Temporarily allow recursion on locks to deal with a double lock in the
[dragonfly.git] / sys / vfs / gnu / ext2fs / ext2_balloc.c
1 /*
2  *  modified for Lites 1.1
3  *
4  *  Aug 1995, Godmar Back (gback@cs.utah.edu)
5  *  University of Utah, Department of Computer Science
6  */
7 /*
8  * Copyright (c) 1982, 1986, 1989, 1993
9  *      The Regents of the University of California.  All rights reserved.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 3. All advertising materials mentioning features or use of this software
20  *    must display the following acknowledgement:
21  *      This product includes software developed by the University of
22  *      California, Berkeley and its contributors.
23  * 4. Neither the name of the University nor the names of its contributors
24  *    may be used to endorse or promote products derived from this software
25  *    without specific prior written permission.
26  *
27  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
28  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
29  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
30  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
31  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
32  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
33  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
34  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
35  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
36  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37  * SUCH DAMAGE.
38  *
39  *      @(#)ffs_balloc.c        8.4 (Berkeley) 9/23/93
40  * $FreeBSD: src/sys/gnu/ext2fs/ext2_balloc.c,v 1.9.2.1 2000/08/03 00:52:57 peter Exp $
41  * $DragonFly: src/sys/vfs/gnu/ext2fs/ext2_balloc.c,v 1.6 2004/04/08 20:57:52 cpressey Exp $
42  */
43
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/buf.h>
47 #include <sys/lock.h>
48 #include <sys/ucred.h>
49 #include <sys/vnode.h>
50
51 #include <vfs/ufs/quota.h>
52 #include <vfs/ufs/inode.h>
53 #include <vfs/ufs/ufs_extern.h>
54
55 #include "ext2_fs.h"
56 #include "ext2_fs_sb.h"
57 #include "fs.h"
58 #include "ext2_extern.h"
59
60 /*
61  * Balloc defines the structure of file system storage
62  * by allocating the physical blocks on a device given
63  * the inode and the logical block number in a file.
64  */
65 int
66 ext2_balloc(struct inode *ip, daddr_t bn, int size, struct ucred *cred,
67             struct buf **bpp, int flags)
68 {
69         struct ext2_sb_info *fs;
70         daddr_t nb;
71         struct buf *bp, *nbp;
72         struct vnode *vp = ITOV(ip);
73         struct indir indirs[NIADDR + 2];
74         daddr_t newb, lbn, *bap, pref;
75         int osize, nsize, num, i, error;
76 /*
77 ext2_debug("ext2_balloc called (%d, %d, %d)\n", 
78         ip->i_number, (int)bn, (int)size);
79 */
80         *bpp = NULL;
81         if (bn < 0)
82                 return (EFBIG);
83         fs = ip->i_e2fs;
84         lbn = bn;
85
86         /*
87          * check if this is a sequential block allocation. 
88          * If so, increment next_alloc fields to allow ext2_blkpref 
89          * to make a good guess
90          */
91         if (lbn == ip->i_next_alloc_block + 1) {
92                 ip->i_next_alloc_block++;
93                 ip->i_next_alloc_goal++;
94         }
95
96         /*
97          * The first NDADDR blocks are direct blocks
98          */
99         if (bn < NDADDR) {
100                 nb = ip->i_db[bn];
101                 /* no new block is to be allocated, and no need to expand
102                    the file */
103                 if (nb != 0 && ip->i_size >= (bn + 1) * fs->s_blocksize) {
104                         error = bread(vp, bn, fs->s_blocksize, &bp);
105                         if (error) {
106                                 brelse(bp);
107                                 return (error);
108                         }
109                         *bpp = bp;
110                         return (0);
111                 }
112                 if (nb != 0) {
113                         /*
114                          * Consider need to reallocate a fragment.
115                          */
116                         osize = fragroundup(fs, blkoff(fs, ip->i_size));
117                         nsize = fragroundup(fs, size);
118                         if (nsize <= osize) {
119                                 error = bread(vp, bn, osize, &bp);
120                                 if (error) {
121                                         brelse(bp);
122                                         return (error);
123                                 }
124                         } else {
125                         /* Godmar thinks: this shouldn't happen w/o fragments */
126                                 printf("nsize %d(%d) > osize %d(%d) nb %d\n", 
127                                         (int)nsize, (int)size, (int)osize, 
128                                         (int)ip->i_size, (int)nb);
129                                 panic(
130                                     "ext2_balloc: Something is terribly wrong");
131 /*
132  * please note there haven't been any changes from here on -
133  * FFS seems to work.
134  */
135                         }
136                 } else {
137                         if (ip->i_size < (bn + 1) * fs->s_blocksize)
138                                 nsize = fragroundup(fs, size);
139                         else
140                                 nsize = fs->s_blocksize;
141                         error = ext2_alloc(ip, bn,
142                             ext2_blkpref(ip, bn, (int)bn, &ip->i_db[0], 0),
143                             nsize, cred, &newb);
144                         if (error)
145                                 return (error);
146                         bp = getblk(vp, bn, nsize, 0, 0);
147                         bp->b_blkno = fsbtodb(fs, newb);
148                         if (flags & B_CLRBUF)
149                                 vfs_bio_clrbuf(bp);
150                 }
151                 ip->i_db[bn] = dbtofsb(fs, bp->b_blkno);
152                 ip->i_flag |= IN_CHANGE | IN_UPDATE;
153                 *bpp = bp;
154                 return (0);
155         }
156         /*
157          * Determine the number of levels of indirection.
158          */
159         pref = 0;
160         if ((error = ufs_getlbns(vp, bn, indirs, &num)) != 0)
161                 return(error);
162 #if DIAGNOSTIC
163         if (num < 1)
164                 panic ("ext2_balloc: ufs_bmaparray returned indirect block");
165 #endif
166         /*
167          * Fetch the first indirect block allocating if necessary.
168          */
169         --num;
170         nb = ip->i_ib[indirs[0].in_off];
171         if (nb == 0) {
172 #if 0
173                 pref = ext2_blkpref(ip, lbn, 0, (daddr_t *)0, 0);
174 #else
175                 /* see the comment by ext2_blkpref. What we do here is
176                    to pretend that it'd be good for a block holding indirect
177                    pointers to be allocated near its predecessor in terms 
178                    of indirection, or the last direct block. 
179                    We shamelessly exploit the fact that i_ib immediately
180                    follows i_db. 
181                    Godmar thinks it make sense to allocate i_ib[0] immediately
182                    after i_db[11], but it's not utterly clear whether this also
183                    applies to i_ib[1] and i_ib[0]
184                 */
185
186                 pref = ext2_blkpref(ip, lbn, indirs[0].in_off + 
187                                              EXT2_NDIR_BLOCKS, &ip->i_db[0], 0);
188 #endif
189                 if ((error = ext2_alloc(ip, lbn, pref, (int)fs->s_blocksize,
190                     cred, &newb)) != 0)
191                         return (error);
192                 nb = newb;
193                 bp = getblk(vp, indirs[1].in_lbn, fs->s_blocksize, 0, 0);
194                 bp->b_blkno = fsbtodb(fs, newb);
195                 vfs_bio_clrbuf(bp);
196                 /*
197                  * Write synchronously so that indirect blocks
198                  * never point at garbage.
199                  */
200                 if ((error = bwrite(bp)) != 0) {
201                         ext2_blkfree(ip, nb, fs->s_blocksize);
202                         return (error);
203                 }
204                 ip->i_ib[indirs[0].in_off] = newb;
205                 ip->i_flag |= IN_CHANGE | IN_UPDATE;
206         }
207         /*
208          * Fetch through the indirect blocks, allocating as necessary.
209          */
210         for (i = 1;;) {
211                 error = bread(vp,
212                     indirs[i].in_lbn, (int)fs->s_blocksize, &bp);
213                 if (error) {
214                         brelse(bp);
215                         return (error);
216                 }
217                 bap = (daddr_t *)bp->b_data;
218                 nb = bap[indirs[i].in_off];
219                 if (i == num)
220                         break;
221                 i += 1;
222                 if (nb != 0) {
223                         brelse(bp);
224                         continue;
225                 }
226                 if (pref == 0) 
227 #if 1
228                         /* see the comment above and by ext2_blkpref
229                          * I think this implements Linux policy, but
230                          * does it really make sense to allocate to
231                          * block containing pointers together ?
232                          * Also, will it ever succeed ?
233                          */
234                         pref = ext2_blkpref(ip, lbn, indirs[i].in_off, bap,
235                                                 bp->b_lblkno);
236 #else
237                         pref = ext2_blkpref(ip, lbn, 0, (daddr_t *)0, 0);
238 #endif
239                 if ((error =
240                     ext2_alloc(ip, lbn, pref, (int)fs->s_blocksize, cred, &newb)) != 0) {
241                         brelse(bp);
242                         return (error);
243                 }
244                 nb = newb;
245                 nbp = getblk(vp, indirs[i].in_lbn, fs->s_blocksize, 0, 0);
246                 nbp->b_blkno = fsbtodb(fs, nb);
247                 vfs_bio_clrbuf(nbp);
248                 /*
249                  * Write synchronously so that indirect blocks
250                  * never point at garbage.
251                  */
252                 if ((error = bwrite(nbp)) != 0) {
253                         ext2_blkfree(ip, nb, fs->s_blocksize);
254                         brelse(bp);
255                         return (error);
256                 }
257                 bap[indirs[i - 1].in_off] = nb;
258                 /*
259                  * If required, write synchronously, otherwise use
260                  * delayed write.
261                  */
262                 if (flags & B_SYNC) {
263                         bwrite(bp);
264                 } else {
265                         bdwrite(bp);
266                 }
267         }
268         /*
269          * Get the data block, allocating if necessary.
270          */
271         if (nb == 0) {
272                 pref = ext2_blkpref(ip, lbn, indirs[i].in_off, &bap[0], 
273                                 bp->b_lblkno);
274                 if ((error = ext2_alloc(ip,
275                     lbn, pref, (int)fs->s_blocksize, cred, &newb)) != 0) {
276                         brelse(bp);
277                         return (error);
278                 }
279                 nb = newb;
280                 nbp = getblk(vp, lbn, fs->s_blocksize, 0, 0);
281                 nbp->b_blkno = fsbtodb(fs, nb);
282                 if (flags & B_CLRBUF)
283                         vfs_bio_clrbuf(nbp);
284                 bap[indirs[i].in_off] = nb;
285                 /*
286                  * If required, write synchronously, otherwise use
287                  * delayed write.
288                  */
289                 if (flags & B_SYNC) {
290                         bwrite(bp);
291                 } else {
292                         bdwrite(bp);
293                 }
294                 *bpp = nbp;
295                 return (0);
296         }
297         brelse(bp);
298         if (flags & B_CLRBUF) {
299                 error = bread(vp, lbn, (int)fs->s_blocksize, &nbp);
300                 if (error) {
301                         brelse(nbp);
302                         return (error);
303                 }
304         } else {
305                 nbp = getblk(vp, lbn, fs->s_blocksize, 0, 0);
306                 nbp->b_blkno = fsbtodb(fs, nb);
307         }
308         *bpp = nbp;
309         return (0);
310 }