fdfs.c revision 1.1 1 /* $NetBSD: fdfs.c,v 1.1 2006/03/30 19:10:13 perseant Exp $ */
2
3 /*-
4 * Copyright (c) 2005 The NetBSD Foundation, Inc.
5 * All rights reserved.
6 *
7 * This code is derived from software contributed to The NetBSD Foundation
8 * by Konrad E. Schroder <perseant (at) hhhh.org>.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by the NetBSD
21 * Foundation, Inc. and its contributors.
22 * 4. Neither the name of The NetBSD Foundation nor the names of its
23 * contributors may be used to endorse or promote products derived
24 * from this software without specific prior written permission.
25 *
26 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
27 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
28 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
29 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
30 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36 * POSSIBILITY OF SUCH DAMAGE.
37 */
38
39 /*
40 * Buffer cache routines for a file-descriptor backed filesystem.
41 * This is part of lfs_cleanerd so there is also a "segment pointer" that
42 * we can make buffers out of without duplicating memory or reading the data
43 * again.
44 */
45
46 #include <err.h>
47 #include <fcntl.h>
48 #include <time.h>
49 #include <stdio.h>
50 #include <stdlib.h>
51 #include <string.h>
52 #include <unistd.h>
53
54 #include <sys/syslog.h>
55 #include <sys/param.h>
56 #include <sys/mount.h>
57 #include <sys/stat.h>
58
59 #include "vnode.h"
60 #include "bufcache.h"
61 #include "fdfs.h"
62
63 /*
64 * Return a "vnode" interface to a given file descriptor.
65 */
66 struct uvnode *
67 fd_vget(int fd, int bsize, int segsize, int nseg)
68 {
69 struct fdfs *fs;
70 struct uvnode *vp;
71 int i;
72
73 fs = (struct fdfs *)malloc(sizeof(*fs));
74 if (fs == NULL)
75 return NULL;
76 if (segsize > 0) {
77 fs->fd_bufp = (struct fd_buf *)malloc(nseg *
78 sizeof(struct fd_buf));
79 if (fs->fd_bufp == NULL) {
80 free(fs);
81 return NULL;
82 }
83 for (i = 0; i < nseg; i++) {
84 fs->fd_bufp[i].start = 0x0;
85 fs->fd_bufp[i].end = 0x0;
86 fs->fd_bufp[i].buf = (char *)malloc(segsize);
87 if (fs->fd_bufp[i].buf == NULL) {
88 while (--i >= 0)
89 free(fs->fd_bufp[i].buf);
90 free(fs->fd_bufp);
91 free(fs);
92 return NULL;
93 }
94 }
95 } else
96 fs->fd_bufp = NULL;
97
98 fs->fd_fd = fd;
99 fs->fd_bufc = nseg;
100 fs->fd_bufi = 0;
101 fs->fd_bsize = bsize;
102 fs->fd_ssize = segsize;
103
104 vp = (struct uvnode *) malloc(sizeof(*vp));
105 memset(vp, 0, sizeof(*vp));
106 vp->v_fd = fd;
107 vp->v_fs = fs;
108 vp->v_usecount = 0;
109 vp->v_strategy_op = fd_vop_strategy;
110 vp->v_bwrite_op = fd_vop_bwrite;
111 vp->v_bmap_op = fd_vop_bmap;
112 LIST_INIT(&vp->v_cleanblkhd);
113 LIST_INIT(&vp->v_dirtyblkhd);
114 vp->v_data = NULL;
115
116 return vp;
117 }
118
119 /*
120 * Deallocate a vnode.
121 */
122 void
123 fd_reclaim(struct uvnode *vp)
124 {
125 int i;
126 struct ubuf *bp;
127 struct fdfs *fs;
128
129 LIST_FOREACH(bp, &vp->v_dirtyblkhd, b_vnbufs) {
130 bremfree(bp);
131 buf_destroy(bp);
132 }
133 LIST_FOREACH(bp, &vp->v_cleanblkhd, b_vnbufs) {
134 bremfree(bp);
135 buf_destroy(bp);
136 }
137
138 fs = (struct fdfs *)vp->v_fs;
139 for (i = 0; i < fs->fd_bufc; i++)
140 free(fs->fd_bufp[i].buf);
141 free(fs->fd_bufp);
142 free(fs);
143 memset(vp, 0, sizeof(vp));
144 }
145
146 /*
147 * We won't be using that last segment after all.
148 */
149 void
150 fd_release(struct uvnode *vp)
151 {
152 --((struct fdfs *)vp->v_fs)->fd_bufi;
153 }
154
155 /*
156 * Reset buffer pointer to first buffer.
157 */
158 void
159 fd_release_all(struct uvnode *vp)
160 {
161 ((struct fdfs *)vp->v_fs)->fd_bufi = 0;
162 }
163
164 /*
165 * Prepare a segment buffer which we will expect to read from.
166 */
167 int
168 fd_preload(struct uvnode *vp, daddr_t start)
169 {
170 struct fdfs *fs = (struct fdfs *)vp->v_fs;
171 struct fd_buf *t;
172 int r;
173
174 /* We might need to allocate more buffers. */
175 if (fs->fd_bufi == fs->fd_bufc) {
176 fs->fd_bufc = fs->fd_bufi + 1;
177 syslog(LOG_DEBUG, "increasing number of segment buffers to %d",
178 fs->fd_bufc);
179 t = realloc(fs->fd_bufp, fs->fd_bufc * sizeof(struct fd_buf));
180 if (t == NULL)
181 return -1;
182 fs->fd_bufp = t;
183 fs->fd_bufp[fs->fd_bufi].start = 0x0;
184 fs->fd_bufp[fs->fd_bufi].end = 0x0;
185 fs->fd_bufp[fs->fd_bufi].buf = (char *)malloc(fs->fd_ssize);
186 if (fs->fd_bufp[fs->fd_bufi].buf == NULL) {
187 --fs->fd_bufc;
188 free(fs->fd_bufp[fs->fd_bufc].buf);
189 }
190 }
191
192 /* Read the current buffer. */
193 fs->fd_bufp[fs->fd_bufi].start = start;
194 fs->fd_bufp[fs->fd_bufi].end = start + fs->fd_ssize / fs->fd_bsize;
195
196 if ((r = pread(fs->fd_fd, fs->fd_bufp[fs->fd_bufi].buf,
197 (size_t)fs->fd_ssize, start * fs->fd_bsize)) < 0) {
198 syslog(LOG_ERR, "preload to segment buffer %d", fs->fd_bufi);
199 return r;
200 }
201
202 fs->fd_bufi = fs->fd_bufi + 1;
203 return 0;
204 }
205
206 /*
207 * Get a pointer to a block contained in one of the segment buffers,
208 * as if from bread() but avoiding the buffer cache.
209 */
210 char *
211 fd_ptrget(struct uvnode *vp, daddr_t bn)
212 {
213 int i;
214 struct fdfs *fs;
215
216 fs = (struct fdfs *)vp->v_fs;
217 for (i = 0; i < fs->fd_bufc; i++) {
218 if (bn >= fs->fd_bufp[i].start && bn < fs->fd_bufp[i].end) {
219 return fs->fd_bufp[i].buf +
220 (bn - fs->fd_bufp[i].start) * fs->fd_bsize;
221 }
222 }
223 return NULL;
224 }
225
226 /*
227 * Strategy routine. We can read from the segment buffer if requested.
228 */
229 int
230 fd_vop_strategy(struct ubuf * bp)
231 {
232 struct fdfs *fs;
233 char *cp;
234 int count;
235
236 fs = (struct fdfs *)bp->b_vp->v_fs;
237 if (bp->b_flags & B_READ) {
238 if ((cp = fd_ptrget(bp->b_vp, bp->b_blkno)) != NULL) {
239 free(bp->b_data);
240 bp->b_data = cp;
241 bp->b_flags |= (B_DONTFREE | B_DONE);
242 return 0;
243 }
244 count = pread(bp->b_vp->v_fd, bp->b_data, bp->b_bcount,
245 bp->b_blkno * fs->fd_bsize);
246 if (count == bp->b_bcount)
247 bp->b_flags |= B_DONE;
248 } else {
249 count = pwrite(bp->b_vp->v_fd, bp->b_data, bp->b_bcount,
250 bp->b_blkno * fs->fd_bsize);
251 if (count == 0) {
252 perror("pwrite");
253 return -1;
254 }
255 bp->b_flags &= ~B_DELWRI;
256 reassignbuf(bp, bp->b_vp);
257 }
258 return 0;
259 }
260
261 /*
262 * Delayed write.
263 */
264 int
265 fd_vop_bwrite(struct ubuf * bp)
266 {
267 bp->b_flags |= B_DELWRI;
268 reassignbuf(bp, bp->b_vp);
269 brelse(bp);
270 return 0;
271 }
272
273 /*
274 * Map lbn to disk address. Since we are using the file
275 * descriptor as the "disk", the disk address is meaningless
276 * and we just return the block address.
277 */
278 int
279 fd_vop_bmap(struct uvnode * vp, daddr_t lbn, daddr_t * daddrp)
280 {
281 *daddrp = lbn;
282 return 0;
283 }
284