Commit | Line | Data |
---|---|---|
8adf0536 JB |
1 | /* -*- mode: c; c-basic-offset: 8; -*- |
2 | * vim: noexpandtab sw=8 ts=8 sts=0: | |
3 | * | |
4 | * stack_user.c | |
5 | * | |
6 | * Code which interfaces ocfs2 with fs/dlm and a userspace stack. | |
7 | * | |
8 | * Copyright (C) 2007 Oracle. All rights reserved. | |
9 | * | |
10 | * This program is free software; you can redistribute it and/or | |
11 | * modify it under the terms of the GNU General Public | |
12 | * License as published by the Free Software Foundation, version 2. | |
13 | * | |
14 | * This program is distributed in the hope that it will be useful, | |
15 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
16 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
17 | * General Public License for more details. | |
18 | */ | |
19 | ||
20 | #include <linux/module.h> | |
6427a727 JB |
21 | #include <linux/fs.h> |
22 | #include <linux/miscdevice.h> | |
23 | #include <linux/mutex.h> | |
24 | #include <linux/reboot.h> | |
462c7e6a | 25 | #include <asm/uaccess.h> |
8adf0536 JB |
26 | |
27 | #include "stackglue.h" | |
28 | ||
29 | ||
6427a727 JB |
30 | /* |
31 | * The control protocol starts with a handshake. Until the handshake | |
32 | * is complete, the control device will fail all write(2)s. | |
33 | * | |
34 | * The handshake is simple. First, the client reads until EOF. Each line | |
35 | * of output is a supported protocol tag. All protocol tags are a single | |
36 | * character followed by a two hex digit version number. Currently the | |
37 | * only things supported is T01, for "Text-base version 0x01". Next, the | |
de870ef0 JB |
38 | * client writes the version they would like to use, including the newline. |
39 | * Thus, the protocol tag is 'T01\n'. If the version tag written is | |
40 | * unknown, -EINVAL is returned. Once the negotiation is complete, the | |
41 | * client can start sending messages. | |
42 | * | |
43 | * The T01 protocol only has one message, "DOWN". It has the following | |
44 | * syntax: | |
45 | * | |
46 | * DOWN<space><32-char-cap-hex-uuid><space><8-char-hex-nodenum><newline> | |
47 | * | |
48 | * eg: | |
49 | * | |
50 | * DOWN 632A924FDD844190BDA93C0DF6B94899 00000001\n | |
51 | * | |
52 | * This is 47 characters. | |
6427a727 JB |
53 | */ |
54 | ||
462c7e6a JB |
55 | /* |
56 | * Whether or not the client has done the handshake. | |
57 | * For now, we have just one protocol version. | |
58 | */ | |
59 | #define OCFS2_CONTROL_PROTO "T01\n" | |
60 | #define OCFS2_CONTROL_PROTO_LEN 4 | |
61 | #define OCFS2_CONTROL_HANDSHAKE_INVALID (0) | |
62 | #define OCFS2_CONTROL_HANDSHAKE_READ (1) | |
63 | #define OCFS2_CONTROL_HANDSHAKE_VALID (2) | |
de870ef0 JB |
64 | #define OCFS2_CONTROL_MESSAGE_DOWN "DOWN" |
65 | #define OCFS2_CONTROL_MESSAGE_DOWN_LEN 4 | |
66 | #define OCFS2_CONTROL_MESSAGE_DOWN_TOTAL_LEN 47 | |
67 | #define OCFS2_TEXT_UUID_LEN 32 | |
68 | #define OCFS2_CONTROL_MESSAGE_NODENUM_LEN 8 | |
462c7e6a | 69 | |
6427a727 JB |
70 | /* |
71 | * ocfs2_live_connection is refcounted because the filesystem and | |
72 | * miscdevice sides can detach in different order. Let's just be safe. | |
73 | */ | |
74 | struct ocfs2_live_connection { | |
75 | struct list_head oc_list; | |
76 | struct ocfs2_cluster_connection *oc_conn; | |
77 | }; | |
78 | ||
462c7e6a JB |
79 | struct ocfs2_control_private { |
80 | struct list_head op_list; | |
81 | int op_state; | |
82 | }; | |
83 | ||
6427a727 JB |
84 | static atomic_t ocfs2_control_opened; |
85 | ||
86 | static LIST_HEAD(ocfs2_live_connection_list); | |
462c7e6a | 87 | static LIST_HEAD(ocfs2_control_private_list); |
6427a727 JB |
88 | static DEFINE_MUTEX(ocfs2_control_lock); |
89 | ||
462c7e6a JB |
90 | static inline void ocfs2_control_set_handshake_state(struct file *file, |
91 | int state) | |
92 | { | |
93 | struct ocfs2_control_private *p = file->private_data; | |
94 | p->op_state = state; | |
95 | } | |
96 | ||
97 | static inline int ocfs2_control_get_handshake_state(struct file *file) | |
98 | { | |
99 | struct ocfs2_control_private *p = file->private_data; | |
100 | return p->op_state; | |
101 | } | |
102 | ||
6427a727 JB |
103 | static struct ocfs2_live_connection *ocfs2_connection_find(const char *name) |
104 | { | |
105 | size_t len = strlen(name); | |
106 | struct ocfs2_live_connection *c; | |
107 | ||
108 | BUG_ON(!mutex_is_locked(&ocfs2_control_lock)); | |
109 | ||
110 | list_for_each_entry(c, &ocfs2_live_connection_list, oc_list) { | |
111 | if ((c->oc_conn->cc_namelen == len) && | |
112 | !strncmp(c->oc_conn->cc_name, name, len)) | |
113 | return c; | |
114 | } | |
115 | ||
116 | return c; | |
117 | } | |
118 | ||
119 | /* | |
120 | * ocfs2_live_connection structures are created underneath the ocfs2 | |
121 | * mount path. Since the VFS prevents multiple calls to | |
122 | * fill_super(), we can't get dupes here. | |
123 | */ | |
124 | static int ocfs2_live_connection_new(struct ocfs2_cluster_connection *conn, | |
125 | struct ocfs2_live_connection **c_ret) | |
126 | { | |
127 | int rc = 0; | |
128 | struct ocfs2_live_connection *c; | |
129 | ||
130 | c = kzalloc(sizeof(struct ocfs2_live_connection), GFP_KERNEL); | |
131 | if (!c) | |
132 | return -ENOMEM; | |
133 | ||
134 | mutex_lock(&ocfs2_control_lock); | |
135 | c->oc_conn = conn; | |
136 | ||
137 | if (atomic_read(&ocfs2_control_opened)) | |
138 | list_add(&c->oc_list, &ocfs2_live_connection_list); | |
139 | else { | |
140 | printk(KERN_ERR | |
141 | "ocfs2: Userspace control daemon is not present\n"); | |
142 | rc = -ESRCH; | |
143 | } | |
144 | ||
145 | mutex_unlock(&ocfs2_control_lock); | |
146 | ||
147 | if (!rc) | |
148 | *c_ret = c; | |
149 | else | |
150 | kfree(c); | |
151 | ||
152 | return rc; | |
153 | } | |
154 | ||
155 | /* | |
156 | * This function disconnects the cluster connection from ocfs2_control. | |
157 | * Afterwards, userspace can't affect the cluster connection. | |
158 | */ | |
159 | static void ocfs2_live_connection_drop(struct ocfs2_live_connection *c) | |
160 | { | |
161 | mutex_lock(&ocfs2_control_lock); | |
162 | list_del_init(&c->oc_list); | |
163 | c->oc_conn = NULL; | |
164 | mutex_unlock(&ocfs2_control_lock); | |
165 | ||
166 | kfree(c); | |
167 | } | |
168 | ||
de870ef0 | 169 | static ssize_t ocfs2_control_cfu(void *target, size_t target_len, |
462c7e6a JB |
170 | const char __user *buf, size_t count) |
171 | { | |
172 | /* The T01 expects write(2) calls to have exactly one command */ | |
173 | if (count != target_len) | |
174 | return -EINVAL; | |
175 | ||
176 | if (copy_from_user(target, buf, target_len)) | |
177 | return -EFAULT; | |
178 | ||
179 | return count; | |
180 | } | |
181 | ||
182 | static ssize_t ocfs2_control_validate_handshake(struct file *file, | |
183 | const char __user *buf, | |
184 | size_t count) | |
185 | { | |
186 | ssize_t ret; | |
187 | char kbuf[OCFS2_CONTROL_PROTO_LEN]; | |
188 | ||
189 | ret = ocfs2_control_cfu(kbuf, OCFS2_CONTROL_PROTO_LEN, | |
190 | buf, count); | |
191 | if (ret != count) | |
192 | return ret; | |
193 | ||
194 | if (strncmp(kbuf, OCFS2_CONTROL_PROTO, OCFS2_CONTROL_PROTO_LEN)) | |
195 | return -EINVAL; | |
196 | ||
197 | atomic_inc(&ocfs2_control_opened); | |
198 | ocfs2_control_set_handshake_state(file, | |
199 | OCFS2_CONTROL_HANDSHAKE_VALID); | |
200 | ||
201 | ||
202 | return count; | |
203 | } | |
204 | ||
de870ef0 JB |
205 | static void ocfs2_control_send_down(const char *uuid, |
206 | int nodenum) | |
207 | { | |
208 | struct ocfs2_live_connection *c; | |
209 | ||
210 | mutex_lock(&ocfs2_control_lock); | |
211 | ||
212 | c = ocfs2_connection_find(uuid); | |
213 | if (c) { | |
214 | BUG_ON(c->oc_conn == NULL); | |
215 | c->oc_conn->cc_recovery_handler(nodenum, | |
216 | c->oc_conn->cc_recovery_data); | |
217 | } | |
218 | ||
219 | mutex_unlock(&ocfs2_control_lock); | |
220 | } | |
221 | ||
222 | /* DOWN<space><32-char-cap-hex-uuid><space><8-char-hex-nodenum><newline> */ | |
223 | struct ocfs2_control_message_down { | |
224 | char tag[OCFS2_CONTROL_MESSAGE_DOWN_LEN]; | |
225 | char space1; | |
226 | char uuid[OCFS2_TEXT_UUID_LEN]; | |
227 | char space2; | |
228 | char nodestr[OCFS2_CONTROL_MESSAGE_NODENUM_LEN]; | |
229 | char newline; | |
230 | }; | |
231 | ||
232 | static ssize_t ocfs2_control_message(struct file *file, | |
233 | const char __user *buf, | |
234 | size_t count) | |
235 | { | |
236 | ssize_t ret; | |
237 | char *p = NULL; | |
238 | long nodenum; | |
239 | struct ocfs2_control_message_down msg; | |
240 | ||
241 | /* Try to catch padding issues */ | |
242 | WARN_ON(offsetof(struct ocfs2_control_message_down, uuid) != | |
243 | (sizeof(msg.tag) + sizeof(msg.space1))); | |
244 | ||
245 | memset(&msg, 0, sizeof(struct ocfs2_control_message_down)); | |
246 | ret = ocfs2_control_cfu(&msg, OCFS2_CONTROL_MESSAGE_DOWN_TOTAL_LEN, | |
247 | buf, count); | |
248 | if (ret != count) | |
249 | return ret; | |
250 | ||
251 | if (strncmp(msg.tag, OCFS2_CONTROL_MESSAGE_DOWN, | |
252 | strlen(OCFS2_CONTROL_MESSAGE_DOWN))) | |
253 | return -EINVAL; | |
254 | ||
255 | if ((msg.space1 != ' ') || (msg.space2 != ' ') || | |
256 | (msg.newline != '\n')) | |
257 | return -EINVAL; | |
258 | msg.space1 = msg.space2 = msg.newline = '\0'; | |
259 | ||
260 | nodenum = simple_strtol(msg.nodestr, &p, 16); | |
261 | if (!p || *p) | |
262 | return -EINVAL; | |
263 | ||
264 | if ((nodenum == LONG_MIN) || (nodenum == LONG_MAX) || | |
265 | (nodenum > INT_MAX) || (nodenum < 0)) | |
266 | return -ERANGE; | |
267 | ||
268 | ocfs2_control_send_down(msg.uuid, nodenum); | |
269 | ||
270 | return count; | |
271 | } | |
6427a727 JB |
272 | |
273 | static ssize_t ocfs2_control_write(struct file *file, | |
274 | const char __user *buf, | |
275 | size_t count, | |
276 | loff_t *ppos) | |
8adf0536 | 277 | { |
462c7e6a JB |
278 | ssize_t ret; |
279 | ||
280 | switch (ocfs2_control_get_handshake_state(file)) { | |
281 | case OCFS2_CONTROL_HANDSHAKE_INVALID: | |
282 | ret = -EINVAL; | |
283 | break; | |
284 | ||
285 | case OCFS2_CONTROL_HANDSHAKE_READ: | |
286 | ret = ocfs2_control_validate_handshake(file, buf, | |
287 | count); | |
288 | break; | |
289 | ||
290 | case OCFS2_CONTROL_HANDSHAKE_VALID: | |
de870ef0 | 291 | ret = ocfs2_control_message(file, buf, count); |
462c7e6a JB |
292 | break; |
293 | ||
294 | default: | |
295 | BUG(); | |
296 | ret = -EIO; | |
297 | break; | |
298 | } | |
299 | ||
300 | return ret; | |
8adf0536 JB |
301 | } |
302 | ||
462c7e6a JB |
303 | /* |
304 | * This is a naive version. If we ever have a new protocol, we'll expand | |
305 | * it. Probably using seq_file. | |
306 | */ | |
6427a727 JB |
307 | static ssize_t ocfs2_control_read(struct file *file, |
308 | char __user *buf, | |
309 | size_t count, | |
310 | loff_t *ppos) | |
311 | { | |
462c7e6a JB |
312 | char *proto_string = OCFS2_CONTROL_PROTO; |
313 | size_t to_write = 0; | |
314 | ||
315 | if (*ppos >= OCFS2_CONTROL_PROTO_LEN) | |
316 | return 0; | |
317 | ||
318 | to_write = OCFS2_CONTROL_PROTO_LEN - *ppos; | |
319 | if (to_write > count) | |
320 | to_write = count; | |
321 | if (copy_to_user(buf, proto_string + *ppos, to_write)) | |
322 | return -EFAULT; | |
323 | ||
324 | *ppos += to_write; | |
325 | ||
326 | /* Have we read the whole protocol list? */ | |
327 | if (*ppos >= OCFS2_CONTROL_PROTO_LEN) | |
328 | ocfs2_control_set_handshake_state(file, | |
329 | OCFS2_CONTROL_HANDSHAKE_READ); | |
330 | ||
331 | return to_write; | |
6427a727 JB |
332 | } |
333 | ||
334 | static int ocfs2_control_release(struct inode *inode, struct file *file) | |
335 | { | |
462c7e6a JB |
336 | struct ocfs2_control_private *p = file->private_data; |
337 | ||
338 | mutex_lock(&ocfs2_control_lock); | |
339 | ||
340 | if (ocfs2_control_get_handshake_state(file) != | |
341 | OCFS2_CONTROL_HANDSHAKE_VALID) | |
342 | goto out; | |
343 | ||
6427a727 | 344 | if (atomic_dec_and_test(&ocfs2_control_opened)) { |
6427a727 JB |
345 | if (!list_empty(&ocfs2_live_connection_list)) { |
346 | /* XXX: Do bad things! */ | |
347 | printk(KERN_ERR | |
348 | "ocfs2: Unexpected release of ocfs2_control!\n" | |
349 | " Loss of cluster connection requires " | |
350 | "an emergency restart!\n"); | |
351 | emergency_restart(); | |
352 | } | |
6427a727 JB |
353 | } |
354 | ||
462c7e6a JB |
355 | out: |
356 | list_del_init(&p->op_list); | |
357 | file->private_data = NULL; | |
358 | ||
359 | mutex_unlock(&ocfs2_control_lock); | |
360 | ||
361 | kfree(p); | |
362 | ||
6427a727 JB |
363 | return 0; |
364 | } | |
365 | ||
366 | static int ocfs2_control_open(struct inode *inode, struct file *file) | |
367 | { | |
462c7e6a JB |
368 | struct ocfs2_control_private *p; |
369 | ||
370 | p = kzalloc(sizeof(struct ocfs2_control_private), GFP_KERNEL); | |
371 | if (!p) | |
372 | return -ENOMEM; | |
373 | ||
374 | mutex_lock(&ocfs2_control_lock); | |
375 | file->private_data = p; | |
376 | list_add(&p->op_list, &ocfs2_control_private_list); | |
377 | mutex_unlock(&ocfs2_control_lock); | |
6427a727 JB |
378 | |
379 | return 0; | |
380 | } | |
381 | ||
382 | static const struct file_operations ocfs2_control_fops = { | |
383 | .open = ocfs2_control_open, | |
384 | .release = ocfs2_control_release, | |
385 | .read = ocfs2_control_read, | |
386 | .write = ocfs2_control_write, | |
387 | .owner = THIS_MODULE, | |
388 | }; | |
389 | ||
390 | struct miscdevice ocfs2_control_device = { | |
391 | .minor = MISC_DYNAMIC_MINOR, | |
392 | .name = "ocfs2_control", | |
393 | .fops = &ocfs2_control_fops, | |
394 | }; | |
395 | ||
396 | static int ocfs2_control_init(void) | |
397 | { | |
398 | int rc; | |
399 | ||
400 | atomic_set(&ocfs2_control_opened, 0); | |
401 | ||
402 | rc = misc_register(&ocfs2_control_device); | |
403 | if (rc) | |
404 | printk(KERN_ERR | |
405 | "ocfs2: Unable to register ocfs2_control device " | |
406 | "(errno %d)\n", | |
407 | -rc); | |
408 | ||
409 | return rc; | |
410 | } | |
411 | ||
412 | static void ocfs2_control_exit(void) | |
413 | { | |
414 | int rc; | |
415 | ||
416 | rc = misc_deregister(&ocfs2_control_device); | |
417 | if (rc) | |
418 | printk(KERN_ERR | |
419 | "ocfs2: Unable to deregister ocfs2_control device " | |
420 | "(errno %d)\n", | |
421 | -rc); | |
422 | } | |
423 | ||
424 | static int __init user_stack_init(void) | |
425 | { | |
426 | return ocfs2_control_init(); | |
427 | } | |
428 | ||
8adf0536 JB |
429 | static void __exit user_stack_exit(void) |
430 | { | |
6427a727 | 431 | ocfs2_control_exit(); |
8adf0536 JB |
432 | } |
433 | ||
434 | MODULE_AUTHOR("Oracle"); | |
435 | MODULE_DESCRIPTION("ocfs2 driver for userspace cluster stacks"); | |
436 | MODULE_LICENSE("GPL"); | |
437 | module_init(user_stack_init); | |
438 | module_exit(user_stack_exit); |