xref: /csrg-svn/sys/ufs/ffs/ffs_inode.c (revision 56619)
1 /*
2  * Copyright (c) 1982, 1986, 1989 Regents of the University of California.
3  * All rights reserved.
4  *
5  * %sccs.include.redist.c%
6  *
7  *	@(#)ffs_inode.c	7.63 (Berkeley) 10/23/92
8  */
9 
10 #include <sys/param.h>
11 #include <sys/systm.h>
12 #include <sys/mount.h>
13 #include <sys/proc.h>
14 #include <sys/file.h>
15 #include <sys/buf.h>
16 #include <sys/vnode.h>
17 #include <sys/kernel.h>
18 #include <sys/malloc.h>
19 #include <sys/trace.h>
20 #include <sys/resourcevar.h>
21 
22 #include <vm/vm.h>
23 
24 #include <ufs/ufs/quota.h>
25 #include <ufs/ufs/inode.h>
26 #include <ufs/ufs/ufsmount.h>
27 #include <ufs/ufs/ufs_extern.h>
28 
29 #include <ufs/ffs/fs.h>
30 #include <ufs/ffs/ffs_extern.h>
31 
32 static int ffs_indirtrunc __P((struct inode *, daddr_t, daddr_t, daddr_t, int,
33 	    long *));
34 
35 int
36 ffs_init()
37 {
38 	return (ufs_init());
39 }
40 
41 /*
42  * Update the access, modified, and inode change times as specified
43  * by the IACC, IUPD, and ICHG flags respectively. The IMOD flag
44  * is used to specify that the inode needs to be updated but that
45  * the times have already been set. The access and modified times
46  * are taken from the second and third parameters; the inode change
47  * time is always taken from the current time. If waitfor is set,
48  * then wait for the disk write of the inode to complete.
49  */
50 int
51 ffs_update(ap)
52 	struct vop_update_args /* {
53 		struct vnode *a_vp;
54 		struct timeval *a_ta;
55 		struct timeval *a_tm;
56 		int a_waitfor;
57 	} */ *ap;
58 {
59 	struct buf *bp;
60 	struct inode *ip;
61 	struct dinode *dp;
62 	register struct fs *fs;
63 	int error;
64 
65 	if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
66 		return (0);
67 	ip = VTOI(ap->a_vp);
68 	if ((ip->i_flag & (IUPD|IACC|ICHG|IMOD)) == 0)
69 		return (0);
70 	if (ip->i_flag&IACC)
71 		ip->i_atime.ts_sec = ap->a_ta->tv_sec;
72 	if (ip->i_flag&IUPD) {
73 		ip->i_mtime.ts_sec = ap->a_tm->tv_sec;
74 		ip->i_modrev++;
75 	}
76 	if (ip->i_flag&ICHG)
77 		ip->i_ctime.ts_sec = time.tv_sec;
78 	ip->i_flag &= ~(IUPD|IACC|ICHG|IMOD);
79 	fs = ip->i_fs;
80 	/*
81 	 * Ensure that uid and gid are correct. This is a temporary
82 	 * fix until fsck has been changed to do the update.
83 	 */
84 	if (fs->fs_inodefmt < FS_44INODEFMT) {		/* XXX */
85 		ip->i_din.di_ouid = ip->i_uid;		/* XXX */
86 		ip->i_din.di_ogid = ip->i_gid;		/* XXX */
87 	}						/* XXX */
88 	if (error = bread(ip->i_devvp, fsbtodb(fs, itod(fs, ip->i_number)),
89 		(int)fs->fs_bsize, NOCRED, &bp)) {
90 		brelse(bp);
91 		return (error);
92 	}
93 	dp = bp->b_un.b_dino + itoo(fs, ip->i_number);
94 	*dp = ip->i_din;
95 	if (ap->a_waitfor)
96 		return (bwrite(bp));
97 	else {
98 		bdwrite(bp);
99 		return (0);
100 	}
101 }
102 
103 #define	SINGLE	0	/* index of single indirect block */
104 #define	DOUBLE	1	/* index of double indirect block */
105 #define	TRIPLE	2	/* index of triple indirect block */
106 /*
107  * Truncate the inode ip to at most length size.  Free affected disk
108  * blocks -- the blocks of the file are removed in reverse order.
109  */
110 ffs_truncate(ap)
111 	struct vop_truncate_args /* {
112 		struct vnode *a_vp;
113 		off_t a_length;
114 		int a_flags;
115 		struct ucred *a_cred;
116 		struct proc *a_p;
117 	} */ *ap;
118 {
119 	register struct vnode *ovp = ap->a_vp;
120 	register daddr_t lastblock;
121 	register struct inode *oip;
122 	daddr_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR];
123 	off_t length = ap->a_length;
124 	register struct fs *fs;
125 	register struct inode *ip;
126 	struct buf *bp;
127 	int offset, size, level;
128 	long count, nblocks, vflags, blocksreleased = 0;
129 	struct timeval tv;
130 	register int i;
131 	int aflags, error, allerror;
132 	off_t osize;
133 
134 	oip = VTOI(ovp);
135 	tv = time;
136 	if (ovp->v_type == VLNK && ovp->v_mount->mnt_maxsymlinklen > 0) {
137 #ifdef DIAGNOSTIC
138 		if (length != 0)
139 			panic("ffs_truncate: partial truncate of symlink");
140 #endif
141 		bzero((char *)&oip->i_shortlink, (u_int)oip->i_size);
142 		oip->i_size = 0;
143 		oip->i_flag |= ICHG|IUPD;
144 		return (VOP_UPDATE(ovp, &tv, &tv, 1));
145 	}
146 	if (oip->i_size <= length) {
147 		oip->i_flag |= ICHG|IUPD;
148 		return (VOP_UPDATE(ovp, &tv, &tv, 1));
149 	}
150 	vnode_pager_setsize(ovp, (u_long)length);
151 	/*
152 	 * Calculate index into inode's block list of
153 	 * last direct and indirect blocks (if any)
154 	 * which we want to keep.  Lastblock is -1 when
155 	 * the file is truncated to 0.
156 	 */
157 	fs = oip->i_fs;
158 	lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1;
159 	lastiblock[SINGLE] = lastblock - NDADDR;
160 	lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
161 	lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
162 	nblocks = btodb(fs->fs_bsize);
163 	/*
164 	 * Update the size of the file. If the file is not being
165 	 * truncated to a block boundry, the contents of the
166 	 * partial block following the end of the file must be
167 	 * zero'ed in case it ever become accessable again because
168 	 * of subsequent file growth.
169 	 */
170 	osize = oip->i_size;
171 	offset = blkoff(fs, length);
172 	if (offset == 0) {
173 		oip->i_size = length;
174 	} else {
175 		lbn = lblkno(fs, length);
176 		aflags = B_CLRBUF;
177 		if (ap->a_flags & IO_SYNC)
178 			aflags |= B_SYNC;
179 #ifdef QUOTA
180 		if (error = getinoquota(oip))
181 			return (error);
182 #endif
183 		if (error = ffs_balloc(oip, lbn, offset, ap->a_cred, &bp, aflags))
184 			return (error);
185 		oip->i_size = length;
186 		size = blksize(fs, oip, lbn);
187 		(void) vnode_pager_uncache(ovp);
188 		bzero(bp->b_un.b_addr + offset, (unsigned)(size - offset));
189 		allocbuf(bp, size);
190 		if (ap->a_flags & IO_SYNC)
191 			bwrite(bp);
192 		else
193 			bdwrite(bp);
194 	}
195 	/*
196 	 * Update file and block pointers on disk before we start freeing
197 	 * blocks.  If we crash before free'ing blocks below, the blocks
198 	 * will be returned to the free list.  lastiblock values are also
199 	 * normalized to -1 for calls to ffs_indirtrunc below.
200 	 */
201 	MALLOC(ip, struct inode *, sizeof(*ip), M_FFSNODE, M_WAITOK);
202 	*ip = *oip;
203 	ip->i_size = osize;
204 	for (level = TRIPLE; level >= SINGLE; level--)
205 		if (lastiblock[level] < 0) {
206 			oip->i_ib[level] = 0;
207 			lastiblock[level] = -1;
208 		}
209 	for (i = NDADDR - 1; i > lastblock; i--)
210 		oip->i_db[i] = 0;
211 	oip->i_flag |= ICHG|IUPD;
212 	vflags = ((length > 0) ? V_SAVE : 0) | V_SAVEMETA;
213 	allerror = vinvalbuf(ovp, vflags, ap->a_cred, ap->a_p);
214 	if (error = VOP_UPDATE(ovp, &tv, &tv, MNT_WAIT))
215 		allerror = error;
216 
217 	/*
218 	 * Indirect blocks first.
219 	 */
220 	indir_lbn[SINGLE] = -NDADDR;
221 	indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
222 	indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
223 	ITOV(ip)->v_data = ip;
224 	for (level = TRIPLE; level >= SINGLE; level--) {
225 		bn = ip->i_ib[level];
226 		if (bn != 0) {
227 			error = ffs_indirtrunc(ip, indir_lbn[level],
228 			    fsbtodb(fs, bn), lastiblock[level], level, &count);
229 			if (error)
230 				allerror = error;
231 			blocksreleased += count;
232 			if (lastiblock[level] < 0) {
233 				ip->i_ib[level] = 0;
234 				ffs_blkfree(ip, bn, fs->fs_bsize);
235 				blocksreleased += nblocks;
236 			}
237 		}
238 		if (lastiblock[level] >= 0)
239 			goto done;
240 	}
241 
242 	/*
243 	 * All whole direct blocks or frags.
244 	 */
245 	for (i = NDADDR - 1; i > lastblock; i--) {
246 		register long bsize;
247 
248 		bn = ip->i_db[i];
249 		if (bn == 0)
250 			continue;
251 		ip->i_db[i] = 0;
252 		bsize = blksize(fs, ip, i);
253 		ffs_blkfree(ip, bn, bsize);
254 		blocksreleased += btodb(bsize);
255 	}
256 	if (lastblock < 0)
257 		goto done;
258 
259 	/*
260 	 * Finally, look for a change in size of the
261 	 * last direct block; release any frags.
262 	 */
263 	bn = ip->i_db[lastblock];
264 	if (bn != 0) {
265 		long oldspace, newspace;
266 
267 		/*
268 		 * Calculate amount of space we're giving
269 		 * back as old block size minus new block size.
270 		 */
271 		oldspace = blksize(fs, ip, lastblock);
272 		ip->i_size = length;
273 		newspace = blksize(fs, ip, lastblock);
274 		if (newspace == 0)
275 			panic("itrunc: newspace");
276 		if (oldspace - newspace > 0) {
277 			/*
278 			 * Block number of space to be free'd is
279 			 * the old block # plus the number of frags
280 			 * required for the storage we're keeping.
281 			 */
282 			bn += numfrags(fs, newspace);
283 			ffs_blkfree(ip, bn, oldspace - newspace);
284 			blocksreleased += btodb(oldspace - newspace);
285 		}
286 	}
287 done:
288 #ifdef DIAGNOSTIC
289 	for (level = SINGLE; level <= TRIPLE; level++)
290 		if (ip->i_ib[level] != oip->i_ib[level])
291 			panic("itrunc1");
292 	for (i = 0; i < NDADDR; i++)
293 		if (ip->i_db[i] != oip->i_db[i])
294 			panic("itrunc2");
295 	if (length == 0 &&
296 	    (ovp->v_dirtyblkhd.le_next || ovp->v_cleanblkhd.le_next))
297 		panic("itrunc3");
298 #endif /* DIAGNOSTIC */
299 	ITOV(ip)->v_data = oip;
300 	FREE(ip, M_FFSNODE);
301 	oip->i_blocks -= blocksreleased;
302 	if (oip->i_blocks < 0)			/* sanity */
303 		oip->i_blocks = 0;
304 	oip->i_flag |= ICHG;
305 #ifdef QUOTA
306 	if (!getinoquota(oip))
307 		(void) chkdq(oip, -blocksreleased, NOCRED, 0);
308 #endif
309 	return (allerror);
310 }
311 
312 /*
313  * Release blocks associated with the inode ip and stored in the indirect
314  * block bn.  Blocks are free'd in LIFO order up to (but not including)
315  * lastbn.  If level is greater than SINGLE, the block is an indirect block
316  * and recursive calls to indirtrunc must be used to cleanse other indirect
317  * blocks.
318  *
319  * NB: triple indirect blocks are untested.
320  */
321 static int
322 ffs_indirtrunc(ip, lbn, dbn, lastbn, level, countp)
323 	register struct inode *ip;
324 	daddr_t lbn, lastbn;
325 	daddr_t dbn;
326 	int level;
327 	long *countp;
328 {
329 	register int i;
330 	struct buf *bp;
331 	register struct fs *fs = ip->i_fs;
332 	register daddr_t *bap;
333 	struct vnode *vp;
334 	daddr_t *copy, nb, nlbn, last;
335 	long blkcount, factor;
336 	int nblocks, blocksreleased = 0;
337 	int error = 0, allerror = 0;
338 
339 	/*
340 	 * Calculate index in current block of last
341 	 * block to be kept.  -1 indicates the entire
342 	 * block so we need not calculate the index.
343 	 */
344 	factor = 1;
345 	for (i = SINGLE; i < level; i++)
346 		factor *= NINDIR(fs);
347 	last = lastbn;
348 	if (lastbn > 0)
349 		last /= factor;
350 	nblocks = btodb(fs->fs_bsize);
351 	/*
352 	 * Get buffer of block pointers, zero those entries corresponding
353 	 * to blocks to be free'd, and update on disk copy first.  Since
354 	 * double(triple) indirect before single(double) indirect, calls
355 	 * to bmap on these blocks will fail.  However, we already have
356 	 * the on disk address, so we have to set the b_blkno field
357 	 * explicitly instead of letting bread do everything for us.
358 	 */
359 	vp = ITOV(ip);
360 	bp = getblk(vp, lbn, (int)fs->fs_bsize);
361 	if (bp->b_flags & (B_DONE | B_DELWRI)) {
362 		/* Braces must be here in case trace evaluates to nothing. */
363 		trace(TR_BREADHIT, pack(vp, fs->fs_bsize), lbn);
364 	} else {
365 		trace(TR_BREADMISS, pack(vp, fs->fs_bsize), lbn);
366 		curproc->p_stats->p_ru.ru_inblock++;	/* pay for read */
367 		bp->b_flags |= B_READ;
368 		if (bp->b_bcount > bp->b_bufsize)
369 			panic("ffs_indirtrunc: bad buffer size");
370 		bp->b_blkno = dbn;
371 		VOP_STRATEGY(bp);
372 		error = biowait(bp);
373 	}
374 	if (error) {
375 		brelse(bp);
376 		*countp = 0;
377 		return (error);
378 	}
379 
380 	bap = bp->b_un.b_daddr;
381 	MALLOC(copy, daddr_t *, fs->fs_bsize, M_TEMP, M_WAITOK);
382 	bcopy((caddr_t)bap, (caddr_t)copy, (u_int)fs->fs_bsize);
383 	bzero((caddr_t)&bap[last + 1],
384 	  (u_int)(NINDIR(fs) - (last + 1)) * sizeof (daddr_t));
385 	if (last == -1)
386 		bp->b_flags |= B_INVAL;
387 	error = bwrite(bp);
388 	if (error)
389 		allerror = error;
390 	bap = copy;
391 
392 	/*
393 	 * Recursively free totally unused blocks.
394 	 */
395 	for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
396 	    i--, nlbn += factor) {
397 		nb = bap[i];
398 		if (nb == 0)
399 			continue;
400 		if (level > SINGLE) {
401 			if (error = ffs_indirtrunc(ip, nlbn,
402 			    fsbtodb(fs, nb), (daddr_t)-1, level - 1, &blkcount))
403 				allerror = error;
404 			blocksreleased += blkcount;
405 		}
406 		ffs_blkfree(ip, nb, fs->fs_bsize);
407 		blocksreleased += nblocks;
408 	}
409 
410 	/*
411 	 * Recursively free last partial block.
412 	 */
413 	if (level > SINGLE && lastbn >= 0) {
414 		last = lastbn % factor;
415 		nb = bap[i];
416 		if (nb != 0) {
417 			if (error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
418 			    last, level - 1, &blkcount))
419 				allerror = error;
420 			blocksreleased += blkcount;
421 		}
422 	}
423 	FREE(copy, M_TEMP);
424 	*countp = blocksreleased;
425 	return (allerror);
426 }
427