xref: /csrg-svn/sys/ufs/ffs/ffs_inode.c (revision 53865)
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	7.53 (Berkeley) 06/04/92
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 (ap)
48 	struct vop_vget_args *ap;
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(ap->a_mp);
61 	dev = ump->um_dev;
62 	if ((*ap->a_vpp = ufs_ihashget(dev, ap->a_ino)) != NULL)
63 		return (0);
64 
65 	/* Allocate a new vnode/inode. */
66 	if (error = getnewvnode(VT_UFS, ap->a_mp, ffs_vnodeop_p, &vp)) {
67 		*ap->a_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 = ap->a_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, ap->a_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 		*ap->a_vpp = NULL;
110 		return (error);
111 	}
112 	dp = bp->b_un.b_dino;
113 	dp += itoo(fs, ap->a_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(ap->a_mp, ffs_specop_p, FFS_FIFOOPS, &vp)) {
122 		ufs_iput(ip);
123 		*ap->a_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 	/*
143 	 * Ensure that uid and gid are correct. This is a temporary
144 	 * fix until fsck has been changed to do the update.
145 	 */
146 	ip->i_uid = ip->i_din.di_ouid;
147 	ip->i_gid = ip->i_din.di_ogid;
148 
149 	*ap->a_vpp = vp;
150 	return (0);
151 }
152 
153 /*
154  * Update the access, modified, and inode change times as specified
155  * by the IACC, IUPD, and ICHG flags respectively. The IMOD flag
156  * is used to specify that the inode needs to be updated but that
157  * the times have already been set. The access and modified times
158  * are taken from the second and third parameters; the inode change
159  * time is always taken from the current time. If waitfor is set,
160  * then wait for the disk write of the inode to complete.
161  */
162 int
163 ffs_update (ap)
164 	struct vop_update_args *ap;
165 {
166 	struct buf *bp;
167 	struct inode *ip;
168 	struct dinode *dp;
169 	register struct fs *fs;
170 	int error;
171 
172 	if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
173 		return (0);
174 	ip = VTOI(ap->a_vp);
175 	if ((ip->i_flag & (IUPD|IACC|ICHG|IMOD)) == 0)
176 		return (0);
177 	if (ip->i_flag&IACC)
178 		ip->i_atime.tv_sec = ap->a_ta->tv_sec;
179 	if (ip->i_flag&IUPD) {
180 		ip->i_mtime.tv_sec = ap->a_tm->tv_sec;
181 		INCRQUAD(ip->i_modrev);
182 	}
183 	if (ip->i_flag&ICHG)
184 		ip->i_ctime.tv_sec = time.tv_sec;
185 	ip->i_flag &= ~(IUPD|IACC|ICHG|IMOD);
186 	/*
187 	 * Ensure that uid and gid are correct. This is a temporary
188 	 * fix until fsck has been changed to do the update.
189 	 */
190 	ip->i_din.di_ouid = ip->i_uid;
191 	ip->i_din.di_ogid = ip->i_gid;
192 
193 	fs = ip->i_fs;
194 	if (error = bread(ip->i_devvp, fsbtodb(fs, itod(fs, ip->i_number)),
195 		(int)fs->fs_bsize, NOCRED, &bp)) {
196 		brelse(bp);
197 		return (error);
198 	}
199 	dp = bp->b_un.b_dino + itoo(fs, ip->i_number);
200 	*dp = ip->i_din;
201 	if (ap->a_waitfor)
202 		return (bwrite(bp));
203 	else {
204 		bdwrite(bp);
205 		return (0);
206 	}
207 }
208 
209 #define	SINGLE	0	/* index of single indirect block */
210 #define	DOUBLE	1	/* index of double indirect block */
211 #define	TRIPLE	2	/* index of triple indirect block */
212 /*
213  * Truncate the inode ip to at most length size.  Free affected disk
214  * blocks -- the blocks of the file are removed in reverse order.
215  *
216  * NB: triple indirect blocks are untested.
217  */
218 ffs_truncate (ap)
219 	struct vop_truncate_args *ap;
220 {
221 	USES_VOP_UPDATE;
222 	register struct vnode *ovp = ap->a_vp;
223 	register daddr_t lastblock;
224 	register struct inode *oip;
225 	daddr_t bn, lbn, lastiblock[NIADDR];
226 	register struct fs *fs;
227 	register struct inode *ip;
228 	struct buf *bp;
229 	int offset, size, level;
230 	long count, nblocks, blocksreleased = 0;
231 	register int i;
232 	int aflags, error, allerror;
233 	struct inode tip;
234 	off_t osize;
235 
236 	vnode_pager_setsize(ovp, (u_long)ap->a_length);
237 	oip = VTOI(ovp);
238 	if (oip->i_size <= ap->a_length) {
239 		oip->i_flag |= ICHG|IUPD;
240 		error = VOP_UPDATE(ovp, &time, &time, 1);
241 		return (error);
242 	}
243 	/*
244 	 * Calculate index into inode's block list of
245 	 * last direct and indirect blocks (if any)
246 	 * which we want to keep.  Lastblock is -1 when
247 	 * the file is truncated to 0.
248 	 */
249 	fs = oip->i_fs;
250 	lastblock = lblkno(fs, ap->a_length + fs->fs_bsize - 1) - 1;
251 	lastiblock[SINGLE] = lastblock - NDADDR;
252 	lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
253 	lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
254 	nblocks = btodb(fs->fs_bsize);
255 	/*
256 	 * Update the size of the file. If the file is not being
257 	 * truncated to a block boundry, the contents of the
258 	 * partial block following the end of the file must be
259 	 * zero'ed in case it ever become accessable again because
260 	 * of subsequent file growth.
261 	 */
262 	osize = oip->i_size;
263 	offset = blkoff(fs, ap->a_length);
264 	if (offset == 0) {
265 		oip->i_size = ap->a_length;
266 	} else {
267 		lbn = lblkno(fs, ap->a_length);
268 		aflags = B_CLRBUF;
269 		if (ap->a_flags & IO_SYNC)
270 			aflags |= B_SYNC;
271 #ifdef QUOTA
272 		if (error = getinoquota(oip))
273 			return (error);
274 #endif
275 		if (error = ffs_balloc(oip, lbn, offset, ap->a_cred, &bp, aflags))
276 			return (error);
277 		oip->i_size = ap->a_length;
278 		size = blksize(fs, oip, lbn);
279 		(void) vnode_pager_uncache(ovp);
280 		bzero(bp->b_un.b_addr + offset, (unsigned)(size - offset));
281 		allocbuf(bp, size);
282 		if (ap->a_flags & IO_SYNC)
283 			bwrite(bp);
284 		else
285 			bdwrite(bp);
286 	}
287 	/*
288 	 * Update file and block pointers on disk before we start freeing
289 	 * blocks.  If we crash before free'ing blocks below, the blocks
290 	 * will be returned to the free list.  lastiblock values are also
291 	 * normalized to -1 for calls to ffs_indirtrunc below.
292 	 */
293 	tip = *oip;
294 	tip.i_size = osize;
295 	for (level = TRIPLE; level >= SINGLE; level--)
296 		if (lastiblock[level] < 0) {
297 			oip->i_ib[level] = 0;
298 			lastiblock[level] = -1;
299 		}
300 	for (i = NDADDR - 1; i > lastblock; i--)
301 		oip->i_db[i] = 0;
302 	oip->i_flag |= ICHG|IUPD;
303 	vinvalbuf(ovp, (ap->a_length > 0));
304 	allerror = VOP_UPDATE(ovp, &time, &time, MNT_WAIT);
305 
306 	/*
307 	 * Indirect blocks first.
308 	 */
309 	ip = &tip;
310 	for (level = TRIPLE; level >= SINGLE; level--) {
311 		bn = ip->i_ib[level];
312 		if (bn != 0) {
313 			error = ffs_indirtrunc(ip,
314 			    bn, lastiblock[level], level, &count);
315 			if (error)
316 				allerror = error;
317 			blocksreleased += count;
318 			if (lastiblock[level] < 0) {
319 				ip->i_ib[level] = 0;
320 				ffs_blkfree(ip, bn, fs->fs_bsize);
321 				blocksreleased += nblocks;
322 			}
323 		}
324 		if (lastiblock[level] >= 0)
325 			goto done;
326 	}
327 
328 	/*
329 	 * All whole direct blocks or frags.
330 	 */
331 	for (i = NDADDR - 1; i > lastblock; i--) {
332 		register long bsize;
333 
334 		bn = ip->i_db[i];
335 		if (bn == 0)
336 			continue;
337 		ip->i_db[i] = 0;
338 		bsize = blksize(fs, ip, i);
339 		ffs_blkfree(ip, bn, bsize);
340 		blocksreleased += btodb(bsize);
341 	}
342 	if (lastblock < 0)
343 		goto done;
344 
345 	/*
346 	 * Finally, look for a change in size of the
347 	 * last direct block; release any frags.
348 	 */
349 	bn = ip->i_db[lastblock];
350 	if (bn != 0) {
351 		long oldspace, newspace;
352 
353 		/*
354 		 * Calculate amount of space we're giving
355 		 * back as old block size minus new block size.
356 		 */
357 		oldspace = blksize(fs, ip, lastblock);
358 		ip->i_size = ap->a_length;
359 		newspace = blksize(fs, ip, lastblock);
360 		if (newspace == 0)
361 			panic("itrunc: newspace");
362 		if (oldspace - newspace > 0) {
363 			/*
364 			 * Block number of space to be free'd is
365 			 * the old block # plus the number of frags
366 			 * required for the storage we're keeping.
367 			 */
368 			bn += numfrags(fs, newspace);
369 			ffs_blkfree(ip, bn, oldspace - newspace);
370 			blocksreleased += btodb(oldspace - newspace);
371 		}
372 	}
373 done:
374 /* BEGIN PARANOIA */
375 	for (level = SINGLE; level <= TRIPLE; level++)
376 		if (ip->i_ib[level] != oip->i_ib[level])
377 			panic("itrunc1");
378 	for (i = 0; i < NDADDR; i++)
379 		if (ip->i_db[i] != oip->i_db[i])
380 			panic("itrunc2");
381 /* END PARANOIA */
382 	oip->i_blocks -= blocksreleased;
383 	if (oip->i_blocks < 0)			/* sanity */
384 		oip->i_blocks = 0;
385 	oip->i_flag |= ICHG;
386 #ifdef QUOTA
387 	if (!getinoquota(oip))
388 		(void) chkdq(oip, -blocksreleased, NOCRED, 0);
389 #endif
390 	return (allerror);
391 }
392 
393 /*
394  * Release blocks associated with the inode ip and stored in the indirect
395  * block bn.  Blocks are free'd in LIFO order up to (but not including)
396  * lastbn.  If level is greater than SINGLE, the block is an indirect block
397  * and recursive calls to indirtrunc must be used to cleanse other indirect
398  * blocks.
399  *
400  * NB: triple indirect blocks are untested.
401  */
402 static int
403 ffs_indirtrunc(ip, bn, lastbn, level, countp)
404 	register struct inode *ip;
405 	daddr_t bn, lastbn;
406 	int level;
407 	long *countp;
408 {
409 	register int i;
410 	struct buf *bp;
411 	register struct fs *fs = ip->i_fs;
412 	register daddr_t *bap;
413 	daddr_t *copy, nb, last;
414 	long blkcount, factor;
415 	int nblocks, blocksreleased = 0;
416 	int error, allerror = 0;
417 
418 	/*
419 	 * Calculate index in current block of last
420 	 * block to be kept.  -1 indicates the entire
421 	 * block so we need not calculate the index.
422 	 */
423 	factor = 1;
424 	for (i = SINGLE; i < level; i++)
425 		factor *= NINDIR(fs);
426 	last = lastbn;
427 	if (lastbn > 0)
428 		last /= factor;
429 	nblocks = btodb(fs->fs_bsize);
430 	/*
431 	 * Get buffer of block pointers, zero those
432 	 * entries corresponding to blocks to be free'd,
433 	 * and update on disk copy first.
434 	 */
435 	error = bread(ip->i_devvp, fsbtodb(fs, bn), (int)fs->fs_bsize,
436 		NOCRED, &bp);
437 	if (error) {
438 		brelse(bp);
439 		*countp = 0;
440 		return (error);
441 	}
442 	bap = bp->b_un.b_daddr;
443 	MALLOC(copy, daddr_t *, fs->fs_bsize, M_TEMP, M_WAITOK);
444 	bcopy((caddr_t)bap, (caddr_t)copy, (u_int)fs->fs_bsize);
445 	bzero((caddr_t)&bap[last + 1],
446 	  (u_int)(NINDIR(fs) - (last + 1)) * sizeof (daddr_t));
447 	if (last == -1)
448 		bp->b_flags |= B_INVAL;
449 	error = bwrite(bp);
450 	if (error)
451 		allerror = error;
452 	bap = copy;
453 
454 	/*
455 	 * Recursively free totally unused blocks.
456 	 */
457 	for (i = NINDIR(fs) - 1; i > last; i--) {
458 		nb = bap[i];
459 		if (nb == 0)
460 			continue;
461 		if (level > SINGLE) {
462 			if (error = ffs_indirtrunc(ip,
463 			    nb, (daddr_t)-1, level - 1, &blkcount))
464 				allerror = error;
465 			blocksreleased += blkcount;
466 		}
467 		ffs_blkfree(ip, nb, fs->fs_bsize);
468 		blocksreleased += nblocks;
469 	}
470 
471 	/*
472 	 * Recursively free last partial block.
473 	 */
474 	if (level > SINGLE && lastbn >= 0) {
475 		last = lastbn % factor;
476 		nb = bap[i];
477 		if (nb != 0) {
478 			if (error =
479 			    ffs_indirtrunc(ip, nb, last, level - 1, &blkcount))
480 				allerror = error;
481 			blocksreleased += blkcount;
482 		}
483 	}
484 	FREE(copy, M_TEMP);
485 	*countp = blocksreleased;
486 	return (allerror);
487 }
488