xref: /openbsd/sys/ufs/ffs/ffs_inode.c (revision 3bef86f7)
1 /*	$OpenBSD: ffs_inode.c,v 1.82 2024/01/09 03:15:59 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/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 #ifdef DIAGNOSTIC
157 		if (length != 0)
158 			panic("ffs_truncate: partial truncate of symlink");
159 #endif
160 		memset(SHORTLINK(oip), 0, (size_t) DIP(oip, size));
161 		DIP_ASSIGN(oip, size, 0);
162 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
163 		return (UFS_UPDATE(oip, 1));
164 	}
165 
166 	if ((error = getinoquota(oip)) != 0)
167 		return (error);
168 
169 	fs = oip->i_fs;
170 	if (length > fs->fs_maxfilesize)
171 		return (EFBIG);
172 
173 	uvm_vnp_setsize(ovp, length);
174 	oip->i_ci.ci_lasta = oip->i_ci.ci_clen
175 	    = oip->i_ci.ci_cstart = oip->i_ci.ci_lastw = 0;
176 
177 	if (DOINGSOFTDEP(ovp)) {
178 		if (length > 0 || softdep_slowdown(ovp)) {
179 			/*
180 			 * If a file is only partially truncated, then
181 			 * we have to clean up the data structures
182 			 * describing the allocation past the truncation
183 			 * point. Finding and deallocating those structures
184 			 * is a lot of work. Since partial truncation occurs
185 			 * rarely, we solve the problem by syncing the file
186 			 * so that it will have no data structures left.
187 			 */
188 			if ((error = VOP_FSYNC(ovp, cred, MNT_WAIT,
189 					       curproc)) != 0)
190 				return (error);
191 		} else {
192 			(void)ufs_quota_free_blocks(oip, DIP(oip, blocks),
193 			    NOCRED);
194 			softdep_setup_freeblocks(oip, length);
195 			vinvalbuf(ovp, 0, cred, curproc, 0, INFSLP);
196 			oip->i_flag |= IN_CHANGE | IN_UPDATE;
197 			return (UFS_UPDATE(oip, 0));
198 		}
199 	}
200 
201 	osize = DIP(oip, size);
202 	/*
203 	 * Lengthen the size of the file. We must ensure that the
204 	 * last byte of the file is allocated. Since the smallest
205 	 * value of osize is 0, length will be at least 1.
206 	 */
207 	if (osize < length) {
208 		aflags = B_CLRBUF;
209 		if (flags & IO_SYNC)
210 			aflags |= B_SYNC;
211 		error = UFS_BUF_ALLOC(oip, length - 1, 1,
212 				   cred, aflags, &bp);
213 		if (error)
214 			return (error);
215 		DIP_ASSIGN(oip, size, length);
216 		uvm_vnp_setsize(ovp, length);
217 		(void) uvm_vnp_uncache(ovp);
218 		if (aflags & B_SYNC)
219 			bwrite(bp);
220 		else
221 			bawrite(bp);
222 		oip->i_flag |= IN_CHANGE | IN_UPDATE;
223 		return (UFS_UPDATE(oip, 1));
224 	}
225 	uvm_vnp_setsize(ovp, length);
226 
227 	/*
228 	 * Shorten the size of the file. If the file is not being
229 	 * truncated to a block boundary, the contents of the
230 	 * partial block following the end of the file must be
231 	 * zero'ed in case it ever becomes accessible again because
232 	 * of subsequent file growth. Directories however are not
233 	 * zero'ed as they should grow back initialized to empty.
234 	 */
235 	offset = blkoff(fs, length);
236 	if (offset == 0) {
237 		DIP_ASSIGN(oip, size, length);
238 	} else {
239 		lbn = lblkno(fs, length);
240 		aflags = B_CLRBUF;
241 		if (flags & IO_SYNC)
242 			aflags |= B_SYNC;
243 		error = UFS_BUF_ALLOC(oip, length - 1, 1,
244 				   cred, aflags, &bp);
245 		if (error)
246 			return (error);
247 		/*
248 		 * When we are doing soft updates and the UFS_BALLOC
249 		 * above fills in a direct block hole with a full sized
250 		 * block that will be truncated down to a fragment below,
251 		 * we must flush out the block dependency with an FSYNC
252 		 * so that we do not get a soft updates inconsistency
253 		 * when we create the fragment below.
254 		 */
255 		if (DOINGSOFTDEP(ovp) && lbn < NDADDR &&
256 		    fragroundup(fs, blkoff(fs, length)) < fs->fs_bsize &&
257 		    (error = VOP_FSYNC(ovp, cred, MNT_WAIT, curproc)) != 0)
258 			return (error);
259 		DIP_ASSIGN(oip, size, length);
260 		size = blksize(fs, oip, lbn);
261 		(void) uvm_vnp_uncache(ovp);
262 		if (ovp->v_type != VDIR)
263 			memset(bp->b_data + offset, 0, size - offset);
264 		buf_adjcnt(bp, size);
265 		if (aflags & B_SYNC)
266 			bwrite(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 	/*
283 	 * Update file and block pointers on disk before we start freeing
284 	 * blocks.  If we crash before free'ing blocks below, the blocks
285 	 * will be returned to the free list.  lastiblock values are also
286 	 * normalized to -1 for calls to ffs_indirtrunc below.
287 	 */
288 	for (level = TRIPLE; level >= SINGLE; level--) {
289 		oldblks[NDADDR + level] = DIP(oip, ib[level]);
290 		if (lastiblock[level] < 0) {
291 			DIP_ASSIGN(oip, ib[level], 0);
292 			lastiblock[level] = -1;
293 		}
294 	}
295 
296 	for (i = 0; i < NDADDR; i++) {
297 		oldblks[i] = DIP(oip, db[i]);
298 		if (i > lastblock)
299 			DIP_ASSIGN(oip, db[i], 0);
300 	}
301 
302 	oip->i_flag |= IN_CHANGE | IN_UPDATE;
303 	if ((error = UFS_UPDATE(oip, 1)) != 0)
304 		allerror = error;
305 
306 	/*
307 	 * Having written the new inode to disk, save its new configuration
308 	 * and put back the old block pointers long enough to process them.
309 	 * Note that we save the new block configuration so we can check it
310 	 * when we are done.
311 	 */
312 	for (i = 0; i < NDADDR; i++) {
313 		newblks[i] = DIP(oip, db[i]);
314 		DIP_ASSIGN(oip, db[i], oldblks[i]);
315 	}
316 
317 	for (i = 0; i < NIADDR; i++) {
318 		newblks[NDADDR + i] = DIP(oip, ib[i]);
319 		DIP_ASSIGN(oip, ib[i], oldblks[NDADDR + i]);
320 	}
321 
322 	DIP_ASSIGN(oip, size, osize);
323 	vflags = ((length > 0) ? V_SAVE : 0) | V_SAVEMETA;
324 	allerror = vinvalbuf(ovp, vflags, cred, curproc, 0, INFSLP);
325 
326 	/*
327 	 * Indirect blocks first.
328 	 */
329 	indir_lbn[SINGLE] = -NDADDR;
330 	indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
331 	indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
332 	for (level = TRIPLE; level >= SINGLE; level--) {
333 		bn = DIP(oip, ib[level]);
334 		if (bn != 0) {
335 			error = ffs_indirtrunc(oip, indir_lbn[level],
336 			    fsbtodb(fs, bn), lastiblock[level], level, &count);
337 			if (error)
338 				allerror = error;
339 			blocksreleased += count;
340 			if (lastiblock[level] < 0) {
341 				DIP_ASSIGN(oip, ib[level], 0);
342 				ffs_blkfree(oip, bn, fs->fs_bsize);
343 				blocksreleased += nblocks;
344 			}
345 		}
346 		if (lastiblock[level] >= 0)
347 			goto done;
348 	}
349 
350 	/*
351 	 * All whole direct blocks or frags.
352 	 */
353 	for (i = NDADDR - 1; i > lastblock; i--) {
354 		long bsize;
355 
356 		bn = DIP(oip, db[i]);
357 		if (bn == 0)
358 			continue;
359 
360 		DIP_ASSIGN(oip, db[i], 0);
361 		bsize = blksize(fs, oip, i);
362 		ffs_blkfree(oip, bn, bsize);
363 		blocksreleased += btodb(bsize);
364 	}
365 	if (lastblock < 0)
366 		goto done;
367 
368 	/*
369 	 * Finally, look for a change in size of the
370 	 * last direct block; release any frags.
371 	 */
372 	bn = DIP(oip, db[lastblock]);
373 	if (bn != 0) {
374 		long oldspace, newspace;
375 
376 		/*
377 		 * Calculate amount of space we're giving
378 		 * back as old block size minus new block size.
379 		 */
380 		oldspace = blksize(fs, oip, lastblock);
381 		DIP_ASSIGN(oip, size, length);
382 		newspace = blksize(fs, oip, lastblock);
383 		if (newspace == 0)
384 			panic("ffs_truncate: newspace");
385 		if (oldspace - newspace > 0) {
386 			/*
387 			 * Block number of space to be free'd is
388 			 * the old block # plus the number of frags
389 			 * required for the storage we're keeping.
390 			 */
391 			bn += numfrags(fs, newspace);
392 			ffs_blkfree(oip, bn, oldspace - newspace);
393 			blocksreleased += btodb(oldspace - newspace);
394 		}
395 	}
396 done:
397 #ifdef DIAGNOSTIC
398 	for (level = SINGLE; level <= TRIPLE; level++)
399 		if (newblks[NDADDR + level] != DIP(oip, ib[level]))
400 			panic("ffs_truncate1");
401 	for (i = 0; i < NDADDR; i++)
402 		if (newblks[i] != DIP(oip, db[i]))
403 			panic("ffs_truncate2");
404 #endif /* DIAGNOSTIC */
405 	/*
406 	 * Put back the real size.
407 	 */
408 	DIP_ASSIGN(oip, size, length);
409 	if (DIP(oip, blocks) >= blocksreleased)
410 		DIP_ADD(oip, blocks, -blocksreleased);
411 	else	/* sanity */
412 		DIP_ASSIGN(oip, blocks, 0);
413 	oip->i_flag |= IN_CHANGE;
414 	(void)ufs_quota_free_blocks(oip, blocksreleased, NOCRED);
415 	return (allerror);
416 }
417 
418 #ifdef FFS2
419 #define BAP(ip, i) (((ip)->i_ump->um_fstype == UM_UFS2) ? bap2[i] : bap1[i])
420 #define BAP_ASSIGN(ip, i, value)					\
421 	do {								\
422 		if ((ip)->i_ump->um_fstype == UM_UFS2)			\
423 			bap2[i] = (value);				\
424 		else							\
425 			bap1[i] = (value);				\
426 	} while (0)
427 #else
428 #define BAP(ip, i) bap1[i]
429 #define BAP_ASSIGN(ip, i, value) do { bap1[i] = (value); } while (0)
430 #endif /* FFS2 */
431 
432 /*
433  * Release blocks associated with the inode ip and stored in the indirect
434  * block bn.  Blocks are free'd in LIFO order up to (but not including)
435  * lastbn.  If level is greater than SINGLE, the block is an indirect block
436  * and recursive calls to indirtrunc must be used to cleanse other indirect
437  * blocks.
438  *
439  * NB: triple indirect blocks are untested.
440  */
441 int
442 ffs_indirtrunc(struct inode *ip, daddr_t lbn, daddr_t dbn,
443     daddr_t lastbn, int level, long *countp)
444 {
445 	int i;
446 	struct buf *bp;
447 	struct fs *fs = ip->i_fs;
448 	struct vnode *vp;
449 	void *copy = NULL;
450 	daddr_t nb, nlbn, last;
451 	long blkcount, factor;
452 	int nblocks, blocksreleased = 0;
453 	int error = 0, allerror = 0;
454 	int32_t *bap1 = NULL;
455 #ifdef FFS2
456 	int64_t *bap2 = NULL;
457 #endif
458 
459 	/*
460 	 * Calculate index in current block of last
461 	 * block to be kept.  -1 indicates the entire
462 	 * block so we need not calculate the index.
463 	 */
464 	factor = 1;
465 	for (i = SINGLE; i < level; i++)
466 		factor *= NINDIR(fs);
467 	last = lastbn;
468 	if (lastbn > 0)
469 		last /= factor;
470 	nblocks = btodb(fs->fs_bsize);
471 	/*
472 	 * Get buffer of block pointers, zero those entries corresponding
473 	 * to blocks to be free'd, and update on disk copy first.  Since
474 	 * double(triple) indirect before single(double) indirect, calls
475 	 * to bmap on these blocks will fail.  However, we already have
476 	 * the on disk address, so we have to set the b_blkno field
477 	 * explicitly instead of letting bread do everything for us.
478 	 */
479 	vp = ITOV(ip);
480 	bp = getblk(vp, lbn, (int)fs->fs_bsize, 0, INFSLP);
481 	if (!(bp->b_flags & (B_DONE | B_DELWRI))) {
482 		curproc->p_ru.ru_inblock++;		/* pay for read */
483 		bcstats.pendingreads++;
484 		bcstats.numreads++;
485 		bp->b_flags |= B_READ;
486 		if (bp->b_bcount > bp->b_bufsize)
487 			panic("ffs_indirtrunc: bad buffer size");
488 		bp->b_blkno = dbn;
489 		VOP_STRATEGY(bp->b_vp, bp);
490 		error = biowait(bp);
491 	}
492 	if (error) {
493 		brelse(bp);
494 		*countp = 0;
495 		return (error);
496 	}
497 
498 #ifdef FFS2
499 	if (ip->i_ump->um_fstype == UM_UFS2)
500 		bap2 = (int64_t *)bp->b_data;
501 	else
502 #endif
503 		bap1 = (int32_t *)bp->b_data;
504 
505 	if (lastbn != -1) {
506 		copy = malloc(fs->fs_bsize, M_TEMP, M_WAITOK);
507 		memcpy(copy, bp->b_data, fs->fs_bsize);
508 
509 		for (i = last + 1; i < NINDIR(fs); i++)
510 			BAP_ASSIGN(ip, i, 0);
511 
512 		if (!DOINGASYNC(vp)) {
513 			error = bwrite(bp);
514 			if (error)
515 				allerror = error;
516 		} else {
517 			bawrite(bp);
518 		}
519 
520 #ifdef FFS2
521 		if (ip->i_ump->um_fstype == UM_UFS2)
522 			bap2 = (int64_t *)copy;
523 		else
524 #endif
525 			bap1 = (int32_t *)copy;
526 	}
527 
528 	/*
529 	 * Recursively free totally unused blocks.
530 	 */
531 	for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
532 	    i--, nlbn += factor) {
533 		nb = BAP(ip, i);
534 		if (nb == 0)
535 			continue;
536 		if (level > SINGLE) {
537 			error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
538 					       -1, level - 1, &blkcount);
539 			if (error)
540 				allerror = error;
541 			blocksreleased += blkcount;
542 		}
543 		ffs_blkfree(ip, nb, fs->fs_bsize);
544 		blocksreleased += nblocks;
545 	}
546 
547 	/*
548 	 * Recursively free last partial block.
549 	 */
550 	if (level > SINGLE && lastbn >= 0) {
551 		last = lastbn % factor;
552 		nb = BAP(ip, i);
553 		if (nb != 0) {
554 			error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
555 					       last, level - 1, &blkcount);
556 			if (error)
557 				allerror = error;
558 			blocksreleased += blkcount;
559 		}
560 	}
561 	if (copy != NULL) {
562 		free(copy, M_TEMP, fs->fs_bsize);
563 	} else {
564 		bp->b_flags |= B_INVAL;
565 		brelse(bp);
566 	}
567 
568 	*countp = blocksreleased;
569 	return (allerror);
570 }
571