xref: /csrg-svn/sys/ufs/ffs/ffs_inode.c (revision 53234)
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	7.47 (Berkeley) 04/21/92
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 
20 #include <ufs/ufs/quota.h>
21 #include <ufs/ufs/inode.h>
22 #include <ufs/ufs/ufsmount.h>
23 #include <ufs/ufs/ufs_extern.h>
24 
25 #include <ufs/ffs/fs.h>
26 #include <ufs/ffs/ffs_extern.h>
27 
28 static int ffs_indirtrunc __P((struct inode *, daddr_t, daddr_t, int, long *));
29 
30 extern u_long nextgennumber;
31 
32 int
33 ffs_init()
34 {
35 	return (ufs_init());
36 }
37 
38 /*
39  * Look up a UFS dinode number to find its incore vnode.
40  * If it is not in core, read it in from the specified device.
41  * If it is in core, wait for the lock bit to clear, then
42  * return the inode locked. Detection and handling of mount
43  * points must be done by the calling routine.
44  */
45 ffs_vget(mntp, ino, vpp)
46 	struct mount *mntp;
47 	ino_t ino;
48 	struct vnode **vpp;
49 {
50 	register struct fs *fs;
51 	register struct inode *ip;
52 	struct ufsmount *ump;
53 	struct buf *bp;
54 	struct dinode *dp;
55 	struct vnode *vp;
56 	union ihead *ih;
57 	dev_t dev;
58 	int i, type, error;
59 
60 	ump = VFSTOUFS(mntp);
61 	dev = ump->um_dev;
62 	if ((*vpp = ufs_ihashget(dev, ino)) != NULL)
63 		return (0);
64 
65 	/* Allocate a new vnode/inode. */
66 	if (error = getnewvnode(VT_UFS, mntp, &ffs_vnodeops, &vp)) {
67 		*vpp = NULL;
68 		return (error);
69 	}
70 	type = ump->um_devvp->v_tag == VT_MFS ? M_MFSNODE : M_FFSNODE; /* XXX */
71 	MALLOC(ip, struct inode *, sizeof(struct inode), type, M_WAITOK);
72 	vp->v_data = ip;
73 	ip->i_vnode = vp;
74 	ip->i_flag = 0;
75 	ip->i_devvp = 0;
76 	ip->i_mode = 0;
77 	ip->i_diroff = 0;
78 	ip->i_lockf = 0;
79 	ip->i_fs = fs = ump->um_fs;
80 	ip->i_dev = dev;
81 	ip->i_number = ino;
82 #ifdef QUOTA
83 	for (i = 0; i < MAXQUOTAS; i++)
84 		ip->i_dquot[i] = NODQUOT;
85 #endif
86 	/*
87 	 * Put it onto its hash chain and lock it so that other requests for
88 	 * this inode will block if they arrive while we are sleeping waiting
89 	 * for old data structures to be purged or for the contents of the
90 	 * disk portion of this inode to be read.
91 	 */
92 	ufs_ihashins(ip);
93 
94 	/* Read in the disk contents for the inode, copy into the inode. */
95 	if (error = bread(ump->um_devvp, fsbtodb(fs, itod(fs, ino)),
96 	    (int)fs->fs_bsize, NOCRED, &bp)) {
97 		/*
98 		 * The inode does not contain anything useful, so it would
99 		 * be misleading to leave it on its hash chain. It will be
100 		 * returned to the free list by ufs_iput().
101 		 */
102 		remque(ip);
103 		ip->i_forw = ip;
104 		ip->i_back = ip;
105 
106 		/* Unlock and discard unneeded inode. */
107 		ufs_iput(ip);
108 		brelse(bp);
109 		*vpp = NULL;
110 		return (error);
111 	}
112 	dp = bp->b_un.b_dino;
113 	dp += itoo(fs, ino);
114 	ip->i_din = *dp;
115 	brelse(bp);
116 
117 	/*
118 	 * Initialize the vnode from the inode, check for aliases.
119 	 * Note that the underlying vnode may have changed.
120 	 */
121 	if (error = ufs_vinit(mntp, &ffs_specops, FFS_FIFOOPS, &vp)) {
122 		ufs_iput(ip);
123 		*vpp = NULL;
124 		return (error);
125 	}
126 	/*
127 	 * Finish inode initialization now that aliasing has been resolved.
128 	 */
129 	ip->i_devvp = ump->um_devvp;
130 	VREF(ip->i_devvp);
131 	/*
132 	 * Set up a generation number for this inode if it does not
133 	 * already have one. This should only happen on old filesystems.
134 	 */
135 	if (ip->i_gen == 0) {
136 		if (++nextgennumber < (u_long)time.tv_sec)
137 			nextgennumber = time.tv_sec;
138 		ip->i_gen = nextgennumber;
139 		if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0)
140 			ip->i_flag |= IMOD;
141 	}
142 	*vpp = vp;
143 	return (0);
144 }
145 
146 /*
147  * Update the access, modified, and inode change times as specified
148  * by the IACC, IUPD, and ICHG flags respectively. The IMOD flag
149  * is used to specify that the inode needs to be updated but that
150  * the times have already been set. The access and modified times
151  * are taken from the second and third parameters; the inode change
152  * time is always taken from the current time. If waitfor is set,
153  * then wait for the disk write of the inode to complete.
154  */
155 int
156 ffs_update(vp, ta, tm, waitfor)
157 	register struct vnode *vp;
158 	struct timeval *ta, *tm;
159 	int waitfor;
160 {
161 	struct buf *bp;
162 	struct inode *ip;
163 	struct dinode *dp;
164 	register struct fs *fs;
165 	int error;
166 
167 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
168 		return (0);
169 	ip = VTOI(vp);
170 	if ((ip->i_flag & (IUPD|IACC|ICHG|IMOD)) == 0)
171 		return (0);
172 	if (ip->i_flag&IACC)
173 		ip->i_atime = ta->tv_sec;
174 	if (ip->i_flag&IUPD) {
175 		ip->i_mtime = tm->tv_sec;
176 		INCRQUAD(ip->i_modrev);
177 	}
178 	if (ip->i_flag&ICHG)
179 		ip->i_ctime = time.tv_sec;
180 	ip->i_flag &= ~(IUPD|IACC|ICHG|IMOD);
181 
182 	fs = ip->i_fs;
183 	if (error = bread(ip->i_devvp, fsbtodb(fs, itod(fs, ip->i_number)),
184 		(int)fs->fs_bsize, NOCRED, &bp)) {
185 		brelse(bp);
186 		return (error);
187 	}
188 	dp = bp->b_un.b_dino + itoo(fs, ip->i_number);
189 	*dp = ip->i_din;
190 	if (waitfor)
191 		return (bwrite(bp));
192 	else {
193 		bdwrite(bp);
194 		return (0);
195 	}
196 }
197 
198 #define	SINGLE	0	/* index of single indirect block */
199 #define	DOUBLE	1	/* index of double indirect block */
200 #define	TRIPLE	2	/* index of triple indirect block */
201 /*
202  * Truncate the inode ip to at most length size.  Free affected disk
203  * blocks -- the blocks of the file are removed in reverse order.
204  *
205  * NB: triple indirect blocks are untested.
206  */
207 ffs_truncate(ovp, length, flags)
208 	register struct vnode *ovp;
209 	off_t length;
210 	int flags;
211 {
212 	register daddr_t lastblock;
213 	register struct inode *oip;
214 	daddr_t bn, lbn, lastiblock[NIADDR];
215 	register struct fs *fs;
216 	register struct inode *ip;
217 	struct buf *bp;
218 	int offset, size, level;
219 	long count, nblocks, blocksreleased = 0;
220 	register int i;
221 	int aflags, error, allerror;
222 	struct inode tip;
223 	off_t osize;
224 
225 	vnode_pager_setsize(ovp, (u_long)length);
226 	oip = VTOI(ovp);
227 	if (oip->i_size <= length) {
228 		oip->i_flag |= ICHG|IUPD;
229 		error = ffs_update(ovp, &time, &time, 1);
230 		return (error);
231 	}
232 	/*
233 	 * Calculate index into inode's block list of
234 	 * last direct and indirect blocks (if any)
235 	 * which we want to keep.  Lastblock is -1 when
236 	 * the file is truncated to 0.
237 	 */
238 	fs = oip->i_fs;
239 	lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1;
240 	lastiblock[SINGLE] = lastblock - NDADDR;
241 	lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
242 	lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
243 	nblocks = btodb(fs->fs_bsize);
244 	/*
245 	 * Update the size of the file. If the file is not being
246 	 * truncated to a block boundry, the contents of the
247 	 * partial block following the end of the file must be
248 	 * zero'ed in case it ever become accessable again because
249 	 * of subsequent file growth.
250 	 */
251 	osize = oip->i_size;
252 	offset = blkoff(fs, length);
253 	if (offset == 0) {
254 		oip->i_size = length;
255 	} else {
256 		lbn = lblkno(fs, length);
257 		aflags = B_CLRBUF;
258 		if (flags & IO_SYNC)
259 			aflags |= B_SYNC;
260 #ifdef QUOTA
261 		if (error = getinoquota(oip))
262 			return (error);
263 #endif
264 		if (error = ffs_balloc(oip, lbn, offset, &bp, aflags))
265 			return (error);
266 		oip->i_size = length;
267 		size = blksize(fs, oip, lbn);
268 		(void) vnode_pager_uncache(ovp);
269 		bzero(bp->b_un.b_addr + offset, (unsigned)(size - offset));
270 		allocbuf(bp, size);
271 		if (flags & IO_SYNC)
272 			bwrite(bp);
273 		else
274 			bdwrite(bp);
275 	}
276 	/*
277 	 * Update file and block pointers on disk before we start freeing
278 	 * blocks.  If we crash before free'ing blocks below, the blocks
279 	 * will be returned to the free list.  lastiblock values are also
280 	 * normalized to -1 for calls to ffs_indirtrunc below.
281 	 */
282 	tip = *oip;
283 	tip.i_size = osize;
284 	for (level = TRIPLE; level >= SINGLE; level--)
285 		if (lastiblock[level] < 0) {
286 			oip->i_ib[level] = 0;
287 			lastiblock[level] = -1;
288 		}
289 	for (i = NDADDR - 1; i > lastblock; i--)
290 		oip->i_db[i] = 0;
291 	oip->i_flag |= ICHG|IUPD;
292 	vinvalbuf(ovp, (length > 0));
293 	allerror = ffs_update(ovp, &time, &time, MNT_WAIT);
294 
295 	/*
296 	 * Indirect blocks first.
297 	 */
298 	ip = &tip;
299 	for (level = TRIPLE; level >= SINGLE; level--) {
300 		bn = ip->i_ib[level];
301 		if (bn != 0) {
302 			error = ffs_indirtrunc(ip,
303 			    bn, lastiblock[level], level, &count);
304 			if (error)
305 				allerror = error;
306 			blocksreleased += count;
307 			if (lastiblock[level] < 0) {
308 				ip->i_ib[level] = 0;
309 				ffs_blkfree(ip, bn, fs->fs_bsize);
310 				blocksreleased += nblocks;
311 			}
312 		}
313 		if (lastiblock[level] >= 0)
314 			goto done;
315 	}
316 
317 	/*
318 	 * All whole direct blocks or frags.
319 	 */
320 	for (i = NDADDR - 1; i > lastblock; i--) {
321 		register long bsize;
322 
323 		bn = ip->i_db[i];
324 		if (bn == 0)
325 			continue;
326 		ip->i_db[i] = 0;
327 		bsize = blksize(fs, ip, i);
328 		ffs_blkfree(ip, bn, bsize);
329 		blocksreleased += btodb(bsize);
330 	}
331 	if (lastblock < 0)
332 		goto done;
333 
334 	/*
335 	 * Finally, look for a change in size of the
336 	 * last direct block; release any frags.
337 	 */
338 	bn = ip->i_db[lastblock];
339 	if (bn != 0) {
340 		long oldspace, newspace;
341 
342 		/*
343 		 * Calculate amount of space we're giving
344 		 * back as old block size minus new block size.
345 		 */
346 		oldspace = blksize(fs, ip, lastblock);
347 		ip->i_size = length;
348 		newspace = blksize(fs, ip, lastblock);
349 		if (newspace == 0)
350 			panic("itrunc: newspace");
351 		if (oldspace - newspace > 0) {
352 			/*
353 			 * Block number of space to be free'd is
354 			 * the old block # plus the number of frags
355 			 * required for the storage we're keeping.
356 			 */
357 			bn += numfrags(fs, newspace);
358 			ffs_blkfree(ip, bn, oldspace - newspace);
359 			blocksreleased += btodb(oldspace - newspace);
360 		}
361 	}
362 done:
363 /* BEGIN PARANOIA */
364 	for (level = SINGLE; level <= TRIPLE; level++)
365 		if (ip->i_ib[level] != oip->i_ib[level])
366 			panic("itrunc1");
367 	for (i = 0; i < NDADDR; i++)
368 		if (ip->i_db[i] != oip->i_db[i])
369 			panic("itrunc2");
370 /* END PARANOIA */
371 	oip->i_blocks -= blocksreleased;
372 	if (oip->i_blocks < 0)			/* sanity */
373 		oip->i_blocks = 0;
374 	oip->i_flag |= ICHG;
375 #ifdef QUOTA
376 	if (!getinoquota(oip))
377 		(void) chkdq(oip, -blocksreleased, NOCRED, 0);
378 #endif
379 	return (allerror);
380 }
381 
382 /*
383  * Release blocks associated with the inode ip and stored in the indirect
384  * block bn.  Blocks are free'd in LIFO order up to (but not including)
385  * lastbn.  If level is greater than SINGLE, the block is an indirect block
386  * and recursive calls to indirtrunc must be used to cleanse other indirect
387  * blocks.
388  *
389  * NB: triple indirect blocks are untested.
390  */
391 static int
392 ffs_indirtrunc(ip, bn, lastbn, level, countp)
393 	register struct inode *ip;
394 	daddr_t bn, lastbn;
395 	int level;
396 	long *countp;
397 {
398 	register int i;
399 	struct buf *bp;
400 	register struct fs *fs = ip->i_fs;
401 	register daddr_t *bap;
402 	daddr_t *copy, nb, last;
403 	long blkcount, factor;
404 	int nblocks, blocksreleased = 0;
405 	int error, allerror = 0;
406 
407 	/*
408 	 * Calculate index in current block of last
409 	 * block to be kept.  -1 indicates the entire
410 	 * block so we need not calculate the index.
411 	 */
412 	factor = 1;
413 	for (i = SINGLE; i < level; i++)
414 		factor *= NINDIR(fs);
415 	last = lastbn;
416 	if (lastbn > 0)
417 		last /= factor;
418 	nblocks = btodb(fs->fs_bsize);
419 	/*
420 	 * Get buffer of block pointers, zero those
421 	 * entries corresponding to blocks to be free'd,
422 	 * and update on disk copy first.
423 	 */
424 	error = bread(ip->i_devvp, fsbtodb(fs, bn), (int)fs->fs_bsize,
425 		NOCRED, &bp);
426 	if (error) {
427 		brelse(bp);
428 		*countp = 0;
429 		return (error);
430 	}
431 	bap = bp->b_un.b_daddr;
432 	MALLOC(copy, daddr_t *, fs->fs_bsize, M_TEMP, M_WAITOK);
433 	bcopy((caddr_t)bap, (caddr_t)copy, (u_int)fs->fs_bsize);
434 	bzero((caddr_t)&bap[last + 1],
435 	  (u_int)(NINDIR(fs) - (last + 1)) * sizeof (daddr_t));
436 	if (last == -1)
437 		bp->b_flags |= B_INVAL;
438 	error = bwrite(bp);
439 	if (error)
440 		allerror = error;
441 	bap = copy;
442 
443 	/*
444 	 * Recursively free totally unused blocks.
445 	 */
446 	for (i = NINDIR(fs) - 1; i > last; i--) {
447 		nb = bap[i];
448 		if (nb == 0)
449 			continue;
450 		if (level > SINGLE) {
451 			if (error = ffs_indirtrunc(ip,
452 			    nb, (daddr_t)-1, level - 1, &blkcount))
453 				allerror = error;
454 			blocksreleased += blkcount;
455 		}
456 		ffs_blkfree(ip, nb, fs->fs_bsize);
457 		blocksreleased += nblocks;
458 	}
459 
460 	/*
461 	 * Recursively free last partial block.
462 	 */
463 	if (level > SINGLE && lastbn >= 0) {
464 		last = lastbn % factor;
465 		nb = bap[i];
466 		if (nb != 0) {
467 			if (error =
468 			    ffs_indirtrunc(ip, nb, last, level - 1, &blkcount))
469 				allerror = error;
470 			blocksreleased += blkcount;
471 		}
472 	}
473 	FREE(copy, M_TEMP);
474 	*countp = blocksreleased;
475 	return (allerror);
476 }
477