		size = blksize(fs, oip, lbn);
262 		(void) uvm_vnp_uncache(ovp);
263 		if (ovp->v_type != VDIR)
264 			memset(bp->b_data + offset, 0, size - offset);
265 		buf_adjcnt(bp, size);
266 		if (aflags & B_SYNC)
267 			bwrite(bp);
268 		else
269 			bawrite(bp);
270 	}
271 	/*
272 	 * Calculate index into inode's block list of
273 	 * last direct and indirect blocks (if any)
274 	 * which we want to keep.  Lastblock is -1 when
275 	 * the file is truncated to 0.
276 	 */
277 	lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1;
278 	lastiblock[SINGLE] = lastblock - NDADDR;
279 	lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
280 	lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
281 	nblocks = btodb(fs->fs_bsize);
282 
283 	/*
284 	 * Update file and block pointers on disk before we start freeing
285 	 * blocks.  If we crash before free'ing blocks below, the blocks
286 	 * will be returned to the free list.  lastiblock values are also
287 	 * normalized to -1 for calls to ffs_indirtrunc below.
288 	 */
289 	for (level = TRIPLE; level >= SINGLE; level--) {
290 		oldblks[NDADDR + level] = DIP(oip, ib[level]);
291 		if (lastiblock[level] < 0) {
292 			DIP_ASSIGN(oip, ib[level], 0);
293 			lastiblock[level] = -1;
294 		}
295 	}
296 
297 	for (i = 0; i < NDADDR; i++) {
298 		oldblks[i] = DIP(oip, db[i]);
299 		if (i > lastblock)
300 			DIP_ASSIGN(oip, db[i], 0);
301 	}
302 
303 	oip->i_flag |= IN_CHANGE | IN_UPDATE;
304 	if ((error = UFS_UPDATE(oip, 1)) != 0)
305 		allerror = error;
306 
307 	/*
308 	 * Having written the new inode to disk, save its new configuration
309 	 * and put back the old block pointers long enough to process them.
310 	 * Note that we save the new block configuration so we can check it
311 	 * when we are done.
312 	 */
313 	for (i = 0; i < NDADDR; i++) {
314 		newblks[i] = DIP(oip, db[i]);
315 		DIP_ASSIGN(oip, db[i], oldblks[i]);
316 	}
317 
318 	for (i = 0; i < NIADDR; i++) {
319 		newblks[NDADDR + i] = DIP(oip, ib[i]);
320 		DIP_ASSIGN(oip, ib[i], oldblks[NDADDR + i]);
321 	}
322 
323 	DIP_ASSIGN(oip, size, osize);
324 	vflags = ((length > 0) ? V_SAVE : 0) | V_SAVEMETA;
325 	allerror = vinvalbuf(ovp, vflags, cred, curproc, 0, 0);
326 
327 	/*
328 	 * Indirect blocks first.
329 	 */
330 	indir_lbn[SINGLE] = -NDADDR;
331 	indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
332 	indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
333 	for (level = TRIPLE; level >= SINGLE; level--) {
334 		bn = DIP(oip, ib[level]);
335 		if (bn != 0) {
336 			error = ffs_indirtrunc(oip, indir_lbn[level],
337 			    fsbtodb(fs, bn), lastiblock[level], level, &count);
338 			if (error)
339 				allerror = error;
340 			blocksreleased += count;
341 			if (lastiblock[level] < 0) {
342 				DIP_ASSIGN(oip, ib[level], 0);
343 				ffs_blkfree(oip, bn, fs->fs_bsize);
344 				blocksreleased += nblocks;
345 			}
346 		}
347 		if (lastiblock[level] >= 0)
348 			goto done;
349 	}
350 
351 	/*
352 	 * All whole direct blocks or frags.
353 	 */
354 	for (i = NDADDR - 1; i > lastblock; i--) {
355 		long bsize;
356 
357 		bn = DIP(oip, db[i]);
358 		if (bn == 0)
359 			continue;
360 
361 		DIP_ASSIGN(oip, db[i], 0);
362 		bsize = blksize(fs, oip, i);
363 		ffs_blkfree(oip, bn, bsize);
364 		blocksreleased += btodb(bsize);
365 	}
366 	if (lastblock < 0)
367 		goto done;
368 
369 	/*
370 	 * Finally, look for a change in size of the
371 	 * last direct block; release any frags.
372 	 */
373 	bn = DIP(oip, db[lastblock]);
374 	if (bn != 0) {
375 		long oldspace, newspace;
376 
377 		/*
378 		 * Calculate amount of space we're giving
379 		 * back as old block size minus new block size.
380 		 */
381 		oldspace = blksize(fs, oip, lastblock);
382 		DIP_ASSIGN(oip, size, length);
383 		newspace = blksize(fs, oip, lastblock);
384 		if (newspace == 0)
385 			panic("ffs_truncate: newspace");
386 		if (oldspace - newspace > 0) {
387 			/*
388 			 * Block number of space to be free'd is
389 			 * the old block # plus the number of frags
390 			 * required for the storage we're keeping.
391 			 */
392 			bn += numfrags(fs, newspace);
393 			ffs_blkfree(oip, bn, oldspace - newspace);
394 			blocksreleased += btodb(oldspace - newspace);
395 		}
396 	}
397 done:
398 #ifdef DIAGNOSTIC
399 	for (level = SINGLE; level <= TRIPLE; level++)
400 		if (newblks[NDADDR + level] != DIP(oip, ib[level]))
401 			panic("ffs_truncate1");
402 	for (i = 0; i < NDADDR; i++)
403 		if (newblks[i] != DIP(oip, db[i]))
404 			panic("ffs_truncate2");
405 #endif /* DIAGNOSTIC */
406 	/*
407 	 * Put back the real size.
408 	 */
409 	DIP_ASSIGN(oip, size, length);
410 	if (DIP(oip, blocks) >= blocksreleased)
411 		DIP_ADD(oip, blocks, -blocksreleased);
412 	else	/* sanity */
413 		DIP_ASSIGN(oip, blocks, 0);
414 	oip->i_flag |= IN_CHANGE;
415 	(void)ufs_quota_free_blocks(oip, blocksreleased, NOCRED);
416 	return (allerror);
417 }
418 
419 #ifdef FFS2
420 #define BAP(ip, i) (((ip)->i_ump->um_fstype == UM_UFS2) ? bap2[i] : bap1[i])
421 #define BAP_ASSIGN(ip, i, value)					\
422 	do {								\
423 		if ((ip)->i_ump->um_fstype == UM_UFS2)			\
424 			bap2[i] = (value);				\
425 		else							\
426 			bap1[i] = (value);				\
427 	} while (0)
428 #else
429 #define BAP(ip, i) bap1[i]
430 #define BAP_ASSIGN(ip, i, value) do { bap1[i] = (value); } while (0)
431 #endif /* FFS2 */
432 
433 /*
434  * Release blocks associated with the inode ip and stored in the indirect
435  * block bn.  Blocks are free'd in LIFO order up to (but not including)
436  * lastbn.  If level is greater than SINGLE, the block is an indirect block
437  * and recursive calls to indirtrunc must be used to cleanse other indirect
438  * blocks.
439  *
440  * NB: triple indirect blocks are untested.
441  */
442 int
443 ffs_indirtrunc(struct inode *ip, daddr_t lbn, daddr_t dbn,
444     daddr_t lastbn, int level, long *countp)
445 {
446 	int i;
447 	struct buf *bp;
448 	struct fs *fs = ip->i_fs;
449 	struct vnode *vp;
450 	void *copy = NULL;
451 	daddr_t nb, nlbn, last;
452 	long blkcount, factor;
453 	int nblocks, blocksreleased = 0;
454 	int error = 0, allerror = 0;
455 	int32_t *bap1 = NULL;
456 #ifdef FFS2
457 	int64_t *bap2 = NULL;
458 #endif
459 
460 	/*
461 	 * Calculate index in current block of last
462 	 * block to be kept.  -1 indicates the entire
463 	 * block so we need not calculate the index.
464 	 */
465 	factor = 1;
466 	for (i = SINGLE; i < level; i++)
467 		factor *= NINDIR(fs);
468 	last = lastbn;
469 	if (lastbn > 0)
470 		last /= factor;
471 	nblocks = btodb(fs->fs_bsize);
472 	/*
473 	 * Get buffer of block pointers, zero those entries corresponding
474 	 * to blocks to be free'd, and update on disk copy first.  Since
475 	 * double(triple) indirect before single(double) indirect, calls
476 	 * to bmap on these blocks will fail.  However, we already have
477 	 * the on disk address, so we have to set the b_blkno field
478 	 * explicitly instead of letting bread do everything for us.
479 	 */
480 	vp = ITOV(ip);
481 	bp = getblk(vp, lbn, (int)fs->fs_bsize, 0, 0);
482 	if (!(bp->b_flags & (B_DONE | B_DELWRI))) {
483 		curproc->p_ru.ru_inblock++;		/* pay for read */
484 		bcstats.pendingreads++;
485 		bcstats.numreads++;
486 		bp->b_flags |= B_READ;
487 		if (bp->b_bcount > bp->b_bufsize)
488 			panic("ffs_indirtrunc: bad buffer size");
489 		bp->b_blkno = dbn;
490 		VOP_STRATEGY(bp);
491 		error = biowait(bp);
492 	}
493 	if (error) {
494 		brelse(bp);
495 		*countp = 0;
496 		return (error);
497 	}
498 
499 #ifdef FFS2
500 	if (ip->i_ump->um_fstype == UM_UFS2)
501 		bap2 = (int64_t *)bp->b_data;
502 	else
503 #endif
504 		bap1 = (int32_t *)bp->b_data;
505 
506 	if (lastbn != -1) {
507 		copy = malloc(fs->fs_bsize, M_TEMP, M_WAITOK);
508 		memcpy(copy, bp->b_data, fs->fs_bsize);
509 
510 		for (i = last + 1; i < NINDIR(fs); i++)
511 			BAP_ASSIGN(ip, i, 0);
512 
513 		if (!DOINGASYNC(vp)) {
514 			error = bwrite(bp);
515 			if (error)
516 				allerror = error;
517 		} else {
518 			bawrite(bp);
519 		}
520 
521 #ifdef FFS2
522 		if (ip->i_ump->um_fstype == UM_UFS2)
523 			bap2 = (int64_t *)copy;
524 		else
525 #endif
526 			bap1 = (int32_t *)copy;
527 	}
528 
529 	/*
530 	 * Recursively free totally unused blocks.
531 	 */
532 	for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
533 	    i--, nlbn += factor) {
534 		nb = BAP(ip, i);
535 		if (nb == 0)
536 			continue;
537 		if (level > SINGLE) {
538 			error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
539 					       -1, level - 1, &blkcount);
540 			if (error)
541 				allerror = error;
542 			blocksreleased += blkcount;
543 		}
544 		ffs_blkfree(ip, nb, fs->fs_bsize);
545 		blocksreleased += nblocks;
546 	}
547 
548 	/*
549 	 * Recursively free last partial block.
550 	 */
551 	if (level > SINGLE && lastbn >= 0) {
552 		last = lastbn % factor;
553 		nb = BAP(ip, i);
554 		if (nb != 0) {
555 			error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
556 					       last, level - 1, &blkcount);
557 			if (error)
558 				allerror = error;
559 			blocksreleased += blkcount;
560 		}
561 	}
562 	if (copy != NULL) {
563 		free(copy, M_TEMP, 0);
564 	} else {
565 		bp->b_flags |= B_INVAL;
566 		brelse(bp);
567 	}
568 
569 	*countp = blocksreleased;
570 	return (allerror);
571 }
572