Commit | Line | Data |
---|---|---|
597d0cae | 1 | /* |
0f8e0d9a | 2 | * Copyright (C) 2006-2008 Red Hat, Inc. All rights reserved. |
597d0cae DT |
3 | * |
4 | * This copyrighted material is made available to anyone wishing to use, | |
5 | * modify, copy, or redistribute it subject to the terms and conditions | |
6 | * of the GNU General Public License v.2. | |
7 | */ | |
8 | ||
9 | #include <linux/miscdevice.h> | |
10 | #include <linux/init.h> | |
11 | #include <linux/wait.h> | |
12 | #include <linux/module.h> | |
13 | #include <linux/file.h> | |
14 | #include <linux/fs.h> | |
15 | #include <linux/poll.h> | |
16 | #include <linux/signal.h> | |
17 | #include <linux/spinlock.h> | |
18 | #include <linux/dlm.h> | |
19 | #include <linux/dlm_device.h> | |
20 | ||
21 | #include "dlm_internal.h" | |
22 | #include "lockspace.h" | |
23 | #include "lock.h" | |
24 | #include "lvb_table.h" | |
84c6e8cd | 25 | #include "user.h" |
597d0cae | 26 | |
0fe410d3 | 27 | static const char name_prefix[] = "dlm"; |
00977a59 | 28 | static const struct file_operations device_fops; |
dc68c7ed DT |
29 | static atomic_t dlm_monitor_opened; |
30 | static int dlm_monitor_unused = 1; | |
597d0cae DT |
31 | |
32 | #ifdef CONFIG_COMPAT | |
33 | ||
34 | struct dlm_lock_params32 { | |
35 | __u8 mode; | |
36 | __u8 namelen; | |
d7db923e DT |
37 | __u16 unused; |
38 | __u32 flags; | |
597d0cae DT |
39 | __u32 lkid; |
40 | __u32 parent; | |
d7db923e DT |
41 | __u64 xid; |
42 | __u64 timeout; | |
597d0cae DT |
43 | __u32 castparam; |
44 | __u32 castaddr; | |
45 | __u32 bastparam; | |
46 | __u32 bastaddr; | |
47 | __u32 lksb; | |
597d0cae DT |
48 | char lvb[DLM_USER_LVB_LEN]; |
49 | char name[0]; | |
50 | }; | |
51 | ||
52 | struct dlm_write_request32 { | |
53 | __u32 version[3]; | |
54 | __u8 cmd; | |
55 | __u8 is64bit; | |
56 | __u8 unused[2]; | |
57 | ||
58 | union { | |
59 | struct dlm_lock_params32 lock; | |
60 | struct dlm_lspace_params lspace; | |
72c2be77 | 61 | struct dlm_purge_params purge; |
597d0cae DT |
62 | } i; |
63 | }; | |
64 | ||
65 | struct dlm_lksb32 { | |
66 | __u32 sb_status; | |
67 | __u32 sb_lkid; | |
68 | __u8 sb_flags; | |
69 | __u32 sb_lvbptr; | |
70 | }; | |
71 | ||
72 | struct dlm_lock_result32 { | |
d7db923e | 73 | __u32 version[3]; |
597d0cae DT |
74 | __u32 length; |
75 | __u32 user_astaddr; | |
76 | __u32 user_astparam; | |
77 | __u32 user_lksb; | |
78 | struct dlm_lksb32 lksb; | |
79 | __u8 bast_mode; | |
80 | __u8 unused[3]; | |
81 | /* Offsets may be zero if no data is present */ | |
82 | __u32 lvb_offset; | |
83 | }; | |
84 | ||
85 | static void compat_input(struct dlm_write_request *kb, | |
2a79289e | 86 | struct dlm_write_request32 *kb32, |
cb79f199 | 87 | size_t count) |
597d0cae DT |
88 | { |
89 | kb->version[0] = kb32->version[0]; | |
90 | kb->version[1] = kb32->version[1]; | |
91 | kb->version[2] = kb32->version[2]; | |
92 | ||
93 | kb->cmd = kb32->cmd; | |
94 | kb->is64bit = kb32->is64bit; | |
95 | if (kb->cmd == DLM_USER_CREATE_LOCKSPACE || | |
96 | kb->cmd == DLM_USER_REMOVE_LOCKSPACE) { | |
97 | kb->i.lspace.flags = kb32->i.lspace.flags; | |
98 | kb->i.lspace.minor = kb32->i.lspace.minor; | |
cb79f199 AV |
99 | memcpy(kb->i.lspace.name, kb32->i.lspace.name, count - |
100 | offsetof(struct dlm_write_request32, i.lspace.name)); | |
72c2be77 DT |
101 | } else if (kb->cmd == DLM_USER_PURGE) { |
102 | kb->i.purge.nodeid = kb32->i.purge.nodeid; | |
103 | kb->i.purge.pid = kb32->i.purge.pid; | |
597d0cae DT |
104 | } else { |
105 | kb->i.lock.mode = kb32->i.lock.mode; | |
106 | kb->i.lock.namelen = kb32->i.lock.namelen; | |
107 | kb->i.lock.flags = kb32->i.lock.flags; | |
108 | kb->i.lock.lkid = kb32->i.lock.lkid; | |
109 | kb->i.lock.parent = kb32->i.lock.parent; | |
d7db923e DT |
110 | kb->i.lock.xid = kb32->i.lock.xid; |
111 | kb->i.lock.timeout = kb32->i.lock.timeout; | |
597d0cae DT |
112 | kb->i.lock.castparam = (void *)(long)kb32->i.lock.castparam; |
113 | kb->i.lock.castaddr = (void *)(long)kb32->i.lock.castaddr; | |
114 | kb->i.lock.bastparam = (void *)(long)kb32->i.lock.bastparam; | |
115 | kb->i.lock.bastaddr = (void *)(long)kb32->i.lock.bastaddr; | |
116 | kb->i.lock.lksb = (void *)(long)kb32->i.lock.lksb; | |
117 | memcpy(kb->i.lock.lvb, kb32->i.lock.lvb, DLM_USER_LVB_LEN); | |
cb79f199 AV |
118 | memcpy(kb->i.lock.name, kb32->i.lock.name, count - |
119 | offsetof(struct dlm_write_request32, i.lock.name)); | |
597d0cae DT |
120 | } |
121 | } | |
122 | ||
123 | static void compat_output(struct dlm_lock_result *res, | |
124 | struct dlm_lock_result32 *res32) | |
125 | { | |
d7db923e DT |
126 | res32->version[0] = res->version[0]; |
127 | res32->version[1] = res->version[1]; | |
128 | res32->version[2] = res->version[2]; | |
129 | ||
597d0cae DT |
130 | res32->user_astaddr = (__u32)(long)res->user_astaddr; |
131 | res32->user_astparam = (__u32)(long)res->user_astparam; | |
132 | res32->user_lksb = (__u32)(long)res->user_lksb; | |
133 | res32->bast_mode = res->bast_mode; | |
134 | ||
135 | res32->lvb_offset = res->lvb_offset; | |
136 | res32->length = res->length; | |
137 | ||
138 | res32->lksb.sb_status = res->lksb.sb_status; | |
139 | res32->lksb.sb_flags = res->lksb.sb_flags; | |
140 | res32->lksb.sb_lkid = res->lksb.sb_lkid; | |
141 | res32->lksb.sb_lvbptr = (__u32)(long)res->lksb.sb_lvbptr; | |
142 | } | |
143 | #endif | |
144 | ||
84d8cd69 DT |
145 | /* Figure out if this lock is at the end of its life and no longer |
146 | available for the application to use. The lkb still exists until | |
147 | the final ast is read. A lock becomes EOL in three situations: | |
148 | 1. a noqueue request fails with EAGAIN | |
149 | 2. an unlock completes with EUNLOCK | |
150 | 3. a cancel of a waiting request completes with ECANCEL/EDEADLK | |
151 | An EOL lock needs to be removed from the process's list of locks. | |
152 | And we can't allow any new operation on an EOL lock. This is | |
153 | not related to the lifetime of the lkb struct which is managed | |
154 | entirely by refcount. */ | |
155 | ||
156 | static int lkb_is_endoflife(struct dlm_lkb *lkb, int sb_status, int type) | |
157 | { | |
158 | switch (sb_status) { | |
159 | case -DLM_EUNLOCK: | |
160 | return 1; | |
161 | case -DLM_ECANCEL: | |
162 | case -ETIMEDOUT: | |
8b4021fa | 163 | case -EDEADLK: |
84d8cd69 DT |
164 | if (lkb->lkb_grmode == DLM_LOCK_IV) |
165 | return 1; | |
166 | break; | |
167 | case -EAGAIN: | |
168 | if (type == AST_COMP && lkb->lkb_grmode == DLM_LOCK_IV) | |
169 | return 1; | |
170 | break; | |
171 | } | |
172 | return 0; | |
173 | } | |
174 | ||
ef0c2bb0 DT |
175 | /* we could possibly check if the cancel of an orphan has resulted in the lkb |
176 | being removed and then remove that lkb from the orphans list and free it */ | |
597d0cae | 177 | |
fd22a51b | 178 | void dlm_user_add_ast(struct dlm_lkb *lkb, int type, int bastmode) |
597d0cae DT |
179 | { |
180 | struct dlm_ls *ls; | |
181 | struct dlm_user_args *ua; | |
182 | struct dlm_user_proc *proc; | |
ef0c2bb0 | 183 | int eol = 0, ast_type; |
597d0cae | 184 | |
ef0c2bb0 | 185 | if (lkb->lkb_flags & (DLM_IFL_ORPHAN | DLM_IFL_DEAD)) |
597d0cae | 186 | return; |
597d0cae DT |
187 | |
188 | ls = lkb->lkb_resource->res_ls; | |
189 | mutex_lock(&ls->ls_clear_proc_locks); | |
190 | ||
191 | /* If ORPHAN/DEAD flag is set, it means the process is dead so an ast | |
192 | can't be delivered. For ORPHAN's, dlm_clear_proc_locks() freed | |
ef0c2bb0 DT |
193 | lkb->ua so we can't try to use it. This second check is necessary |
194 | for cases where a completion ast is received for an operation that | |
195 | began before clear_proc_locks did its cancel/unlock. */ | |
597d0cae | 196 | |
ef0c2bb0 | 197 | if (lkb->lkb_flags & (DLM_IFL_ORPHAN | DLM_IFL_DEAD)) |
597d0cae | 198 | goto out; |
597d0cae | 199 | |
d292c0cc DT |
200 | DLM_ASSERT(lkb->lkb_ua, dlm_print_lkb(lkb);); |
201 | ua = lkb->lkb_ua; | |
597d0cae DT |
202 | proc = ua->proc; |
203 | ||
204 | if (type == AST_BAST && ua->bastaddr == NULL) | |
205 | goto out; | |
206 | ||
207 | spin_lock(&proc->asts_spin); | |
ef0c2bb0 DT |
208 | |
209 | ast_type = lkb->lkb_ast_type; | |
210 | lkb->lkb_ast_type |= type; | |
fd22a51b DT |
211 | if (bastmode) |
212 | lkb->lkb_bastmode = bastmode; | |
ef0c2bb0 DT |
213 | |
214 | if (!ast_type) { | |
597d0cae DT |
215 | kref_get(&lkb->lkb_ref); |
216 | list_add_tail(&lkb->lkb_astqueue, &proc->asts); | |
597d0cae DT |
217 | wake_up_interruptible(&proc->wait); |
218 | } | |
ef0c2bb0 DT |
219 | if (type == AST_COMP && (ast_type & AST_COMP)) |
220 | log_debug(ls, "ast overlap %x status %x %x", | |
221 | lkb->lkb_id, ua->lksb.sb_status, lkb->lkb_flags); | |
222 | ||
84d8cd69 | 223 | eol = lkb_is_endoflife(lkb, ua->lksb.sb_status, type); |
ef0c2bb0 DT |
224 | if (eol) { |
225 | lkb->lkb_ast_type &= ~AST_BAST; | |
226 | lkb->lkb_flags |= DLM_IFL_ENDOFLIFE; | |
227 | } | |
a1bc86e6 | 228 | |
597d0cae DT |
229 | /* We want to copy the lvb to userspace when the completion |
230 | ast is read if the status is 0, the lock has an lvb and | |
231 | lvb_ops says we should. We could probably have set_lvb_lock() | |
232 | set update_user_lvb instead and not need old_mode */ | |
233 | ||
234 | if ((lkb->lkb_ast_type & AST_COMP) && | |
235 | (lkb->lkb_lksb->sb_status == 0) && | |
236 | lkb->lkb_lksb->sb_lvbptr && | |
237 | dlm_lvb_operations[ua->old_mode + 1][lkb->lkb_grmode + 1]) | |
238 | ua->update_user_lvb = 1; | |
239 | else | |
240 | ua->update_user_lvb = 0; | |
241 | ||
242 | spin_unlock(&proc->asts_spin); | |
34e22bed | 243 | |
ef0c2bb0 | 244 | if (eol) { |
ce5246b9 | 245 | spin_lock(&proc->locks_spin); |
ef0c2bb0 DT |
246 | if (!list_empty(&lkb->lkb_ownqueue)) { |
247 | list_del_init(&lkb->lkb_ownqueue); | |
248 | dlm_put_lkb(lkb); | |
249 | } | |
ce5246b9 | 250 | spin_unlock(&proc->locks_spin); |
34e22bed | 251 | } |
597d0cae DT |
252 | out: |
253 | mutex_unlock(&ls->ls_clear_proc_locks); | |
254 | } | |
255 | ||
256 | static int device_user_lock(struct dlm_user_proc *proc, | |
257 | struct dlm_lock_params *params) | |
258 | { | |
259 | struct dlm_ls *ls; | |
260 | struct dlm_user_args *ua; | |
261 | int error = -ENOMEM; | |
262 | ||
263 | ls = dlm_find_lockspace_local(proc->lockspace); | |
264 | if (!ls) | |
265 | return -ENOENT; | |
266 | ||
267 | if (!params->castaddr || !params->lksb) { | |
268 | error = -EINVAL; | |
269 | goto out; | |
270 | } | |
271 | ||
272 | ua = kzalloc(sizeof(struct dlm_user_args), GFP_KERNEL); | |
273 | if (!ua) | |
274 | goto out; | |
275 | ua->proc = proc; | |
276 | ua->user_lksb = params->lksb; | |
277 | ua->castparam = params->castparam; | |
278 | ua->castaddr = params->castaddr; | |
279 | ua->bastparam = params->bastparam; | |
280 | ua->bastaddr = params->bastaddr; | |
d7db923e | 281 | ua->xid = params->xid; |
597d0cae DT |
282 | |
283 | if (params->flags & DLM_LKF_CONVERT) | |
284 | error = dlm_user_convert(ls, ua, | |
285 | params->mode, params->flags, | |
d7db923e DT |
286 | params->lkid, params->lvb, |
287 | (unsigned long) params->timeout); | |
597d0cae DT |
288 | else { |
289 | error = dlm_user_request(ls, ua, | |
290 | params->mode, params->flags, | |
291 | params->name, params->namelen, | |
d7db923e | 292 | (unsigned long) params->timeout); |
597d0cae DT |
293 | if (!error) |
294 | error = ua->lksb.sb_lkid; | |
295 | } | |
296 | out: | |
297 | dlm_put_lockspace(ls); | |
298 | return error; | |
299 | } | |
300 | ||
301 | static int device_user_unlock(struct dlm_user_proc *proc, | |
302 | struct dlm_lock_params *params) | |
303 | { | |
304 | struct dlm_ls *ls; | |
305 | struct dlm_user_args *ua; | |
306 | int error = -ENOMEM; | |
307 | ||
308 | ls = dlm_find_lockspace_local(proc->lockspace); | |
309 | if (!ls) | |
310 | return -ENOENT; | |
311 | ||
312 | ua = kzalloc(sizeof(struct dlm_user_args), GFP_KERNEL); | |
313 | if (!ua) | |
314 | goto out; | |
315 | ua->proc = proc; | |
316 | ua->user_lksb = params->lksb; | |
317 | ua->castparam = params->castparam; | |
318 | ua->castaddr = params->castaddr; | |
319 | ||
320 | if (params->flags & DLM_LKF_CANCEL) | |
321 | error = dlm_user_cancel(ls, ua, params->flags, params->lkid); | |
322 | else | |
323 | error = dlm_user_unlock(ls, ua, params->flags, params->lkid, | |
324 | params->lvb); | |
325 | out: | |
326 | dlm_put_lockspace(ls); | |
327 | return error; | |
328 | } | |
329 | ||
8b4021fa DT |
330 | static int device_user_deadlock(struct dlm_user_proc *proc, |
331 | struct dlm_lock_params *params) | |
332 | { | |
333 | struct dlm_ls *ls; | |
334 | int error; | |
335 | ||
336 | ls = dlm_find_lockspace_local(proc->lockspace); | |
337 | if (!ls) | |
338 | return -ENOENT; | |
339 | ||
340 | error = dlm_user_deadlock(ls, params->flags, params->lkid); | |
341 | ||
342 | dlm_put_lockspace(ls); | |
343 | return error; | |
344 | } | |
345 | ||
0f8e0d9a | 346 | static int dlm_device_register(struct dlm_ls *ls, char *name) |
254da030 PC |
347 | { |
348 | int error, len; | |
349 | ||
0f8e0d9a DT |
350 | /* The device is already registered. This happens when the |
351 | lockspace is created multiple times from userspace. */ | |
352 | if (ls->ls_device.name) | |
353 | return 0; | |
354 | ||
254da030 PC |
355 | error = -ENOMEM; |
356 | len = strlen(name) + strlen(name_prefix) + 2; | |
357 | ls->ls_device.name = kzalloc(len, GFP_KERNEL); | |
358 | if (!ls->ls_device.name) | |
359 | goto fail; | |
360 | ||
361 | snprintf((char *)ls->ls_device.name, len, "%s_%s", name_prefix, | |
362 | name); | |
363 | ls->ls_device.fops = &device_fops; | |
364 | ls->ls_device.minor = MISC_DYNAMIC_MINOR; | |
365 | ||
366 | error = misc_register(&ls->ls_device); | |
367 | if (error) { | |
368 | kfree(ls->ls_device.name); | |
369 | } | |
370 | fail: | |
371 | return error; | |
372 | } | |
373 | ||
0f8e0d9a DT |
374 | int dlm_device_deregister(struct dlm_ls *ls) |
375 | { | |
376 | int error; | |
377 | ||
378 | /* The device is not registered. This happens when the lockspace | |
379 | was never used from userspace, or when device_create_lockspace() | |
380 | calls dlm_release_lockspace() after the register fails. */ | |
381 | if (!ls->ls_device.name) | |
382 | return 0; | |
383 | ||
384 | error = misc_deregister(&ls->ls_device); | |
385 | if (!error) | |
386 | kfree(ls->ls_device.name); | |
387 | return error; | |
388 | } | |
389 | ||
72c2be77 DT |
390 | static int device_user_purge(struct dlm_user_proc *proc, |
391 | struct dlm_purge_params *params) | |
392 | { | |
393 | struct dlm_ls *ls; | |
394 | int error; | |
395 | ||
396 | ls = dlm_find_lockspace_local(proc->lockspace); | |
397 | if (!ls) | |
398 | return -ENOENT; | |
399 | ||
400 | error = dlm_user_purge(ls, proc, params->nodeid, params->pid); | |
401 | ||
402 | dlm_put_lockspace(ls); | |
403 | return error; | |
404 | } | |
405 | ||
597d0cae DT |
406 | static int device_create_lockspace(struct dlm_lspace_params *params) |
407 | { | |
408 | dlm_lockspace_t *lockspace; | |
409 | struct dlm_ls *ls; | |
254da030 | 410 | int error; |
597d0cae DT |
411 | |
412 | if (!capable(CAP_SYS_ADMIN)) | |
413 | return -EPERM; | |
414 | ||
415 | error = dlm_new_lockspace(params->name, strlen(params->name), | |
3ae1acf9 | 416 | &lockspace, params->flags, DLM_USER_LVB_LEN); |
597d0cae DT |
417 | if (error) |
418 | return error; | |
419 | ||
420 | ls = dlm_find_lockspace_local(lockspace); | |
421 | if (!ls) | |
422 | return -ENOENT; | |
423 | ||
0f8e0d9a | 424 | error = dlm_device_register(ls, params->name); |
597d0cae | 425 | dlm_put_lockspace(ls); |
597d0cae | 426 | |
254da030 PC |
427 | if (error) |
428 | dlm_release_lockspace(lockspace, 0); | |
429 | else | |
430 | error = ls->ls_device.minor; | |
431 | ||
597d0cae DT |
432 | return error; |
433 | } | |
434 | ||
435 | static int device_remove_lockspace(struct dlm_lspace_params *params) | |
436 | { | |
437 | dlm_lockspace_t *lockspace; | |
438 | struct dlm_ls *ls; | |
c6e6f0ba | 439 | int error, force = 0; |
597d0cae DT |
440 | |
441 | if (!capable(CAP_SYS_ADMIN)) | |
442 | return -EPERM; | |
443 | ||
444 | ls = dlm_find_lockspace_device(params->minor); | |
445 | if (!ls) | |
446 | return -ENOENT; | |
447 | ||
c6e6f0ba DT |
448 | if (params->flags & DLM_USER_LSFLG_FORCEFREE) |
449 | force = 2; | |
450 | ||
597d0cae | 451 | lockspace = ls->ls_local_handle; |
0f8e0d9a | 452 | dlm_put_lockspace(ls); |
597d0cae | 453 | |
0f8e0d9a DT |
454 | /* The final dlm_release_lockspace waits for references to go to |
455 | zero, so all processes will need to close their device for the | |
456 | ls before the release will proceed. release also calls the | |
457 | device_deregister above. Converting a positive return value | |
458 | from release to zero means that userspace won't know when its | |
459 | release was the final one, but it shouldn't need to know. */ | |
597d0cae | 460 | |
c6e6f0ba | 461 | error = dlm_release_lockspace(lockspace, force); |
0f8e0d9a DT |
462 | if (error > 0) |
463 | error = 0; | |
597d0cae DT |
464 | return error; |
465 | } | |
466 | ||
467 | /* Check the user's version matches ours */ | |
468 | static int check_version(struct dlm_write_request *req) | |
469 | { | |
470 | if (req->version[0] != DLM_DEVICE_VERSION_MAJOR || | |
471 | (req->version[0] == DLM_DEVICE_VERSION_MAJOR && | |
472 | req->version[1] > DLM_DEVICE_VERSION_MINOR)) { | |
473 | ||
474 | printk(KERN_DEBUG "dlm: process %s (%d) version mismatch " | |
475 | "user (%d.%d.%d) kernel (%d.%d.%d)\n", | |
476 | current->comm, | |
ba25f9dc | 477 | task_pid_nr(current), |
597d0cae DT |
478 | req->version[0], |
479 | req->version[1], | |
480 | req->version[2], | |
481 | DLM_DEVICE_VERSION_MAJOR, | |
482 | DLM_DEVICE_VERSION_MINOR, | |
483 | DLM_DEVICE_VERSION_PATCH); | |
484 | return -EINVAL; | |
485 | } | |
486 | return 0; | |
487 | } | |
488 | ||
489 | /* | |
490 | * device_write | |
491 | * | |
492 | * device_user_lock | |
493 | * dlm_user_request -> request_lock | |
494 | * dlm_user_convert -> convert_lock | |
495 | * | |
496 | * device_user_unlock | |
497 | * dlm_user_unlock -> unlock_lock | |
498 | * dlm_user_cancel -> cancel_lock | |
499 | * | |
500 | * device_create_lockspace | |
501 | * dlm_new_lockspace | |
502 | * | |
503 | * device_remove_lockspace | |
504 | * dlm_release_lockspace | |
505 | */ | |
506 | ||
507 | /* a write to a lockspace device is a lock or unlock request, a write | |
508 | to the control device is to create/remove a lockspace */ | |
509 | ||
510 | static ssize_t device_write(struct file *file, const char __user *buf, | |
511 | size_t count, loff_t *ppos) | |
512 | { | |
513 | struct dlm_user_proc *proc = file->private_data; | |
514 | struct dlm_write_request *kbuf; | |
515 | sigset_t tmpsig, allsigs; | |
516 | int error; | |
517 | ||
518 | #ifdef CONFIG_COMPAT | |
519 | if (count < sizeof(struct dlm_write_request32)) | |
520 | #else | |
521 | if (count < sizeof(struct dlm_write_request)) | |
522 | #endif | |
523 | return -EINVAL; | |
524 | ||
cb79f199 | 525 | kbuf = kzalloc(count + 1, GFP_KERNEL); |
597d0cae DT |
526 | if (!kbuf) |
527 | return -ENOMEM; | |
528 | ||
529 | if (copy_from_user(kbuf, buf, count)) { | |
530 | error = -EFAULT; | |
531 | goto out_free; | |
532 | } | |
533 | ||
534 | if (check_version(kbuf)) { | |
535 | error = -EBADE; | |
536 | goto out_free; | |
537 | } | |
538 | ||
539 | #ifdef CONFIG_COMPAT | |
540 | if (!kbuf->is64bit) { | |
541 | struct dlm_write_request32 *k32buf; | |
542 | k32buf = (struct dlm_write_request32 *)kbuf; | |
cb79f199 | 543 | kbuf = kmalloc(count + 1 + (sizeof(struct dlm_write_request) - |
597d0cae | 544 | sizeof(struct dlm_write_request32)), GFP_KERNEL); |
cb980d9a DT |
545 | if (!kbuf) { |
546 | kfree(k32buf); | |
597d0cae | 547 | return -ENOMEM; |
cb980d9a | 548 | } |
597d0cae DT |
549 | |
550 | if (proc) | |
551 | set_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags); | |
cb79f199 | 552 | compat_input(kbuf, k32buf, count + 1); |
597d0cae DT |
553 | kfree(k32buf); |
554 | } | |
555 | #endif | |
556 | ||
557 | /* do we really need this? can a write happen after a close? */ | |
558 | if ((kbuf->cmd == DLM_USER_LOCK || kbuf->cmd == DLM_USER_UNLOCK) && | |
cb980d9a DT |
559 | (proc && test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))) { |
560 | error = -EINVAL; | |
561 | goto out_free; | |
562 | } | |
597d0cae DT |
563 | |
564 | sigfillset(&allsigs); | |
565 | sigprocmask(SIG_BLOCK, &allsigs, &tmpsig); | |
566 | ||
567 | error = -EINVAL; | |
568 | ||
569 | switch (kbuf->cmd) | |
570 | { | |
571 | case DLM_USER_LOCK: | |
572 | if (!proc) { | |
573 | log_print("no locking on control device"); | |
574 | goto out_sig; | |
575 | } | |
576 | error = device_user_lock(proc, &kbuf->i.lock); | |
577 | break; | |
578 | ||
579 | case DLM_USER_UNLOCK: | |
580 | if (!proc) { | |
581 | log_print("no locking on control device"); | |
582 | goto out_sig; | |
583 | } | |
584 | error = device_user_unlock(proc, &kbuf->i.lock); | |
585 | break; | |
586 | ||
8b4021fa DT |
587 | case DLM_USER_DEADLOCK: |
588 | if (!proc) { | |
589 | log_print("no locking on control device"); | |
590 | goto out_sig; | |
591 | } | |
592 | error = device_user_deadlock(proc, &kbuf->i.lock); | |
593 | break; | |
594 | ||
597d0cae DT |
595 | case DLM_USER_CREATE_LOCKSPACE: |
596 | if (proc) { | |
597 | log_print("create/remove only on control device"); | |
598 | goto out_sig; | |
599 | } | |
600 | error = device_create_lockspace(&kbuf->i.lspace); | |
601 | break; | |
602 | ||
603 | case DLM_USER_REMOVE_LOCKSPACE: | |
604 | if (proc) { | |
605 | log_print("create/remove only on control device"); | |
606 | goto out_sig; | |
607 | } | |
608 | error = device_remove_lockspace(&kbuf->i.lspace); | |
609 | break; | |
610 | ||
72c2be77 DT |
611 | case DLM_USER_PURGE: |
612 | if (!proc) { | |
613 | log_print("no locking on control device"); | |
614 | goto out_sig; | |
615 | } | |
616 | error = device_user_purge(proc, &kbuf->i.purge); | |
617 | break; | |
618 | ||
597d0cae DT |
619 | default: |
620 | log_print("Unknown command passed to DLM device : %d\n", | |
621 | kbuf->cmd); | |
622 | } | |
623 | ||
624 | out_sig: | |
625 | sigprocmask(SIG_SETMASK, &tmpsig, NULL); | |
626 | recalc_sigpending(); | |
627 | out_free: | |
628 | kfree(kbuf); | |
629 | return error; | |
630 | } | |
631 | ||
632 | /* Every process that opens the lockspace device has its own "proc" structure | |
633 | hanging off the open file that's used to keep track of locks owned by the | |
634 | process and asts that need to be delivered to the process. */ | |
635 | ||
636 | static int device_open(struct inode *inode, struct file *file) | |
637 | { | |
638 | struct dlm_user_proc *proc; | |
639 | struct dlm_ls *ls; | |
640 | ||
641 | ls = dlm_find_lockspace_device(iminor(inode)); | |
f9f2ed48 | 642 | if (!ls) |
597d0cae DT |
643 | return -ENOENT; |
644 | ||
645 | proc = kzalloc(sizeof(struct dlm_user_proc), GFP_KERNEL); | |
646 | if (!proc) { | |
647 | dlm_put_lockspace(ls); | |
648 | return -ENOMEM; | |
649 | } | |
650 | ||
651 | proc->lockspace = ls->ls_local_handle; | |
652 | INIT_LIST_HEAD(&proc->asts); | |
653 | INIT_LIST_HEAD(&proc->locks); | |
a1bc86e6 | 654 | INIT_LIST_HEAD(&proc->unlocking); |
597d0cae DT |
655 | spin_lock_init(&proc->asts_spin); |
656 | spin_lock_init(&proc->locks_spin); | |
657 | init_waitqueue_head(&proc->wait); | |
658 | file->private_data = proc; | |
659 | ||
660 | return 0; | |
661 | } | |
662 | ||
663 | static int device_close(struct inode *inode, struct file *file) | |
664 | { | |
665 | struct dlm_user_proc *proc = file->private_data; | |
666 | struct dlm_ls *ls; | |
667 | sigset_t tmpsig, allsigs; | |
668 | ||
669 | ls = dlm_find_lockspace_local(proc->lockspace); | |
670 | if (!ls) | |
671 | return -ENOENT; | |
672 | ||
673 | sigfillset(&allsigs); | |
674 | sigprocmask(SIG_BLOCK, &allsigs, &tmpsig); | |
675 | ||
676 | set_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags); | |
677 | ||
678 | dlm_clear_proc_locks(ls, proc); | |
679 | ||
680 | /* at this point no more lkb's should exist for this lockspace, | |
681 | so there's no chance of dlm_user_add_ast() being called and | |
682 | looking for lkb->ua->proc */ | |
683 | ||
684 | kfree(proc); | |
685 | file->private_data = NULL; | |
686 | ||
687 | dlm_put_lockspace(ls); | |
688 | dlm_put_lockspace(ls); /* for the find in device_open() */ | |
689 | ||
690 | /* FIXME: AUTOFREE: if this ls is no longer used do | |
691 | device_remove_lockspace() */ | |
692 | ||
693 | sigprocmask(SIG_SETMASK, &tmpsig, NULL); | |
694 | recalc_sigpending(); | |
695 | ||
696 | return 0; | |
697 | } | |
698 | ||
699 | static int copy_result_to_user(struct dlm_user_args *ua, int compat, int type, | |
700 | int bmode, char __user *buf, size_t count) | |
701 | { | |
702 | #ifdef CONFIG_COMPAT | |
703 | struct dlm_lock_result32 result32; | |
704 | #endif | |
705 | struct dlm_lock_result result; | |
706 | void *resultptr; | |
707 | int error=0; | |
708 | int len; | |
709 | int struct_len; | |
710 | ||
711 | memset(&result, 0, sizeof(struct dlm_lock_result)); | |
d7db923e DT |
712 | result.version[0] = DLM_DEVICE_VERSION_MAJOR; |
713 | result.version[1] = DLM_DEVICE_VERSION_MINOR; | |
714 | result.version[2] = DLM_DEVICE_VERSION_PATCH; | |
597d0cae DT |
715 | memcpy(&result.lksb, &ua->lksb, sizeof(struct dlm_lksb)); |
716 | result.user_lksb = ua->user_lksb; | |
717 | ||
718 | /* FIXME: dlm1 provides for the user's bastparam/addr to not be updated | |
719 | in a conversion unless the conversion is successful. See code | |
720 | in dlm_user_convert() for updating ua from ua_tmp. OpenVMS, though, | |
721 | notes that a new blocking AST address and parameter are set even if | |
722 | the conversion fails, so maybe we should just do that. */ | |
723 | ||
724 | if (type == AST_BAST) { | |
725 | result.user_astaddr = ua->bastaddr; | |
726 | result.user_astparam = ua->bastparam; | |
727 | result.bast_mode = bmode; | |
728 | } else { | |
729 | result.user_astaddr = ua->castaddr; | |
730 | result.user_astparam = ua->castparam; | |
731 | } | |
732 | ||
733 | #ifdef CONFIG_COMPAT | |
734 | if (compat) | |
735 | len = sizeof(struct dlm_lock_result32); | |
736 | else | |
737 | #endif | |
738 | len = sizeof(struct dlm_lock_result); | |
739 | struct_len = len; | |
740 | ||
741 | /* copy lvb to userspace if there is one, it's been updated, and | |
742 | the user buffer has space for it */ | |
743 | ||
744 | if (ua->update_user_lvb && ua->lksb.sb_lvbptr && | |
745 | count >= len + DLM_USER_LVB_LEN) { | |
746 | if (copy_to_user(buf+len, ua->lksb.sb_lvbptr, | |
747 | DLM_USER_LVB_LEN)) { | |
748 | error = -EFAULT; | |
749 | goto out; | |
750 | } | |
751 | ||
752 | result.lvb_offset = len; | |
753 | len += DLM_USER_LVB_LEN; | |
754 | } | |
755 | ||
756 | result.length = len; | |
757 | resultptr = &result; | |
758 | #ifdef CONFIG_COMPAT | |
759 | if (compat) { | |
760 | compat_output(&result, &result32); | |
761 | resultptr = &result32; | |
762 | } | |
763 | #endif | |
764 | ||
765 | if (copy_to_user(buf, resultptr, struct_len)) | |
766 | error = -EFAULT; | |
767 | else | |
768 | error = len; | |
769 | out: | |
770 | return error; | |
771 | } | |
772 | ||
d7db923e DT |
773 | static int copy_version_to_user(char __user *buf, size_t count) |
774 | { | |
775 | struct dlm_device_version ver; | |
776 | ||
777 | memset(&ver, 0, sizeof(struct dlm_device_version)); | |
778 | ver.version[0] = DLM_DEVICE_VERSION_MAJOR; | |
779 | ver.version[1] = DLM_DEVICE_VERSION_MINOR; | |
780 | ver.version[2] = DLM_DEVICE_VERSION_PATCH; | |
781 | ||
782 | if (copy_to_user(buf, &ver, sizeof(struct dlm_device_version))) | |
783 | return -EFAULT; | |
784 | return sizeof(struct dlm_device_version); | |
785 | } | |
786 | ||
597d0cae DT |
787 | /* a read returns a single ast described in a struct dlm_lock_result */ |
788 | ||
789 | static ssize_t device_read(struct file *file, char __user *buf, size_t count, | |
790 | loff_t *ppos) | |
791 | { | |
792 | struct dlm_user_proc *proc = file->private_data; | |
793 | struct dlm_lkb *lkb; | |
597d0cae DT |
794 | DECLARE_WAITQUEUE(wait, current); |
795 | int error, type=0, bmode=0, removed = 0; | |
796 | ||
d7db923e DT |
797 | if (count == sizeof(struct dlm_device_version)) { |
798 | error = copy_version_to_user(buf, count); | |
799 | return error; | |
800 | } | |
801 | ||
802 | if (!proc) { | |
803 | log_print("non-version read from control device %zu", count); | |
804 | return -EINVAL; | |
805 | } | |
806 | ||
597d0cae DT |
807 | #ifdef CONFIG_COMPAT |
808 | if (count < sizeof(struct dlm_lock_result32)) | |
809 | #else | |
810 | if (count < sizeof(struct dlm_lock_result)) | |
811 | #endif | |
812 | return -EINVAL; | |
813 | ||
814 | /* do we really need this? can a read happen after a close? */ | |
815 | if (test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags)) | |
816 | return -EINVAL; | |
817 | ||
818 | spin_lock(&proc->asts_spin); | |
819 | if (list_empty(&proc->asts)) { | |
820 | if (file->f_flags & O_NONBLOCK) { | |
821 | spin_unlock(&proc->asts_spin); | |
822 | return -EAGAIN; | |
823 | } | |
824 | ||
825 | add_wait_queue(&proc->wait, &wait); | |
826 | ||
827 | repeat: | |
828 | set_current_state(TASK_INTERRUPTIBLE); | |
829 | if (list_empty(&proc->asts) && !signal_pending(current)) { | |
830 | spin_unlock(&proc->asts_spin); | |
831 | schedule(); | |
832 | spin_lock(&proc->asts_spin); | |
833 | goto repeat; | |
834 | } | |
835 | set_current_state(TASK_RUNNING); | |
836 | remove_wait_queue(&proc->wait, &wait); | |
837 | ||
838 | if (signal_pending(current)) { | |
839 | spin_unlock(&proc->asts_spin); | |
840 | return -ERESTARTSYS; | |
841 | } | |
842 | } | |
843 | ||
597d0cae DT |
844 | /* there may be both completion and blocking asts to return for |
845 | the lkb, don't remove lkb from asts list unless no asts remain */ | |
846 | ||
847 | lkb = list_entry(proc->asts.next, struct dlm_lkb, lkb_astqueue); | |
848 | ||
849 | if (lkb->lkb_ast_type & AST_COMP) { | |
850 | lkb->lkb_ast_type &= ~AST_COMP; | |
851 | type = AST_COMP; | |
852 | } else if (lkb->lkb_ast_type & AST_BAST) { | |
853 | lkb->lkb_ast_type &= ~AST_BAST; | |
854 | type = AST_BAST; | |
855 | bmode = lkb->lkb_bastmode; | |
856 | } | |
857 | ||
858 | if (!lkb->lkb_ast_type) { | |
859 | list_del(&lkb->lkb_astqueue); | |
860 | removed = 1; | |
861 | } | |
862 | spin_unlock(&proc->asts_spin); | |
863 | ||
d292c0cc | 864 | error = copy_result_to_user(lkb->lkb_ua, |
597d0cae DT |
865 | test_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags), |
866 | type, bmode, buf, count); | |
867 | ||
868 | /* removes reference for the proc->asts lists added by | |
869 | dlm_user_add_ast() and may result in the lkb being freed */ | |
870 | if (removed) | |
871 | dlm_put_lkb(lkb); | |
872 | ||
873 | return error; | |
874 | } | |
875 | ||
876 | static unsigned int device_poll(struct file *file, poll_table *wait) | |
877 | { | |
878 | struct dlm_user_proc *proc = file->private_data; | |
879 | ||
880 | poll_wait(file, &proc->wait, wait); | |
881 | ||
882 | spin_lock(&proc->asts_spin); | |
883 | if (!list_empty(&proc->asts)) { | |
884 | spin_unlock(&proc->asts_spin); | |
885 | return POLLIN | POLLRDNORM; | |
886 | } | |
887 | spin_unlock(&proc->asts_spin); | |
888 | return 0; | |
889 | } | |
890 | ||
dc68c7ed DT |
891 | int dlm_user_daemon_available(void) |
892 | { | |
893 | /* dlm_controld hasn't started (or, has started, but not | |
894 | properly populated configfs) */ | |
895 | ||
896 | if (!dlm_our_nodeid()) | |
897 | return 0; | |
898 | ||
899 | /* This is to deal with versions of dlm_controld that don't | |
900 | know about the monitor device. We assume that if the | |
901 | dlm_controld was started (above), but the monitor device | |
902 | was never opened, that it's an old version. dlm_controld | |
903 | should open the monitor device before populating configfs. */ | |
904 | ||
905 | if (dlm_monitor_unused) | |
906 | return 1; | |
907 | ||
908 | return atomic_read(&dlm_monitor_opened) ? 1 : 0; | |
909 | } | |
910 | ||
597d0cae DT |
911 | static int ctl_device_open(struct inode *inode, struct file *file) |
912 | { | |
913 | file->private_data = NULL; | |
914 | return 0; | |
915 | } | |
916 | ||
917 | static int ctl_device_close(struct inode *inode, struct file *file) | |
918 | { | |
919 | return 0; | |
920 | } | |
921 | ||
dc68c7ed DT |
922 | static int monitor_device_open(struct inode *inode, struct file *file) |
923 | { | |
924 | atomic_inc(&dlm_monitor_opened); | |
925 | dlm_monitor_unused = 0; | |
926 | return 0; | |
927 | } | |
928 | ||
929 | static int monitor_device_close(struct inode *inode, struct file *file) | |
930 | { | |
931 | if (atomic_dec_and_test(&dlm_monitor_opened)) | |
932 | dlm_stop_lockspaces(); | |
933 | return 0; | |
934 | } | |
935 | ||
00977a59 | 936 | static const struct file_operations device_fops = { |
597d0cae DT |
937 | .open = device_open, |
938 | .release = device_close, | |
939 | .read = device_read, | |
940 | .write = device_write, | |
941 | .poll = device_poll, | |
942 | .owner = THIS_MODULE, | |
943 | }; | |
944 | ||
00977a59 | 945 | static const struct file_operations ctl_device_fops = { |
597d0cae DT |
946 | .open = ctl_device_open, |
947 | .release = ctl_device_close, | |
d7db923e | 948 | .read = device_read, |
597d0cae DT |
949 | .write = device_write, |
950 | .owner = THIS_MODULE, | |
951 | }; | |
952 | ||
0fe410d3 DC |
953 | static struct miscdevice ctl_device = { |
954 | .name = "dlm-control", | |
955 | .fops = &ctl_device_fops, | |
956 | .minor = MISC_DYNAMIC_MINOR, | |
957 | }; | |
958 | ||
dc68c7ed DT |
959 | static const struct file_operations monitor_device_fops = { |
960 | .open = monitor_device_open, | |
961 | .release = monitor_device_close, | |
962 | .owner = THIS_MODULE, | |
963 | }; | |
964 | ||
965 | static struct miscdevice monitor_device = { | |
966 | .name = "dlm-monitor", | |
967 | .fops = &monitor_device_fops, | |
968 | .minor = MISC_DYNAMIC_MINOR, | |
969 | }; | |
970 | ||
30727174 | 971 | int __init dlm_user_init(void) |
597d0cae DT |
972 | { |
973 | int error; | |
974 | ||
dc68c7ed DT |
975 | atomic_set(&dlm_monitor_opened, 0); |
976 | ||
597d0cae | 977 | error = misc_register(&ctl_device); |
dc68c7ed | 978 | if (error) { |
597d0cae | 979 | log_print("misc_register failed for control device"); |
dc68c7ed DT |
980 | goto out; |
981 | } | |
597d0cae | 982 | |
dc68c7ed DT |
983 | error = misc_register(&monitor_device); |
984 | if (error) { | |
985 | log_print("misc_register failed for monitor device"); | |
986 | misc_deregister(&ctl_device); | |
987 | } | |
988 | out: | |
597d0cae DT |
989 | return error; |
990 | } | |
991 | ||
992 | void dlm_user_exit(void) | |
993 | { | |
994 | misc_deregister(&ctl_device); | |
dc68c7ed | 995 | misc_deregister(&monitor_device); |
597d0cae DT |
996 | } |
997 |