xref: /csrg-svn/sys/ufs/ffs/ffs_inode.c (revision 53471)
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	7.49 (Berkeley) 05/13/92
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 <vm/vm.h>
21 
22 #include <ufs/ufs/quota.h>
23 #include <ufs/ufs/inode.h>
24 #include <ufs/ufs/ufsmount.h>
25 #include <ufs/ufs/ufs_extern.h>
26 
27 #include <ufs/ffs/fs.h>
28 #include <ufs/ffs/ffs_extern.h>
29 
30 static int ffs_indirtrunc __P((struct inode *, daddr_t, daddr_t, int, long *));
31 
32 extern u_long nextgennumber;
33 
34 int
35 ffs_init()
36 {
37 	return (ufs_init());
38 }
39 
40 /*
41  * Look up a UFS dinode number to find its incore vnode.
42  * If it is not in core, read it in from the specified device.
43  * If it is in core, wait for the lock bit to clear, then
44  * return the inode locked. Detection and handling of mount
45  * points must be done by the calling routine.
46  */
47 ffs_vget(mntp, ino, vpp)
48 	struct mount *mntp;
49 	ino_t ino;
50 	struct vnode **vpp;
51 {
52 	register struct fs *fs;
53 	register struct inode *ip;
54 	struct ufsmount *ump;
55 	struct buf *bp;
56 	struct dinode *dp;
57 	struct vnode *vp;
58 	union ihead *ih;
59 	dev_t dev;
60 	int i, type, error;
61 
62 	ump = VFSTOUFS(mntp);
63 	dev = ump->um_dev;
64 	if ((*vpp = ufs_ihashget(dev, ino)) != NULL)
65 		return (0);
66 
67 	/* Allocate a new vnode/inode. */
68 	if (error = getnewvnode(VT_UFS, mntp, &ffs_vnodeops, &vp)) {
69 		*vpp = NULL;
70 		return (error);
71 	}
72 	type = ump->um_devvp->v_tag == VT_MFS ? M_MFSNODE : M_FFSNODE; /* XXX */
73 	MALLOC(ip, struct inode *, sizeof(struct inode), type, M_WAITOK);
74 	vp->v_data = ip;
75 	ip->i_vnode = vp;
76 	ip->i_flag = 0;
77 	ip->i_devvp = 0;
78 	ip->i_mode = 0;
79 	ip->i_diroff = 0;
80 	ip->i_lockf = 0;
81 	ip->i_fs = fs = ump->um_fs;
82 	ip->i_dev = dev;
83 	ip->i_number = ino;
84 #ifdef QUOTA
85 	for (i = 0; i < MAXQUOTAS; i++)
86 		ip->i_dquot[i] = NODQUOT;
87 #endif
88 	/*
89 	 * Put it onto its hash chain and lock it so that other requests for
90 	 * this inode will block if they arrive while we are sleeping waiting
91 	 * for old data structures to be purged or for the contents of the
92 	 * disk portion of this inode to be read.
93 	 */
94 	ufs_ihashins(ip);
95 
96 	/* Read in the disk contents for the inode, copy into the inode. */
97 	if (error = bread(ump->um_devvp, fsbtodb(fs, itod(fs, ino)),
98 	    (int)fs->fs_bsize, NOCRED, &bp)) {
99 		/*
100 		 * The inode does not contain anything useful, so it would
101 		 * be misleading to leave it on its hash chain. It will be
102 		 * returned to the free list by ufs_iput().
103 		 */
104 		remque(ip);
105 		ip->i_forw = ip;
106 		ip->i_back = ip;
107 
108 		/* Unlock and discard unneeded inode. */
109 		ufs_iput(ip);
110 		brelse(bp);
111 		*vpp = NULL;
112 		return (error);
113 	}
114 	dp = bp->b_un.b_dino;
115 	dp += itoo(fs, ino);
116 	ip->i_din = *dp;
117 	brelse(bp);
118 
119 	/*
120 	 * Initialize the vnode from the inode, check for aliases.
121 	 * Note that the underlying vnode may have changed.
122 	 */
123 	if (error = ufs_vinit(mntp, &ffs_specops, FFS_FIFOOPS, &vp)) {
124 		ufs_iput(ip);
125 		*vpp = NULL;
126 		return (error);
127 	}
128 	/*
129 	 * Finish inode initialization now that aliasing has been resolved.
130 	 */
131 	ip->i_devvp = ump->um_devvp;
132 	VREF(ip->i_devvp);
133 	/*
134 	 * Set up a generation number for this inode if it does not
135 	 * already have one. This should only happen on old filesystems.
136 	 */
137 	if (ip->i_gen == 0) {
138 		if (++nextgennumber < (u_long)time.tv_sec)
139 			nextgennumber = time.tv_sec;
140 		ip->i_gen = nextgennumber;
141 		if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0)
142 			ip->i_flag |= IMOD;
143 	}
144 	*vpp = vp;
145 	return (0);
146 }
147 
148 /*
149  * Update the access, modified, and inode change times as specified
150  * by the IACC, IUPD, and ICHG flags respectively. The IMOD flag
151  * is used to specify that the inode needs to be updated but that
152  * the times have already been set. The access and modified times
153  * are taken from the second and third parameters; the inode change
154  * time is always taken from the current time. If waitfor is set,
155  * then wait for the disk write of the inode to complete.
156  */
157 int
158 ffs_update(vp, ta, tm, waitfor)
159 	register struct vnode *vp;
160 	struct timeval *ta, *tm;
161 	int waitfor;
162 {
163 	struct buf *bp;
164 	struct inode *ip;
165 	struct dinode *dp;
166 	register struct fs *fs;
167 	int error;
168 
169 	if (vp->v_mount->mnt_flag & MNT_RDONLY)
170 		return (0);
171 	ip = VTOI(vp);
172 	if ((ip->i_flag & (IUPD|IACC|ICHG|IMOD)) == 0)
173 		return (0);
174 	if (ip->i_flag&IACC)
175 		ip->i_atime.tv_sec = ta->tv_sec;
176 	if (ip->i_flag&IUPD) {
177 		ip->i_mtime.tv_sec = tm->tv_sec;
178 		INCRQUAD(ip->i_modrev);
179 	}
180 	if (ip->i_flag&ICHG)
181 		ip->i_ctime.tv_sec = time.tv_sec;
182 	ip->i_flag &= ~(IUPD|IACC|ICHG|IMOD);
183 
184 	fs = ip->i_fs;
185 	if (error = bread(ip->i_devvp, fsbtodb(fs, itod(fs, ip->i_number)),
186 		(int)fs->fs_bsize, NOCRED, &bp)) {
187 		brelse(bp);
188 		return (error);
189 	}
190 	dp = bp->b_un.b_dino + itoo(fs, ip->i_number);
191 	*dp = ip->i_din;
192 	if (waitfor)
193 		return (bwrite(bp));
194 	else {
195 		bdwrite(bp);
196 		return (0);
197 	}
198 }
199 
200 #define	SINGLE	0	/* index of single indirect block */
201 #define	DOUBLE	1	/* index of double indirect block */
202 #define	TRIPLE	2	/* index of triple indirect block */
203 /*
204  * Truncate the inode ip to at most length size.  Free affected disk
205  * blocks -- the blocks of the file are removed in reverse order.
206  *
207  * NB: triple indirect blocks are untested.
208  */
209 ffs_truncate(ovp, length, flags, cred)
210 	register struct vnode *ovp;
211 	off_t length;
212 	int flags;
213 	struct ucred *cred;
214 {
215 	register daddr_t lastblock;
216 	register struct inode *oip;
217 	daddr_t bn, lbn, lastiblock[NIADDR];
218 	register struct fs *fs;
219 	register struct inode *ip;
220 	struct buf *bp;
221 	int offset, size, level;
222 	long count, nblocks, blocksreleased = 0;
223 	register int i;
224 	int aflags, error, allerror;
225 	struct inode tip;
226 	off_t osize;
227 
228 	vnode_pager_setsize(ovp, (u_long)length);
229 	oip = VTOI(ovp);
230 	if (oip->i_size <= length) {
231 		oip->i_flag |= ICHG|IUPD;
232 		error = ffs_update(ovp, &time, &time, 1);
233 		return (error);
234 	}
235 	/*
236 	 * Calculate index into inode's block list of
237 	 * last direct and indirect blocks (if any)
238 	 * which we want to keep.  Lastblock is -1 when
239 	 * the file is truncated to 0.
240 	 */
241 	fs = oip->i_fs;
242 	lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1;
243 	lastiblock[SINGLE] = lastblock - NDADDR;
244 	lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
245 	lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
246 	nblocks = btodb(fs->fs_bsize);
247 	/*
248 	 * Update the size of the file. If the file is not being
249 	 * truncated to a block boundry, the contents of the
250 	 * partial block following the end of the file must be
251 	 * zero'ed in case it ever become accessable again because
252 	 * of subsequent file growth.
253 	 */
254 	osize = oip->i_size;
255 	offset = blkoff(fs, length);
256 	if (offset == 0) {
257 		oip->i_size = length;
258 	} else {
259 		lbn = lblkno(fs, length);
260 		aflags = B_CLRBUF;
261 		if (flags & IO_SYNC)
262 			aflags |= B_SYNC;
263 #ifdef QUOTA
264 		if (error = getinoquota(oip))
265 			return (error);
266 #endif
267 		if (error = ffs_balloc(oip, lbn, offset, cred, &bp, aflags))
268 			return (error);
269 		oip->i_size = length;
270 		size = blksize(fs, oip, lbn);
271 		(void) vnode_pager_uncache(ovp);
272 		bzero(bp->b_un.b_addr + offset, (unsigned)(size - offset));
273 		allocbuf(bp, size);
274 		if (flags & IO_SYNC)
275 			bwrite(bp);
276 		else
277 			bdwrite(bp);
278 	}
279 	/*
280 	 * Update file and block pointers on disk before we start freeing
281 	 * blocks.  If we crash before free'ing blocks below, the blocks
282 	 * will be returned to the free list.  lastiblock values are also
283 	 * normalized to -1 for calls to ffs_indirtrunc below.
284 	 */
285 	tip = *oip;
286 	tip.i_size = osize;
287 	for (level = TRIPLE; level >= SINGLE; level--)
288 		if (lastiblock[level] < 0) {
289 			oip->i_ib[level] = 0;
290 			lastiblock[level] = -1;
291 		}
292 	for (i = NDADDR - 1; i > lastblock; i--)
293 		oip->i_db[i] = 0;
294 	oip->i_flag |= ICHG|IUPD;
295 	vinvalbuf(ovp, (length > 0));
296 	allerror = ffs_update(ovp, &time, &time, MNT_WAIT);
297 
298 	/*
299 	 * Indirect blocks first.
300 	 */
301 	ip = &tip;
302 	for (level = TRIPLE; level >= SINGLE; level--) {
303 		bn = ip->i_ib[level];
304 		if (bn != 0) {
305 			error = ffs_indirtrunc(ip,
306 			    bn, lastiblock[level], level, &count);
307 			if (error)
308 				allerror = error;
309 			blocksreleased += count;
310 			if (lastiblock[level] < 0) {
311 				ip->i_ib[level] = 0;
312 				ffs_blkfree(ip, bn, fs->fs_bsize);
313 				blocksreleased += nblocks;
314 			}
315 		}
316 		if (lastiblock[level] >= 0)
317 			goto done;
318 	}
319 
320 	/*
321 	 * All whole direct blocks or frags.
322 	 */
323 	for (i = NDADDR - 1; i > lastblock; i--) {
324 		register long bsize;
325 
326 		bn = ip->i_db[i];
327 		if (bn == 0)
328 			continue;
329 		ip->i_db[i] = 0;
330 		bsize = blksize(fs, ip, i);
331 		ffs_blkfree(ip, bn, bsize);
332 		blocksreleased += btodb(bsize);
333 	}
334 	if (lastblock < 0)
335 		goto done;
336 
337 	/*
338 	 * Finally, look for a change in size of the
339 	 * last direct block; release any frags.
340 	 */
341 	bn = ip->i_db[lastblock];
342 	if (bn != 0) {
343 		long oldspace, newspace;
344 
345 		/*
346 		 * Calculate amount of space we're giving
347 		 * back as old block size minus new block size.
348 		 */
349 		oldspace = blksize(fs, ip, lastblock);
350 		ip->i_size = length;
351 		newspace = blksize(fs, ip, lastblock);
352 		if (newspace == 0)
353 			panic("itrunc: newspace");
354 		if (oldspace - newspace > 0) {
355 			/*
356 			 * Block number of space to be free'd is
357 			 * the old block # plus the number of frags
358 			 * required for the storage we're keeping.
359 			 */
360 			bn += numfrags(fs, newspace);
361 			ffs_blkfree(ip, bn, oldspace - newspace);
362 			blocksreleased += btodb(oldspace - newspace);
363 		}
364 	}
365 done:
366 /* BEGIN PARANOIA */
367 	for (level = SINGLE; level <= TRIPLE; level++)
368 		if (ip->i_ib[level] != oip->i_ib[level])
369 			panic("itrunc1");
370 	for (i = 0; i < NDADDR; i++)
371 		if (ip->i_db[i] != oip->i_db[i])
372 			panic("itrunc2");
373 /* END PARANOIA */
374 	oip->i_blocks -= blocksreleased;
375 	if (oip->i_blocks < 0)			/* sanity */
376 		oip->i_blocks = 0;
377 	oip->i_flag |= ICHG;
378 #ifdef QUOTA
379 	if (!getinoquota(oip))
380 		(void) chkdq(oip, -blocksreleased, NOCRED, 0);
381 #endif
382 	return (allerror);
383 }
384 
385 /*
386  * Release blocks associated with the inode ip and stored in the indirect
387  * block bn.  Blocks are free'd in LIFO order up to (but not including)
388  * lastbn.  If level is greater than SINGLE, the block is an indirect block
389  * and recursive calls to indirtrunc must be used to cleanse other indirect
390  * blocks.
391  *
392  * NB: triple indirect blocks are untested.
393  */
394 static int
395 ffs_indirtrunc(ip, bn, lastbn, level, countp)
396 	register struct inode *ip;
397 	daddr_t bn, lastbn;
398 	int level;
399 	long *countp;
400 {
401 	register int i;
402 	struct buf *bp;
403 	register struct fs *fs = ip->i_fs;
404 	register daddr_t *bap;
405 	daddr_t *copy, nb, last;
406 	long blkcount, factor;
407 	int nblocks, blocksreleased = 0;
408 	int error, allerror = 0;
409 
410 	/*
411 	 * Calculate index in current block of last
412 	 * block to be kept.  -1 indicates the entire
413 	 * block so we need not calculate the index.
414 	 */
415 	factor = 1;
416 	for (i = SINGLE; i < level; i++)
417 		factor *= NINDIR(fs);
418 	last = lastbn;
419 	if (lastbn > 0)
420 		last /= factor;
421 	nblocks = btodb(fs->fs_bsize);
422 	/*
423 	 * Get buffer of block pointers, zero those
424 	 * entries corresponding to blocks to be free'd,
425 	 * and update on disk copy first.
426 	 */
427 	error = bread(ip->i_devvp, fsbtodb(fs, bn), (int)fs->fs_bsize,
428 		NOCRED, &bp);
429 	if (error) {
430 		brelse(bp);
431 		*countp = 0;
432 		return (error);
433 	}
434 	bap = bp->b_un.b_daddr;
435 	MALLOC(copy, daddr_t *, fs->fs_bsize, M_TEMP, M_WAITOK);
436 	bcopy((caddr_t)bap, (caddr_t)copy, (u_int)fs->fs_bsize);
437 	bzero((caddr_t)&bap[last + 1],
438 	  (u_int)(NINDIR(fs) - (last + 1)) * sizeof (daddr_t));
439 	if (last == -1)
440 		bp->b_flags |= B_INVAL;
441 	error = bwrite(bp);
442 	if (error)
443 		allerror = error;
444 	bap = copy;
445 
446 	/*
447 	 * Recursively free totally unused blocks.
448 	 */
449 	for (i = NINDIR(fs) - 1; i > last; i--) {
450 		nb = bap[i];
451 		if (nb == 0)
452 			continue;
453 		if (level > SINGLE) {
454 			if (error = ffs_indirtrunc(ip,
455 			    nb, (daddr_t)-1, level - 1, &blkcount))
456 				allerror = error;
457 			blocksreleased += blkcount;
458 		}
459 		ffs_blkfree(ip, nb, fs->fs_bsize);
460 		blocksreleased += nblocks;
461 	}
462 
463 	/*
464 	 * Recursively free last partial block.
465 	 */
466 	if (level > SINGLE && lastbn >= 0) {
467 		last = lastbn % factor;
468 		nb = bap[i];
469 		if (nb != 0) {
470 			if (error =
471 			    ffs_indirtrunc(ip, nb, last, level - 1, &blkcount))
472 				allerror = error;
473 			blocksreleased += blkcount;
474 		}
475 	}
476 	FREE(copy, M_TEMP);
477 	*countp = blocksreleased;
478 	return (allerror);
479 }
480