1 |
1275 |
phoenix |
/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
|
2 |
|
|
* vim:expandtab:shiftwidth=8:tabstop=8:
|
3 |
|
|
*
|
4 |
|
|
* Copyright (C) 1999 Peter J. Braam <braam@clusterfs.com>
|
5 |
|
|
*
|
6 |
|
|
* This file is part of InterMezzo, http://www.inter-mezzo.org.
|
7 |
|
|
*
|
8 |
|
|
* InterMezzo is free software; you can redistribute it and/or
|
9 |
|
|
* modify it under the terms of version 2 of the GNU General Public
|
10 |
|
|
* License as published by the Free Software Foundation.
|
11 |
|
|
*
|
12 |
|
|
* InterMezzo is distributed in the hope that it will be useful,
|
13 |
|
|
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
14 |
|
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
15 |
|
|
* GNU General Public License for more details.
|
16 |
|
|
*
|
17 |
|
|
* You should have received a copy of the GNU General Public License
|
18 |
|
|
* along with InterMezzo; if not, write to the Free Software
|
19 |
|
|
* Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
|
20 |
|
|
*
|
21 |
|
|
* Sysctrl entries for Intermezzo!
|
22 |
|
|
*/
|
23 |
|
|
|
24 |
|
|
#define __NO_VERSION__
|
25 |
|
|
#include <linux/config.h> /* for CONFIG_PROC_FS */
|
26 |
|
|
#include <linux/module.h>
|
27 |
|
|
#include <linux/sched.h>
|
28 |
|
|
#include <linux/mm.h>
|
29 |
|
|
#include <linux/sysctl.h>
|
30 |
|
|
#include <linux/swapctl.h>
|
31 |
|
|
#include <linux/proc_fs.h>
|
32 |
|
|
#include <linux/slab.h>
|
33 |
|
|
#include <linux/vmalloc.h>
|
34 |
|
|
#include <linux/stat.h>
|
35 |
|
|
#include <linux/ctype.h>
|
36 |
|
|
#include <linux/init.h>
|
37 |
|
|
#include <asm/bitops.h>
|
38 |
|
|
#include <asm/segment.h>
|
39 |
|
|
#include <asm/uaccess.h>
|
40 |
|
|
#include <linux/utsname.h>
|
41 |
|
|
#include <linux/blk.h>
|
42 |
|
|
|
43 |
|
|
|
44 |
|
|
#include <linux/intermezzo_fs.h>
|
45 |
|
|
#include <linux/intermezzo_psdev.h>
|
46 |
|
|
|
47 |
|
|
/* /proc entries */
|
48 |
|
|
|
49 |
|
|
#ifdef CONFIG_PROC_FS
|
50 |
|
|
struct proc_dir_entry *proc_fs_intermezzo;
|
51 |
|
|
int intermezzo_mount_get_info( char * buffer, char ** start, off_t offset,
|
52 |
|
|
int length)
|
53 |
|
|
{
|
54 |
|
|
int len=0;
|
55 |
|
|
|
56 |
|
|
/* this works as long as we are below 1024 characters! */
|
57 |
|
|
*start = buffer + offset;
|
58 |
|
|
len -= offset;
|
59 |
|
|
|
60 |
|
|
if ( len < 0 )
|
61 |
|
|
return -EINVAL;
|
62 |
|
|
|
63 |
|
|
return len;
|
64 |
|
|
}
|
65 |
|
|
|
66 |
|
|
#endif
|
67 |
|
|
|
68 |
|
|
|
69 |
|
|
/* SYSCTL below */
|
70 |
|
|
|
71 |
|
|
static struct ctl_table_header *intermezzo_table_header = NULL;
|
72 |
|
|
/* 0x100 to avoid any chance of collisions at any point in the tree with
|
73 |
|
|
* non-directories
|
74 |
|
|
*/
|
75 |
|
|
#define PSDEV_INTERMEZZO (0x100)
|
76 |
|
|
|
77 |
|
|
#define PSDEV_DEBUG 1 /* control debugging */
|
78 |
|
|
#define PSDEV_TRACE 2 /* control enter/leave pattern */
|
79 |
|
|
#define PSDEV_TIMEOUT 3 /* timeout on upcalls to become intrble */
|
80 |
|
|
#define PSDEV_HARD 4 /* mount type "hard" or "soft" */
|
81 |
|
|
#define PSDEV_NO_FILTER 5 /* controls presto_chk */
|
82 |
|
|
#define PSDEV_NO_JOURNAL 6 /* controls presto_chk */
|
83 |
|
|
#define PSDEV_NO_UPCALL 7 /* controls lento_upcall */
|
84 |
|
|
#define PSDEV_ERRORVAL 8 /* controls presto_debug_fail_blkdev */
|
85 |
|
|
#define PSDEV_EXCL_GID 9 /* which GID is ignored by presto */
|
86 |
|
|
#define PSDEV_BYTES_TO_CLOSE 11 /* bytes to write before close */
|
87 |
|
|
|
88 |
|
|
/* These are global presto control options */
|
89 |
|
|
#define PRESTO_PRIMARY_CTLCNT 2
|
90 |
|
|
static struct ctl_table presto_table[ PRESTO_PRIMARY_CTLCNT + MAX_CHANNEL + 1] =
|
91 |
|
|
{
|
92 |
|
|
{PSDEV_DEBUG, "debug", &presto_debug, sizeof(int), 0644, NULL, &proc_dointvec},
|
93 |
|
|
{PSDEV_TRACE, "trace", &presto_print_entry, sizeof(int), 0644, NULL, &proc_dointvec},
|
94 |
|
|
};
|
95 |
|
|
|
96 |
|
|
/*
|
97 |
|
|
* Intalling the sysctl entries: strategy
|
98 |
|
|
* - have templates for each /proc/sys/intermezzo/ entry
|
99 |
|
|
* such an entry exists for each /dev/presto
|
100 |
|
|
* (proto_channel_entry)
|
101 |
|
|
* - have a template for the contents of such directories
|
102 |
|
|
* (proto_psdev_table)
|
103 |
|
|
* - have the master table (presto_table)
|
104 |
|
|
*
|
105 |
|
|
* When installing, malloc, memcpy and fix up the pointers to point to
|
106 |
|
|
* the appropriate constants in izo_channels[your_minor]
|
107 |
|
|
*/
|
108 |
|
|
|
109 |
|
|
static ctl_table proto_psdev_table[] = {
|
110 |
|
|
{PSDEV_HARD, "hard", 0, sizeof(int), 0644, NULL, &proc_dointvec},
|
111 |
|
|
{PSDEV_NO_FILTER, "no_filter", 0, sizeof(int), 0644, NULL, &proc_dointvec},
|
112 |
|
|
{PSDEV_NO_JOURNAL, "no_journal", NULL, sizeof(int), 0644, NULL, &proc_dointvec},
|
113 |
|
|
{PSDEV_NO_UPCALL, "no_upcall", NULL, sizeof(int), 0644, NULL, &proc_dointvec},
|
114 |
|
|
{PSDEV_TIMEOUT, "timeout", NULL, sizeof(int), 0644, NULL, &proc_dointvec},
|
115 |
|
|
#ifdef PRESTO_DEBUG
|
116 |
|
|
{PSDEV_ERRORVAL, "errorval", NULL, sizeof(int), 0644, NULL, &proc_dointvec},
|
117 |
|
|
#endif
|
118 |
|
|
{ 0 }
|
119 |
|
|
};
|
120 |
|
|
|
121 |
|
|
static ctl_table proto_channel_entry = {
|
122 |
|
|
PSDEV_INTERMEZZO, 0, NULL, 0, 0555, 0,
|
123 |
|
|
};
|
124 |
|
|
|
125 |
|
|
static ctl_table intermezzo_table[2] = {
|
126 |
|
|
{PSDEV_INTERMEZZO, "intermezzo", NULL, 0, 0555, presto_table},
|
127 |
|
|
{0}
|
128 |
|
|
};
|
129 |
|
|
|
130 |
|
|
/* support for external setting and getting of opts. */
|
131 |
|
|
/* particularly via ioctl. The Right way to do this is via sysctl,
|
132 |
|
|
* but that will have to wait until intermezzo gets its own nice set of
|
133 |
|
|
* sysctl IDs
|
134 |
|
|
*/
|
135 |
|
|
/* we made these separate as setting may in future be more restricted
|
136 |
|
|
* than getting
|
137 |
|
|
*/
|
138 |
|
|
#ifdef RON_MINNICH
|
139 |
|
|
int dosetopt(int minor, struct psdev_opt *opt)
|
140 |
|
|
{
|
141 |
|
|
int retval = 0;
|
142 |
|
|
int newval = opt->optval;
|
143 |
|
|
|
144 |
|
|
ENTRY;
|
145 |
|
|
|
146 |
|
|
switch(opt->optname) {
|
147 |
|
|
|
148 |
|
|
case PSDEV_TIMEOUT:
|
149 |
|
|
izo_channels[minor].uc_timeout = newval;
|
150 |
|
|
break;
|
151 |
|
|
|
152 |
|
|
case PSDEV_HARD:
|
153 |
|
|
izo_channels[minor].uc_hard = newval;
|
154 |
|
|
break;
|
155 |
|
|
|
156 |
|
|
case PSDEV_NO_FILTER:
|
157 |
|
|
izo_channels[minor].uc_no_filter = newval;
|
158 |
|
|
break;
|
159 |
|
|
|
160 |
|
|
case PSDEV_NO_JOURNAL:
|
161 |
|
|
izo_channels[minor].uc_no_journal = newval;
|
162 |
|
|
break;
|
163 |
|
|
|
164 |
|
|
case PSDEV_NO_UPCALL:
|
165 |
|
|
izo_channels[minor].uc_no_upcall = newval;
|
166 |
|
|
break;
|
167 |
|
|
|
168 |
|
|
#ifdef PRESTO_DEBUG
|
169 |
|
|
case PSDEV_ERRORVAL: {
|
170 |
|
|
/* If we have a positive arg, set a breakpoint for that
|
171 |
|
|
* value. If we have a negative arg, make that device
|
172 |
|
|
* read-only. FIXME It would be much better to only
|
173 |
|
|
* allow setting the underlying device read-only for the
|
174 |
|
|
* current presto cache.
|
175 |
|
|
*/
|
176 |
|
|
int errorval = izo_channels[minor].uc_errorval;
|
177 |
|
|
if (errorval < 0) {
|
178 |
|
|
if (newval == 0)
|
179 |
|
|
set_device_ro(-errorval, 0);
|
180 |
|
|
else
|
181 |
|
|
CERROR("device %s already read only\n",
|
182 |
|
|
kdevname(-errorval));
|
183 |
|
|
} else {
|
184 |
|
|
if (newval < 0)
|
185 |
|
|
set_device_ro(-newval, 1);
|
186 |
|
|
izo_channels[minor].uc_errorval = newval;
|
187 |
|
|
CDEBUG(D_PSDEV, "setting errorval to %d\n", newval);
|
188 |
|
|
}
|
189 |
|
|
|
190 |
|
|
break;
|
191 |
|
|
}
|
192 |
|
|
#endif
|
193 |
|
|
|
194 |
|
|
case PSDEV_TRACE:
|
195 |
|
|
case PSDEV_DEBUG:
|
196 |
|
|
case PSDEV_BYTES_TO_CLOSE:
|
197 |
|
|
default:
|
198 |
|
|
CDEBUG(D_PSDEV,
|
199 |
|
|
"ioctl: dosetopt: minor %d, bad optname 0x%x, \n",
|
200 |
|
|
minor, opt->optname);
|
201 |
|
|
|
202 |
|
|
retval = -EINVAL;
|
203 |
|
|
}
|
204 |
|
|
|
205 |
|
|
EXIT;
|
206 |
|
|
return retval;
|
207 |
|
|
}
|
208 |
|
|
|
209 |
|
|
int dogetopt(int minor, struct psdev_opt *opt)
|
210 |
|
|
{
|
211 |
|
|
int retval = 0;
|
212 |
|
|
|
213 |
|
|
ENTRY;
|
214 |
|
|
|
215 |
|
|
switch(opt->optname) {
|
216 |
|
|
|
217 |
|
|
case PSDEV_TIMEOUT:
|
218 |
|
|
opt->optval = izo_channels[minor].uc_timeout;
|
219 |
|
|
break;
|
220 |
|
|
|
221 |
|
|
case PSDEV_HARD:
|
222 |
|
|
opt->optval = izo_channels[minor].uc_hard;
|
223 |
|
|
break;
|
224 |
|
|
|
225 |
|
|
case PSDEV_NO_FILTER:
|
226 |
|
|
opt->optval = izo_channels[minor].uc_no_filter;
|
227 |
|
|
break;
|
228 |
|
|
|
229 |
|
|
case PSDEV_NO_JOURNAL:
|
230 |
|
|
opt->optval = izo_channels[minor].uc_no_journal;
|
231 |
|
|
break;
|
232 |
|
|
|
233 |
|
|
case PSDEV_NO_UPCALL:
|
234 |
|
|
opt->optval = izo_channels[minor].uc_no_upcall;
|
235 |
|
|
break;
|
236 |
|
|
|
237 |
|
|
#ifdef PSDEV_DEBUG
|
238 |
|
|
case PSDEV_ERRORVAL: {
|
239 |
|
|
int errorval = izo_channels[minor].uc_errorval;
|
240 |
|
|
if (errorval < 0 && is_read_only(-errorval))
|
241 |
|
|
CERROR("device %s has been set read-only\n",
|
242 |
|
|
kdevname(-errorval));
|
243 |
|
|
opt->optval = izo_channels[minor].uc_errorval;
|
244 |
|
|
break;
|
245 |
|
|
}
|
246 |
|
|
#endif
|
247 |
|
|
|
248 |
|
|
case PSDEV_TRACE:
|
249 |
|
|
case PSDEV_DEBUG:
|
250 |
|
|
case PSDEV_BYTES_TO_CLOSE:
|
251 |
|
|
default:
|
252 |
|
|
CDEBUG(D_PSDEV,
|
253 |
|
|
"ioctl: dogetopt: minor %d, bad optval 0x%x, \n",
|
254 |
|
|
minor, opt->optname);
|
255 |
|
|
|
256 |
|
|
retval = -EINVAL;
|
257 |
|
|
}
|
258 |
|
|
|
259 |
|
|
EXIT;
|
260 |
|
|
return retval;
|
261 |
|
|
}
|
262 |
|
|
#endif
|
263 |
|
|
|
264 |
|
|
|
265 |
|
|
/* allocate the tables for the presto devices. We need
|
266 |
|
|
* sizeof(proto_channel_table)/sizeof(proto_channel_table[0])
|
267 |
|
|
* entries for each dev
|
268 |
|
|
*/
|
269 |
|
|
int /* __init */ init_intermezzo_sysctl(void)
|
270 |
|
|
{
|
271 |
|
|
int i;
|
272 |
|
|
int total_dev = MAX_CHANNEL;
|
273 |
|
|
int entries_per_dev = sizeof(proto_psdev_table) /
|
274 |
|
|
sizeof(proto_psdev_table[0]);
|
275 |
|
|
int total_entries = entries_per_dev * total_dev;
|
276 |
|
|
ctl_table *dev_ctl_table;
|
277 |
|
|
|
278 |
|
|
PRESTO_ALLOC(dev_ctl_table, sizeof(ctl_table) * total_entries);
|
279 |
|
|
|
280 |
|
|
if (! dev_ctl_table) {
|
281 |
|
|
CERROR("WARNING: presto couldn't allocate dev_ctl_table\n");
|
282 |
|
|
EXIT;
|
283 |
|
|
return -ENOMEM;
|
284 |
|
|
}
|
285 |
|
|
|
286 |
|
|
/* now fill in the entries ... we put the individual presto<x>
|
287 |
|
|
* entries at the end of the table, and the per-presto stuff
|
288 |
|
|
* starting at the front. We assume that the compiler makes
|
289 |
|
|
* this code more efficient, but really, who cares ... it
|
290 |
|
|
* happens once per reboot.
|
291 |
|
|
*/
|
292 |
|
|
for(i = 0; i < total_dev; i++) {
|
293 |
|
|
/* entry for this /proc/sys/intermezzo/intermezzo"i" */
|
294 |
|
|
ctl_table *psdev = &presto_table[i + PRESTO_PRIMARY_CTLCNT];
|
295 |
|
|
/* entries for the individual "files" in this "directory" */
|
296 |
|
|
ctl_table *psdev_entries = &dev_ctl_table[i * entries_per_dev];
|
297 |
|
|
/* init the psdev and psdev_entries with the prototypes */
|
298 |
|
|
*psdev = proto_channel_entry;
|
299 |
|
|
memcpy(psdev_entries, proto_psdev_table,
|
300 |
|
|
sizeof(proto_psdev_table));
|
301 |
|
|
/* now specialize them ... */
|
302 |
|
|
/* the psdev has to point to psdev_entries, and fix the number */
|
303 |
|
|
psdev->ctl_name = psdev->ctl_name + i + 1; /* sorry */
|
304 |
|
|
|
305 |
|
|
PRESTO_ALLOC((void*)psdev->procname, PROCNAME_SIZE);
|
306 |
|
|
if (!psdev->procname) {
|
307 |
|
|
PRESTO_FREE(dev_ctl_table,
|
308 |
|
|
sizeof(ctl_table) * total_entries);
|
309 |
|
|
return -ENOMEM;
|
310 |
|
|
}
|
311 |
|
|
sprintf((char *) psdev->procname, "intermezzo%d", i);
|
312 |
|
|
/* hook presto into */
|
313 |
|
|
psdev->child = psdev_entries;
|
314 |
|
|
|
315 |
|
|
/* now for each psdev entry ... */
|
316 |
|
|
psdev_entries[0].data = &(izo_channels[i].uc_hard);
|
317 |
|
|
psdev_entries[1].data = &(izo_channels[i].uc_no_filter);
|
318 |
|
|
psdev_entries[2].data = &(izo_channels[i].uc_no_journal);
|
319 |
|
|
psdev_entries[3].data = &(izo_channels[i].uc_no_upcall);
|
320 |
|
|
psdev_entries[4].data = &(izo_channels[i].uc_timeout);
|
321 |
|
|
#ifdef PRESTO_DEBUG
|
322 |
|
|
psdev_entries[5].data = &(izo_channels[i].uc_errorval);
|
323 |
|
|
#endif
|
324 |
|
|
}
|
325 |
|
|
|
326 |
|
|
|
327 |
|
|
#ifdef CONFIG_SYSCTL
|
328 |
|
|
if ( !intermezzo_table_header )
|
329 |
|
|
intermezzo_table_header =
|
330 |
|
|
register_sysctl_table(intermezzo_table, 0);
|
331 |
|
|
#endif
|
332 |
|
|
#ifdef CONFIG_PROC_FS
|
333 |
|
|
proc_fs_intermezzo = proc_mkdir("intermezzo", proc_root_fs);
|
334 |
|
|
proc_fs_intermezzo->owner = THIS_MODULE;
|
335 |
|
|
create_proc_info_entry("mounts", 0, proc_fs_intermezzo,
|
336 |
|
|
intermezzo_mount_get_info);
|
337 |
|
|
#endif
|
338 |
|
|
return 0;
|
339 |
|
|
}
|
340 |
|
|
|
341 |
|
|
void cleanup_intermezzo_sysctl(void)
|
342 |
|
|
{
|
343 |
|
|
int total_dev = MAX_CHANNEL;
|
344 |
|
|
int entries_per_dev = sizeof(proto_psdev_table) /
|
345 |
|
|
sizeof(proto_psdev_table[0]);
|
346 |
|
|
int total_entries = entries_per_dev * total_dev;
|
347 |
|
|
int i;
|
348 |
|
|
|
349 |
|
|
#ifdef CONFIG_SYSCTL
|
350 |
|
|
if ( intermezzo_table_header )
|
351 |
|
|
unregister_sysctl_table(intermezzo_table_header);
|
352 |
|
|
intermezzo_table_header = NULL;
|
353 |
|
|
#endif
|
354 |
|
|
for(i = 0; i < total_dev; i++) {
|
355 |
|
|
/* entry for this /proc/sys/intermezzo/intermezzo"i" */
|
356 |
|
|
ctl_table *psdev = &presto_table[i + PRESTO_PRIMARY_CTLCNT];
|
357 |
|
|
PRESTO_FREE(psdev->procname, PROCNAME_SIZE);
|
358 |
|
|
}
|
359 |
|
|
/* presto_table[PRESTO_PRIMARY_CTLCNT].child points to the
|
360 |
|
|
* dev_ctl_table previously allocated in init_intermezzo_psdev()
|
361 |
|
|
*/
|
362 |
|
|
PRESTO_FREE(presto_table[PRESTO_PRIMARY_CTLCNT].child, sizeof(ctl_table) * total_entries);
|
363 |
|
|
|
364 |
|
|
#if CONFIG_PROC_FS
|
365 |
|
|
remove_proc_entry("mounts", proc_fs_intermezzo);
|
366 |
|
|
remove_proc_entry("intermezzo", proc_root_fs);
|
367 |
|
|
#endif
|
368 |
|
|
}
|
369 |
|
|
|