xref: /csrg-svn/sys/ufs/lfs/lfs_balloc.c (revision 55807)
1 /*
2  * Copyright (c) 1989, 1991 Regents of the University of California.
3  * All rights reserved.
4  *
5  * %sccs.include.redist.c%
6  *
7  *	@(#)lfs_balloc.c	7.35 (Berkeley) 08/01/92
8  */
9 
10 #include <sys/param.h>
11 #include <sys/buf.h>
12 #include <sys/proc.h>
13 #include <sys/vnode.h>
14 #include <sys/mount.h>
15 #include <sys/resourcevar.h>
16 #include <sys/trace.h>
17 
18 #include <miscfs/specfs/specdev.h>
19 
20 #include <ufs/ufs/quota.h>
21 #include <ufs/ufs/inode.h>
22 #include <ufs/ufs/ufsmount.h>
23 
24 #include <ufs/lfs/lfs.h>
25 #include <ufs/lfs/lfs_extern.h>
26 
27 int lfs_getlbns __P((struct vnode *, daddr_t, INDIR *, int *));
28 
29 /*
30  * Bmap converts a the logical block number of a file to its physical block
31  * number on the disk. The conversion is done by using the logical block
32  * number to index into the array of block pointers described by the dinode.
33  */
34 int
35 lfs_bmap(ap)
36 	struct vop_bmap_args /* {
37 		struct vnode *a_vp;
38 		daddr_t  a_bn;
39 		struct vnode **a_vpp;
40 		daddr_t *a_bnp;
41 	} */ *ap;
42 {
43 	/*
44 	 * Check for underlying vnode requests and ensure that logical
45 	 * to physical mapping is requested.
46 	 */
47 	if (ap->a_vpp != NULL)
48 		*ap->a_vpp = VTOI(ap->a_vp)->i_devvp;
49 	if (ap->a_bnp == NULL)
50 		return (0);
51 
52 	return (lfs_bmaparray(ap->a_vp, ap->a_bn, ap->a_bnp, NULL, NULL));
53 }
54 
55 /*
56  * LFS has a different version of bmap from FFS because of a naming conflict.
57  * In FFS, meta blocks are given real disk addresses at allocation time, and
58  * are linked into the device vnode, using a logical block number which is
59  * the same as the physical block number.  This can't be done by LFS because
60  * blocks aren't given disk addresses until they're written, so there's no
61  * way to distinguish the meta-data blocks for one file from any other file.
62  * This means that meta-data blocks have to be on the vnode for the file so
63  * they can be found, and have to have "names" different from the standard
64  * data blocks.  To do this, we divide the name space into positive and
65  * negative block numbers, and give the meta-data blocks negative logical
66  * numbers.  Indirect blocks are addressed by the negative address of the
67  * first data block to which they point.  Double indirect blocks are addressed
68  * by one less than the address of the first indirect block to which they
69  * point.  Triple indirect blocks are addressed by one less than the address
70  * of the first double indirect block to which they point.
71  */
72 int
73 lfs_bmaparray(vp, bn, bnp, ap, nump)
74 	struct vnode *vp;
75 	register daddr_t bn;
76 	daddr_t *bnp;
77 	INDIR *ap;
78 	int *nump;
79 {
80 	register struct inode *ip;
81 	struct buf *bp;
82 	struct lfs *fs;
83 	struct vnode *devvp;
84 	INDIR a[NIADDR], *xap;
85 	daddr_t *bap, daddr;
86 	long metalbn;
87 	int error, num, off;
88 	struct vop_strategy_args vop_strategy_a;
89 
90 	ip = VTOI(vp);
91 #ifdef DIAGNOSTIC
92 	if (ap != NULL && nump == NULL || ap == NULL && nump != NULL)
93 		panic("lfs_bmaparray: invalid arguments");
94 #endif
95 
96 	xap = ap == NULL ? a : ap;
97 	if (!nump)
98 		nump = &num;
99 	if (error = lfs_getlbns(vp, bn, xap, nump))
100 		return (error);
101 
102 	num = *nump;
103 	if (num == 0) {
104 		*bnp = ip->i_db[bn];
105 		if (*bnp == 0)
106 			*bnp = UNASSIGNED;
107 		return (0);
108 	}
109 
110 
111 	/* Get disk address out of indirect block array */
112 	daddr = ip->i_ib[xap->in_off];
113 
114 	/* Fetch through the indirect blocks. */
115 	fs = ip->i_lfs;
116 	devvp = VFSTOUFS(vp->v_mount)->um_devvp;
117 
118 	for (bp = NULL, ++xap; daddr && --num; ++xap) {
119 		/* If looking for a meta-block, break out when we find it. */
120 		metalbn = xap->in_lbn;
121 		if (metalbn == bn)
122 			break;
123 
124 		/*
125 		 * Read in the appropriate indirect block.  LFS can't do a
126 		 * bread because bread knows that FFS will hand it the device
127 		 * vnode, not the file vnode, so the b_dev and b_blkno would
128 		 * be wrong.
129 		 *
130 		 * XXX
131 		 * This REALLY needs to be fixed, at the very least it needs
132 		 * to be rethought when the buffer cache goes away.  When it's
133 		 * fixed, change lfs_bmaparray and lfs_getlbns to take an ip,
134 		 * not a vp.
135 		 */
136 		if (bp)
137 			brelse(bp);
138 		bp = getblk(vp, metalbn, fs->lfs_bsize);
139 		if (bp->b_flags & (B_DONE | B_DELWRI)) {
140 			trace(TR_BREADHIT, pack(vp, size), metalbn);
141 		} else {
142 			trace(TR_BREADMISS, pack(vp, size), metalbn);
143 			bp->b_blkno = daddr;
144 			bp->b_flags |= B_READ;
145 			bp->b_dev = devvp->v_rdev;
146 			/*
147 			 * Call a strategy VOP by hand.
148 			 */
149 			vop_strategy_a.a_desc = VDESC(vop_strategy);
150 			vop_strategy_a.a_bp=bp;
151 			VOCALL(devvp->v_op, VOFFSET(vop_strategy), \
152 			       &vop_strategy_a);
153 			curproc->p_stats->p_ru.ru_inblock++;	/* XXX */
154 			if (error = biowait(bp)) {
155 				brelse(bp);
156 				return (error);
157 			}
158 		}
159 		daddr = bp->b_un.b_daddr[xap->in_off];
160 	}
161 	if (bp)
162 		brelse(bp);
163 
164 	*bnp = daddr == 0 ? UNASSIGNED : daddr;
165 	return (0);
166 }
167 
168 /*
169  * Create an array of logical block number/offset pairs which represent the
170  * path of indirect blocks required to access a data block.  The first "pair"
171  * contains the logical block number of the appropriate single, double or
172  * triple indirect block and the offset into the inode indirect block array.
173  * Note, the logical block number of the inode single/double/triple indirect
174  * block appears twice in the array, once with the offset into the i_ib and
175  * once with the offset into the page itself.
176  */
177 int
178 lfs_getlbns(vp, bn, ap, nump)
179 	struct vnode *vp;
180 	register daddr_t bn;
181 	INDIR *ap;
182 	int *nump;
183 {
184 	struct lfs *fs;
185 	long metalbn, realbn;
186 	int j, numlevels, off, sh;
187 
188 	if (nump)
189 		*nump = 0;
190 	numlevels = 0;
191 	realbn = bn;
192 	if ((long)bn < 0)
193 		bn = -(long)bn;
194 
195 	/* The first NDADDR blocks are direct blocks. */
196 	if (bn < NDADDR)
197 		return (0);
198 
199 	/*
200 	 * Determine the number of levels of indirection.  After this loop
201 	 * is done, sh indicates the number of data blocks possible at the
202 	 * given level of indirection, and NIADDR - j is the number of levels
203 	 * of indirection needed to locate the requested block.
204 	 */
205 	bn -= NDADDR;
206 	fs = VTOI(vp)->i_lfs;
207 	sh = 1;
208 	for (j = NIADDR; j > 0; j--) {
209 		sh *= NINDIR(fs);
210 		if (bn < sh)
211 			break;
212 		bn -= sh;
213 	}
214 	if (j == 0)
215 		return (EFBIG);
216 
217 	/* Calculate the address of the first meta-block. */
218 	if (realbn >= 0)
219 		metalbn = -(realbn - bn + NIADDR - j);
220 	else
221 		metalbn = -(-realbn - bn + NIADDR - j);
222 
223 	/*
224 	 * At each iteration, off is the offset into the bap array which is
225 	 * an array of disk addresses at the current level of indirection.
226 	 * The logical block number and the offset in that block are stored
227 	 * into the argument array.
228 	 */
229 	++numlevels;
230 	ap->in_lbn = metalbn;
231 	ap->in_off = off = NIADDR - j;
232 	ap++;
233 	for (; j <= NIADDR; j++) {
234 		/* If searching for a meta-data block, quit when found. */
235 		if (metalbn == realbn)
236 			break;
237 
238 		sh /= NINDIR(fs);
239 		off = (bn / sh) % NINDIR(fs);
240 
241 		++numlevels;
242 		ap->in_lbn = metalbn;
243 		ap->in_off = off;
244 		++ap;
245 
246 		metalbn -= -1 + off * sh;
247 	}
248 	if (nump)
249 		*nump = numlevels;
250 	return (0);
251 }
252 
253 int
254 lfs_balloc(vp, iosize, lbn, bpp)
255 	struct vnode *vp;
256 	u_long iosize;
257 	daddr_t lbn;
258 	struct buf **bpp;
259 {
260 	struct buf *bp;
261 	struct inode *ip;
262 	struct lfs *fs;
263 	daddr_t daddr;
264 	int error, newblock;
265 
266 	ip = VTOI(vp);
267 	fs = ip->i_lfs;
268 
269 	/*
270 	 * Three cases: it's a block beyond the end of file, it's a block in
271 	 * the file that may or may not have been assigned a disk address or
272 	 * we're writing an entire block.  Note, if the daddr is unassigned,
273 	 * the block might still have existed in the cache.  If it did, make
274 	 * sure we don't count it as a new block or zero out its contents.
275 	 */
276 	newblock = ip->i_size <= lbn << fs->lfs_bshift;
277 	if (!newblock && (error = VOP_BMAP(vp, lbn, NULL, &daddr)))
278 		return (error);
279 
280 	if (newblock || daddr == UNASSIGNED || iosize == fs->lfs_bsize) {
281 		*bpp = bp = getblk(vp, lbn, fs->lfs_bsize);
282 		if (newblock ||
283 		    daddr == UNASSIGNED && !(bp->b_flags & B_CACHE)) {
284 			ip->i_blocks += btodb(fs->lfs_bsize);
285 			fs->lfs_bfree -= btodb(fs->lfs_bsize);
286 			if (iosize != fs->lfs_bsize)
287 				clrbuf(bp);
288 		}
289 		return (0);
290 	}
291 	return (bread(vp, lbn, fs->lfs_bsize, NOCRED, bpp));
292 
293 }
294