xref: /openbsd/sys/ufs/ffs/ffs_inode.c (revision 09467b48)
1 /*	$OpenBSD: ffs_inode.c,v 1.80 2019/07/25 01:43:21 cheloha 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 	fs = oip->i_fs;
172 	if (length > fs->fs_maxfilesize)
173 		return (EFBIG);
174 
175 	uvm_vnp_setsize(ovp, length);
176 	oip->i_ci.ci_lasta = oip->i_ci.ci_clen
177 	    = oip->i_ci.ci_cstart = oip->i_ci.ci_lastw = 0;
178 
179 	if (DOINGSOFTDEP(ovp)) {
180 		if (length > 0 || softdep_slowdown(ovp)) {
181 			/*
182 			 * If a file is only partially truncated, then
183 			 * we have to clean up the data structures
184 			 * describing the allocation past the truncation
185 			 * point. Finding and deallocating those structures
186 			 * is a lot of work. Since partial truncation occurs
187 			 * rarely, we solve the problem by syncing the file
188 			 * so that it will have no data structures left.
189 			 */
190 			if ((error = VOP_FSYNC(ovp, cred, MNT_WAIT,
191 					       curproc)) != 0)
192 				return (error);
193 		} else {
194 			(void)ufs_quota_free_blocks(oip, DIP(oip, blocks),
195 			    NOCRED);
196 			softdep_setup_freeblocks(oip, length);
197 			vinvalbuf(ovp, 0, cred, curproc, 0, INFSLP);
198 			oip->i_flag |= IN_CHANGE | IN_UPDATE;
199 			return (UFS_UPDATE(oip, 0));
200 		}
201 	}
202 
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 		aflags = B_CLRBUF;
211 		if (flags & IO_SYNC)
212 			aflags |= B_SYNC;
213 		error = UFS_BUF_ALLOC(oip, length - 1, 1,
214 				   cred, aflags, &bp);
215 		if (error)
216 			return (error);
217 		DIP_ASSIGN(oip, size, length);
218 		uvm_vnp_setsize(ovp, length);
219 		(void) uvm_vnp_uncache(ovp);
220 		if (aflags & B_SYNC)
221 			bwrite(bp);
222 		else
223 			bawrite(bp);
224 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
225 		return (UFS_UPDATE(oip, 1));
226 	}
227 	uvm_vnp_setsize(ovp, length);
228 
229 	/*
230 	 * Shorten the size of the file. If the file is not being
231 	 * truncated to a block boundary, the contents of the
232 	 * partial block following the end of the file must be
233 	 * zero'ed in case it ever becomes accessible again because
234 	 * of subsequent file growth. Directories however are not
235 	 * zero'ed as they should grow back initialized to empty.
236 	 */
237 	offset = blkoff(fs, length);
238 	if (offset == 0) {
239 		DIP_ASSIGN(oip, size, length);
240 	} else {
241 		lbn = lblkno(fs, length);
242 		aflags = B_CLRBUF;
243 		if (flags & IO_SYNC)
244 			aflags |= B_SYNC;
245 		error = UFS_BUF_ALLOC(oip, length - 1, 1,
246 				   cred, aflags, &bp);
247 		if (error)
248 			return (error);
249 		/*
250 		 * When we are doing soft updates and the UFS_BALLOC
251 		 * above fills in a direct block hole with a full sized
252 		 * block that will be truncated down to a fragment below,
253 		 * we must flush out the block dependency with an FSYNC
254 		 * so that we do not get a soft updates inconsistency
255 		 * when we create the fragment below.
256 		 */
257 		if (DOINGSOFTDEP(ovp) && lbn < NDADDR &&
258 		    fragroundup(fs, blkoff(fs, length)) < fs->fs_bsize &&
259 		    (error = VOP_FSYNC(ovp, cred, MNT_WAIT, curproc)) != 0)
260 			return (error);
261 		DIP_ASSIGN(oip, size, length);
262 		size = blksize(fs, oip, lbn);
263 		(void) uvm_vnp_uncache(ovp);
264 		if (ovp->v_type != VDIR)
265 			memset(bp->b_data + offset, 0, size - offset);
266 		buf_adjcnt(bp, size);
267 		if (aflags & B_SYNC)
268 			bwrite(bp);
269 		else
270 			bawrite(bp);
271 	}
272 	/*
273 	 * Calculate index into inode's block list of
274 	 * last direct and indirect blocks (if any)
275 	 * which we want to keep.  Lastblock is -1 when
276 	 * the file is truncated to 0.
277 	 */
278 	lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1;
279 	lastiblock[SINGLE] = lastblock - NDADDR;
280 	lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
281 	lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
282 	nblocks = btodb(fs->fs_bsize);
283 
284 	/*
285 	 * Update file and block pointers on disk before we start freeing
286 	 * blocks.  If we crash before free'ing blocks below, the blocks
287 	 * will be returned to the free list.  lastiblock values are also
288 	 * normalized to -1 for calls to ffs_indirtrunc below.
289 	 */
290 	for (level = TRIPLE; level >= SINGLE; level--) {
291 		oldblks[NDADDR + level] = DIP(oip, ib[level]);
292 		if (lastiblock[level] < 0) {
293 			DIP_ASSIGN(oip, ib[level], 0);
294 			lastiblock[level] = -1;
295 		}
296 	}
297 
298 	for (i = 0; i < NDADDR; i++) {
299 		oldblks[i] = DIP(oip, db[i]);
300 		if (i > lastblock)
301 			DIP_ASSIGN(oip, db[i], 0);
302 	}
303 
304 	oip->i_flag |= IN_CHANGE | IN_UPDATE;
305 	if ((error = UFS_UPDATE(oip, 1)) != 0)
306 		allerror = error;
307 
308 	/*
309 	 * Having written the new inode to disk, save its new configuration
310 	 * and put back the old block pointers long enough to process them.
311 	 * Note that we save the new block configuration so we can check it
312 	 * when we are done.
313 	 */
314 	for (i = 0; i < NDADDR; i++) {
315 		newblks[i] = DIP(oip, db[i]);
316 		DIP_ASSIGN(oip, db[i], oldblks[i]);
317 	}
318 
319 	for (i = 0; i < NIADDR; i++) {
320 		newblks[NDADDR + i] = DIP(oip, ib[i]);
321 		DIP_ASSIGN(oip, ib[i], oldblks[NDADDR + i]);
322 	}
323 
324 	DIP_ASSIGN(oip, size, osize);
325 	vflags = ((length > 0) ? V_SAVE : 0) | V_SAVEMETA;
326 	allerror = vinvalbuf(ovp, vflags, cred, curproc, 0, INFSLP);
327 
328 	/*
329 	 * Indirect blocks first.
330 	 */
331 	indir_lbn[SINGLE] = -NDADDR;
332 	indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
333 	indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
334 	for (level = TRIPLE; level >= SINGLE; level--) {
335 		bn = DIP(oip, ib[level]);
336 		if (bn != 0) {
337 			error = ffs_indirtrunc(oip, indir_lbn[level],
338 			    fsbtodb(fs, bn), lastiblock[level], level, &count);
339 			if (error)
340 				allerror = error;
341 			blocksreleased += count;
342 			if (lastiblock[level] < 0) {
343 				DIP_ASSIGN(oip, ib[level], 0);
344 				ffs_blkfree(oip, bn, fs->fs_bsize);
345 				blocksreleased += nblocks;
346 			}
347 		}
348 		if (lastiblock[level] >= 0)
349 			goto done;
350 	}
351 
352 	/*
353 	 * All whole direct blocks or frags.
354 	 */
355 	for (i = NDADDR - 1; i > lastblock; i--) {
356 		long bsize;
357 
358 		bn = DIP(oip, db[i]);
359 		if (bn == 0)
360 			continue;
361 
362 		DIP_ASSIGN(oip, db[i], 0);
363 		bsize = blksize(fs, oip, i);
364 		ffs_blkfree(oip, bn, bsize);
365 		blocksreleased += btodb(bsize);
366 	}
367 	if (lastblock < 0)
368 		goto done;
369 
370 	/*
371 	 * Finally, look for a change in size of the
372 	 * last direct block; release any frags.
373 	 */
374 	bn = DIP(oip, db[lastblock]);
375 	if (bn != 0) {
376 		long oldspace, newspace;
377 
378 		/*
379 		 * Calculate amount of space we're giving
380 		 * back as old block size minus new block size.
381 		 */
382 		oldspace = blksize(fs, oip, lastblock);
383 		DIP_ASSIGN(oip, size, length);
384 		newspace = blksize(fs, oip, lastblock);
385 		if (newspace == 0)
386 			panic("ffs_truncate: newspace");
387 		if (oldspace - newspace > 0) {
388 			/*
389 			 * Block number of space to be free'd is
390 			 * the old block # plus the number of frags
391 			 * required for the storage we're keeping.
392 			 */
393 			bn += numfrags(fs, newspace);
394 			ffs_blkfree(oip, bn, oldspace - newspace);
395 			blocksreleased += btodb(oldspace - newspace);
396 		}
397 	}
398 done:
399 #ifdef DIAGNOSTIC
400 	for (level = SINGLE; level <= TRIPLE; level++)
401 		if (newblks[NDADDR + level] != DIP(oip, ib[level]))
402 			panic("ffs_truncate1");
403 	for (i = 0; i < NDADDR; i++)
404 		if (newblks[i] != DIP(oip, db[i]))
405 			panic("ffs_truncate2");
406 #endif /* DIAGNOSTIC */
407 	/*
408 	 * Put back the real size.
409 	 */
410 	DIP_ASSIGN(oip, size, length);
411 	if (DIP(oip, blocks) >= blocksreleased)
412 		DIP_ADD(oip, blocks, -blocksreleased);
413 	else	/* sanity */
414 		DIP_ASSIGN(oip, blocks, 0);
415 	oip->i_flag |= IN_CHANGE;
416 	(void)ufs_quota_free_blocks(oip, blocksreleased, NOCRED);
417 	return (allerror);
418 }
419 
420 #ifdef FFS2
421 #define BAP(ip, i) (((ip)->i_ump->um_fstype == UM_UFS2) ? bap2[i] : bap1[i])
422 #define BAP_ASSIGN(ip, i, value)					\
423 	do {								\
424 		if ((ip)->i_ump->um_fstype == UM_UFS2)			\
425 			bap2[i] = (value);				\
426 		else							\
427 			bap1[i] = (value);				\
428 	} while (0)
429 #else
430 #define BAP(ip, i) bap1[i]
431 #define BAP_ASSIGN(ip, i, value) do { bap1[i] = (value); } while (0)
432 #endif /* FFS2 */
433 
434 /*
435  * Release blocks associated with the inode ip and stored in the indirect
436  * block bn.  Blocks are free'd in LIFO order up to (but not including)
437  * lastbn.  If level is greater than SINGLE, the block is an indirect block
438  * and recursive calls to indirtrunc must be used to cleanse other indirect
439  * blocks.
440  *
441  * NB: triple indirect blocks are untested.
442  */
443 int
444 ffs_indirtrunc(struct inode *ip, daddr_t lbn, daddr_t dbn,
445     daddr_t lastbn, int level, long *countp)
446 {
447 	int i;
448 	struct buf *bp;
449 	struct fs *fs = ip->i_fs;
450 	struct vnode *vp;
451 	void *copy = NULL;
452 	daddr_t nb, nlbn, last;
453 	long blkcount, factor;
454 	int nblocks, blocksreleased = 0;
455 	int error = 0, allerror = 0;
456 	int32_t *bap1 = NULL;
457 #ifdef FFS2
458 	int64_t *bap2 = NULL;
459 #endif
460 
461 	/*
462 	 * Calculate index in current block of last
463 	 * block to be kept.  -1 indicates the entire
464 	 * block so we need not calculate the index.
465 	 */
466 	factor = 1;
467 	for (i = SINGLE; i < level; i++)
468 		factor *= NINDIR(fs);
469 	last = lastbn;
470 	if (lastbn > 0)
471 		last /= factor;
472 	nblocks = btodb(fs->fs_bsize);
473 	/*
474 	 * Get buffer of block pointers, zero those entries corresponding
475 	 * to blocks to be free'd, and update on disk copy first.  Since
476 	 * double(triple) indirect before single(double) indirect, calls
477 	 * to bmap on these blocks will fail.  However, we already have
478 	 * the on disk address, so we have to set the b_blkno field
479 	 * explicitly instead of letting bread do everything for us.
480 	 */
481 	vp = ITOV(ip);
482 	bp = getblk(vp, lbn, (int)fs->fs_bsize, 0, INFSLP);
483 	if (!(bp->b_flags & (B_DONE | B_DELWRI))) {
484 		curproc->p_ru.ru_inblock++;		/* pay for read */
485 		bcstats.pendingreads++;
486 		bcstats.numreads++;
487 		bp->b_flags |= B_READ;
488 		if (bp->b_bcount > bp->b_bufsize)
489 			panic("ffs_indirtrunc: bad buffer size");
490 		bp->b_blkno = dbn;
491 		VOP_STRATEGY(bp);
492 		error = biowait(bp);
493 	}
494 	if (error) {
495 		brelse(bp);
496 		*countp = 0;
497 		return (error);
498 	}
499 
500 #ifdef FFS2
501 	if (ip->i_ump->um_fstype == UM_UFS2)
502 		bap2 = (int64_t *)bp->b_data;
503 	else
504 #endif
505 		bap1 = (int32_t *)bp->b_data;
506 
507 	if (lastbn != -1) {
508 		copy = malloc(fs->fs_bsize, M_TEMP, M_WAITOK);
509 		memcpy(copy, bp->b_data, fs->fs_bsize);
510 
511 		for (i = last + 1; i < NINDIR(fs); i++)
512 			BAP_ASSIGN(ip, i, 0);
513 
514 		if (!DOINGASYNC(vp)) {
515 			error = bwrite(bp);
516 			if (error)
517 				allerror = error;
518 		} else {
519 			bawrite(bp);
520 		}
521 
522 #ifdef FFS2
523 		if (ip->i_ump->um_fstype == UM_UFS2)
524 			bap2 = (int64_t *)copy;
525 		else
526 #endif
527 			bap1 = (int32_t *)copy;
528 	}
529 
530 	/*
531 	 * Recursively free totally unused blocks.
532 	 */
533 	for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
534 	    i--, nlbn += factor) {
535 		nb = BAP(ip, i);
536 		if (nb == 0)
537 			continue;
538 		if (level > SINGLE) {
539 			error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
540 					       -1, level - 1, &blkcount);
541 			if (error)
542 				allerror = error;
543 			blocksreleased += blkcount;
544 		}
545 		ffs_blkfree(ip, nb, fs->fs_bsize);
546 		blocksreleased += nblocks;
547 	}
548 
549 	/*
550 	 * Recursively free last partial block.
551 	 */
552 	if (level > SINGLE && lastbn >= 0) {
553 		last = lastbn % factor;
554 		nb = BAP(ip, i);
555 		if (nb != 0) {
556 			error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
557 					       last, level - 1, &blkcount);
558 			if (error)
559 				allerror = error;
560 			blocksreleased += blkcount;
561 		}
562 	}
563 	if (copy != NULL) {
564 		free(copy, M_TEMP, fs->fs_bsize);
565 	} else {
566 		bp->b_flags |= B_INVAL;
567 		brelse(bp);
568 	}
569 
570 	*countp = blocksreleased;
571 	return (allerror);
572 }
573