]>
Commit | Line | Data |
---|---|---|
597d0cae | 1 | /* |
7fe2b319 | 2 | * Copyright (C) 2006-2010 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> | |
5a0e3ad6 | 20 | #include <linux/slab.h> |
597d0cae DT |
21 | |
22 | #include "dlm_internal.h" | |
23 | #include "lockspace.h" | |
24 | #include "lock.h" | |
25 | #include "lvb_table.h" | |
84c6e8cd | 26 | #include "user.h" |
597d0cae | 27 | |
0fe410d3 | 28 | static const char name_prefix[] = "dlm"; |
00977a59 | 29 | static const struct file_operations device_fops; |
dc68c7ed DT |
30 | static atomic_t dlm_monitor_opened; |
31 | static int dlm_monitor_unused = 1; | |
597d0cae DT |
32 | |
33 | #ifdef CONFIG_COMPAT | |
34 | ||
35 | struct dlm_lock_params32 { | |
36 | __u8 mode; | |
37 | __u8 namelen; | |
d7db923e DT |
38 | __u16 unused; |
39 | __u32 flags; | |
597d0cae DT |
40 | __u32 lkid; |
41 | __u32 parent; | |
d7db923e DT |
42 | __u64 xid; |
43 | __u64 timeout; | |
597d0cae DT |
44 | __u32 castparam; |
45 | __u32 castaddr; | |
46 | __u32 bastparam; | |
47 | __u32 bastaddr; | |
48 | __u32 lksb; | |
597d0cae DT |
49 | char lvb[DLM_USER_LVB_LEN]; |
50 | char name[0]; | |
51 | }; | |
52 | ||
53 | struct dlm_write_request32 { | |
54 | __u32 version[3]; | |
55 | __u8 cmd; | |
56 | __u8 is64bit; | |
57 | __u8 unused[2]; | |
58 | ||
59 | union { | |
60 | struct dlm_lock_params32 lock; | |
61 | struct dlm_lspace_params lspace; | |
72c2be77 | 62 | struct dlm_purge_params purge; |
597d0cae DT |
63 | } i; |
64 | }; | |
65 | ||
66 | struct dlm_lksb32 { | |
67 | __u32 sb_status; | |
68 | __u32 sb_lkid; | |
69 | __u8 sb_flags; | |
70 | __u32 sb_lvbptr; | |
71 | }; | |
72 | ||
73 | struct dlm_lock_result32 { | |
d7db923e | 74 | __u32 version[3]; |
597d0cae DT |
75 | __u32 length; |
76 | __u32 user_astaddr; | |
77 | __u32 user_astparam; | |
78 | __u32 user_lksb; | |
79 | struct dlm_lksb32 lksb; | |
80 | __u8 bast_mode; | |
81 | __u8 unused[3]; | |
82 | /* Offsets may be zero if no data is present */ | |
83 | __u32 lvb_offset; | |
84 | }; | |
85 | ||
86 | static void compat_input(struct dlm_write_request *kb, | |
2a79289e | 87 | struct dlm_write_request32 *kb32, |
1fecb1c4 | 88 | int namelen) |
597d0cae DT |
89 | { |
90 | kb->version[0] = kb32->version[0]; | |
91 | kb->version[1] = kb32->version[1]; | |
92 | kb->version[2] = kb32->version[2]; | |
93 | ||
94 | kb->cmd = kb32->cmd; | |
95 | kb->is64bit = kb32->is64bit; | |
96 | if (kb->cmd == DLM_USER_CREATE_LOCKSPACE || | |
97 | kb->cmd == DLM_USER_REMOVE_LOCKSPACE) { | |
98 | kb->i.lspace.flags = kb32->i.lspace.flags; | |
99 | kb->i.lspace.minor = kb32->i.lspace.minor; | |
1fecb1c4 | 100 | memcpy(kb->i.lspace.name, kb32->i.lspace.name, namelen); |
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); | |
1fecb1c4 | 118 | memcpy(kb->i.lock.name, kb32->i.lock.name, namelen); |
597d0cae DT |
119 | } |
120 | } | |
121 | ||
122 | static void compat_output(struct dlm_lock_result *res, | |
123 | struct dlm_lock_result32 *res32) | |
124 | { | |
d7db923e DT |
125 | res32->version[0] = res->version[0]; |
126 | res32->version[1] = res->version[1]; | |
127 | res32->version[2] = res->version[2]; | |
128 | ||
597d0cae DT |
129 | res32->user_astaddr = (__u32)(long)res->user_astaddr; |
130 | res32->user_astparam = (__u32)(long)res->user_astparam; | |
131 | res32->user_lksb = (__u32)(long)res->user_lksb; | |
132 | res32->bast_mode = res->bast_mode; | |
133 | ||
134 | res32->lvb_offset = res->lvb_offset; | |
135 | res32->length = res->length; | |
136 | ||
137 | res32->lksb.sb_status = res->lksb.sb_status; | |
138 | res32->lksb.sb_flags = res->lksb.sb_flags; | |
139 | res32->lksb.sb_lkid = res->lksb.sb_lkid; | |
140 | res32->lksb.sb_lvbptr = (__u32)(long)res->lksb.sb_lvbptr; | |
141 | } | |
142 | #endif | |
143 | ||
84d8cd69 DT |
144 | /* Figure out if this lock is at the end of its life and no longer |
145 | available for the application to use. The lkb still exists until | |
146 | the final ast is read. A lock becomes EOL in three situations: | |
147 | 1. a noqueue request fails with EAGAIN | |
148 | 2. an unlock completes with EUNLOCK | |
149 | 3. a cancel of a waiting request completes with ECANCEL/EDEADLK | |
150 | An EOL lock needs to be removed from the process's list of locks. | |
151 | And we can't allow any new operation on an EOL lock. This is | |
152 | not related to the lifetime of the lkb struct which is managed | |
153 | entirely by refcount. */ | |
154 | ||
155 | static int lkb_is_endoflife(struct dlm_lkb *lkb, int sb_status, int type) | |
156 | { | |
157 | switch (sb_status) { | |
158 | case -DLM_EUNLOCK: | |
159 | return 1; | |
160 | case -DLM_ECANCEL: | |
161 | case -ETIMEDOUT: | |
8b4021fa | 162 | case -EDEADLK: |
84d8cd69 DT |
163 | if (lkb->lkb_grmode == DLM_LOCK_IV) |
164 | return 1; | |
165 | break; | |
166 | case -EAGAIN: | |
167 | if (type == AST_COMP && lkb->lkb_grmode == DLM_LOCK_IV) | |
168 | return 1; | |
169 | break; | |
170 | } | |
171 | return 0; | |
172 | } | |
173 | ||
ef0c2bb0 DT |
174 | /* we could possibly check if the cancel of an orphan has resulted in the lkb |
175 | being removed and then remove that lkb from the orphans list and free it */ | |
597d0cae | 176 | |
7fe2b319 | 177 | void dlm_user_add_ast(struct dlm_lkb *lkb, int type, int mode) |
597d0cae DT |
178 | { |
179 | struct dlm_ls *ls; | |
180 | struct dlm_user_args *ua; | |
181 | struct dlm_user_proc *proc; | |
ef0c2bb0 | 182 | int eol = 0, ast_type; |
597d0cae | 183 | |
ef0c2bb0 | 184 | if (lkb->lkb_flags & (DLM_IFL_ORPHAN | DLM_IFL_DEAD)) |
597d0cae | 185 | return; |
597d0cae DT |
186 | |
187 | ls = lkb->lkb_resource->res_ls; | |
188 | mutex_lock(&ls->ls_clear_proc_locks); | |
189 | ||
190 | /* If ORPHAN/DEAD flag is set, it means the process is dead so an ast | |
191 | can't be delivered. For ORPHAN's, dlm_clear_proc_locks() freed | |
ef0c2bb0 DT |
192 | lkb->ua so we can't try to use it. This second check is necessary |
193 | for cases where a completion ast is received for an operation that | |
194 | began before clear_proc_locks did its cancel/unlock. */ | |
597d0cae | 195 | |
ef0c2bb0 | 196 | if (lkb->lkb_flags & (DLM_IFL_ORPHAN | DLM_IFL_DEAD)) |
597d0cae | 197 | goto out; |
597d0cae | 198 | |
d292c0cc DT |
199 | DLM_ASSERT(lkb->lkb_ua, dlm_print_lkb(lkb);); |
200 | ua = lkb->lkb_ua; | |
597d0cae DT |
201 | proc = ua->proc; |
202 | ||
203 | if (type == AST_BAST && ua->bastaddr == NULL) | |
204 | goto out; | |
205 | ||
206 | spin_lock(&proc->asts_spin); | |
ef0c2bb0 DT |
207 | |
208 | ast_type = lkb->lkb_ast_type; | |
209 | lkb->lkb_ast_type |= type; | |
7fe2b319 DT |
210 | if (type == AST_BAST) |
211 | lkb->lkb_bastmode = mode; | |
212 | else | |
213 | lkb->lkb_castmode = mode; | |
ef0c2bb0 DT |
214 | |
215 | if (!ast_type) { | |
597d0cae DT |
216 | kref_get(&lkb->lkb_ref); |
217 | list_add_tail(&lkb->lkb_astqueue, &proc->asts); | |
89d799d0 | 218 | lkb->lkb_ast_first = type; |
597d0cae DT |
219 | wake_up_interruptible(&proc->wait); |
220 | } | |
ef0c2bb0 DT |
221 | if (type == AST_COMP && (ast_type & AST_COMP)) |
222 | log_debug(ls, "ast overlap %x status %x %x", | |
223 | lkb->lkb_id, ua->lksb.sb_status, lkb->lkb_flags); | |
224 | ||
84d8cd69 | 225 | eol = lkb_is_endoflife(lkb, ua->lksb.sb_status, type); |
ef0c2bb0 | 226 | if (eol) { |
ef0c2bb0 DT |
227 | lkb->lkb_flags |= DLM_IFL_ENDOFLIFE; |
228 | } | |
a1bc86e6 | 229 | |
597d0cae DT |
230 | /* We want to copy the lvb to userspace when the completion |
231 | ast is read if the status is 0, the lock has an lvb and | |
232 | lvb_ops says we should. We could probably have set_lvb_lock() | |
233 | set update_user_lvb instead and not need old_mode */ | |
234 | ||
235 | if ((lkb->lkb_ast_type & AST_COMP) && | |
236 | (lkb->lkb_lksb->sb_status == 0) && | |
237 | lkb->lkb_lksb->sb_lvbptr && | |
238 | dlm_lvb_operations[ua->old_mode + 1][lkb->lkb_grmode + 1]) | |
239 | ua->update_user_lvb = 1; | |
240 | else | |
241 | ua->update_user_lvb = 0; | |
242 | ||
243 | spin_unlock(&proc->asts_spin); | |
34e22bed | 244 | |
ef0c2bb0 | 245 | if (eol) { |
ce5246b9 | 246 | spin_lock(&proc->locks_spin); |
ef0c2bb0 DT |
247 | if (!list_empty(&lkb->lkb_ownqueue)) { |
248 | list_del_init(&lkb->lkb_ownqueue); | |
249 | dlm_put_lkb(lkb); | |
250 | } | |
ce5246b9 | 251 | spin_unlock(&proc->locks_spin); |
34e22bed | 252 | } |
597d0cae DT |
253 | out: |
254 | mutex_unlock(&ls->ls_clear_proc_locks); | |
255 | } | |
256 | ||
257 | static int device_user_lock(struct dlm_user_proc *proc, | |
258 | struct dlm_lock_params *params) | |
259 | { | |
260 | struct dlm_ls *ls; | |
261 | struct dlm_user_args *ua; | |
262 | int error = -ENOMEM; | |
263 | ||
264 | ls = dlm_find_lockspace_local(proc->lockspace); | |
265 | if (!ls) | |
266 | return -ENOENT; | |
267 | ||
268 | if (!params->castaddr || !params->lksb) { | |
269 | error = -EINVAL; | |
270 | goto out; | |
271 | } | |
272 | ||
573c24c4 | 273 | ua = kzalloc(sizeof(struct dlm_user_args), GFP_NOFS); |
597d0cae DT |
274 | if (!ua) |
275 | goto out; | |
276 | ua->proc = proc; | |
277 | ua->user_lksb = params->lksb; | |
278 | ua->castparam = params->castparam; | |
279 | ua->castaddr = params->castaddr; | |
280 | ua->bastparam = params->bastparam; | |
281 | ua->bastaddr = params->bastaddr; | |
d7db923e | 282 | ua->xid = params->xid; |
597d0cae DT |
283 | |
284 | if (params->flags & DLM_LKF_CONVERT) | |
285 | error = dlm_user_convert(ls, ua, | |
286 | params->mode, params->flags, | |
d7db923e DT |
287 | params->lkid, params->lvb, |
288 | (unsigned long) params->timeout); | |
597d0cae DT |
289 | else { |
290 | error = dlm_user_request(ls, ua, | |
291 | params->mode, params->flags, | |
292 | params->name, params->namelen, | |
d7db923e | 293 | (unsigned long) params->timeout); |
597d0cae DT |
294 | if (!error) |
295 | error = ua->lksb.sb_lkid; | |
296 | } | |
297 | out: | |
298 | dlm_put_lockspace(ls); | |
299 | return error; | |
300 | } | |
301 | ||
302 | static int device_user_unlock(struct dlm_user_proc *proc, | |
303 | struct dlm_lock_params *params) | |
304 | { | |
305 | struct dlm_ls *ls; | |
306 | struct dlm_user_args *ua; | |
307 | int error = -ENOMEM; | |
308 | ||
309 | ls = dlm_find_lockspace_local(proc->lockspace); | |
310 | if (!ls) | |
311 | return -ENOENT; | |
312 | ||
573c24c4 | 313 | ua = kzalloc(sizeof(struct dlm_user_args), GFP_NOFS); |
597d0cae DT |
314 | if (!ua) |
315 | goto out; | |
316 | ua->proc = proc; | |
317 | ua->user_lksb = params->lksb; | |
318 | ua->castparam = params->castparam; | |
319 | ua->castaddr = params->castaddr; | |
320 | ||
321 | if (params->flags & DLM_LKF_CANCEL) | |
322 | error = dlm_user_cancel(ls, ua, params->flags, params->lkid); | |
323 | else | |
324 | error = dlm_user_unlock(ls, ua, params->flags, params->lkid, | |
325 | params->lvb); | |
326 | out: | |
327 | dlm_put_lockspace(ls); | |
328 | return error; | |
329 | } | |
330 | ||
8b4021fa DT |
331 | static int device_user_deadlock(struct dlm_user_proc *proc, |
332 | struct dlm_lock_params *params) | |
333 | { | |
334 | struct dlm_ls *ls; | |
335 | int error; | |
336 | ||
337 | ls = dlm_find_lockspace_local(proc->lockspace); | |
338 | if (!ls) | |
339 | return -ENOENT; | |
340 | ||
341 | error = dlm_user_deadlock(ls, params->flags, params->lkid); | |
342 | ||
343 | dlm_put_lockspace(ls); | |
344 | return error; | |
345 | } | |
346 | ||
0f8e0d9a | 347 | static int dlm_device_register(struct dlm_ls *ls, char *name) |
254da030 PC |
348 | { |
349 | int error, len; | |
350 | ||
0f8e0d9a DT |
351 | /* The device is already registered. This happens when the |
352 | lockspace is created multiple times from userspace. */ | |
353 | if (ls->ls_device.name) | |
354 | return 0; | |
355 | ||
254da030 PC |
356 | error = -ENOMEM; |
357 | len = strlen(name) + strlen(name_prefix) + 2; | |
573c24c4 | 358 | ls->ls_device.name = kzalloc(len, GFP_NOFS); |
254da030 PC |
359 | if (!ls->ls_device.name) |
360 | goto fail; | |
361 | ||
362 | snprintf((char *)ls->ls_device.name, len, "%s_%s", name_prefix, | |
363 | name); | |
364 | ls->ls_device.fops = &device_fops; | |
365 | ls->ls_device.minor = MISC_DYNAMIC_MINOR; | |
366 | ||
367 | error = misc_register(&ls->ls_device); | |
368 | if (error) { | |
369 | kfree(ls->ls_device.name); | |
370 | } | |
371 | fail: | |
372 | return error; | |
373 | } | |
374 | ||
0f8e0d9a DT |
375 | int dlm_device_deregister(struct dlm_ls *ls) |
376 | { | |
377 | int error; | |
378 | ||
379 | /* The device is not registered. This happens when the lockspace | |
380 | was never used from userspace, or when device_create_lockspace() | |
381 | calls dlm_release_lockspace() after the register fails. */ | |
382 | if (!ls->ls_device.name) | |
383 | return 0; | |
384 | ||
385 | error = misc_deregister(&ls->ls_device); | |
386 | if (!error) | |
387 | kfree(ls->ls_device.name); | |
388 | return error; | |
389 | } | |
390 | ||
72c2be77 DT |
391 | static int device_user_purge(struct dlm_user_proc *proc, |
392 | struct dlm_purge_params *params) | |
393 | { | |
394 | struct dlm_ls *ls; | |
395 | int error; | |
396 | ||
397 | ls = dlm_find_lockspace_local(proc->lockspace); | |
398 | if (!ls) | |
399 | return -ENOENT; | |
400 | ||
401 | error = dlm_user_purge(ls, proc, params->nodeid, params->pid); | |
402 | ||
403 | dlm_put_lockspace(ls); | |
404 | return error; | |
405 | } | |
406 | ||
597d0cae DT |
407 | static int device_create_lockspace(struct dlm_lspace_params *params) |
408 | { | |
409 | dlm_lockspace_t *lockspace; | |
410 | struct dlm_ls *ls; | |
254da030 | 411 | int error; |
597d0cae DT |
412 | |
413 | if (!capable(CAP_SYS_ADMIN)) | |
414 | return -EPERM; | |
415 | ||
416 | error = dlm_new_lockspace(params->name, strlen(params->name), | |
3ae1acf9 | 417 | &lockspace, params->flags, DLM_USER_LVB_LEN); |
597d0cae DT |
418 | if (error) |
419 | return error; | |
420 | ||
421 | ls = dlm_find_lockspace_local(lockspace); | |
422 | if (!ls) | |
423 | return -ENOENT; | |
424 | ||
0f8e0d9a | 425 | error = dlm_device_register(ls, params->name); |
597d0cae | 426 | dlm_put_lockspace(ls); |
597d0cae | 427 | |
254da030 PC |
428 | if (error) |
429 | dlm_release_lockspace(lockspace, 0); | |
430 | else | |
431 | error = ls->ls_device.minor; | |
432 | ||
597d0cae DT |
433 | return error; |
434 | } | |
435 | ||
436 | static int device_remove_lockspace(struct dlm_lspace_params *params) | |
437 | { | |
438 | dlm_lockspace_t *lockspace; | |
439 | struct dlm_ls *ls; | |
c6e6f0ba | 440 | int error, force = 0; |
597d0cae DT |
441 | |
442 | if (!capable(CAP_SYS_ADMIN)) | |
443 | return -EPERM; | |
444 | ||
445 | ls = dlm_find_lockspace_device(params->minor); | |
446 | if (!ls) | |
447 | return -ENOENT; | |
448 | ||
c6e6f0ba DT |
449 | if (params->flags & DLM_USER_LSFLG_FORCEFREE) |
450 | force = 2; | |
451 | ||
597d0cae | 452 | lockspace = ls->ls_local_handle; |
0f8e0d9a | 453 | dlm_put_lockspace(ls); |
597d0cae | 454 | |
0f8e0d9a DT |
455 | /* The final dlm_release_lockspace waits for references to go to |
456 | zero, so all processes will need to close their device for the | |
457 | ls before the release will proceed. release also calls the | |
458 | device_deregister above. Converting a positive return value | |
459 | from release to zero means that userspace won't know when its | |
460 | release was the final one, but it shouldn't need to know. */ | |
597d0cae | 461 | |
c6e6f0ba | 462 | error = dlm_release_lockspace(lockspace, force); |
0f8e0d9a DT |
463 | if (error > 0) |
464 | error = 0; | |
597d0cae DT |
465 | return error; |
466 | } | |
467 | ||
468 | /* Check the user's version matches ours */ | |
469 | static int check_version(struct dlm_write_request *req) | |
470 | { | |
471 | if (req->version[0] != DLM_DEVICE_VERSION_MAJOR || | |
472 | (req->version[0] == DLM_DEVICE_VERSION_MAJOR && | |
473 | req->version[1] > DLM_DEVICE_VERSION_MINOR)) { | |
474 | ||
475 | printk(KERN_DEBUG "dlm: process %s (%d) version mismatch " | |
476 | "user (%d.%d.%d) kernel (%d.%d.%d)\n", | |
477 | current->comm, | |
ba25f9dc | 478 | task_pid_nr(current), |
597d0cae DT |
479 | req->version[0], |
480 | req->version[1], | |
481 | req->version[2], | |
482 | DLM_DEVICE_VERSION_MAJOR, | |
483 | DLM_DEVICE_VERSION_MINOR, | |
484 | DLM_DEVICE_VERSION_PATCH); | |
485 | return -EINVAL; | |
486 | } | |
487 | return 0; | |
488 | } | |
489 | ||
490 | /* | |
491 | * device_write | |
492 | * | |
493 | * device_user_lock | |
494 | * dlm_user_request -> request_lock | |
495 | * dlm_user_convert -> convert_lock | |
496 | * | |
497 | * device_user_unlock | |
498 | * dlm_user_unlock -> unlock_lock | |
499 | * dlm_user_cancel -> cancel_lock | |
500 | * | |
501 | * device_create_lockspace | |
502 | * dlm_new_lockspace | |
503 | * | |
504 | * device_remove_lockspace | |
505 | * dlm_release_lockspace | |
506 | */ | |
507 | ||
508 | /* a write to a lockspace device is a lock or unlock request, a write | |
509 | to the control device is to create/remove a lockspace */ | |
510 | ||
511 | static ssize_t device_write(struct file *file, const char __user *buf, | |
512 | size_t count, loff_t *ppos) | |
513 | { | |
514 | struct dlm_user_proc *proc = file->private_data; | |
515 | struct dlm_write_request *kbuf; | |
516 | sigset_t tmpsig, allsigs; | |
517 | int error; | |
518 | ||
519 | #ifdef CONFIG_COMPAT | |
520 | if (count < sizeof(struct dlm_write_request32)) | |
521 | #else | |
522 | if (count < sizeof(struct dlm_write_request)) | |
523 | #endif | |
524 | return -EINVAL; | |
525 | ||
573c24c4 | 526 | kbuf = kzalloc(count + 1, GFP_NOFS); |
597d0cae DT |
527 | if (!kbuf) |
528 | return -ENOMEM; | |
529 | ||
530 | if (copy_from_user(kbuf, buf, count)) { | |
531 | error = -EFAULT; | |
532 | goto out_free; | |
533 | } | |
534 | ||
535 | if (check_version(kbuf)) { | |
536 | error = -EBADE; | |
537 | goto out_free; | |
538 | } | |
539 | ||
540 | #ifdef CONFIG_COMPAT | |
541 | if (!kbuf->is64bit) { | |
542 | struct dlm_write_request32 *k32buf; | |
1fecb1c4 DT |
543 | int namelen = 0; |
544 | ||
545 | if (count > sizeof(struct dlm_write_request32)) | |
546 | namelen = count - sizeof(struct dlm_write_request32); | |
547 | ||
597d0cae | 548 | k32buf = (struct dlm_write_request32 *)kbuf; |
1fecb1c4 DT |
549 | |
550 | /* add 1 after namelen so that the name string is terminated */ | |
551 | kbuf = kzalloc(sizeof(struct dlm_write_request) + namelen + 1, | |
573c24c4 | 552 | GFP_NOFS); |
cb980d9a DT |
553 | if (!kbuf) { |
554 | kfree(k32buf); | |
597d0cae | 555 | return -ENOMEM; |
cb980d9a | 556 | } |
597d0cae DT |
557 | |
558 | if (proc) | |
559 | set_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags); | |
1fecb1c4 DT |
560 | |
561 | compat_input(kbuf, k32buf, namelen); | |
597d0cae DT |
562 | kfree(k32buf); |
563 | } | |
564 | #endif | |
565 | ||
566 | /* do we really need this? can a write happen after a close? */ | |
567 | if ((kbuf->cmd == DLM_USER_LOCK || kbuf->cmd == DLM_USER_UNLOCK) && | |
cb980d9a DT |
568 | (proc && test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))) { |
569 | error = -EINVAL; | |
570 | goto out_free; | |
571 | } | |
597d0cae DT |
572 | |
573 | sigfillset(&allsigs); | |
574 | sigprocmask(SIG_BLOCK, &allsigs, &tmpsig); | |
575 | ||
576 | error = -EINVAL; | |
577 | ||
578 | switch (kbuf->cmd) | |
579 | { | |
580 | case DLM_USER_LOCK: | |
581 | if (!proc) { | |
582 | log_print("no locking on control device"); | |
583 | goto out_sig; | |
584 | } | |
585 | error = device_user_lock(proc, &kbuf->i.lock); | |
586 | break; | |
587 | ||
588 | case DLM_USER_UNLOCK: | |
589 | if (!proc) { | |
590 | log_print("no locking on control device"); | |
591 | goto out_sig; | |
592 | } | |
593 | error = device_user_unlock(proc, &kbuf->i.lock); | |
594 | break; | |
595 | ||
8b4021fa DT |
596 | case DLM_USER_DEADLOCK: |
597 | if (!proc) { | |
598 | log_print("no locking on control device"); | |
599 | goto out_sig; | |
600 | } | |
601 | error = device_user_deadlock(proc, &kbuf->i.lock); | |
602 | break; | |
603 | ||
597d0cae DT |
604 | case DLM_USER_CREATE_LOCKSPACE: |
605 | if (proc) { | |
606 | log_print("create/remove only on control device"); | |
607 | goto out_sig; | |
608 | } | |
609 | error = device_create_lockspace(&kbuf->i.lspace); | |
610 | break; | |
611 | ||
612 | case DLM_USER_REMOVE_LOCKSPACE: | |
613 | if (proc) { | |
614 | log_print("create/remove only on control device"); | |
615 | goto out_sig; | |
616 | } | |
617 | error = device_remove_lockspace(&kbuf->i.lspace); | |
618 | break; | |
619 | ||
72c2be77 DT |
620 | case DLM_USER_PURGE: |
621 | if (!proc) { | |
622 | log_print("no locking on control device"); | |
623 | goto out_sig; | |
624 | } | |
625 | error = device_user_purge(proc, &kbuf->i.purge); | |
626 | break; | |
627 | ||
597d0cae DT |
628 | default: |
629 | log_print("Unknown command passed to DLM device : %d\n", | |
630 | kbuf->cmd); | |
631 | } | |
632 | ||
633 | out_sig: | |
634 | sigprocmask(SIG_SETMASK, &tmpsig, NULL); | |
635 | recalc_sigpending(); | |
636 | out_free: | |
637 | kfree(kbuf); | |
638 | return error; | |
639 | } | |
640 | ||
641 | /* Every process that opens the lockspace device has its own "proc" structure | |
642 | hanging off the open file that's used to keep track of locks owned by the | |
643 | process and asts that need to be delivered to the process. */ | |
644 | ||
645 | static int device_open(struct inode *inode, struct file *file) | |
646 | { | |
647 | struct dlm_user_proc *proc; | |
648 | struct dlm_ls *ls; | |
649 | ||
650 | ls = dlm_find_lockspace_device(iminor(inode)); | |
f9f2ed48 | 651 | if (!ls) |
597d0cae DT |
652 | return -ENOENT; |
653 | ||
573c24c4 | 654 | proc = kzalloc(sizeof(struct dlm_user_proc), GFP_NOFS); |
597d0cae DT |
655 | if (!proc) { |
656 | dlm_put_lockspace(ls); | |
657 | return -ENOMEM; | |
658 | } | |
659 | ||
660 | proc->lockspace = ls->ls_local_handle; | |
661 | INIT_LIST_HEAD(&proc->asts); | |
662 | INIT_LIST_HEAD(&proc->locks); | |
a1bc86e6 | 663 | INIT_LIST_HEAD(&proc->unlocking); |
597d0cae DT |
664 | spin_lock_init(&proc->asts_spin); |
665 | spin_lock_init(&proc->locks_spin); | |
666 | init_waitqueue_head(&proc->wait); | |
667 | file->private_data = proc; | |
668 | ||
669 | return 0; | |
670 | } | |
671 | ||
672 | static int device_close(struct inode *inode, struct file *file) | |
673 | { | |
674 | struct dlm_user_proc *proc = file->private_data; | |
675 | struct dlm_ls *ls; | |
676 | sigset_t tmpsig, allsigs; | |
677 | ||
678 | ls = dlm_find_lockspace_local(proc->lockspace); | |
679 | if (!ls) | |
680 | return -ENOENT; | |
681 | ||
682 | sigfillset(&allsigs); | |
683 | sigprocmask(SIG_BLOCK, &allsigs, &tmpsig); | |
684 | ||
685 | set_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags); | |
686 | ||
687 | dlm_clear_proc_locks(ls, proc); | |
688 | ||
689 | /* at this point no more lkb's should exist for this lockspace, | |
690 | so there's no chance of dlm_user_add_ast() being called and | |
691 | looking for lkb->ua->proc */ | |
692 | ||
693 | kfree(proc); | |
694 | file->private_data = NULL; | |
695 | ||
696 | dlm_put_lockspace(ls); | |
697 | dlm_put_lockspace(ls); /* for the find in device_open() */ | |
698 | ||
699 | /* FIXME: AUTOFREE: if this ls is no longer used do | |
700 | device_remove_lockspace() */ | |
701 | ||
702 | sigprocmask(SIG_SETMASK, &tmpsig, NULL); | |
703 | recalc_sigpending(); | |
704 | ||
705 | return 0; | |
706 | } | |
707 | ||
708 | static int copy_result_to_user(struct dlm_user_args *ua, int compat, int type, | |
89d799d0 | 709 | int mode, char __user *buf, size_t count) |
597d0cae DT |
710 | { |
711 | #ifdef CONFIG_COMPAT | |
712 | struct dlm_lock_result32 result32; | |
713 | #endif | |
714 | struct dlm_lock_result result; | |
715 | void *resultptr; | |
716 | int error=0; | |
717 | int len; | |
718 | int struct_len; | |
719 | ||
720 | memset(&result, 0, sizeof(struct dlm_lock_result)); | |
d7db923e DT |
721 | result.version[0] = DLM_DEVICE_VERSION_MAJOR; |
722 | result.version[1] = DLM_DEVICE_VERSION_MINOR; | |
723 | result.version[2] = DLM_DEVICE_VERSION_PATCH; | |
597d0cae DT |
724 | memcpy(&result.lksb, &ua->lksb, sizeof(struct dlm_lksb)); |
725 | result.user_lksb = ua->user_lksb; | |
726 | ||
727 | /* FIXME: dlm1 provides for the user's bastparam/addr to not be updated | |
728 | in a conversion unless the conversion is successful. See code | |
729 | in dlm_user_convert() for updating ua from ua_tmp. OpenVMS, though, | |
730 | notes that a new blocking AST address and parameter are set even if | |
731 | the conversion fails, so maybe we should just do that. */ | |
732 | ||
733 | if (type == AST_BAST) { | |
734 | result.user_astaddr = ua->bastaddr; | |
735 | result.user_astparam = ua->bastparam; | |
89d799d0 | 736 | result.bast_mode = mode; |
597d0cae DT |
737 | } else { |
738 | result.user_astaddr = ua->castaddr; | |
739 | result.user_astparam = ua->castparam; | |
740 | } | |
741 | ||
742 | #ifdef CONFIG_COMPAT | |
743 | if (compat) | |
744 | len = sizeof(struct dlm_lock_result32); | |
745 | else | |
746 | #endif | |
747 | len = sizeof(struct dlm_lock_result); | |
748 | struct_len = len; | |
749 | ||
750 | /* copy lvb to userspace if there is one, it's been updated, and | |
751 | the user buffer has space for it */ | |
752 | ||
753 | if (ua->update_user_lvb && ua->lksb.sb_lvbptr && | |
754 | count >= len + DLM_USER_LVB_LEN) { | |
755 | if (copy_to_user(buf+len, ua->lksb.sb_lvbptr, | |
756 | DLM_USER_LVB_LEN)) { | |
757 | error = -EFAULT; | |
758 | goto out; | |
759 | } | |
760 | ||
761 | result.lvb_offset = len; | |
762 | len += DLM_USER_LVB_LEN; | |
763 | } | |
764 | ||
765 | result.length = len; | |
766 | resultptr = &result; | |
767 | #ifdef CONFIG_COMPAT | |
768 | if (compat) { | |
769 | compat_output(&result, &result32); | |
770 | resultptr = &result32; | |
771 | } | |
772 | #endif | |
773 | ||
774 | if (copy_to_user(buf, resultptr, struct_len)) | |
775 | error = -EFAULT; | |
776 | else | |
777 | error = len; | |
778 | out: | |
779 | return error; | |
780 | } | |
781 | ||
d7db923e DT |
782 | static int copy_version_to_user(char __user *buf, size_t count) |
783 | { | |
784 | struct dlm_device_version ver; | |
785 | ||
786 | memset(&ver, 0, sizeof(struct dlm_device_version)); | |
787 | ver.version[0] = DLM_DEVICE_VERSION_MAJOR; | |
788 | ver.version[1] = DLM_DEVICE_VERSION_MINOR; | |
789 | ver.version[2] = DLM_DEVICE_VERSION_PATCH; | |
790 | ||
791 | if (copy_to_user(buf, &ver, sizeof(struct dlm_device_version))) | |
792 | return -EFAULT; | |
793 | return sizeof(struct dlm_device_version); | |
794 | } | |
795 | ||
597d0cae DT |
796 | /* a read returns a single ast described in a struct dlm_lock_result */ |
797 | ||
798 | static ssize_t device_read(struct file *file, char __user *buf, size_t count, | |
799 | loff_t *ppos) | |
800 | { | |
801 | struct dlm_user_proc *proc = file->private_data; | |
802 | struct dlm_lkb *lkb; | |
597d0cae | 803 | DECLARE_WAITQUEUE(wait, current); |
89d799d0 DT |
804 | int error = 0, removed; |
805 | int ret_type, ret_mode; | |
806 | int bastmode, castmode, do_bast, do_cast; | |
597d0cae | 807 | |
d7db923e DT |
808 | if (count == sizeof(struct dlm_device_version)) { |
809 | error = copy_version_to_user(buf, count); | |
810 | return error; | |
811 | } | |
812 | ||
813 | if (!proc) { | |
814 | log_print("non-version read from control device %zu", count); | |
815 | return -EINVAL; | |
816 | } | |
817 | ||
597d0cae DT |
818 | #ifdef CONFIG_COMPAT |
819 | if (count < sizeof(struct dlm_lock_result32)) | |
820 | #else | |
821 | if (count < sizeof(struct dlm_lock_result)) | |
822 | #endif | |
823 | return -EINVAL; | |
824 | ||
89d799d0 DT |
825 | try_another: |
826 | ||
597d0cae DT |
827 | /* do we really need this? can a read happen after a close? */ |
828 | if (test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags)) | |
829 | return -EINVAL; | |
830 | ||
831 | spin_lock(&proc->asts_spin); | |
832 | if (list_empty(&proc->asts)) { | |
833 | if (file->f_flags & O_NONBLOCK) { | |
834 | spin_unlock(&proc->asts_spin); | |
835 | return -EAGAIN; | |
836 | } | |
837 | ||
838 | add_wait_queue(&proc->wait, &wait); | |
839 | ||
840 | repeat: | |
841 | set_current_state(TASK_INTERRUPTIBLE); | |
842 | if (list_empty(&proc->asts) && !signal_pending(current)) { | |
843 | spin_unlock(&proc->asts_spin); | |
844 | schedule(); | |
845 | spin_lock(&proc->asts_spin); | |
846 | goto repeat; | |
847 | } | |
848 | set_current_state(TASK_RUNNING); | |
849 | remove_wait_queue(&proc->wait, &wait); | |
850 | ||
851 | if (signal_pending(current)) { | |
852 | spin_unlock(&proc->asts_spin); | |
853 | return -ERESTARTSYS; | |
854 | } | |
855 | } | |
856 | ||
597d0cae DT |
857 | /* there may be both completion and blocking asts to return for |
858 | the lkb, don't remove lkb from asts list unless no asts remain */ | |
859 | ||
860 | lkb = list_entry(proc->asts.next, struct dlm_lkb, lkb_astqueue); | |
861 | ||
89d799d0 DT |
862 | removed = 0; |
863 | ret_type = 0; | |
864 | ret_mode = 0; | |
865 | do_bast = lkb->lkb_ast_type & AST_BAST; | |
866 | do_cast = lkb->lkb_ast_type & AST_COMP; | |
867 | bastmode = lkb->lkb_bastmode; | |
868 | castmode = lkb->lkb_castmode; | |
869 | ||
870 | /* when both are queued figure out which to do first and | |
871 | switch first so the other goes in the next read */ | |
872 | ||
873 | if (do_cast && do_bast) { | |
874 | if (lkb->lkb_ast_first == AST_COMP) { | |
875 | ret_type = AST_COMP; | |
876 | ret_mode = castmode; | |
877 | lkb->lkb_ast_type &= ~AST_COMP; | |
878 | lkb->lkb_ast_first = AST_BAST; | |
879 | } else { | |
880 | ret_type = AST_BAST; | |
881 | ret_mode = bastmode; | |
882 | lkb->lkb_ast_type &= ~AST_BAST; | |
883 | lkb->lkb_ast_first = AST_COMP; | |
884 | } | |
885 | } else { | |
886 | ret_type = lkb->lkb_ast_first; | |
887 | ret_mode = (ret_type == AST_COMP) ? castmode : bastmode; | |
888 | lkb->lkb_ast_type &= ~ret_type; | |
889 | lkb->lkb_ast_first = 0; | |
890 | } | |
891 | ||
892 | /* if we're doing a bast but the bast is unnecessary, then | |
893 | switch to do nothing or do a cast if that was needed next */ | |
894 | ||
895 | if ((ret_type == AST_BAST) && | |
896 | dlm_modes_compat(bastmode, lkb->lkb_castmode_done)) { | |
897 | ret_type = 0; | |
898 | ret_mode = 0; | |
899 | ||
900 | if (do_cast) { | |
901 | ret_type = AST_COMP; | |
902 | ret_mode = castmode; | |
903 | lkb->lkb_ast_type &= ~AST_COMP; | |
904 | lkb->lkb_ast_first = 0; | |
905 | } | |
906 | } | |
907 | ||
908 | if (lkb->lkb_ast_first != lkb->lkb_ast_type) { | |
909 | log_print("device_read %x ast_first %x ast_type %x", | |
910 | lkb->lkb_id, lkb->lkb_ast_first, lkb->lkb_ast_type); | |
597d0cae DT |
911 | } |
912 | ||
913 | if (!lkb->lkb_ast_type) { | |
914 | list_del(&lkb->lkb_astqueue); | |
915 | removed = 1; | |
916 | } | |
917 | spin_unlock(&proc->asts_spin); | |
918 | ||
89d799d0 DT |
919 | if (ret_type) { |
920 | error = copy_result_to_user(lkb->lkb_ua, | |
921 | test_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags), | |
922 | ret_type, ret_mode, buf, count); | |
923 | ||
924 | if (ret_type == AST_COMP) | |
925 | lkb->lkb_castmode_done = castmode; | |
926 | if (ret_type == AST_BAST) | |
927 | lkb->lkb_bastmode_done = bastmode; | |
928 | } | |
597d0cae DT |
929 | |
930 | /* removes reference for the proc->asts lists added by | |
931 | dlm_user_add_ast() and may result in the lkb being freed */ | |
89d799d0 | 932 | |
597d0cae DT |
933 | if (removed) |
934 | dlm_put_lkb(lkb); | |
935 | ||
89d799d0 DT |
936 | /* the bast that was queued was eliminated (see unnecessary above), |
937 | leaving nothing to return */ | |
938 | ||
939 | if (!ret_type) | |
940 | goto try_another; | |
941 | ||
597d0cae DT |
942 | return error; |
943 | } | |
944 | ||
945 | static unsigned int device_poll(struct file *file, poll_table *wait) | |
946 | { | |
947 | struct dlm_user_proc *proc = file->private_data; | |
948 | ||
949 | poll_wait(file, &proc->wait, wait); | |
950 | ||
951 | spin_lock(&proc->asts_spin); | |
952 | if (!list_empty(&proc->asts)) { | |
953 | spin_unlock(&proc->asts_spin); | |
954 | return POLLIN | POLLRDNORM; | |
955 | } | |
956 | spin_unlock(&proc->asts_spin); | |
957 | return 0; | |
958 | } | |
959 | ||
dc68c7ed DT |
960 | int dlm_user_daemon_available(void) |
961 | { | |
962 | /* dlm_controld hasn't started (or, has started, but not | |
963 | properly populated configfs) */ | |
964 | ||
965 | if (!dlm_our_nodeid()) | |
966 | return 0; | |
967 | ||
968 | /* This is to deal with versions of dlm_controld that don't | |
969 | know about the monitor device. We assume that if the | |
970 | dlm_controld was started (above), but the monitor device | |
971 | was never opened, that it's an old version. dlm_controld | |
972 | should open the monitor device before populating configfs. */ | |
973 | ||
974 | if (dlm_monitor_unused) | |
975 | return 1; | |
976 | ||
977 | return atomic_read(&dlm_monitor_opened) ? 1 : 0; | |
978 | } | |
979 | ||
597d0cae DT |
980 | static int ctl_device_open(struct inode *inode, struct file *file) |
981 | { | |
982 | file->private_data = NULL; | |
983 | return 0; | |
984 | } | |
985 | ||
986 | static int ctl_device_close(struct inode *inode, struct file *file) | |
987 | { | |
988 | return 0; | |
989 | } | |
990 | ||
dc68c7ed DT |
991 | static int monitor_device_open(struct inode *inode, struct file *file) |
992 | { | |
993 | atomic_inc(&dlm_monitor_opened); | |
994 | dlm_monitor_unused = 0; | |
995 | return 0; | |
996 | } | |
997 | ||
998 | static int monitor_device_close(struct inode *inode, struct file *file) | |
999 | { | |
1000 | if (atomic_dec_and_test(&dlm_monitor_opened)) | |
1001 | dlm_stop_lockspaces(); | |
1002 | return 0; | |
1003 | } | |
1004 | ||
00977a59 | 1005 | static const struct file_operations device_fops = { |
597d0cae DT |
1006 | .open = device_open, |
1007 | .release = device_close, | |
1008 | .read = device_read, | |
1009 | .write = device_write, | |
1010 | .poll = device_poll, | |
1011 | .owner = THIS_MODULE, | |
1012 | }; | |
1013 | ||
00977a59 | 1014 | static const struct file_operations ctl_device_fops = { |
597d0cae DT |
1015 | .open = ctl_device_open, |
1016 | .release = ctl_device_close, | |
d7db923e | 1017 | .read = device_read, |
597d0cae DT |
1018 | .write = device_write, |
1019 | .owner = THIS_MODULE, | |
1020 | }; | |
1021 | ||
0fe410d3 DC |
1022 | static struct miscdevice ctl_device = { |
1023 | .name = "dlm-control", | |
1024 | .fops = &ctl_device_fops, | |
1025 | .minor = MISC_DYNAMIC_MINOR, | |
1026 | }; | |
1027 | ||
dc68c7ed DT |
1028 | static const struct file_operations monitor_device_fops = { |
1029 | .open = monitor_device_open, | |
1030 | .release = monitor_device_close, | |
1031 | .owner = THIS_MODULE, | |
1032 | }; | |
1033 | ||
1034 | static struct miscdevice monitor_device = { | |
1035 | .name = "dlm-monitor", | |
1036 | .fops = &monitor_device_fops, | |
1037 | .minor = MISC_DYNAMIC_MINOR, | |
1038 | }; | |
1039 | ||
30727174 | 1040 | int __init dlm_user_init(void) |
597d0cae DT |
1041 | { |
1042 | int error; | |
1043 | ||
dc68c7ed DT |
1044 | atomic_set(&dlm_monitor_opened, 0); |
1045 | ||
597d0cae | 1046 | error = misc_register(&ctl_device); |
dc68c7ed | 1047 | if (error) { |
597d0cae | 1048 | log_print("misc_register failed for control device"); |
dc68c7ed DT |
1049 | goto out; |
1050 | } | |
597d0cae | 1051 | |
dc68c7ed DT |
1052 | error = misc_register(&monitor_device); |
1053 | if (error) { | |
1054 | log_print("misc_register failed for monitor device"); | |
1055 | misc_deregister(&ctl_device); | |
1056 | } | |
1057 | out: | |
597d0cae DT |
1058 | return error; |
1059 | } | |
1060 | ||
1061 | void dlm_user_exit(void) | |
1062 | { | |
1063 | misc_deregister(&ctl_device); | |
dc68c7ed | 1064 | misc_deregister(&monitor_device); |
597d0cae DT |
1065 | } |
1066 |