Commit | Line | Data |
---|---|---|
597d0cae | 1 | /* |
1fecb1c4 | 2 | * Copyright (C) 2006-2009 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, |
1fecb1c4 | 87 | int namelen) |
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; | |
1fecb1c4 | 99 | memcpy(kb->i.lspace.name, kb32->i.lspace.name, namelen); |
72c2be77 DT |
100 | } else if (kb->cmd == DLM_USER_PURGE) { |
101 | kb->i.purge.nodeid = kb32->i.purge.nodeid; | |
102 | kb->i.purge.pid = kb32->i.purge.pid; | |
597d0cae DT |
103 | } else { |
104 | kb->i.lock.mode = kb32->i.lock.mode; | |
105 | kb->i.lock.namelen = kb32->i.lock.namelen; | |
106 | kb->i.lock.flags = kb32->i.lock.flags; | |
107 | kb->i.lock.lkid = kb32->i.lock.lkid; | |
108 | kb->i.lock.parent = kb32->i.lock.parent; | |
d7db923e DT |
109 | kb->i.lock.xid = kb32->i.lock.xid; |
110 | kb->i.lock.timeout = kb32->i.lock.timeout; | |
597d0cae DT |
111 | kb->i.lock.castparam = (void *)(long)kb32->i.lock.castparam; |
112 | kb->i.lock.castaddr = (void *)(long)kb32->i.lock.castaddr; | |
113 | kb->i.lock.bastparam = (void *)(long)kb32->i.lock.bastparam; | |
114 | kb->i.lock.bastaddr = (void *)(long)kb32->i.lock.bastaddr; | |
115 | kb->i.lock.lksb = (void *)(long)kb32->i.lock.lksb; | |
116 | memcpy(kb->i.lock.lvb, kb32->i.lock.lvb, DLM_USER_LVB_LEN); | |
1fecb1c4 | 117 | memcpy(kb->i.lock.name, kb32->i.lock.name, namelen); |
597d0cae DT |
118 | } |
119 | } | |
120 | ||
121 | static void compat_output(struct dlm_lock_result *res, | |
122 | struct dlm_lock_result32 *res32) | |
123 | { | |
d7db923e DT |
124 | res32->version[0] = res->version[0]; |
125 | res32->version[1] = res->version[1]; | |
126 | res32->version[2] = res->version[2]; | |
127 | ||
597d0cae DT |
128 | res32->user_astaddr = (__u32)(long)res->user_astaddr; |
129 | res32->user_astparam = (__u32)(long)res->user_astparam; | |
130 | res32->user_lksb = (__u32)(long)res->user_lksb; | |
131 | res32->bast_mode = res->bast_mode; | |
132 | ||
133 | res32->lvb_offset = res->lvb_offset; | |
134 | res32->length = res->length; | |
135 | ||
136 | res32->lksb.sb_status = res->lksb.sb_status; | |
137 | res32->lksb.sb_flags = res->lksb.sb_flags; | |
138 | res32->lksb.sb_lkid = res->lksb.sb_lkid; | |
139 | res32->lksb.sb_lvbptr = (__u32)(long)res->lksb.sb_lvbptr; | |
140 | } | |
141 | #endif | |
142 | ||
84d8cd69 DT |
143 | /* Figure out if this lock is at the end of its life and no longer |
144 | available for the application to use. The lkb still exists until | |
145 | the final ast is read. A lock becomes EOL in three situations: | |
146 | 1. a noqueue request fails with EAGAIN | |
147 | 2. an unlock completes with EUNLOCK | |
148 | 3. a cancel of a waiting request completes with ECANCEL/EDEADLK | |
149 | An EOL lock needs to be removed from the process's list of locks. | |
150 | And we can't allow any new operation on an EOL lock. This is | |
151 | not related to the lifetime of the lkb struct which is managed | |
152 | entirely by refcount. */ | |
153 | ||
154 | static int lkb_is_endoflife(struct dlm_lkb *lkb, int sb_status, int type) | |
155 | { | |
156 | switch (sb_status) { | |
157 | case -DLM_EUNLOCK: | |
158 | return 1; | |
159 | case -DLM_ECANCEL: | |
160 | case -ETIMEDOUT: | |
8b4021fa | 161 | case -EDEADLK: |
84d8cd69 DT |
162 | if (lkb->lkb_grmode == DLM_LOCK_IV) |
163 | return 1; | |
164 | break; | |
165 | case -EAGAIN: | |
166 | if (type == AST_COMP && lkb->lkb_grmode == DLM_LOCK_IV) | |
167 | return 1; | |
168 | break; | |
169 | } | |
170 | return 0; | |
171 | } | |
172 | ||
ef0c2bb0 DT |
173 | /* we could possibly check if the cancel of an orphan has resulted in the lkb |
174 | being removed and then remove that lkb from the orphans list and free it */ | |
597d0cae | 175 | |
fd22a51b | 176 | void dlm_user_add_ast(struct dlm_lkb *lkb, int type, int bastmode) |
597d0cae DT |
177 | { |
178 | struct dlm_ls *ls; | |
179 | struct dlm_user_args *ua; | |
180 | struct dlm_user_proc *proc; | |
ef0c2bb0 | 181 | int eol = 0, ast_type; |
597d0cae | 182 | |
ef0c2bb0 | 183 | if (lkb->lkb_flags & (DLM_IFL_ORPHAN | DLM_IFL_DEAD)) |
597d0cae | 184 | return; |
597d0cae DT |
185 | |
186 | ls = lkb->lkb_resource->res_ls; | |
187 | mutex_lock(&ls->ls_clear_proc_locks); | |
188 | ||
189 | /* If ORPHAN/DEAD flag is set, it means the process is dead so an ast | |
190 | can't be delivered. For ORPHAN's, dlm_clear_proc_locks() freed | |
ef0c2bb0 DT |
191 | lkb->ua so we can't try to use it. This second check is necessary |
192 | for cases where a completion ast is received for an operation that | |
193 | began before clear_proc_locks did its cancel/unlock. */ | |
597d0cae | 194 | |
ef0c2bb0 | 195 | if (lkb->lkb_flags & (DLM_IFL_ORPHAN | DLM_IFL_DEAD)) |
597d0cae | 196 | goto out; |
597d0cae | 197 | |
d292c0cc DT |
198 | DLM_ASSERT(lkb->lkb_ua, dlm_print_lkb(lkb);); |
199 | ua = lkb->lkb_ua; | |
597d0cae DT |
200 | proc = ua->proc; |
201 | ||
202 | if (type == AST_BAST && ua->bastaddr == NULL) | |
203 | goto out; | |
204 | ||
205 | spin_lock(&proc->asts_spin); | |
ef0c2bb0 DT |
206 | |
207 | ast_type = lkb->lkb_ast_type; | |
208 | lkb->lkb_ast_type |= type; | |
fd22a51b DT |
209 | if (bastmode) |
210 | lkb->lkb_bastmode = bastmode; | |
ef0c2bb0 DT |
211 | |
212 | if (!ast_type) { | |
597d0cae DT |
213 | kref_get(&lkb->lkb_ref); |
214 | list_add_tail(&lkb->lkb_astqueue, &proc->asts); | |
597d0cae DT |
215 | wake_up_interruptible(&proc->wait); |
216 | } | |
ef0c2bb0 DT |
217 | if (type == AST_COMP && (ast_type & AST_COMP)) |
218 | log_debug(ls, "ast overlap %x status %x %x", | |
219 | lkb->lkb_id, ua->lksb.sb_status, lkb->lkb_flags); | |
220 | ||
84d8cd69 | 221 | eol = lkb_is_endoflife(lkb, ua->lksb.sb_status, type); |
ef0c2bb0 DT |
222 | if (eol) { |
223 | lkb->lkb_ast_type &= ~AST_BAST; | |
224 | lkb->lkb_flags |= DLM_IFL_ENDOFLIFE; | |
225 | } | |
a1bc86e6 | 226 | |
597d0cae DT |
227 | /* We want to copy the lvb to userspace when the completion |
228 | ast is read if the status is 0, the lock has an lvb and | |
229 | lvb_ops says we should. We could probably have set_lvb_lock() | |
230 | set update_user_lvb instead and not need old_mode */ | |
231 | ||
232 | if ((lkb->lkb_ast_type & AST_COMP) && | |
233 | (lkb->lkb_lksb->sb_status == 0) && | |
234 | lkb->lkb_lksb->sb_lvbptr && | |
235 | dlm_lvb_operations[ua->old_mode + 1][lkb->lkb_grmode + 1]) | |
236 | ua->update_user_lvb = 1; | |
237 | else | |
238 | ua->update_user_lvb = 0; | |
239 | ||
240 | spin_unlock(&proc->asts_spin); | |
34e22bed | 241 | |
ef0c2bb0 | 242 | if (eol) { |
ce5246b9 | 243 | spin_lock(&proc->locks_spin); |
ef0c2bb0 DT |
244 | if (!list_empty(&lkb->lkb_ownqueue)) { |
245 | list_del_init(&lkb->lkb_ownqueue); | |
246 | dlm_put_lkb(lkb); | |
247 | } | |
ce5246b9 | 248 | spin_unlock(&proc->locks_spin); |
34e22bed | 249 | } |
597d0cae DT |
250 | out: |
251 | mutex_unlock(&ls->ls_clear_proc_locks); | |
252 | } | |
253 | ||
254 | static int device_user_lock(struct dlm_user_proc *proc, | |
255 | struct dlm_lock_params *params) | |
256 | { | |
257 | struct dlm_ls *ls; | |
258 | struct dlm_user_args *ua; | |
259 | int error = -ENOMEM; | |
260 | ||
261 | ls = dlm_find_lockspace_local(proc->lockspace); | |
262 | if (!ls) | |
263 | return -ENOENT; | |
264 | ||
265 | if (!params->castaddr || !params->lksb) { | |
266 | error = -EINVAL; | |
267 | goto out; | |
268 | } | |
269 | ||
573c24c4 | 270 | ua = kzalloc(sizeof(struct dlm_user_args), GFP_NOFS); |
597d0cae DT |
271 | if (!ua) |
272 | goto out; | |
273 | ua->proc = proc; | |
274 | ua->user_lksb = params->lksb; | |
275 | ua->castparam = params->castparam; | |
276 | ua->castaddr = params->castaddr; | |
277 | ua->bastparam = params->bastparam; | |
278 | ua->bastaddr = params->bastaddr; | |
d7db923e | 279 | ua->xid = params->xid; |
597d0cae DT |
280 | |
281 | if (params->flags & DLM_LKF_CONVERT) | |
282 | error = dlm_user_convert(ls, ua, | |
283 | params->mode, params->flags, | |
d7db923e DT |
284 | params->lkid, params->lvb, |
285 | (unsigned long) params->timeout); | |
597d0cae DT |
286 | else { |
287 | error = dlm_user_request(ls, ua, | |
288 | params->mode, params->flags, | |
289 | params->name, params->namelen, | |
d7db923e | 290 | (unsigned long) params->timeout); |
597d0cae DT |
291 | if (!error) |
292 | error = ua->lksb.sb_lkid; | |
293 | } | |
294 | out: | |
295 | dlm_put_lockspace(ls); | |
296 | return error; | |
297 | } | |
298 | ||
299 | static int device_user_unlock(struct dlm_user_proc *proc, | |
300 | struct dlm_lock_params *params) | |
301 | { | |
302 | struct dlm_ls *ls; | |
303 | struct dlm_user_args *ua; | |
304 | int error = -ENOMEM; | |
305 | ||
306 | ls = dlm_find_lockspace_local(proc->lockspace); | |
307 | if (!ls) | |
308 | return -ENOENT; | |
309 | ||
573c24c4 | 310 | ua = kzalloc(sizeof(struct dlm_user_args), GFP_NOFS); |
597d0cae DT |
311 | if (!ua) |
312 | goto out; | |
313 | ua->proc = proc; | |
314 | ua->user_lksb = params->lksb; | |
315 | ua->castparam = params->castparam; | |
316 | ua->castaddr = params->castaddr; | |
317 | ||
318 | if (params->flags & DLM_LKF_CANCEL) | |
319 | error = dlm_user_cancel(ls, ua, params->flags, params->lkid); | |
320 | else | |
321 | error = dlm_user_unlock(ls, ua, params->flags, params->lkid, | |
322 | params->lvb); | |
323 | out: | |
324 | dlm_put_lockspace(ls); | |
325 | return error; | |
326 | } | |
327 | ||
8b4021fa DT |
328 | static int device_user_deadlock(struct dlm_user_proc *proc, |
329 | struct dlm_lock_params *params) | |
330 | { | |
331 | struct dlm_ls *ls; | |
332 | int error; | |
333 | ||
334 | ls = dlm_find_lockspace_local(proc->lockspace); | |
335 | if (!ls) | |
336 | return -ENOENT; | |
337 | ||
338 | error = dlm_user_deadlock(ls, params->flags, params->lkid); | |
339 | ||
340 | dlm_put_lockspace(ls); | |
341 | return error; | |
342 | } | |
343 | ||
0f8e0d9a | 344 | static int dlm_device_register(struct dlm_ls *ls, char *name) |
254da030 PC |
345 | { |
346 | int error, len; | |
347 | ||
0f8e0d9a DT |
348 | /* The device is already registered. This happens when the |
349 | lockspace is created multiple times from userspace. */ | |
350 | if (ls->ls_device.name) | |
351 | return 0; | |
352 | ||
254da030 PC |
353 | error = -ENOMEM; |
354 | len = strlen(name) + strlen(name_prefix) + 2; | |
573c24c4 | 355 | ls->ls_device.name = kzalloc(len, GFP_NOFS); |
254da030 PC |
356 | if (!ls->ls_device.name) |
357 | goto fail; | |
358 | ||
359 | snprintf((char *)ls->ls_device.name, len, "%s_%s", name_prefix, | |
360 | name); | |
361 | ls->ls_device.fops = &device_fops; | |
362 | ls->ls_device.minor = MISC_DYNAMIC_MINOR; | |
363 | ||
364 | error = misc_register(&ls->ls_device); | |
365 | if (error) { | |
366 | kfree(ls->ls_device.name); | |
367 | } | |
368 | fail: | |
369 | return error; | |
370 | } | |
371 | ||
0f8e0d9a DT |
372 | int dlm_device_deregister(struct dlm_ls *ls) |
373 | { | |
374 | int error; | |
375 | ||
376 | /* The device is not registered. This happens when the lockspace | |
377 | was never used from userspace, or when device_create_lockspace() | |
378 | calls dlm_release_lockspace() after the register fails. */ | |
379 | if (!ls->ls_device.name) | |
380 | return 0; | |
381 | ||
382 | error = misc_deregister(&ls->ls_device); | |
383 | if (!error) | |
384 | kfree(ls->ls_device.name); | |
385 | return error; | |
386 | } | |
387 | ||
72c2be77 DT |
388 | static int device_user_purge(struct dlm_user_proc *proc, |
389 | struct dlm_purge_params *params) | |
390 | { | |
391 | struct dlm_ls *ls; | |
392 | int error; | |
393 | ||
394 | ls = dlm_find_lockspace_local(proc->lockspace); | |
395 | if (!ls) | |
396 | return -ENOENT; | |
397 | ||
398 | error = dlm_user_purge(ls, proc, params->nodeid, params->pid); | |
399 | ||
400 | dlm_put_lockspace(ls); | |
401 | return error; | |
402 | } | |
403 | ||
597d0cae DT |
404 | static int device_create_lockspace(struct dlm_lspace_params *params) |
405 | { | |
406 | dlm_lockspace_t *lockspace; | |
407 | struct dlm_ls *ls; | |
254da030 | 408 | int error; |
597d0cae DT |
409 | |
410 | if (!capable(CAP_SYS_ADMIN)) | |
411 | return -EPERM; | |
412 | ||
413 | error = dlm_new_lockspace(params->name, strlen(params->name), | |
3ae1acf9 | 414 | &lockspace, params->flags, DLM_USER_LVB_LEN); |
597d0cae DT |
415 | if (error) |
416 | return error; | |
417 | ||
418 | ls = dlm_find_lockspace_local(lockspace); | |
419 | if (!ls) | |
420 | return -ENOENT; | |
421 | ||
0f8e0d9a | 422 | error = dlm_device_register(ls, params->name); |
597d0cae | 423 | dlm_put_lockspace(ls); |
597d0cae | 424 | |
254da030 PC |
425 | if (error) |
426 | dlm_release_lockspace(lockspace, 0); | |
427 | else | |
428 | error = ls->ls_device.minor; | |
429 | ||
597d0cae DT |
430 | return error; |
431 | } | |
432 | ||
433 | static int device_remove_lockspace(struct dlm_lspace_params *params) | |
434 | { | |
435 | dlm_lockspace_t *lockspace; | |
436 | struct dlm_ls *ls; | |
c6e6f0ba | 437 | int error, force = 0; |
597d0cae DT |
438 | |
439 | if (!capable(CAP_SYS_ADMIN)) | |
440 | return -EPERM; | |
441 | ||
442 | ls = dlm_find_lockspace_device(params->minor); | |
443 | if (!ls) | |
444 | return -ENOENT; | |
445 | ||
c6e6f0ba DT |
446 | if (params->flags & DLM_USER_LSFLG_FORCEFREE) |
447 | force = 2; | |
448 | ||
597d0cae | 449 | lockspace = ls->ls_local_handle; |
0f8e0d9a | 450 | dlm_put_lockspace(ls); |
597d0cae | 451 | |
0f8e0d9a DT |
452 | /* The final dlm_release_lockspace waits for references to go to |
453 | zero, so all processes will need to close their device for the | |
454 | ls before the release will proceed. release also calls the | |
455 | device_deregister above. Converting a positive return value | |
456 | from release to zero means that userspace won't know when its | |
457 | release was the final one, but it shouldn't need to know. */ | |
597d0cae | 458 | |
c6e6f0ba | 459 | error = dlm_release_lockspace(lockspace, force); |
0f8e0d9a DT |
460 | if (error > 0) |
461 | error = 0; | |
597d0cae DT |
462 | return error; |
463 | } | |
464 | ||
465 | /* Check the user's version matches ours */ | |
466 | static int check_version(struct dlm_write_request *req) | |
467 | { | |
468 | if (req->version[0] != DLM_DEVICE_VERSION_MAJOR || | |
469 | (req->version[0] == DLM_DEVICE_VERSION_MAJOR && | |
470 | req->version[1] > DLM_DEVICE_VERSION_MINOR)) { | |
471 | ||
472 | printk(KERN_DEBUG "dlm: process %s (%d) version mismatch " | |
473 | "user (%d.%d.%d) kernel (%d.%d.%d)\n", | |
474 | current->comm, | |
ba25f9dc | 475 | task_pid_nr(current), |
597d0cae DT |
476 | req->version[0], |
477 | req->version[1], | |
478 | req->version[2], | |
479 | DLM_DEVICE_VERSION_MAJOR, | |
480 | DLM_DEVICE_VERSION_MINOR, | |
481 | DLM_DEVICE_VERSION_PATCH); | |
482 | return -EINVAL; | |
483 | } | |
484 | return 0; | |
485 | } | |
486 | ||
487 | /* | |
488 | * device_write | |
489 | * | |
490 | * device_user_lock | |
491 | * dlm_user_request -> request_lock | |
492 | * dlm_user_convert -> convert_lock | |
493 | * | |
494 | * device_user_unlock | |
495 | * dlm_user_unlock -> unlock_lock | |
496 | * dlm_user_cancel -> cancel_lock | |
497 | * | |
498 | * device_create_lockspace | |
499 | * dlm_new_lockspace | |
500 | * | |
501 | * device_remove_lockspace | |
502 | * dlm_release_lockspace | |
503 | */ | |
504 | ||
505 | /* a write to a lockspace device is a lock or unlock request, a write | |
506 | to the control device is to create/remove a lockspace */ | |
507 | ||
508 | static ssize_t device_write(struct file *file, const char __user *buf, | |
509 | size_t count, loff_t *ppos) | |
510 | { | |
511 | struct dlm_user_proc *proc = file->private_data; | |
512 | struct dlm_write_request *kbuf; | |
513 | sigset_t tmpsig, allsigs; | |
514 | int error; | |
515 | ||
516 | #ifdef CONFIG_COMPAT | |
517 | if (count < sizeof(struct dlm_write_request32)) | |
518 | #else | |
519 | if (count < sizeof(struct dlm_write_request)) | |
520 | #endif | |
521 | return -EINVAL; | |
522 | ||
573c24c4 | 523 | kbuf = kzalloc(count + 1, GFP_NOFS); |
597d0cae DT |
524 | if (!kbuf) |
525 | return -ENOMEM; | |
526 | ||
527 | if (copy_from_user(kbuf, buf, count)) { | |
528 | error = -EFAULT; | |
529 | goto out_free; | |
530 | } | |
531 | ||
532 | if (check_version(kbuf)) { | |
533 | error = -EBADE; | |
534 | goto out_free; | |
535 | } | |
536 | ||
537 | #ifdef CONFIG_COMPAT | |
538 | if (!kbuf->is64bit) { | |
539 | struct dlm_write_request32 *k32buf; | |
1fecb1c4 DT |
540 | int namelen = 0; |
541 | ||
542 | if (count > sizeof(struct dlm_write_request32)) | |
543 | namelen = count - sizeof(struct dlm_write_request32); | |
544 | ||
597d0cae | 545 | k32buf = (struct dlm_write_request32 *)kbuf; |
1fecb1c4 DT |
546 | |
547 | /* add 1 after namelen so that the name string is terminated */ | |
548 | kbuf = kzalloc(sizeof(struct dlm_write_request) + namelen + 1, | |
573c24c4 | 549 | GFP_NOFS); |
cb980d9a DT |
550 | if (!kbuf) { |
551 | kfree(k32buf); | |
597d0cae | 552 | return -ENOMEM; |
cb980d9a | 553 | } |
597d0cae DT |
554 | |
555 | if (proc) | |
556 | set_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags); | |
1fecb1c4 DT |
557 | |
558 | compat_input(kbuf, k32buf, namelen); | |
597d0cae DT |
559 | kfree(k32buf); |
560 | } | |
561 | #endif | |
562 | ||
563 | /* do we really need this? can a write happen after a close? */ | |
564 | if ((kbuf->cmd == DLM_USER_LOCK || kbuf->cmd == DLM_USER_UNLOCK) && | |
cb980d9a DT |
565 | (proc && test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))) { |
566 | error = -EINVAL; | |
567 | goto out_free; | |
568 | } | |
597d0cae DT |
569 | |
570 | sigfillset(&allsigs); | |
571 | sigprocmask(SIG_BLOCK, &allsigs, &tmpsig); | |
572 | ||
573 | error = -EINVAL; | |
574 | ||
575 | switch (kbuf->cmd) | |
576 | { | |
577 | case DLM_USER_LOCK: | |
578 | if (!proc) { | |
579 | log_print("no locking on control device"); | |
580 | goto out_sig; | |
581 | } | |
582 | error = device_user_lock(proc, &kbuf->i.lock); | |
583 | break; | |
584 | ||
585 | case DLM_USER_UNLOCK: | |
586 | if (!proc) { | |
587 | log_print("no locking on control device"); | |
588 | goto out_sig; | |
589 | } | |
590 | error = device_user_unlock(proc, &kbuf->i.lock); | |
591 | break; | |
592 | ||
8b4021fa DT |
593 | case DLM_USER_DEADLOCK: |
594 | if (!proc) { | |
595 | log_print("no locking on control device"); | |
596 | goto out_sig; | |
597 | } | |
598 | error = device_user_deadlock(proc, &kbuf->i.lock); | |
599 | break; | |
600 | ||
597d0cae DT |
601 | case DLM_USER_CREATE_LOCKSPACE: |
602 | if (proc) { | |
603 | log_print("create/remove only on control device"); | |
604 | goto out_sig; | |
605 | } | |
606 | error = device_create_lockspace(&kbuf->i.lspace); | |
607 | break; | |
608 | ||
609 | case DLM_USER_REMOVE_LOCKSPACE: | |
610 | if (proc) { | |
611 | log_print("create/remove only on control device"); | |
612 | goto out_sig; | |
613 | } | |
614 | error = device_remove_lockspace(&kbuf->i.lspace); | |
615 | break; | |
616 | ||
72c2be77 DT |
617 | case DLM_USER_PURGE: |
618 | if (!proc) { | |
619 | log_print("no locking on control device"); | |
620 | goto out_sig; | |
621 | } | |
622 | error = device_user_purge(proc, &kbuf->i.purge); | |
623 | break; | |
624 | ||
597d0cae DT |
625 | default: |
626 | log_print("Unknown command passed to DLM device : %d\n", | |
627 | kbuf->cmd); | |
628 | } | |
629 | ||
630 | out_sig: | |
631 | sigprocmask(SIG_SETMASK, &tmpsig, NULL); | |
632 | recalc_sigpending(); | |
633 | out_free: | |
634 | kfree(kbuf); | |
635 | return error; | |
636 | } | |
637 | ||
638 | /* Every process that opens the lockspace device has its own "proc" structure | |
639 | hanging off the open file that's used to keep track of locks owned by the | |
640 | process and asts that need to be delivered to the process. */ | |
641 | ||
642 | static int device_open(struct inode *inode, struct file *file) | |
643 | { | |
644 | struct dlm_user_proc *proc; | |
645 | struct dlm_ls *ls; | |
646 | ||
647 | ls = dlm_find_lockspace_device(iminor(inode)); | |
f9f2ed48 | 648 | if (!ls) |
597d0cae DT |
649 | return -ENOENT; |
650 | ||
573c24c4 | 651 | proc = kzalloc(sizeof(struct dlm_user_proc), GFP_NOFS); |
597d0cae DT |
652 | if (!proc) { |
653 | dlm_put_lockspace(ls); | |
654 | return -ENOMEM; | |
655 | } | |
656 | ||
657 | proc->lockspace = ls->ls_local_handle; | |
658 | INIT_LIST_HEAD(&proc->asts); | |
659 | INIT_LIST_HEAD(&proc->locks); | |
a1bc86e6 | 660 | INIT_LIST_HEAD(&proc->unlocking); |
597d0cae DT |
661 | spin_lock_init(&proc->asts_spin); |
662 | spin_lock_init(&proc->locks_spin); | |
663 | init_waitqueue_head(&proc->wait); | |
664 | file->private_data = proc; | |
665 | ||
666 | return 0; | |
667 | } | |
668 | ||
669 | static int device_close(struct inode *inode, struct file *file) | |
670 | { | |
671 | struct dlm_user_proc *proc = file->private_data; | |
672 | struct dlm_ls *ls; | |
673 | sigset_t tmpsig, allsigs; | |
674 | ||
675 | ls = dlm_find_lockspace_local(proc->lockspace); | |
676 | if (!ls) | |
677 | return -ENOENT; | |
678 | ||
679 | sigfillset(&allsigs); | |
680 | sigprocmask(SIG_BLOCK, &allsigs, &tmpsig); | |
681 | ||
682 | set_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags); | |
683 | ||
684 | dlm_clear_proc_locks(ls, proc); | |
685 | ||
686 | /* at this point no more lkb's should exist for this lockspace, | |
687 | so there's no chance of dlm_user_add_ast() being called and | |
688 | looking for lkb->ua->proc */ | |
689 | ||
690 | kfree(proc); | |
691 | file->private_data = NULL; | |
692 | ||
693 | dlm_put_lockspace(ls); | |
694 | dlm_put_lockspace(ls); /* for the find in device_open() */ | |
695 | ||
696 | /* FIXME: AUTOFREE: if this ls is no longer used do | |
697 | device_remove_lockspace() */ | |
698 | ||
699 | sigprocmask(SIG_SETMASK, &tmpsig, NULL); | |
700 | recalc_sigpending(); | |
701 | ||
702 | return 0; | |
703 | } | |
704 | ||
705 | static int copy_result_to_user(struct dlm_user_args *ua, int compat, int type, | |
706 | int bmode, char __user *buf, size_t count) | |
707 | { | |
708 | #ifdef CONFIG_COMPAT | |
709 | struct dlm_lock_result32 result32; | |
710 | #endif | |
711 | struct dlm_lock_result result; | |
712 | void *resultptr; | |
713 | int error=0; | |
714 | int len; | |
715 | int struct_len; | |
716 | ||
717 | memset(&result, 0, sizeof(struct dlm_lock_result)); | |
d7db923e DT |
718 | result.version[0] = DLM_DEVICE_VERSION_MAJOR; |
719 | result.version[1] = DLM_DEVICE_VERSION_MINOR; | |
720 | result.version[2] = DLM_DEVICE_VERSION_PATCH; | |
597d0cae DT |
721 | memcpy(&result.lksb, &ua->lksb, sizeof(struct dlm_lksb)); |
722 | result.user_lksb = ua->user_lksb; | |
723 | ||
724 | /* FIXME: dlm1 provides for the user's bastparam/addr to not be updated | |
725 | in a conversion unless the conversion is successful. See code | |
726 | in dlm_user_convert() for updating ua from ua_tmp. OpenVMS, though, | |
727 | notes that a new blocking AST address and parameter are set even if | |
728 | the conversion fails, so maybe we should just do that. */ | |
729 | ||
730 | if (type == AST_BAST) { | |
731 | result.user_astaddr = ua->bastaddr; | |
732 | result.user_astparam = ua->bastparam; | |
733 | result.bast_mode = bmode; | |
734 | } else { | |
735 | result.user_astaddr = ua->castaddr; | |
736 | result.user_astparam = ua->castparam; | |
737 | } | |
738 | ||
739 | #ifdef CONFIG_COMPAT | |
740 | if (compat) | |
741 | len = sizeof(struct dlm_lock_result32); | |
742 | else | |
743 | #endif | |
744 | len = sizeof(struct dlm_lock_result); | |
745 | struct_len = len; | |
746 | ||
747 | /* copy lvb to userspace if there is one, it's been updated, and | |
748 | the user buffer has space for it */ | |
749 | ||
750 | if (ua->update_user_lvb && ua->lksb.sb_lvbptr && | |
751 | count >= len + DLM_USER_LVB_LEN) { | |
752 | if (copy_to_user(buf+len, ua->lksb.sb_lvbptr, | |
753 | DLM_USER_LVB_LEN)) { | |
754 | error = -EFAULT; | |
755 | goto out; | |
756 | } | |
757 | ||
758 | result.lvb_offset = len; | |
759 | len += DLM_USER_LVB_LEN; | |
760 | } | |
761 | ||
762 | result.length = len; | |
763 | resultptr = &result; | |
764 | #ifdef CONFIG_COMPAT | |
765 | if (compat) { | |
766 | compat_output(&result, &result32); | |
767 | resultptr = &result32; | |
768 | } | |
769 | #endif | |
770 | ||
771 | if (copy_to_user(buf, resultptr, struct_len)) | |
772 | error = -EFAULT; | |
773 | else | |
774 | error = len; | |
775 | out: | |
776 | return error; | |
777 | } | |
778 | ||
d7db923e DT |
779 | static int copy_version_to_user(char __user *buf, size_t count) |
780 | { | |
781 | struct dlm_device_version ver; | |
782 | ||
783 | memset(&ver, 0, sizeof(struct dlm_device_version)); | |
784 | ver.version[0] = DLM_DEVICE_VERSION_MAJOR; | |
785 | ver.version[1] = DLM_DEVICE_VERSION_MINOR; | |
786 | ver.version[2] = DLM_DEVICE_VERSION_PATCH; | |
787 | ||
788 | if (copy_to_user(buf, &ver, sizeof(struct dlm_device_version))) | |
789 | return -EFAULT; | |
790 | return sizeof(struct dlm_device_version); | |
791 | } | |
792 | ||
597d0cae DT |
793 | /* a read returns a single ast described in a struct dlm_lock_result */ |
794 | ||
795 | static ssize_t device_read(struct file *file, char __user *buf, size_t count, | |
796 | loff_t *ppos) | |
797 | { | |
798 | struct dlm_user_proc *proc = file->private_data; | |
799 | struct dlm_lkb *lkb; | |
597d0cae DT |
800 | DECLARE_WAITQUEUE(wait, current); |
801 | int error, type=0, bmode=0, removed = 0; | |
802 | ||
d7db923e DT |
803 | if (count == sizeof(struct dlm_device_version)) { |
804 | error = copy_version_to_user(buf, count); | |
805 | return error; | |
806 | } | |
807 | ||
808 | if (!proc) { | |
809 | log_print("non-version read from control device %zu", count); | |
810 | return -EINVAL; | |
811 | } | |
812 | ||
597d0cae DT |
813 | #ifdef CONFIG_COMPAT |
814 | if (count < sizeof(struct dlm_lock_result32)) | |
815 | #else | |
816 | if (count < sizeof(struct dlm_lock_result)) | |
817 | #endif | |
818 | return -EINVAL; | |
819 | ||
820 | /* do we really need this? can a read happen after a close? */ | |
821 | if (test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags)) | |
822 | return -EINVAL; | |
823 | ||
824 | spin_lock(&proc->asts_spin); | |
825 | if (list_empty(&proc->asts)) { | |
826 | if (file->f_flags & O_NONBLOCK) { | |
827 | spin_unlock(&proc->asts_spin); | |
828 | return -EAGAIN; | |
829 | } | |
830 | ||
831 | add_wait_queue(&proc->wait, &wait); | |
832 | ||
833 | repeat: | |
834 | set_current_state(TASK_INTERRUPTIBLE); | |
835 | if (list_empty(&proc->asts) && !signal_pending(current)) { | |
836 | spin_unlock(&proc->asts_spin); | |
837 | schedule(); | |
838 | spin_lock(&proc->asts_spin); | |
839 | goto repeat; | |
840 | } | |
841 | set_current_state(TASK_RUNNING); | |
842 | remove_wait_queue(&proc->wait, &wait); | |
843 | ||
844 | if (signal_pending(current)) { | |
845 | spin_unlock(&proc->asts_spin); | |
846 | return -ERESTARTSYS; | |
847 | } | |
848 | } | |
849 | ||
597d0cae DT |
850 | /* there may be both completion and blocking asts to return for |
851 | the lkb, don't remove lkb from asts list unless no asts remain */ | |
852 | ||
853 | lkb = list_entry(proc->asts.next, struct dlm_lkb, lkb_astqueue); | |
854 | ||
855 | if (lkb->lkb_ast_type & AST_COMP) { | |
856 | lkb->lkb_ast_type &= ~AST_COMP; | |
857 | type = AST_COMP; | |
858 | } else if (lkb->lkb_ast_type & AST_BAST) { | |
859 | lkb->lkb_ast_type &= ~AST_BAST; | |
860 | type = AST_BAST; | |
861 | bmode = lkb->lkb_bastmode; | |
862 | } | |
863 | ||
864 | if (!lkb->lkb_ast_type) { | |
865 | list_del(&lkb->lkb_astqueue); | |
866 | removed = 1; | |
867 | } | |
868 | spin_unlock(&proc->asts_spin); | |
869 | ||
d292c0cc | 870 | error = copy_result_to_user(lkb->lkb_ua, |
597d0cae DT |
871 | test_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags), |
872 | type, bmode, buf, count); | |
873 | ||
874 | /* removes reference for the proc->asts lists added by | |
875 | dlm_user_add_ast() and may result in the lkb being freed */ | |
876 | if (removed) | |
877 | dlm_put_lkb(lkb); | |
878 | ||
879 | return error; | |
880 | } | |
881 | ||
882 | static unsigned int device_poll(struct file *file, poll_table *wait) | |
883 | { | |
884 | struct dlm_user_proc *proc = file->private_data; | |
885 | ||
886 | poll_wait(file, &proc->wait, wait); | |
887 | ||
888 | spin_lock(&proc->asts_spin); | |
889 | if (!list_empty(&proc->asts)) { | |
890 | spin_unlock(&proc->asts_spin); | |
891 | return POLLIN | POLLRDNORM; | |
892 | } | |
893 | spin_unlock(&proc->asts_spin); | |
894 | return 0; | |
895 | } | |
896 | ||
dc68c7ed DT |
897 | int dlm_user_daemon_available(void) |
898 | { | |
899 | /* dlm_controld hasn't started (or, has started, but not | |
900 | properly populated configfs) */ | |
901 | ||
902 | if (!dlm_our_nodeid()) | |
903 | return 0; | |
904 | ||
905 | /* This is to deal with versions of dlm_controld that don't | |
906 | know about the monitor device. We assume that if the | |
907 | dlm_controld was started (above), but the monitor device | |
908 | was never opened, that it's an old version. dlm_controld | |
909 | should open the monitor device before populating configfs. */ | |
910 | ||
911 | if (dlm_monitor_unused) | |
912 | return 1; | |
913 | ||
914 | return atomic_read(&dlm_monitor_opened) ? 1 : 0; | |
915 | } | |
916 | ||
597d0cae DT |
917 | static int ctl_device_open(struct inode *inode, struct file *file) |
918 | { | |
919 | file->private_data = NULL; | |
920 | return 0; | |
921 | } | |
922 | ||
923 | static int ctl_device_close(struct inode *inode, struct file *file) | |
924 | { | |
925 | return 0; | |
926 | } | |
927 | ||
dc68c7ed DT |
928 | static int monitor_device_open(struct inode *inode, struct file *file) |
929 | { | |
930 | atomic_inc(&dlm_monitor_opened); | |
931 | dlm_monitor_unused = 0; | |
932 | return 0; | |
933 | } | |
934 | ||
935 | static int monitor_device_close(struct inode *inode, struct file *file) | |
936 | { | |
937 | if (atomic_dec_and_test(&dlm_monitor_opened)) | |
938 | dlm_stop_lockspaces(); | |
939 | return 0; | |
940 | } | |
941 | ||
00977a59 | 942 | static const struct file_operations device_fops = { |
597d0cae DT |
943 | .open = device_open, |
944 | .release = device_close, | |
945 | .read = device_read, | |
946 | .write = device_write, | |
947 | .poll = device_poll, | |
948 | .owner = THIS_MODULE, | |
949 | }; | |
950 | ||
00977a59 | 951 | static const struct file_operations ctl_device_fops = { |
597d0cae DT |
952 | .open = ctl_device_open, |
953 | .release = ctl_device_close, | |
d7db923e | 954 | .read = device_read, |
597d0cae DT |
955 | .write = device_write, |
956 | .owner = THIS_MODULE, | |
957 | }; | |
958 | ||
0fe410d3 DC |
959 | static struct miscdevice ctl_device = { |
960 | .name = "dlm-control", | |
961 | .fops = &ctl_device_fops, | |
962 | .minor = MISC_DYNAMIC_MINOR, | |
963 | }; | |
964 | ||
dc68c7ed DT |
965 | static const struct file_operations monitor_device_fops = { |
966 | .open = monitor_device_open, | |
967 | .release = monitor_device_close, | |
968 | .owner = THIS_MODULE, | |
969 | }; | |
970 | ||
971 | static struct miscdevice monitor_device = { | |
972 | .name = "dlm-monitor", | |
973 | .fops = &monitor_device_fops, | |
974 | .minor = MISC_DYNAMIC_MINOR, | |
975 | }; | |
976 | ||
30727174 | 977 | int __init dlm_user_init(void) |
597d0cae DT |
978 | { |
979 | int error; | |
980 | ||
dc68c7ed DT |
981 | atomic_set(&dlm_monitor_opened, 0); |
982 | ||
597d0cae | 983 | error = misc_register(&ctl_device); |
dc68c7ed | 984 | if (error) { |
597d0cae | 985 | log_print("misc_register failed for control device"); |
dc68c7ed DT |
986 | goto out; |
987 | } | |
597d0cae | 988 | |
dc68c7ed DT |
989 | error = misc_register(&monitor_device); |
990 | if (error) { | |
991 | log_print("misc_register failed for monitor device"); | |
992 | misc_deregister(&ctl_device); | |
993 | } | |
994 | out: | |
597d0cae DT |
995 | return error; |
996 | } | |
997 | ||
998 | void dlm_user_exit(void) | |
999 | { | |
1000 | misc_deregister(&ctl_device); | |
dc68c7ed | 1001 | misc_deregister(&monitor_device); |
597d0cae DT |
1002 | } |
1003 |